Artificial Intelligence and Renewables Towards an Energy Transition 3030638456, 9783030638450

Citation preview

Lecture Notes in Networks and Systems 174

Mustapha Hatti   Editor

Artificial Intelligence and Renewables Towards an Energy Transition

Lecture Notes in Networks and Systems Volume 174

Series Editor Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland Advisory Editors Fernando Gomide, Department of Computer Engineering and Automation—DCA, School of Electrical and Computer Engineering—FEEC, University of Campinas— UNICAMP, São Paulo, Brazil Okyay Kaynak, Department of Electrical and Electronic Engineering, Bogazici University, Istanbul, Turkey Derong Liu, Department of Electrical and Computer Engineering, University of Illinois at Chicago, Chicago, USA; Institute of Automation, Chinese Academy of Sciences, Beijing, China Witold Pedrycz, Department of Electrical and Computer Engineering, University of Alberta, Alberta, Canada; Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland Marios M. Polycarpou, Department of Electrical and Computer Engineering, KIOS Research Center for Intelligent Systems and Networks, University of Cyprus, Nicosia, Cyprus Imre J. Rudas, Óbuda University, Budapest, Hungary Jun Wang, Department of Computer Science, City University of Hong Kong, Kowloon, Hong Kong

The series “Lecture Notes in Networks and Systems” publishes the latest developments in Networks and Systems—quickly, informally and with high quality. Original research reported in proceedings and post-proceedings represents the core of LNNS. Volumes published in LNNS embrace all aspects and subfields of, as well as new challenges in, Networks and Systems. The series contains proceedings and edited volumes in systems and networks, spanning the areas of Cyber-Physical Systems, Autonomous Systems, Sensor Networks, Control Systems, Energy Systems, Automotive Systems, Biological Systems, Vehicular Networking and Connected Vehicles, Aerospace Systems, Automation, Manufacturing, Smart Grids, Nonlinear Systems, Power Systems, Robotics, Social Systems, Economic Systems and other. Of particular value to both the contributors and the readership are the short publication timeframe and the world-wide distribution and exposure which enable both a wide and rapid dissemination of research output. The series covers the theory, applications, and perspectives on the state of the art and future developments relevant to systems and networks, decision making, control, complex processes and related areas, as embedded in the fields of interdisciplinary and applied sciences, engineering, computer science, physics, economics, social, and life sciences, as well as the paradigms and methodologies behind them. Indexed by SCOPUS, INSPEC, WTI Frankfurt eG, zbMATH, SCImago. All books published in the series are submitted for consideration in Web of Science.

More information about this series at http://www.springer.com/series/15179

Mustapha Hatti Editor

Artificial Intelligence and Renewables Towards an Energy Transition

123

Editor Mustapha Hatti EPST-CDER Unité de Développement des Equipements Solaires Bou Ismaïl, Algeria

ISSN 2367-3370 ISSN 2367-3389 (electronic) Lecture Notes in Networks and Systems ISBN 978-3-030-63845-0 ISBN 978-3-030-63846-7 (eBook) https://doi.org/10.1007/978-3-030-63846-7 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Contents

Hybrid Energetic Systems PV-WIND, with Energy Storage Control and Management Solar-Wind-Storage Hybrid System . . . . . . . A. Lakhdara, T. Bahi, and A. K. Moussaoui

3

Modeling and Simulation of PV/Wind Hybrid Energy System . . . . . . . . Slama Abdelhamid, Hamouda Messaoud, and Khiat Mounir

15

Energy Management Analysis of a Wind-Diesel-Battery Hybrid Power System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Nacereddine Guettaf, Seif El Islem Guettaf, Zahira Anane, and Hamou Nouri Power Management Strategy Applied on Hybrid Electric Train . . . . . . Imen Mammeri, M. Benidir, Hamza Bouzeria, Meriem Aissaoui, and Amira Chaib Ras Energy Saving Analysis of Municipal Pool Fed from Hybrid Renewable Energy in Batna Algeria . . . . . . . . . . . . . . . . . . . . . . . . . . . . T. Boutabba, M. L. Bechka, F. Menzri, and S. Drid Contribution of Renewable Energies in Existing Building Retrofits . . . . M. Badeche and Y. Bouchahm Robust Control of Grid-Interfaced Wind Energy Conversion System Based on Active Disturbance Rejection Control . . . . . . . . . . . . . . . . . . . Abdeldjabar Benrabah, Farid Khoucha, Fayçal Benyamina, Ali Raza, and Mohamed Benbouzid Wind Speed Forecasting Based on Discrete Wavelet Transform, Moving Average Method and Gated Recurrent Unit . . . . . . . . . . . . . . . K. Zouaidia, S. Ghanemi, and M. S. Rais DTC-DPC of Induction Machine Connected to Wind Generator . . . . . . Hamidia Fethia and Abbadi Amel

26

36

43 55

62

71 79

v

vi

Contents

Detecting Partial Shading in Grid-Connected PV Station Using Random Forest Classifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Abderrezzaq Ziane, Rachid Dabou, Noredine Sahouane, Ammar Necaibia, Mohammed Mostefaoui, Ahmed Bouraiou, and Abdeljalil Slimani Network Reconfiguration Management in Intelligent Distribution System Taking into Account PV Production Variation Using Grey Wolf Optimizer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mustafa Mosbah, Rabie Zine, Mustapha Hatti, Samir Hamid-Oudjana, and Salem Arif

88

96

An Enhanced MPPT Method Combining Fractional-Order and Fuzzy Logic PID Controller for a Photovoltaic-Wire Feeder System (PV-WFS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 N. Hamouda, B. Babes, A. Boutaghane, S. Kahla, and B. Talbi MPPT Charge Regulator and Monitor for Photovoltaic/Battery System Based on Microcontroller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 H. Assem, F. Bouchafaa, T. Azib, N. Allam, and N. Belhaouas Experimental Validation of a Prototype for Fault Detection and Classification of a Photovoltaic System Using dSPACE . . . . . . . . . . 125 A. Hamied, A. Rabhi, N. Rouibah, and A. Mellit Heuristic Optimization, Modeling and Control of Energetic Systems Optimization Assembly Line Balancing Variables Using Genetic Algorithm Based on Desirability Function Approach . . . . . . . . . . . . . . . 135 Samah A. Aufy and Allaeldin H. Kassam Design of Optimal Decentralized Controller Using Overlapping Decomposition for Smart Building System . . . . . . . . . . . . . . . . . . . . . . . 155 Mohamed Z. Doghmane, Madjid Kidouche, S. Eladj, and B. Belahcene Energy Management Strategy and Optimization of a MicroGrid System Based on Echo State Networks . . . . . . . . . . . . . . . . . . . . . . . . . . 168 I. Abadlia, L. Hassaine, F. Abdoune, and A. Beddar Control and Supervision of Water Tower in Smart City . . . . . . . . . . . . 178 L. Miloudi, A. Djenadi, and A. Hadj Youb Preventive Maintenance Optimization of Top Drives in Smart Rotary Drilling Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184 Idir Kessai, Samir Benammar, Mohamed Z. Doghmane, and Sadek Khelifa Stick-Slip Vibrations Control Strategy Design for Smart Rotary Drilling Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197 Mohamed Z. Doghmane, Abdelmoumen Bacetti, and Madjid Kidouche

Contents

vii

Comparison Study Between Improved JAYA and Particle Swarm Optimization PSO Algorithms for Parameter Extraction of Photovoltaic Module Based on Experimental Test . . . . . . . . . . . . . . . 210 Y. Rehouma, Z. Tir, A. Gacem, F. Rehouma, M. A. Hamida, and A. Gougui Performance Improvement of IFOC Based on Particle Swarm Optimization Algorithm in Induction Motor Control . . . . . . . . . . . . . . . 222 S. E. Rezgui, H. Benalla, and B. Nemmouchi Analyze and Fault Diagnosis of Double Star Induction Motor Using Wavelet Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233 Guermit Hossine and Kouzi Katia Modeling of Hydrocarbons Rotary Drilling Systems Under Torsional Vibrations: A Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243 Chafiaa Mendil, Madjid Kidouche, and Mohamed Z. Doghmane Optimal Placement of Distributed Generation Based PV Source in Electrical Power System for LVSI Improvement Using GA Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252 S. H. Oudjana, Rabie Zine, Mustafa Mosbah, Abdelouahab Khattara, and Salem. Arif Implementation and Optimization of PWM Technique for a Three-phase Inverter Associated with an Asynchronous Machine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260 Chafa Mohamed, Kamel Messaoudi, and Lamri Louze Real Time Implementation of Polynomial Control and Nonlinear Backstepping Strategies Integration for Motion Control of a PMSM . . . 270 Y. Mihoub, D. Toumi, S. Moreau, and S. Hassaine Discrete Time Sensorless PMSM Control Using an Extended Kalman Filter for Electric Vehicle Traction Systems Fed by Multi Level Inverter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281 A. Khemis R., T. Boutabba, and S. Drid Modeling and Characteristic Analysis of Harmonics in Railway Traction Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294 Meriem Aissaoui, M. Benidir, Hamza Bouzeria, Imen Mammeri, and Amira Chaib Ras Hierarchical Control of Paralleled Voltage Source Inverters in Islanded Single Phase Microgrids . . . . . . . . . . . . . . . . . . . . . . . . . . . 302 Ilyas Bennia, Yacine Daili, and Abdelghani Harrag Faults Detection and Diagnosis of Concentrated Photovoltaic Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314 N. Kellil, A. Aissat, and A. Mellit

viii

Contents

New Droop Control Technique for Reactive Power Sharing of Parallel Inverters in Islanded Microgrid . . . . . . . . . . . . . . . . . . . . . . 325 Yacine Daili, Abdelghani Harrag, and Ilyas Bennia Improved Performance of a Fuzzy Control of Double Star Induction Motor Fed by Two Three-Phases Matrix Converters . . . . . . . 336 Mebrouk Mennad, Abderrahim Bentaallah, Yousef Djeriri, and Aicha Bessas A Modified Maximum Power Point Tracking Controller Based on the Perturb and Observe Algorithm Used for Solar Energy System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351 Seif El Islem Guettaf, Abdelouahab Bouafia, Abdelmajid Chaoui, and Nacereddine Guettaf Real Time Implementation of Adaptive Discrete Fuzzy-RST Speed Control and Nonlinear Backstepping Currents Control Techniques for PMSM Drive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361 D. Toumi, Y. Mihoub, S. Moreau, and S. Hassaine Adaptive Self-tuning Backstepping - Nonlinear (NLPI) Controller for the Control of Electric Vehicle with Two-Motor-Wheel Drive . . . . . 374 Ahmed Laguidi, Cherif Benoudjafer, and Othmane Boughazi Optimal Allocation of Renewable Energy Source Integrated-Smart Distribution Systems Based on Technical-Economic Analysis Considering Load Demand and DG Uncertainties . . . . . . . . . . . . . . . . . 391 Mohamed Zellagui, Samir Settoul, Adel Lasmari, Claude Ziad El-Bayeh, Rachid Chenni, and Heba Ahmed Hassan People Counter with Area Occupancy Control for Covid-19 . . . . . . . . . 405 E. Khoumeri, H. Fraoucene, El Hadi Khoumeri, C. Hamouda, and R. Cheggou Finite Element Modelling and Analysis for Modal Investigation of a Blade H-Type Darrieus Rotor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 416 F. Ferroudji, L. Saihi, and K. Roummani Engineering Applications of Artificial Intelligence Heating Load Energy Performance of Residential Building: Machine Learning-Cluster K-Nearest Neighbor CKNN (Part I) . . . . . . . 425 Aissa Boudjella and Manal Y. Boudjella Cooling Load Energy Performance of Residential Building: Machine Learning-Cluster K-Nearest Neighbor CKNN (Part I) . . . . . . . 436 Aissa Boudjella and Manal Y. Boudjella

Contents

ix

Fuzzy Traffic Lights Controller Based on PLC . . . . . . . . . . . . . . . . . . . 447 Mounir Bouhedda, Hamza Benyezza, Yassine Toumi, and Samia Rebouh Implementation of a Smart Traffic Light Controller Based on Multi-agent System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 457 Mounir Bouhedda, Abderrezak Aggoun, Samia Rebouh, and Abderrahmane Oudjouadj Intelligent Solar Shunt Active Power Filter Based on Direct Power Control Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 467 Ghania Boudechiche, Mustapha Sarra, Oualid Aissa, and Abderezak Lashab MPPT - Based Improved Salp Swarm Algorithm for Improving Performance and Efficiency of Photovoltaic System Under Partial Shading Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 478 H. Azli, S. Titri, and C. Larbes Multi-agent System for Voltage Regulation in Smart Grid . . . . . . . . . . . 487 Hadjira Belaidi, Hamid Bentarzi, Zakaria Rabiai, and Abdelkader Abdelmoumene Neural Network-Based Attitude Estimation . . . . . . . . . . . . . . . . . . . . . . 500 Djamel Dhahbane, Abdelkrim Nemra, and Samir Sakhi Developing an Improved ANN Algorithm Assisted by a Colony of Foraging Ants for MPP Tracking of Grid Interactive Solar Powered Arc Welding Machine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 512 S. Kahla, B. Babes, N. Hamouda, A. Boutaghane, and A. Bouafassa Optimal Siting and Sizing of DG Units Using a Decomposition Based Multiobjective Evolutionary Algorithm . . . . . . . . . . . . . . . . . . . . 521 Yaaqoub Latreche, Houssem R. E. H. Bouchekara, Muhammad S. Javaid, Mohammad S. Shahriar, Yusuf A. Sha’aban, and Fouad Kerrour Task Scheduling-Energy Efficient in Cloud Computing . . . . . . . . . . . . . 533 Karima Saidi and Ouassila Hioual An Efficient Hybrid Meta-heuristic Approach for Solving the K-Shortest Paths Problem Over Weighted Large Graphs . . . . . . . . 541 Mohamed Yassine Hayi and Chouiref Zahira A New Mutated-Firefly Algorithm for Parameters Extraction of Solar Photovoltaic Cell Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 551 B. Aoufi and O. Hachana A Survey on Cloud-Based Intelligent Transportation System . . . . . . . . . 562 Y. Khair, A. Dennai, and Y. Elmir

x

Contents

Artificial Neural Network Based Solar Radiation Estimation of Algeria Southwest Cities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 573 D. Benatiallah, K Bouchouicha, A Benatiallah, A. Harouz, and B. Nasri ANN-Based Correction Model of Radiation and Temperature for Solar Energy Application in South of Algeria . . . . . . . . . . . . . . . . . 584 K. Bouchouicha, N. Bailek, M. Bellaoui, B. Oulimar, and D. Benatiallah Parameter Extraction of Two-Diode Solar PV Model Using ANN–GA Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 592 A. Aissaoui, N. Belhaouas, F. Hadjrioua, K. Bakria, and I. Aloui Application of the Genetic Algorithm to the Rule Extraction Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 604 Dounia Yedjour Adaptative Neuro-Fuzzy Inference System for Predicting the Microbial Fungicide Release from Microcapsules Based on Alginate and Different Gelatin Proportions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 612 Hadjer Goudjil, Samia Rebouh, and Mounir Bouhedda Application of Artificial Neural Network-Genetic Algorithm Model in the Prediction of Ibuprofen Release from Microcapsules and Tablets Based on Plant Protein and Its Derivatives . . . . . . . . . . . . . 625 Asma Ghennam, Samia Rebouh, and Mounir Bouhedda Multi-class EEG Signal Classification for Epileptic Seizure Diagnosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 635 Dalila Cherifi, Laid Afoun, Zakaria Iloul, Billal Boukerma, Chaouki Adjerid, Larbi Boubchir, and Amine Nait-Ali Internet of Things, Database and Transmission Detection of H2S Gas Concentration in Oil Refinery Stations by Using Drone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 649 Ibtihaj A. Abdulrazzak, Hussain M. Bierk, and Anas F. Ahmed Study and Assembly of Quadrotor UAV for the Inspection of the Cellular Networks Relays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 659 Hicham Megnafi and Walid Yassine Medjati Divide and Conquer Strategy for Trust Evaluation in Wireless Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 669 Ahmed Saidi, Khelifa Benahmed, and Nouredine Seddiki Smart Unidirectional Road Lighting Control Using NodeMCU ESP8266 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 682 Mouaad Yaichi, Mhamed Rebhi, and Bouchiba Bousmaha

Contents

xi

Cloud-IoT Platform for Smart Irrigation Solution Based on NodeMCU . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 689 Dougani Bentabet Design and Implementation of M2M-Smart Home Based on Arduino-UNO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 697 Rania Djehaiche, Salih Aidel, and Nihad Benziouche A Study of the Parametric Variations Influences on Stick-Slip Vibrations in Smart Rotary Drilling Systems . . . . . . . . . . . . . . . . . . . . . 707 Chafiaa Mendil, Madjid Kidouche, and Mohamed Z. Doghmane Algerian Dialect Translation Applied on COVID-19 Social Media Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 716 Amel Slim, Ahlem Melouah, Yousef Faghihi, and Khouloud Sahib Prospective Analysis for a Long-Term Optimal Labor Force Planning in Algeria (PALOLFA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 727 MS. Guellil, S. Ghouali, O. Khedir, D. Benabou, H. Ayad, and SE Sari-Hassoun Electronically Switchable SIW Band-Pass Filter Based on S-CSRR Using PIN Diodes for WI-FI Applications . . . . . . . . . . . . . . . . . . . . . . . 738 Hichem Boubakar, Mehadji Abri, and Mohamed Benaissa Flexible and Automated Watering System Using Solar Energy . . . . . . . 747 Hicham Megnafi, Arezki Abderrahim Chellal, and Abdeldjalil Benhanifia Towards High Performance Full Duplex MAC Protocol in High Efficiency WLANs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 756 Kenza Hocini and Mohand Yazid Monitoring of Isolated Standalone Renewable Energy Systems . . . . . . . 766 M. Tsebia and H. Bentarzi Towards an Automatic Evaluation of the Performance of Physical Unclonable Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 775 Fahem Zerrouki, Samir Ouchani, and Hafida Bouarfa NTSOAP: A Robust Approach for Non-redundancy Tags of SOAP Messages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 782 Nassima Belkacem, Fouzi Semchedine, Djamil Aissani, and Ahmed Al-Shammari Arabizi Chat Alphabet Transliteration to Algerian Dialect . . . . . . . . . . 790 B. Klouche and S. M. Benslimane

xii

Contents

WSN Based Smart Farm Powered by Solar Energy Harvesting Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 798 S. Titri and N. Izeboudjen Power and Materials in Renewable Energetic Systems Visual Degradation of PV Modules After 30 Years of Exposure in Algeria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 811 Fathia Chekired, El Amin Kouadri Boudjelthia, Fatah Mehareb, and Amina Chahtou Variable Step Size Techniques for Conventional MPPT Algorithms . . . 822 Mhamed Rebhi, Othmane Abdelkhalek, Bouchiba Bousmaha, and Mouaad Yaichi Experimental Investigation to Determine the Performance of Solar Thermal Collector with Single and Double Glazing . . . . . . . . . . . . . . . . 831 Djamel Bensahal, Foued Chabane, Ahmed Yousfi, and Mohamed Rahmani Numerical Study of an Air Flow in a Flat Plate Air Solar Collector with Circular Obstacles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 839 M. A. Amraoui Application of Multiple Population Genetic Algorithm in Optimizing Business Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 847 Nadir Mahammed, Souad Bennabi, and Mahmoud Fahsi An Experimental Validation on Mechanical Damages Caused by Air Cannon Projectile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 856 Mossaab Chenine, Samir Benammar, and Mohamed Z. Doghmane Development of Two Axis Solar Tracking System Interface Using Matlab GUI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 868 Abdelghani Harrag and Yacine Daili Extracting Methods of Positive and Negative Voltage Sequences for Unbalanced Three-Phase Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 877 Chelli Seif Elislam, Boussaid Abdelfettah, and Nemmour Ahmed Lokmane Extended Kalman Filter for the Estimation of the State of Charge of Lithium-Ion Batteries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 885 Mohamed Khalfaoui and Aissa Hamlat An Efficient Strategy for Power Quality Conditioner with Half-Bridge for High-Speed Railway . . . . . . . . . . . . . . . . . . . . . . . 894 Amira Chaib Ras, Ramdane Bouzerara, Hamza Bouzeria, Meriem Aissaoui, and Imen Mammeri

Contents

xiii

Study of a Tri-generation System Using Hydrogen and Fuel Cell in Algiers, Algeria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 902 A. Mraoui, A. Belgacem, S. Djellab, and M. Boudiaf Single-Phase Synchronization Systems for Grid-Connected Converters Based on Enhanced Third-Order Sinusoidal Integrator . . . . 911 S. Kalkoul, H. Benalla, K. Nabti, and Houssem R. E. H. Bouchekara Numerical Simulation of a Solar Cell in CdS/CIGS Using the SCAPS-D1 Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 920 Ghalmi Leila and Bensmaine Souhila Thermal Characterization of a SOFC Fuel Cell . . . . . . . . . . . . . . . . . . . 930 M. Mankour, M. Sekour, and L. Boumadien Optical Flow Based on Lucas-Kanade Method for Motion Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 937 H. Yedjour Simulation of the Electric Properties of a Structure Based on Two Gan P-N Junctions Grown on an Undoped ZnO Nanosheet . . . 946 Zehor Allam, Chahrazad Boudaoud, A. Moumene Benahmed, and Aicha Soufi Spectroscopic Measurements of the Electron Temperature in Plasmas Containing Argon Using Line Intensity Ratio Method . . . . . 958 Askri Souhaila, Ferouani Abdel Karim, Guedda El Habib, and Sahlaoui Mohammed Numerical Simulation of the NOx Chemical Kinetic Removal by Under Various Reduced Electric Fields . . . . . . . . . . . . . . . . . . . . . . . 966 Askri Souhaila, Ferouani Abdel Karim, and Mostefa Lemerini Design and Simulation of a DC-DC Boost Converter Embedded for Space Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 975 A. Hadj Dida, M. Bensaada, and M. Bekhti Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 987

About the Editor

Dr. Mustapha Hatti was born in El-Asnam (Chlef), Algeria. He studied at El Khaldounia School and then at El Wancharissi High School, obtained his electronics engineering diplomat from USTHB Algiers, and his postgraduation studies at USTO -Oran. He worked as research engineer, at CDSE, Ain oussera, Djelfa, CRD, Sonatrach, Hassi messaoud, CRNB, Birine, Djelfa, and Director of Research at UDES / EPST-CDER, Bou Ismail, Tipasa. He leads the “Tipasa Smart City” Association and is an IEEE Senior Member; he is the author of several scientific papers, chapter books. Book editor and several journals special sessions guest editor. His area of interest are smart sustainable energy systems, innovative systems, fuel cell, photovoltaic, optimization, intelligent embedded systems, and artificial intelligence.

xv

Hybrid Energetic Systems PV-WIND, with Energy Storage

Control and Management Solar-Wind-Storage Hybrid System A. Lakhdara1(&), T. Bahi2, and A. K. Moussaoui1 1

Department of Electrical and Automatic Engineering, LGEG, Guelma University, Guelma, Algeria [email protected], [email protected] 2 Department of Electrical, LASA Laboratory, Badji Mokhtar Mokhtar University, Annaba, Algeria [email protected]

Abstract. Nowadays, the development concept of renewable energy conversion facilities is adopted by the majority of countries as a property to be promoted urgently in order to provide global and sustainable solutions to environmental challenges and to cope with the undeniable depletion of fossil energy resources. Indeed, renewable energies (solar, wind, etc.) are a promising alternative for achieving an energy transition and sustainable economic development. In view of the report on the accelerated depletion of fossil resources due to the ever increasing energy needs and the challenges of environmental preservation of carbon dioxide emissions, the use of renewable resources for the production of electricity is a promising alternative. However, solar and wind resources are of intermittent types because the wind turbine output power varies with the wind speed at different conditions and the solar energy also varies with the hourly, daily and seasonal variation of solar irradiation, we propose in this work, behavioural study and analysis of an hybrid generation system combining solar and wind energy connected to the grid with a battery (energy storage) to ensure that the system performs under different climatic conditions. The objective of this work is to ensure the best performances of the proposed hybrid configuration under different possible operating scenarios. Energy management between of renewable energy sources (PV-arrays, wind generator and energy storage), component the hybrid generation system and considered in order to meet the sustained load demands during the varying natural conditions. Keywords: Photovoltaic energy system


Wind energy
Storage energy
Hybrid

1 Introduction In last decade, many problems related to energy factors, ecological aspects, electric demand and financial regulatory restrictions of wholesale markets have arisen worldwide [1]. Consequently, It is then urgent to reduce our dependence on fossil fuels by employing renewable technologies that use natural resources such as wind and solar power to produce clean electricity and thus avert a global disaster and maintain fossil fuel reserves underground [2]. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 3–14, 2021. https://doi.org/10.1007/978-3-030-63846-7_1

4

A. Lakhdara et al.

Renewable Energies (RE) include a number of technological sectors depending on the source of energy recovered and the useful energy obtained. Among them, solar and wind energy represent an efficient and sustainable development possibility and that is why scientific research is developing in the sense of generalizing, improving and optimizing the exploitation of hybrid systems. The Photovoltaic (PV) systems have attracted considerable attention thanks to their remarkable advantages compared to the production of electric energy based on fossil fuels. Solar Cells can be classified as a semiconductor device when solar irradiation penetrates to the solar cells surface, DC flow through the PV panels. On the other hand, the wind energy is one of the most promising and fastest growing energy resources in the world in spite of the nature of the wind and its continually changing speed. A wind turbine is a rotary engine that captures power from a fluid flow (wind) using aerodynamically designed blades and convert it into useful mechanical power [3, 4]. However, because of the intermittent availability of these two last energy sources, a storage system is necessary for the continuity of the service. Indeed, a storage system is required to compensate the lack of the power under all conditions and to ensure a performing and a continuous power source for consumers using a bidirectional DC-DC converter which is controlled to satisfy the energy required by the load [5–7]. The objective of this work is the development of a power management mechanism of the variouced components of the hybrid system. For this purpose, it is a question of putting in equations each component for the modeling and the control of the whole of system. At this level, the necessary condition is to know the entry criteria which are the meteorological data of the site and the data relating to the equipment. So, the system considered consists essentially of a photovoltaic system, a wind turbine, a storage battery and a DC load. In that case, the excess of energy produces compared to that which needs the load would be stored (charging the batteries) and when the energy demanded by the load and higher than that can provide the hybrid system, the stored energy would be used as a complement (unloading batteries) so that the load receives the requested energy whatever the conditions. Batteries are storage devices that continually require a lot of effort to improve their operation. Each part of the proposed chain as well as the global chain are simulated under the MatLab/Simulink software thanks to which the behavior of the installation under different environmental conditions is analyzed in order to validate the study. However, in the first part of this work, the operating principle of the main sources (Wind Turbine and photovoltaic panel) and the modeling of the different constituent parts is developed. In the second part, we are particularly interested in the need for the storage system and its mode of operation under the conditions considered. Simulation results and discussions are presented in Sect. 3 to initiate the analysis of the operation of a hybrid system: PV-Wind turbine-Battery and to validate its operation under irradiation profiles, temperature and wind speed. Finally, conclusions are given in Sect. 4.

Control and Management Solar-Wind-Storage Hybrid System

5

2 Modelisation This section presents the modeling of an hybrid solar-wind-storage system. The application applicable to the analysis and also simulation of a real hybrid solar-windstorage system. one of the keys studies is the behavior of the hybrid process which allows to use renewable energies and power adjustable over time while providing continuous power [8, 9] (Fig. 1).

Fig. 1. Hybrid conversion system

2.1

Photovoltaic System

In this section, we present the modeling of the photovoltaic module which is the main component for the conversion of solar energy as well as the operating principle of the photovoltaic module. A particular interest is reserved for the operation of the photovoltaic generator and its operation at the maximum power point (MPPT) despite the variation particularly the temperature and the irradiation. • Photovoltaic cell Figure 2 shows the single diode-equivalent electrical circuit of a photovoltaic cell [10]. with, ICC(A): short-circuit current of the cell depending on the irradiation and the temperature, I(A): cell current; V(V): cell voltage; Id(A): diode Current; IRp(A): Current of parallel resistor; RP(Ω): parallel resistance which characterizes the junction currents; Rs(Ω): series resistance which characterizes the various resistances of the contacts and connection. The law of the nodes makes it possible to write the following relation [11, 12]:

6

A. Lakhdara et al.

Fig. 2. PV cell equivalent electrical circuit model

I ¼ Icc  Id  IRp

ð1Þ

The junction current Id is given by: q

Id ¼ I0 ðenkTc ðV þ I:Rs Þ  1Þ

ð2Þ

The current in the resistor Rp is given by: IR p ¼

V þ I:Rs Rp

ð3Þ

We obtain: I ¼ Icc  Id 

V þ I:Rs Rp

ð4Þ

And considering that the photovoltaic cell is of good quality, the Rp will have a very high value and therefore the third term of the right part of the Eq. (4) can be neglected ðV þRpI:Rs Þ. Thus, we retain from (2) that: q

enkTc ðV þ IRs Þ [[ 1: So the current-voltage equation of the cell is written: q

I ¼ Icc  I0 enkTc ðV þ I:Rs Þ

ð5Þ

I0(A): saturation current of diode; q = 1.602  10−19 (C): electron load; n: nonideality factor of the diode junction; K = 1.381.10−23 J/K: Boltzmann constant; Tc (K): effective cell temperature; T(K) = 273 = T (°C).

Control and Management Solar-Wind-Storage Hybrid System

7

• Boost Converter Model It is necessary for the conversion voltaic chain, to ensure operation in its optimal regime, it plays the role of source-charge adapter while ensuring, thanks to an adequate control strategy, the transfer of the maximum possible power to the maximum power available from that supplied by the GPV [10] (Fig. 3).

Fig. 3. Schematic diagram of a Boost converter

The duty cycle (Dcycle) is then expressed by the following relationship: Dcycle ¼

To 2 ½0; 1 Tc

ð6Þ

Then, depending on the state (Open or Closed) of the semiconductor (K), we distinguish two (2) configurations relative to the two possible operating phases. Applying Kirchhoff’s laws, we obtain: 8 dvi ðtÞ > < ic1 ðtÞ = c1 dt = ii ðtÞ  il ðtÞ ic2 ðtÞ = c2 dvdto tÞ =  io ðtÞ > : vl ðtÞ = l didtl ðtÞ =  vi ðtÞ

ð7Þ

8 dv ðtÞ > < ic1 ðtÞ = c1 dti = ii ðtÞ  il ðtÞ ic2 ðtÞ = c2 dvo ðtÞ = il ðtÞ  i0 ðtÞ > : v ðtÞ = l dildt= v ðtÞ  v ðtÞ l i o dt

ð8Þ

And,

• Maximum Power Point Tracking Techniques The power generated by the GPV is highly dependent on the intensity of solar irradiation (E) and the temperature (T), a control (MPPT: Maximum Power Point Tracking) is provided to extract the maximum power [13, 14] (Fig. 4).

8

A. Lakhdara et al.

Fig. 4. P&O algorithm

2.2

Wind Conversion System

The wind generator consists on a turbine directly coupled to a synchronous generator with permanent magnets. The blades are rotated by the energy of the wind. • WIND TURBINE Wind turbine converts wind energy into mechanical power. The mechanical power is computed as follows [15]: 1 Pm ¼ Cp :Pe ¼ qpR2 Vw3 Cp ðkÞ 2

ð9Þ

1 Pm ¼ qAVw3 Cp ðk; bÞ 2

ð10Þ

The theoretical value is well-known as ‘Betz limit’ which determines the maximum power that can be extracted from a given wind speed and is defined by [16]:   21 116 Cp ðk; bÞ ¼ 0:5176 ð  0:4b  5Þe ki þ 0:0068k kI

ð11Þ

With: k¼

xm :R Vw

ð12Þ

and: 1 1 0:035  3 ¼ ki k þ 0:08b b þ 1

ð13Þ

Control and Management Solar-Wind-Storage Hybrid System

9

Where, q is the air density (kg/m3), A is the blades swept area, Vw is the wind speed (m/sec), and CP (k,b) is the power coefficient, which is defined as the ratio of turbine power to wind power and it is a function of the pitch angle (b) and tip speed ratio (k). • GENERATOR The Asynchronous Machine (AM) is an electromechanical converter which produces an electric current whose frequency is determined by the rotation speed of this machine [17]. It’s defined by the differential equations of the stator voltages where the corresponding fluxes are written by the expressions (14) and (15), respectively: • Voltage equations 8 > < Va ¼ Ria þ Vb ¼ Rib þ > : Vc ¼ Ric þ

d/a dt d/b dt d/c dt

ð14Þ

• Flux equations 8 < /a ¼ Ls ia þ /fa / ¼ Ls ib þ / : /b ¼ L i þ /fb s c c fc

ð15Þ

With, Ls ¼ L  M

ð16Þ

R is the resistance of a stator phase; /a , /b , /c are the fluxes across the stator phases; L is the own stator inductor; M is the mutual inductance between two stator phases; /fa , /fb , /fc are the total fluxes sent by the inductor in the three phases; Ls is the cyclic stator inductance or synchronous inductance. The dynamic equation is: J

2.3

dxr ¼ Tm  Tem dt

ð17Þ

Storage System

The battery can be represented by its equivalent circuit [18], shown in Fig. 5. The excess energy produced with that required by the charge would be stored (battery charge) and used as a supplement (battery discharge) so that the charge receives the required energy under all conditions. The equation of the voltage across the battery (Vbat) is given by the following relation:

10

A. Lakhdara et al.

Fig. 5. R-C battery model

Vbat ðtÞ ¼ Vc  Rv Ibat ðtÞ

ð18Þ

Where, Vc: ideal input voltage source; Rv: variable internal battery resistance and Ibat: current developed by the battery. The state of charge (SOC) of the battery is: SOC ¼ 1 

Qd Cbat

ð19Þ

With, Qd: battery charge; Cbat: nominal battery capacity. The state of charge (SOC) provides information on the state of charge of the battery, expressed as a percentage, in order to avoid deep discharges or excessive charges which would damage the batteries [18]. As indicated in following equation: SOC ¼

AvailabCbat ðAHÞ :100% NominalCbat ðAHÞ

ð20Þ

3 Simulation Results and Discussion This section is reserved for the analysis of the power management of the hybrid system considered. To this effect, a set of input variables (Irradiation, Temperature and Wind Speed) has been developed, taking into account the operation of each element in the production chain. The selected profiles are shown in Fig. 6, where there are five (5) remarkable intervals that are recorded in the Table 1. Furthermore, considering a profile of the load, as shown in Fig. 7 (Table 2).

Control and Management Solar-Wind-Storage Hybrid System

11

Fig. 6. Profiles of input variables Table 1. Input variables Intervals (s) 0  t < 0.25 0.25  t < 0.5 0.5  t < 0.75 0.75  t < 1.5

2

T (°C) E (W/m ) 1000 25 1000 25 Decrease from 1000 to 400 25 400 Increase from 25 to 30

Ws (m/s) Increase from 5 to 10 Decrease from 10 to 4 Increase from 4 to 11 11

Under all these conditions, the Fig. 7 shows the shape of the power produced by the solar panel. It mainly depends on the evolution of available irradiation. The same figure, also shows the shape of the power produced by the wind turbine (PWind). This allows to view the profile of the total power (PTotal) produced simultaneously by the photovoltaic and wind turbine chain. Thus, according to these profiles, it’s noted that thanks to the bidirectional converter, the operation of the battery in charge and discharge mode is carried out as a function, on the one hand, of the difference between the total power and the charge power, and on the other hand, the control signals of the two

12

A. Lakhdara et al.

Fig. 7. Comportment of powers Table 2. Load profile Intervals (s) PLoad (KW) 0  t < 0.5 10 0.5  t < 1 12 1  t < 1.5 6

semiconductors of the Buck-Boost chopper as shown by the signals in the middle of Fig. 8 which corresponds perfectly to the shape of the state of charge (SOC) of the battery. For the same figure, either the charge mode for the intervals (0 s  t < 0.5 s) & (1 s  t < 1.5 s) and the discharge mode for (0.5 s  t < 1 s). This interpretation reflects the good power management for the hybrid system considered, including two main sources (Photovoltaic and Wind energies) and an auxiliary source (Storage Battery).

Control and Management Solar-Wind-Storage Hybrid System

13

Fig. 8. Charge and discharge modes

4 Conclusion The use of renewable energy conversion systems is a promising and indisputable alternative to the use of fossil fuels, thanks to the advantages they present, mainly, by their availability (durable and inexhaustible) and by their cleanliness (no discharge of waste into the atmosphere). However, the intermittent nature of solar and wind energy, the option of hybridization with storage proves to be a configuration of great importance. For this purpose, an hybrid system is of capital interest for isolated sites. In addition of the interest of such installations, the storage system plays a key role for the continuity of service under the available weather conditions.

References 1. John, A.R., Thomas, F., Sunny, A.S., Balakrishnan, K.J., Ashok, A., Pathirikkat, G.: Multiple renewable energy extraction using MISOC topology for residential applications. In: International Conference on Computer Communication and Informatics (ICCCI-2017), 05 January 2007, Coimbatore, India (2017) 2. Balamurugan, T., et al.: Optimal power flow management control for grid connected photovoltaic/wind turbine/diesel generator (GCPWD) hybrid system with batteries. Int. J. Renew. Energy Res. 3(4), 819–826 (2013)

14

A. Lakhdara et al.

3. Soetedjo, A., Lomi, A., Mulayanto, W.P.: Modeling of wind energy system with MPPT control. In: International Conference on Electrical Engineering and Informatics, 17-1 Bandung, Indonesia (2011) 4. Prechanon, K.: Mathematical model of the PMSG based on wind energy conversion system. Int. Res. J. Innov. Eng. 1(3) (2015). ISSN 2395-0560. www.irjie.com 5. Dalia, M., Belhadja, J., Roboamb, X.: Hybrid solar–wind system with battery storage operating in grid-connected and standalone mode: control and energy management – experimental investigation. Energy J. 35(6), 2587–2595 (2010) 6. Singh, R.S.S., Abbod, M., Balachandran, W.: A design scheme of control/optimization system for hybrid solar—wind and battery energy storages system. In: 51st International Universities Power Engineering Conference (UPEC), Coimbra, pp. 1–6 (2016). https://doi. org/10.1109/upec.2016.8114093 7. Badwawi, R.A., Abusara, M., Mallick, T.: A review of hybrid solar PV and wind energy system. Smart Sci. 3(3), 127–138 (2015). https://doi.org/10.1080/23080477.2015.11665647 8. Anilkumar, T.T., Nayak, P.S.R., Simon, S.P.: Experimental investigation on a prototype solar-wind hybrid system with a pico hydro turbine. Int. J. Emerg. Electr. Power Syst. 18(5) (2016). https://doi.org/10.1515/ijeeps-2016-0103 9. Sumathi, S., Ashok Kumar, L., Surekha, P.: Solar PV and wind energy conversion systems. ISSN 1865-3529 ISSN 1865-3537 (electronic) Green Energy and Technology, ISBN 978-3319-14940-0 ISBN 978-3-319-14941-7 (eBook). https://doi.org/10.1007/978-3-319-14941-7. www.springer.com 10. Bhandari, B., Poudel, S.R., Lee, K., et al.: Mathematical modeling of hybrid renewable energy system: a review on small hydro-solar-wind power generation. Int. J. Precis. Eng. Manuf.-Green Technol. 1, 157–173 (2014). https://doi.org/10.1007/s40684-014-0021-4 11. Sachin, C., Shah, K.B.: Solar photovoltaic fed induction motor for water pumping system using MPPT algorithm. Int. J. Electr. Electron. Eng. (IJEEE) 7(3), 31–42 (2018) 12. Reddy, D.C.K., Narayana, S.S., Ganesh, V.: Performance of DQ based controller for solar wind hybrid power system. Recent Adv. Electr. Electron. Eng. 12(2) (2019). https://doi.org/ 10.2174/2352096511666180514111606 13. Esram, T., Chapman, P.L.: Comparison of photovoltaic array maximum power point tracking techniques. IEEE Trans. Energy Convers. 22(2), 439–449 (2007). https://doi.org/10. 1109/TEC.2006.874230 14. Mihir, P., Vimith, S., Diptarka, D., Valunjkar, R.: Designing and implementation of maximum power point tracking(MPPT) solar charge controller. In: International Conference on Nascent Technologies in Engineering (ICNTE) (2017). https://doi.org/10.1109/icnte. 2017.7947928 15. Lakhdara, A., Bahi, T., Abdelkrim, M. Sliding mode control of doubly-fed induction generator in wind energy conversion system. In: 8th International Conference on Smart Grid, icsmartgrid, 17–19 June 2020, Paris/France (2020) 16. Blackwood, M.: Maximum efficiency of a wind turbine. Undergraduate J. Math. Model. 6(2) (2016). http://scholarcommons.usf.edu/ujmm/vol6/iss2/2 17. Dadabaev, T., Toshkhodzhaeva, I., Mirkhalikova, S.: Modeling of starting transition processes of asynchronous motors with reduced voltage of the supply network. Eur. J. Electr. Eng. 22(1), 23–28 (2020). https://doi.org/10.18280/ejee.220103 18. Babazadeh, R., Khiabani, A.G.: Nonlinear observer design for rc battery model for estimating state of charge & state of health based on state-dependent riccati equation. In: Conference: 2018 IEEE Electrical Power and Energy Conference (EPEC), At Toronto, ON, Canada (2019)

Modeling and Simulation of PV/Wind Hybrid Energy System Slama Abdelhamid1(&), Hamouda Messaoud2, and Khiat Mounir1 1

SCAMRE Laboratory, Electrical Engineering Department, ENPO-MA, Es-Sénia Road, B.P. 1523 El M’Naouer, 31000 Oran, Algeria [email protected] 2 DDI Laboratory, Ahmed Draia University, 01000 Adrar, Algeria [email protected]

Abstract. This work provides a description of a hybrid system connected to the grid, including a PV system and Wind turbine system that share a DC bus and without battery. The paper begins with a description of the system; it has been provided a brief overview of each user component of this system. It was also presented how to control this system. The system has been simulated with accurate climatic conditions and according to an operating system; it is assumed that the two systems work together at the same time, with a focus on clarifying the work of each system according to these climatic conditions. The simulation results show how changing climate conditions can affect the energy produced by the hybrid system, and it also shows power exchanges with the grid. Keywords: Hybrid


PV
Wind turbine
Grid
Power

1 Introduction The great demand of electric energy and the fluctuating price of fuels and their rise, prompted nations to race to devise the most effective way to produce electric energy. Renewable energies used in various ways to generate electric energy. It was also necessary to incorporate different structures for the production of electric energy in the form of hybrid systems. Among them is the wind photovoltaic hybrid system. Several studies in this field has different structures proposed for this type of systems, including study [12]. In our study, we conducted a study of the process of combining two renewable systems for generating electric energy that share the DC bus without a battery. The system has also been simulated under specific weather conditions to clarify the effect of these conditions on the work of the hybrid system.

2 System Description In this paper, the proposed hybrid system assemble a turbine, a DFIG, a photovoltaic array [1], and power converters. The turbine captures the energy of the wind and transmits it to the DFIG. Which connects to the Grid, and Photovoltaic array uses an MPPT

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 15–25, 2021. https://doi.org/10.1007/978-3-030-63846-7_2

16

S. Abdelhamid et al.

connected to the DC/DC converter and DC bus. Also, a control system regulates the hybrid system work. Figure 1 shows the block diagram of the proposed hybrid system.

Fig. 1. Block diagram of the proposed hybrid system

3 Modelling of PV System 3.1

Modeling of PV Array

The essential element of the PV array is the solar cell, which used the photovoltaic effect and converts solar irradiation into DC current. PV cells are grouped together in larger units to get a PV array, which are combined in series and parallel to provide the required output voltage and current. Figure 2 shows the equivalent circuit of the solar cell and the PV array is only the cells arranged in series or in parallel or mixed [2].

Fig. 2. Equivalent circuit of a solar cell.

The mathematical model that estimates the power generation by PV array becomes an algebraically simply model, being the current–voltage relationship defined in (1) [2, 3] "
Ipv ¼ Npp :I0  Npp :Icc e

Vpv Nss :Vt

I

Rs pp;Vt

þ Npv



# 1 

  Npp Vpv Ipv :Rs þ RP Nss Npp

Where: Ipv, Vpv: PV array output current and voltage I0: Solar cell photocurrent Nss, Npp: Number of series and parallel connected pv cells Icc: Solar cell reverses saturation current RS: Cell intrinsic series resistance

ð1Þ

Modeling and Simulation of PV/Wind Hybrid Energy System

17

RP: Cell intrinsic shunt or parallel resistance The above model has been implemented using Matlab/Simulink. The parameters of the model are given in Table 1. Table 1. Parameters of Canadian Solar CS5P-230 M solar module at 25 °C; 1000 W/m2 Parameters Cell numbers of a PV module Nominal short-circuit voltage, A Nominal array open-circuit voltage, V Array current maximum power point, A Array voltage maximum power point, V

Value 96 5.25 58.8 4.84 47.5

Figure 3 illustrates the simulated curves (I-V) and (P-V) for a variable temperature of 25 °C to 75 °C with a constant irradiation of 1000 Wh/m2 and the simulated curves (I-V) and (P-V) for a constant temperature of 25 °C and variable irradiation from 200 Wh/m2 to 950 Wh/m2.

Fig. 3. Curves Ipv(Vpv) and Ppv(Vpv) of solar module: a) at 25 °C and variable irradiation. b) at irradiation 1000 Wh/m2 and variable temperature

3.2

DC_DC Converter

The Boost converter is designed to increase the low voltage of the PV array output. It contains an inductor (L), a diode, a capacitor (C), and a high frequency operated switch like IGBT/MOSFET. This switch is managed by the pulse signal output of MPPT controller [5]. It is shown in Fig. 5.

18

3.3

S. Abdelhamid et al.

MPPT Controller

Several methods are used to maximize energy in different climatic conditions of radiation and temperature [8, 9, 11], among which is the P&O method, which generate the control signal to feed the boost converter switch. The objective of the P&O calculation is to find the varying direction of the current load. Figure 4 shows the process flow of the MPPT algorithm for the proposed system.

Fig. 4. Flow chart of the MPPT algorithm with P&O method.[11]

3.4

DC_AC Converter

Photovoltaic systems exchange power with the grid via a boost converter and a voltage inverter. The DC–DC converters are used to balance the system. Inverters are responsible for converting DC power to AC power. This is done by switching the DC input voltage (or current) in a pre-determined sequence to generate AC voltage (or current) output. Three-phase inverters consist of one, two, or three arms of power switching devices. Each arm consists of four switching devices along with their antiparallel diodes and two neutral clamping diodes [6].

4 Modelling of Wind System The mechanical power extracted from the wind is given as follows: [7, 10] Pm ¼

1  Cp ðk; bÞqpR2 V 3 2

ð2Þ

Modeling and Simulation of PV/Wind Hybrid Energy System

k¼

Xt :R V

19

ð3Þ

Where q is the air density (kg/m3), R is the blade radius (m), V is the wind speed (m/s), Cp(k, b) represents the power coefficient, and can have a value between (0.4 and 0.5). It could be expressed as: [7, 10]   21 116 Cp ¼ 0:5872:  0:4:b  5 :e ki þ 0:0085:k ki

ð4Þ

1 1 0:035  ¼ ki k þ 0:08:b 1 þ b3

ð5Þ

Where; b is the pitch angle and (k) is the tip speed ratio. The electrical generator model (DFIG) is designed using the built-in SimPowerSystem library. The rotor shaft is driven by the Wind Turbine, which produces the mechanical torque according to the generator and wind speed values. The electrical power output of the generator (stator winding) is connected directly to the grid.

5 Control Strategy The configuration used is type DC common bus, where, the rotor is controlled by the same DFIG control structure. As well, the Grid side controller is used to keep up DCbus voltage and generate signals synchronized with the Grid by the same structure of control, but with a small difference to the active power, which is the sum of active power of DFIG and PV system. The following Fig. 5 shows the outline of the proposed control

Fig. 5. Configuration of the hybrid WT_PV system and its local control schematic.

20

S. Abdelhamid et al.

For constant power, the maximum power control is overridden above rated wind speeds by the stall regulation. Figure 6 displays the wind-side converter power controller; which uses PI controllers to display the new q and d axis commands i*wq and i*wd.

Fig. 6. Power controller of the rotor side

A low-pass filter averages a total of the power obtained from the wind and solar sources, and then the filtered value is defined as the real power control. The average effect of a low pass filter can be changed by setting specific time constant. Figure 7 Provides Power controller for grid side [4].

Fig. 7. Power controller for grid side

Fig. 8. DC voltage controller

The common dc-bus voltage is set at a constant value, so that real wind turbine and PV array power generation can transfer into the grid. Figure 8 Displays a DC voltage controller.

6 Simulation and Discussion The overall diagram of this system shows in Fig. 9. Based on the above modeling and control studies, we simulated the hybrid system operating state under certain weather conditions. Figure 9 shows the total structure of the simulated system in the Matlab program. These weather conditions are represented by a fixed wind speed with a value 8 m/s, solar radiation that varies between 200 W/m2 and 800 W/m2 and a constant temperature of 25 °C. Figure 10 depicts these atmospheric variables.

Modeling and Simulation of PV/Wind Hybrid Energy System

Fig. 9. The overall diagram of the hybrid system

21

Fig. 10. The weather conditions

To verify the control method applied in the system, we applied a wind speed of 8 m/s to the wind blades of Turbine, which gave a speed corresponding to the DFIG operating speed in the MPPT control of about 188.4 rad/s, and a mechanical torque of about −5000 N.m. as shown in Fig. 14. This system produces a total of 1340 kW. Including 930 kW from wind power and 410 kW from solar power under the weather conditions that mentioned previously. Therefore, we used solar panels contain 85 in Parallel and 32 serially of modules mentioned earlier. Moreover, a wind turbine includes a DFIG. Figure 11 shows the output voltage of the photovoltaic part of the system. This voltage is followed by changes in the irradiation. This change is evident in moments 2.5 s and 4 s, the change in the resulting current is shown in the figure too. The output boost voltage is stable around the value 1200 V. It shows the output boost voltage and the resulting and injected power in the hybrid system. This power is followed by changes in the irradiation. The change is evident in moments 2.5 s and 4 s. The power varies between 102 kW and 410 kW according to the change in the irradiation. The change in the resulting current is shown in the figure too. It is noted in Fig. 12 that the grid, generated and load tensions are regular and sinusoidal. The grid and the generated current values increase according to the change in the work of the PV system, which follows the changes of the temperature and radiation. This is evident in times 2.5 s and 4 s. In this proposed system, and through Fig. 13, we found that the total harmonic distortion for the generated current is around 2.42% because the wind turbine response is slower than that of the PV array; it has relatively larger dc bus ripple voltage. This is even smaller than IEEE 512-1992 standard.

22

S. Abdelhamid et al.

Fig. 11. The PV result

Fig. 12. The aspects of tensions and currents

Figure 14 shows the components of the quadrature and direct rotor currents of the DFIG. These are well follow the reference. The direct component of the rotor currents is zero, which also gives zero stator reactive power. The quadratic component of the rotor currents starts with a positive value and then changes to a negative value of −250A until the speed has reached the reference value. From 1.9 s, it is becoming positive, and it has value 1000A.

Modeling and Simulation of PV/Wind Hybrid Energy System

23

Fig. 13. The THD of generated currents

Fig. 14. The results of rotor side

The electromagnetic torque curve takes the opposite form of the Iqr curve, and ends with a negative value of -5000 N.m. Therefore, we can say that the DFIG is a generator. Figure 15 shows the components of the quadrature and direct grid currents side. These are well follow the reference too. It takes values that change with the increase in speed and the change of solar radiation; this indicates the increase in the values of the currents injected to the grid. As shown in Fig. 16, where we find that after the moment 1.9 s the grid power becomes negative and the generated power is positive, which means that the grid receives the power from the hybrid system.

24

S. Abdelhamid et al.

Fig. 15. The results of grid side

Fig. 16. Active and reactive power of system

7 Conclusion In this study, a Wind turbine and PV hybrid system was simulated. An MPPT controller was used to extract the possible electrical power from the renewable energies used. Depending on the control method used, the ability to operate the system to produce electrical power and the grid became apparent. The electric currents injected to the grid by the hybrid system increase with the change in the value of the weather variables. All simulation results obtained show the control performance and dynamic behavior of grid connected hybrid system provides good results, they show that the control system is efficiency.

Modeling and Simulation of PV/Wind Hybrid Energy System

25

References 1. Palizban, O., Rezaei, M., Mekhilef, S.: Active and reactive power control for a hybrid system with photovoltaic panel, wind turbine, fuel cells, electrolyzer and super capacitor in off-grid mode. In: IEEE International Conference on Control System, Computing and Engineering, pp. 404–408 (2011) 2. Ravi, K., Sakthigokulrajan, C., Shilaja, C.: Modeling of photovoltaic energy conversion system with integration to substation. Glob. J. Pure Appl. Math. 10, 451–464 (2014) 3. Benkhelil, E. Gherbi, A.: Modeling and simulation of grid-connected photovoltaic generation system. Revue des energies renouvelables SIENR, Algeria, pp. 295–306 (2012) 4. Kim, S.K., Jeon, J.H., Cho, C.H.: Dynamic modeling and control of a grid-connected hybrid generation system with versatile power transfer. IEEE trans. Ind. Electron. 55, 1677–1688 (2008) 5. Khatua, P.K., Ramachandaramurthy, V.K., Yong, J.Y., Pasupuleti, J.: Decoupled control of three phase grid connected solar PV system. Int. J. Eng. Adv. Technol. 9(2), 4218–4222 (2019) 6. Mohammed, A.Y., Mohammed, F.L., Ibrahim, M.Y.: Grid connected photovoltaic system. In: International Conference on Communication, Control, Computing and Electronics Engineering (ICCCCEE), Khartoum, Sudan (2017) 7. Bensaadi, H., Harbouche, Y., Abdessmed, R.: Direct torque control (DTC-SVM) of PMSG based in wind energy conversion system. U.P.B. Sci. Bull. Ser. C 81(2), 227–240 (2019) 8. Hussein, K.H.: Maximum photovoltaic power tracking: an algorithm for rapidly changing atmospheric conditions. Proc. Inst. Electr. Eng Gener. Transm. Distrib. 142(1), 59–64 (1995) 9. Ibrahim, T.: Maximum power point tracking for photovoltaic systems in rapidly changing environmental conditions. In: The 8 Jordanian International Electrical and Electronics Engineering Conference (JIEEEC) (2013) 10. Natsheh, E., Albarbar, A., Yazdani, J.: Modeling and control for smart grid integration of solar/wind energy conversion system. In: 2nd IEEE PES International Conference on Innovative Smart Grid Technologies (ISGT Europe), pp. 1–8 (2011) 11. Chen, Y.-M., et al.: Multi-input inverter for grid-connected hybrid pv/wind power system. IEEE Trans. Power Electron. 22(3), 1070–1077 (2007) 12. Radwan, A.A.A., Mohamed, Y.A.I., Jiang, X.: A hybrid wind-photovoltaic generation system modeling and performance evaluation. In: 2019 IEEE 7th International Conference on Smart Energy Grid Engineering (SEGE), Oshawa, ON, Canada, pp. 129–133 (2019)

Energy Management Analysis of a WindDiesel-Battery Hybrid Power System Nacereddine Guettaf1(&), Seif El Islem Guettaf2, Zahira Anane1, and Hamou Nouri1 1

Laboratory of Automatics Setif (LAS), Electrical Engineering Department, University of Ferhat Abbas Setif 1 (UFAS1), Setif, Algeria [email protected] 2 Laboratoire de Qualité de l’Energie dans les Réseaux Electrique (QUERE), Electrical Engineering Department, University of Ferhat Abbas Setif 1 (UFAS1), Setif, Algeria

Abstract. The Wind-Diesel-Battery (WDB) hybrid power system is proposed to satisfy power requirements in some remote areas locating out of national power grid. In this paper presents a study of renewable energy production systems, coupling a wind turbine, a diesel generator, battery storage system and we have also developed a novel strategy for a hybrid energy management system. For the study of the WDB hybrid system, modeling each block of the conversion chain has been done for a standalone unit in remote location. This WDB was applied in the design of an autonomous system that supplies one of the different kinds loads typically encountered in Algeria. The economic dispatch of the diesel generators was also analysed for the system. Keywords: Renewable energy system
Wind-Diesel-Battery hybrid system Modeling
Management
Economic dispatch




1 Introduction Energy is one of the main factors that contribute greatly to the development and civilization of nations and societies. Globally, the demand of energy has increased continuously with world population growth, industrial development and economic consumption [1]. Recently, renewable energies have been incorporated into the rural electrification process promising trend worldwide. The vast distances and lack of capital are among the major’s obstacles to the development of the network system, in exceptionally developing countries such as our country and even developed countries such as the United States of America and United Kingdom [2, 3]. In fact, renewable energy is used to a large extent compared to the conventional energy due to its good impact on the environment and economic benefits [4]. The main cause of the large use of the Hybrid Renewable Energy Systems (HRES) all around the world is due to the high reliability compared to independent individual systems. The purpose of their use is to provide energy to a large number of domestic consumers located in isolated areas far from the supply network [5]. Both references [6, 7] refer to numerous studies carried out in the fields of determining the dimensions of the physical components and © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 26–35, 2021. https://doi.org/10.1007/978-3-030-63846-7_3

Energy Management Analysis of a Wind-Diesel-Battery

27

configuration installation. Several researchers have discussed how to improve energy flow control strategies as an essential part of a controlled system [8, 9]. HRES is suitable for Algeria in terms of abundance wind energy due to the distinctive geographical structure that it possesses such as mountain ranges and coastal strip. These systems can be an effective solution for the development of an electrical network system, in other hand, to participate in the development of our country. This paper presents a study of a hybrid system wind-diesel-battery designed to supply a group of consumers located far from the main grid. As shown in Fig. 1, the proposed system consists of wind turbines as renewable energy sources, a bank battery system and an AC diesel generator. Also, we need to two types of converters: the first is a DC/AC converter (Inverter) and the second is AC/DC converter for the conversion between the two buses DC and AC.

Fig. 1. Structure of the hybrid power system.

2 Structure of the Hybrid Power System The configuration of the proposed Wind-Diesel-Battery hybrid system is illustrated below. It comprises a wind turbine, forty-three lead calcium batteries of 100 Ah which are connected to the same DC Bus, the load and AC diesel generator which are connected to the same AC Bus and an inverter, AC/DC converter which are between the two buses to transfer the power. Our Wind-Diesel-Battery system hybrid allows covering part of the AC load requirement.

28

N. Guettaf et al.

3 System Components Modeling 3.1

Wind Turbine Model

Average wind energy available at a given site per unit of time and per surface A swept by the rotation of the blades of the wind turbine is written [10–12]: 1 PWT ¼ :Cp :q:A:V 3 ; 2

ð1Þ

Where q is the air density (normally 1.225 kg/m3 at 15 °C) (Kg/m3), V is the average wind speed (m/s), A is the area that the blades sweep, Cp is the aero-motor power coefficient. (0.593), PWT is the output power generated by wind turbines (kW). Figure 2 presents the power curves of three models of wind turbines from a single constructor.

Fig. 2. Power curves of 3 considered models of wind turbine.

3.2

Battery Model

The battery charge state (SOC) is expressed as follows: SOC ðt þ 1Þ ¼ SOC ðtÞ:ð1  Ks Þ 

gp:PBat :103 :102 Cn :NBat :UBat

ð2Þ

Where SOC(t) and SOC (t + 1) are the batteries SOC at the start and end, KS is the self-discharge coefficient, KS is the self-discharge coefficient, ±is defined by the state of the charge/discharge process, PBat is the power transferred to or from the batteries (kW), Cn is the battery capacity (Ah), UBat is the battery voltage (V) and Ƞp is the process efficiency for the system condition (charge or discharge). The discharge

Energy Management Analysis of a Wind-Diesel-Battery

29

efficiency curve of the batteries used in this system shows in Fig. 3. Reference [13] describes the rain flow model which is used to simulate the lifetime of the battery. Figure 4 displays the number of cycles as a function of the depth of discharge for the type of battery studied.

Fig. 3. Discharge efficiency curve of the studied battery.

3.3

Diesel Generator Model

A simple model of the fuel consumption of diesel generator is expressed by the reference [14, 15] as: Qfuel ¼ B  PNGen þ A  PGen ;

ð3Þ

Where A = 0.246 1/kWh and B = 0.08415 1/kWh are coefficients used to calculate the fuel consumed, PN Gen is the nominal power of the generator in (kW), PGen is the power generated by the generator in (kW), Qfuel is the amount of fuel consumed in (L).

4 Management Strategy 4.1

Definition of the System

The energy control system applied in this study is almost the same as described in [9], the produced renewable energy at time t and available on the DC bus, named PDC (t) in (kW) is given as follows:

30

N. Guettaf et al.

Fig. 4. Battery cycle life.

PDC ðtÞ ¼ PWT ðtÞ;

ð4Þ

The apparent power on the AC bus, named S in (kVA) is written as follows: SðtÞ ¼ PLoad ðtÞ= cos u;

ð5Þ

Where Cos u is the load power factor, PLoad is the net power load at time t in (kW). The maximum active power at the input of the inverter, noted PInv_max in (kW), is given by the following equation: PInv

max

¼ SInv : cos u=gInv ;

ð6Þ

Where SInv is the inverter nominal apparent power (kVA). ȠInv is the inverter efficiency of the inverter. 4.2

Power Management

The wind-diesel-battery power management system includes by an inverter located between the DC and AC bus in order to supply the load. It can be operated in two modes. Figure 5 illustrates the power management system chart.

Energy Management Analysis of a Wind-Diesel-Battery

31

Fig. 5. Power system management chart.

4.2.1 Operational Mode 1 In the operational mode 1, the wind turbine PWT is adapted to supply power to the load and charges the batteries when the PWT more than the load demand PLoad i.e. PWT > PLoad, the diesel generator is turned off. 4.2.2 Operational Mode 2 In this operational mode, when the PWT < PLoad, the power source used here to deliver the power load demand, either by a battery bank or diesel generator.

5 Simulation Results and Discussion To improve the performance of the hybrid power system wind-diesel-battery, simulation results are implemented at the MATLAB environment level for a load power of 2.4 kWh hourly with average demand and 5 kW maximum demand during this study. Figure 6 presents the variation of the load during one day when the load is almost the same as every day of the year. Table 1 shows the parameters how to control the energy flow system Figure 7 and Fig. 8 present a profile of wind speed and the output power wind turbine used in this study during one day. Figure 9 shows the power delivered by the WDB system, the output power wind turbine and the load demand. Note that when PWT more than the PLoad, the batteries will be charge by a surplus of the wind turbine (in orange color). In the opposite case, note that when the load demand more than of the power of wind turbine, the batteries help to achieve the power required by the load (in blue color). For the economic dispatch of the diesel generator, we can see that the diesel generator works 3 h (in black color).

32

N. Guettaf et al.

Fig. 6. The load variations during one day. Table 1. The parameters control of energy flow system. Battery

Number of batteries NBat Capacity (Ah) Voltage UBat (V) Inverter Inverter rated power (kW) Efficiency Diesel generator Generator rated power (kW) Generator min power

43 100 12 6 0.9 1.5 0.4

Fig. 7. The wind speed variation during one day.

Energy Management Analysis of a Wind-Diesel-Battery

33

Fig. 8. The output power wind turbine during one day.

Fig. 9. The power delivered by the WDB system.

6 Conclusion In this paper, an effective energy management system has been treated for a hybrid power system comprising a wind turbine, diesel generator and a battery bank to power loads located at isolated sites far from the main grid. In this work, the simulation results

34

N. Guettaf et al.

demonstrate that the energy management system works correctly and reliably, this reliability is shown in the control of the power of each source of the system. The wind turbine supplies the load via an inverter and charges the batteries when the wind turbine has an excess of energy. In the other hand, the batteries help to achieve the power required by the load and reduce the use of the diesel generator to 3 h because it has so many problems such as maintenance, pollution, …etc. Also, the study system has proved its efficiency concerning the economic dispatch of generator, we can see that the load demands are met and the fuel consumption of the diesel generator reduced. It can be concluded that (Wind/Diesel/Battery) are economical and reliable during its lifetime.

References 1. McLellan, B., Zhang, Q., Farzaneh, H., Utama, N.A., Ishihara, K.N.: Resilience, sustainability and risk management: a focus on energy. Challenges 3, 153–182 (2012) 2. Fodhil, F., Hamidat, A., Nadjemi1, O., Alliche, Z., Berkani, L.: Optimum Design of a Hybrid Photovoltaic/Diesel/Battery/System Using Water Cycle Algorithm. In: 3th International Conference on Artificial Intelligence in Renewable Energetic Systems (IC-AIRES 2019), Springer Nature Switzerland AG 2020, pp. 82–93 (2019) 3. Bing, Z., Henerica, T., Xiaohua, X.: August. Model Predictive Control for Energy Dispatch of a Photovoltaic-Diesel-Battery Hybrid Power System. Preprints of the 19th World Congress: The International Federation of Automatic Control, pp. 11135–11140 (2014) 4. Ganesan, E., Dash, S.S., Samanta, C.: Modeling, control, and power management for a gridintegrated photovoltaic, fuel cell, and wind hybrid system. Turk. J. Elec. Eng. Comp. Sci. 24, 4804–4823 (2016) 5. Saib, S., Gherbi, A., Bayindir, R.: Optimization study of hybrid renewable energy system in autonomous site. In: 1th International Conference on Artificial Intelligence in Renewable Energetic Systems (IC-AIRES), Springer International Publishing AG 2018, pp. 431–438 (2017) 6. Yang, H., Zhou, W., Lu, L., Fang, Z.: Optimal sizing method for stand-alone hybrid solarwind system with LPSP technology by using genetic algorithm. Solar Energy 82, 354–367 (2008) 7. Koutroulis, E., Kolokotsa, D., Potirakis, A., Kalaitzakis, K.: Methodology for optimal sizing of stand-alone photovoltaic/wind generator systems using genetic algorithms. Solar Energy 80, 1072–1088 (2006) 8. Dufo-Lopez, R., Bernal-Agustin, J.L.: Design and control strategies of PV-Diesel systems using genetic algorithms. Solar Energy 79, 33–46 (2005) 9. Dufo-Lopez, R., Bernal-Agustin, J.L., Contreras, J.: Optimization of control strategies for stand-alone renewable energy systems with hydrogen storage. Renewable Energy 32, 1102– 1126 (2007) 10. Johnson, G.L.: Wind Energy Systems. book, Electronic Edition Manhattan, 61–70-157 (2006) 11. Merzouk, N.K.: Evaluation du gisement énergétique éolien contribution à la détermination du profil vertical de la vitesse du vent en Algérie. Thèse de doctorat, université Abou Bekr Belkaid de Tlemcen (2006) 12. Bencherif, M.: Modélisation de systèmes énergétiques photovoltaïques et éoliens intégration dans un système hybride basse tension. Thèse de doctorat, université Abou Bekr Belkaid de Tlemcen (2014)

Energy Management Analysis of a Wind-Diesel-Battery

35

13. Manwell, J.F., Rogers, A., Hayman, G., Avelar, C.T., McGowan, J.G., Abdulwahid, U., Wu, K.: Hybrid2- A hybrid system simulation model. Theory manual (2006) 14. Skarstein, O., Ulhen, K.: Design consideration with respect to long-term diesel saving in Wind/Diesel plants. Wind Engineering 13, No. 2 (1989) 15. Raji, A., Noboru, Y.: Optimization of a PV-wind-diesel system using a hybrid genetic algorithm. In: 2012 IEEE Electrical Power and Energy Conference, pp. 80–85 (2012)

Power Management Strategy Applied on Hybrid Electric Train Imen Mammeri(&), M. Benidir, Hamza Bouzeria, Meriem Aissaoui, and Amira Chaib Ras Laboratory of Transportation Engineering and Environment - LITE, Mentouri University, Constantine 1, Constantine, Algeria [email protected], {mohamed.benidir,bouzeria.hamza}@umc.edu.dz

Abstract. Today, railways are developing, especially the hybrid electric train. In this context, interest is given to energy storage, power management and compensation strategy is important. Most of the trams are electrified and usually connected to a continuous electrical network. They capture their energy via an overhead electrical line (catenary) connected to the distribution network. The electric distribution system is a 750 V direct current source. This work deals mainly with the energy optimization, energy storage method and compensation due to voltage drop and that by exploiting the wasted energy. The energy compensation of railways using the DC/DC converter system and storage battery is studied. Results given by MATLAB simulation program show that this strategy can be the best solution. Keywords: Electrical hybrid train
Energy storage Battery
Optimization
Compensation


DC/DC converter


1 Introduction It results from fears that seek to conserve energy and protect the environment around the world to think carefully about making electric vehicles the best solution [1]. Therefore, ways in which energy can be improved, replaced or fully exploited are found. The study deals mainly on the hybrid electric train which is connected to the national electricity grid via irreversible substations that are located along the railway. An AC/DC static converter is used between the main transformer and the overhead catenaries. Factually, this energy consumption by the electric motor results in a decrease in voltage [2]. Numerous researchers use a constant energy source alimentation for their electric vehicles; using the DC-DC converters and the battery [3]. In order to keep the motor voltage constant, DC-DC converters are used for the propulsion system to raise voltage and ensure control and power management in case of the load increase. Using a chopper in electric vehicles makes it possible to maintain the motor current at the desired value while ensuring the gradual change without noticeable loss of the motor voltage. It also makes possible adjusting torque and speed of the motor vehicle in traction as well as for electric braking. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 36–42, 2021. https://doi.org/10.1007/978-3-030-63846-7_4

Power Management Strategy Applied on Hybrid Electric Train

37

The battery is placed for reliable energy storage in the event of charging and discharging and it is linked with the DC-DC transformers (Fig. 1). The purpose of this battery is to cover dynamic needs such as acceleration or recovery [10]. This method aims at decreasing voltage drops and line losses so as to increase active power that leads to the electricity bill saving and consequently the consumption decrease of reactive energy. The objective of this work is to show how to optimize and compensate energy by constant source supply in an electrical hybrid train and by changing the resistance torque in three stages.

Fig. 1. Architecture of the hybrid electric train.

The paper may be divided as follows: Part II Describes the traction chain and the main components of the electrical hybrid train. Section 3 deals with the modeling of the system. In Sect. 4 results and discussions related to simulation results of the studied case. Finally, conclusion is drawn in the last part.

2 Traction Chain Description The main parts of the traction chain are: the transformers, the AC-DC and the catenaries. Due the decrease in voltage below the threshold value, the train becomes unable to provide the necessary energy and slows down. The traction chain of the propulsion system is represented in Fig. 1. Energy storage devices or power supplies and their output voltage vary according to the charging states. The latter is considered as a challenge when integrating storage devices or power supplies with a traction motor. Therefore, their uses are limited due to the size, efficiency and cost of a DC-DC converter [4]. Transformer from dc-dc that adapts to the voltage levels between the ac-dc converter and the battery to DC bus. The traction chain and electrical system studied in Fig. 1 include: • The DC-DC converter, interposed between the battery (Table 1 in appendix shows parameters of the battery model) and the DC bus, is bidirectional reversible current; • A three phase inverter connected to the DC bus ensures the motor supply; • An asynchronous wheel’s motor connected to the vehicle wheel.

38

I. Mammeri et al.

3 System Modeling 3.1

DC - AC Inverter

The inverter losses depend on the machine current and voltage, power factor, switching frequency or machine speed, and semiconductor temperature. In the simulation, the DC-link voltage is 750 V source. Therefore, the inverter is modeled using fourdimensional with voltage, current, speed, and power factor of the machine as inputs. The influence of the power factor and motor speed on losses is also displayed. Three alternate current sources with a phase shift of 120° to each other are applied as the symmetrical load. Through modulation control, the amplitude of phase voltage and power factor can be adjusted. Besides that, the switching frequency is adjusted proportionally to the motor speed to reduce the switching loss of the converter at low motor speed. The switching frequency is ten times the ground frequency of the current supply f. 3.2

DC/DC Converter

For the DC/DC converter, the coupling between thermal and electrical modeling is considered. The converter loss is dependent on the load current, DC-link voltage, battery’s voltage and semiconductor temperature. The effect of the temperature and battery voltage on the loss can be identified, based on an operational point with the DClink voltage and load current equal to 750 V. The semiconductor module is the same as that in the DC/AC converter [5]. DC-DC converters find applications in areas where battery charging, regenerative braking and backup power are required. The power flow in a bidirectional converter is usually from a low voltage end, such as a battery, to a high voltage side. This is referred to as boost operation. An electric train buck-boost converter provides an output voltage, which can be higher or lower than the battery input voltage. Output voltage polarity is opposite to that of the input voltage [6]. It consists of a battery DC input voltage source (Vinput), DC output voltage (Voutput) delivered to the electrical train drive, a controlled switch (Cswitch), a diode (D), a filter inductor (L) and a filter capacitor (C). A capacitance “Cp” is connected in parallel. In fact, this ability helps protect the substation against overvoltage at a high-power demand. When the converter is related to battery, it operates in Boost mode. 3.3

The Battery Model

According to references, many electrical models of batteries are available, to describe and to present the battery dynamic behavior, [8]. In this work, the lithium-ion battery has been used and can be expressed by following mathematical model: VB ¼ VZ:IB Where; Z ¼ RX þ

Rc 1þsþS

and s ¼ Cc :Rc .

ð1Þ

Power Management Strategy Applied on Hybrid Electric Train

39

The capacity of the battery is given by the following equation as: Cbatt ¼ C10

1:67  ð1 þ 0:005 þ DT Þ
I 1 þ 0:67  I10

ð2Þ

Where: DT is the accumulator heating, C10 is the rate capacity. I10. Moreover, the battery state of charge can be given as: SOC ¼ 1 

Q Cbatt

ð3Þ

With: Q ¼ Ibatt  t

ð4Þ

Where: t is the discharging time and with a current Ibatt. Table 1. Parameters of the battery model

3.4

Energy Management Strategy

The energy management system can improve the dynamic response of the electric train and improve its efficiency as well as its autonomy without forgetting the main objective of reversible the energy or the current [7]. • The classical PI energy management The battery charging state is controlled by using the PI controller to obtain the battery power at which it from the demanded power to obtain. Therefore, the controller operation is related to the battery charge state (SOC). When the battery SOC is lower than the reference value (SOCmin), the battery gives its full power and is charged to 80% of its autonomy to ensure battery safety. When the SOC is below the reference value, the substation provides almost the required electrical train power. In order to

40

I. Mammeri et al.

have the best response time and the stability during use the PI controller, the PI gains or parameters are tuned using the MATLAB control system toolbox [9].

4 Results and Discussions In order to describe the behavior of the hybrid energy source management, simulations were carried out using the model in Fig. 1. The following results were simulated in MATLAB/SIMULINK. For testing and seeing the electrical hybrid train vehicle behavior when operating such as the power changes according to different phases, the battery SOC must make a proposed resistance torque to apply different power management. Hence, three torque values were proposed as follows: [0 to 1] second: resistance torque is zero, i.e. the motor not loaded; [1 to 2] second: resistance torque is 800 [Nm], in normal operating phase; [2 to 3] second: resistance torque is −800 [Nm], i.e. the braking phase. The traction chain behavior simulation results such as speed, electromagnetic torque and stator current are illustrated in Figs. 2, 3 and 4. The engine speed is 157 rad/s after simulation. For that, the resistance torque affects the electromagnetic torque as well as the current and speed. The stator current in [1 to 3] phase rises up almost to 450 A.

Fig. 2. Electomagnetic torque response.

Fig. 3. Speed electrical hybrid train.

Note that the battery is discharged [0 to 2] and [2 to 3] when it is being charged, and this indicates that there is a current recovery in Fig. 5. Figures 6 and 7 describe the evolution of the different power values in the battery (Pbatt), grid required power (Pgrid) and railway (electrical train) power (Prailway) in all periods for each strategy. So, it can be concluded from this figure that the grid provides the main demand power and the battery intervene at any negative or positive overshoot power demand. Furthermore, this battery intervenes during any lack of power due to the low response according to the used energy management strategy.

Power Management Strategy Applied on Hybrid Electric Train

Fig. 4. Response of phase A stator current.

Fig. 5. Battery state of charge.

Fig. 6. Electric train power evolution.

Fig. 7. Power evolution

41

42

I. Mammeri et al.

5 Conclusion In this paper, energy management strategies, compensation and energy/current recovery was studied. Using the dc-dc converter and battery has been to make the voltage constant at dc bus. The SOC was utilized for battery and used as a monitoring program and made to improve energy significantly using PI controlled system. According to the results shown, it can be concluded that the utilization of DC/DC and the battery improves energy, especially in the braking phase and storing energy in the battery.

References 1. Allaoua, B., Asnoune, K., Mebarki, B.: Energy management of pem fuel cell/super capacitor hybrid power sources for an electric vehicle. Int. J. Hydrogen Energy 42, 21158–21166 (2017) 2. Poline, M., Gerbaud, L., Pouget, J., Chauvet, F., Castaing, A.: Sizing method by optimization with energy management–application to electrical hybrid train. Int. J. Appl. Electromagn. Mech. 60(S1), S133–S148 (2019) 3. Han, X., Li, F., Zhang, T., Zhang, T., Song, K.: Economic energy management strategy design and simulation for a dual-stack fuel cell electric vehicle. Int. J. Hydrogen Energy 42, 11584–11595 (2017) 4. Pei, X., Nie, S., Kang, Y.: Switch short-circuits fault diagnosis and remedial strategy for fullbridge dc–dc converters. IEEE Trans. Power Electron. 30, 996–1004 (2014) 5. Peng, H., Li, J., Löwenstein, L., Hameyer, K.: A scalable, causal, adaptive energy management strategy based on optimal control theory for a fuel cell hybrid railway vehicle. Appl. Energy 267, 114987 (2020) 6. Guilbert, D., Gaillard, A., N’diaye, A., Djerdir, A.: Power switch failures tolerance and remedial strategies of a 4-leg floating interleaved dc/dc boost converter for photovoltaic/fuel cell applications. Renew. Energy 90, 14–27 (2016) 7. Soumeur, M.A., Gasbaoui, B., Abdelkhalek, O., Ghouili, J., Toumi, T., Chakar, A.: Comparative study of energy management strategies for hybrid proton exchange membrane fuel cell four wheel drive electric vehicle. J. Power source 462, 228167 (2020) 8. Rekioua, D., Bensmail, S., Bettar, N.: Development of hybrid photovoltaic-fuel cell system for stand-alone application. Int. J. Hydrogen Energy 39, 1604–1611 (2014) 9. Hwang, J.J., Chen, Y.J., Kuo, J.K.: The study on the power management system in a fuel cell hybrid vehicle. Int. J. Hydrogen Energy 37, 4476–4489 (2012) 10. Unlubayir, C., Nemeth, T., Meishner, F., Peng, H., Deng, K., Sauer, D.U.: Simulation model with an optimal operation strategy for a hybrid train powered by a battery and a fuel cell. In: 2019 IEEE Vehicle Power and Propulsion Conference (VPPC), pp. 1–5 (2019)

Energy Saving Analysis of Municipal Pool Fed from Hybrid Renewable Energy in Batna Algeria T. Boutabba1(&), M. L. Bechka2(&), F. Menzri3, and S. Drid2

3

1 University of Abbès Laghrour Khenchela, LSPIE Batna Laboratory Algeria, Batna, Algeria [email protected] 2 University of Batna 2, Mostefa Ben Boulaïd), University of LSPIE Laboratory Algeria, Batna, Algeria [email protected], [email protected] MoDERNa Laboratory Mentouri, University of Constantine1, Constantine 25000, Algeria [email protected]

Abstract. This paper presents an energy saving analysis of municipal pool fed from hybrid renewable energy in Batna, Algeria. The study is based on data collected from the municipal pool of new city HAMLA in Batna from January to December 2016. The data consists of all the consumption of electricity and gas in the building. For this purpose, an Energy Saving analysis of the building has been done on the equipment and the behaviour of its occupants in order to determine the origin of the high cost on power and consequently the improvements that should be adopted, so that we can effectively reduce the cost and also to determine a solution for local renewable energy production, with particular focus on hybrid solutions, the software HOMER was used with the intention of creating a system that can minimize the problem of intermittency of renewable energies, in this case, of solar and wind origin. Keywords: Energy Saving pool


Homer software
Hybrid energy
Swimming

1 Introduction The world has been confronted with increasing energy consumption for several decades now. This increase fundamentally challenges the economic model that depends on a colossal amount of energy for its development. This energy remains overwhelmingly fossil-based and therefore non-renewable in the short and medium term, and is the main source of greenhouse gas emissions in a world already suffering from the effects of human activity on its environment. Algeria largest country in Africa and tenth-largest proven reserves of natural gas in the world is the sixth-largest gas exporter and has the third largest reserves of shale gas. It also ranks sixteenth in proven oil reserves. Has been engaged for a few years in a policy energy transition whose objective is both to ensure a energy security by opting for a diversification of its energy resources and also to honor its contractual commitments to mitigate greenhouse gas emissions. Today, this © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 43–54, 2021. https://doi.org/10.1007/978-3-030-63846-7_5

44

T. Boutabba et al.

energy transition policy is available at through the national energy efficiency program (2015–2030) which sets itself the objective of saving 30 million TOE by 2030 and the national energy development program renewable which aims to produce 22,000 MW from energies renewable by 2030 [3]. The rising energy costs and budgetary constraints, particularly in the public sector, has created great constraints on public budgets, resulting in the need to reduce operating costs with the public buildings, without thereby diminish the quality of service offered to its citizens [1, 2]. This study aims to reduce energy consumption in the municipal pool of new city HAMLA (southern exit of the city of Batna), since this facility presents significant expenses to the municipality. For this purpose, a survey of the state of the building has been done on of its equipment and the behavior of its occupants in order to determine the origin of the high cost on power and consequently the improvements that should be adopted, so that we can effectively reduce cost. In order to meets the objective of cost reduction, the software HOMER was used, to determine a solution for local renewable energy production, with particular focus on hybrid solutions, e.g. two or more sources of renewable energy, with the intention of creating a system that can minimize the problem of intermittency of renewable energies, in this case, of solar and wind origin [5–12]. Finally, this study is seen as a comprehensive approach to the demand for energy efficiency and local renewable production.

2 Description of the Municipal Pool The mission of this facility is to contribute to the improvement of the quality of life of the population and to serve the citizens by offering sports services and complementary health and training services “aquatic and recreational activities”. This facility has been designed to serve 6 to 7 session by day (100 users per session); it can serve up to 700 users per day (Fig. 1).

Fig. 1. Site of the municipal pool of new city HAMLA, Batna.

Energy Saving Analysis of Municipal Pool Fed

45

3 Consumption of Municipal Pool The pool spends large amounts of money on energy management, particularly in the form of electricity and natural gas. The electricity bill accounts for 57% of the annual consumption, and gas consumption accounts for 43% of the consumption due to the heating of domestic water and water change in the pool. In what follows we will analyze these data to identify the reasons for these expenses. The access to the facility’s electricity and gas bills for the year 2016 allows the data to be exposed for analysis shown in Fig. 2.

Fig. 2. Cost and distribution of consumption by energy source.

The challenge in this study lies essentially in the absence of any control and monitoring system (Registration) with only electricity and gas meters at the entrance of the circuits. Temperature measurements are manual, and there is no humidity recording. The acquired data are based on interviews with the unit manager.

4 Identification of Equipment In order to identify the various equipments in the facility including lighting, heating and air conditioning, several visits was carried out in 2017 to determine the energy behavior of the pool. 4.1

Lighting

As expected, during the census, different types and technologies of lighting are used depending on the space to be illuminated. Among these types are mainly T8 fluorescent tubes, and HPIT projectors with E40 lamps as shown in Fig. 2 (Table 1).

46

T. Boutabba et al. Table 1. Different lighting technology used in the municipal pool Unit Pool

N° Power (W) Total power (W) 13 400 5200 2 1000 2000 Engine room Ionizing eco. lamp 10 50 500 Corridor 2 18 36 Admin. Wc Fluorescent TubeT8 3 75 225 Entry administration 8 58 464 Hall 4 232 Office head unit 4 232 Reception 2 116 Men’s locker rooms 6 348 Men’s shower 6 348 Locker rooms women 4 232 Women’s shower 4 232 WC of the pool room 2 116 Technical room 1 4 232 Hall public entrance 12 696 Cafeteria 8 464 Hom. publ. toilet. 2 116 WC publ. Fem. 2 116 Technical room 2 4 232 Technical room 3 4 232 Total 12369

4.2

Lamp type E40

Office Equipment

Only one computer with its printer has been identified, and used for the administrative management and archiving of all digital documents. In order to simplify the analysis, this load is assumed to be fixed and equal to 300 W (Fig. 3). 4.3

Heating Boiler System

The pool and domestic water heating system consists of two gas boilers and two generators for water treatment involving 4 Riello brand burners with an electric power of 0.6 Kw each. Figure 6 below shows the two boilers and generators (Fig. 4). The volume of heated water is not counted. The heating system is not equipped with any electronic control system and the water temperature is adjusted manually in order to keep a fixed value set by the operator.

Energy Saving Analysis of Municipal Pool Fed

Fig. 3. Type of lighting used for the pool

Fig. 4. Boilers and generators

Fig. 5. a) Lighting profile, b) Profile of other equipment.

47

48

T. Boutabba et al.

Fig. 6. Annual load distribution according to use

4.4

Load Profile

The load profile of the facility and to the equipment operation periods dependent partly on the opening hours presented in Table 2. Table 2. Municipal pool working hours Journées

Sund Mond Tuesd Wednesd Thursd Frid Saturd

8 10 h30– h30– 10 h 12 h Maintenance Open Open Open Open Open Open Open Open Open Open Open Open

12 h30–13 h30

14 h– 15–30

16 h– 17 h30

18 h– 19 h30

20 h– 21–30

Maintenance Maintenance Maintenance Maintenance Maintenance Maintenance

Open Open Open Open Open Open

Open Open Open Open Open Open

Open Open Open Open Open Open

Open Open Open Open Open Open

To economize and to minimize the electricity consumption, the use of lighting has been reduced to the minimum possible, the use of equipment is limited to the minimum required, many lamps are non-existent, and the use of extractors is manual and has visual judgment to remedy the increase in humidity that sometimes leads to condensation. The only equipment that is in constant use is the electric pumps to ensure a good water temperature in the pool and shower. It should be noted that dehumidification is done through the extractors, and that the profile of the latter varies throughout the year according to the psychometric characteristics. The load profiles in the following Fig. 5 are developed to determine the use of electrical energy.

Energy Saving Analysis of Municipal Pool Fed

4.5

49

Energy Performance

The search for solutions to reduce the consumption of electrical energy requires the development of calculation models in order to be able to optimize the use of this energy. 4.6

Electricity Consumption

The annual consumption of electrical energy is shown in the figure according to use, whether or not it is dependent on opening hours. It should be noted that the most important consumption is that of the equipment including all the pumps in the system.

5 Renewable Energy Supply 5.1

Generation System

Renewable energy resources tend to be widely available while being more environmentally friendly and sustainable than fossil fuels. Some renewable energy sources are particularly useful in remote areas where other energy sources are not available. The benefits of renewable energy face several challenges. An example of this is the standalone solar and wind powered systems, which do not provide continuous power, mainly due to seasonal variations and intermittent availability of solar and wind radiation. One approach to minimizing intermittency of renewable is to combine these two or other energy sources, using the strengths of one power source to balance the weaknesses of another [13]. In the past, hybrid generation systems for solar and wind energy have been applied in remote locations, far away from conventional generation systems, for example, for communication stations. Currently, there is a trend to use systems from renewable sources, such as wind, solar and hydro, and combine them in hybrid systems for gridconnected applications [4]. In order to reduce costs, as well as seeking to promote self-consumption, simulations have been carried out for a basic model for this purpose. The purpose of these tests is to understand whether there is technical and economic feasibility for the adoption of local production systems. 5.2

Equipment

In order to get a true sense of the costs of the system, we chose to look for equipment already existing on the domestic market and available from the Condor Company. The choice of equipment is made by the gain (power) they produce for each Dinar invested, adopting a logic of choosing the least expensive equipment for the equipment that produces more for the same cost (Fig. 7).

50

T. Boutabba et al.

Fig. 7. Basic model generated by software “HOMER”

5.3

Photovoltaic Panels

The choice of the best equipment for the case study is based on a comparison of the different equipment available. The choice involves the use of solar panel Monocrystalline 72 cells 310/315/320Wc from Condor Electronics Company. It is always a good idea to bear in mind that when planning to use photovoltaic panels it is necessary to integrate the current converters, these should be chosen according to the solar energy calculated by HOMER, in order to adjust the inverter to the real needs. Due to the lack of obtaining annual maintenance values for the panels, a value of 5000 DA/year/panel is assumed [14]. 5.4

Wind Turbine

The choice of wind turbines is limited by an important factor, the transmission capacity due to its size. Therefore, it is necessary to choose a model that can be easily transported and that also allows installation on site without the need for large machines for this purpose. 5.5

Climatic Data

• Solar radiation The HOMER software relies on the NASA climate database to import the climatic data of solar radiation availability at the study site. This requires the geographic coordinate’s latitude and longitude to retrieve the data as shown in Fig. 8. The solar radiation data allow the software to predict an annual solar energy distribution that can be seen in Fig. 9. As expected, the largest gains are in the summer months. It should also be noted that the software does not predict whether it will be possible to reach the maximum production capacity of the panel.

Energy Saving Analysis of Municipal Pool Fed

51

Fig. 8. Solar radiation by HOMER software

Fig. 9. Annual distribution of solar energy

• Wind speed It is difficult to obtain measured wind speed data. Autonomous wind speed measurement is expensive and time-consuming, and average wind speeds can vary considerably over short distances due to terrain effects. For these reasons, it is often necessary to synthesize wind data from estimated average monthly wind speeds. From the web site in [15] provides the average monthly wind speed data for many cities around the world including the city of BATNA (Fig. 10). The Fig. 11 represents the availability of wind on an hourly basis throughout the year. In Figs. 9 and 11, the intermittency of the system is clearly visible with mainly a decrease in solar radiation during winter. The main objective of this analysis is to locate these weaknesses.

52

T. Boutabba et al.

Fig. 10. Wind speed by HOMER software

Fig. 11. Wind availability throughout the year – HOMER

5.6

Energy Consumption

The energy demand by the loads is determined by the power absorbed each hour, this demand is expressed by the load profile of the daily consumption as indicated in the previous chapter, on the basis of these values, the program defines the consumption base and the daily average consumption, and can later put a new electricity consumption based on the established base consumption and a new daily average value. 5.7

Basic Model

At this stage, the option to cover current electricity needs from hybrid systems is tested; the user equipment is defined on the draft shield. The load profiles are the same as the basic model. The duration of the project is 15 years, with an inflation rate of 0.6%. Consumption is estimated at 554 kW/day in the basic model (Fig. 12).

Energy Saving Analysis of Municipal Pool Fed

53

Fig. 12. The feasibility of a hybrid system in HOMER software

6 Simulation and Results Simulation results show that the best solution to power the facility is the use of a photovoltaic system alone. A second alternative to the first one is the use of a photovoltaic-wind hybrid system with a photovoltaic field with a capacity of 300 Kw and a 5.6 kw wind turbine. This second option does not seem very interesting considering the climatic conditions of the region, and therefore is only used for the purpose of studying the feasibility of a hybrid system.

7 Conclusion This paper presents the evaluation of energy saving of municipal pool fed from hybrid renewable energy in Batna. The built environment in general and swimming pools in particular are important in terms of energy consumption, which leads us to seek to ensure their energy efficiency (optimization of energy consumption. This work seeks to improve the efficiency of the building and equipment. The energetic improvement of an aquatic sports establishment requires a very deep and detailed study, this work allowed us to make an inventory of the energetic equipments and then the acquisition of an order of importance, the evaluation of the consumption of such an establishment and finally to determine the efficiency and profitability of the installation of an energy production system based on renewable energies. The analysis given by Homer software shows that the use of a photovoltaic system is more interesting considering the solar energy potential of the region.

References 1. Cruz, I., Arias, F., Avia, F., Arribas, L.M., Fiffe, R.P.: Assessment of different energy storage systems for wind energy integration. In: European Wind Energy Conference CiematDer, EWEC, Copenhage (2001) 2. Rekioua, D., Roumila, Z., Rekioua, T.: Etude d’une centrale hybride photovoltaïque - éolien – diesel. Renew. Energy Rev. 11(4), 623–633 (2008) 3. Saheb-Koussa, D., Belhamel, M.: Production d’électricité sans interruption moyennant un système hybride (éolien – photovoltaïque – diesel). Renew. Energy Rev. ICRESD-07 Tlemcen 121–128 (2007) 4. Ming, Y., Yu, X.: Energy Efficiency: Benefits for Environment and Society (Green Energy and Technology). Springer, London (2015)

54

T. Boutabba et al.

5. Yang, H.X., Zhou, W., Lu, L.L., Fang, Z.H.: Optimal sizing method for stand-alone hybrid solar wind system with LPSP technology by using genetic algorithm. Solar Energy 82(4), 354–367 (2008) (Hong Kong, República Popular da China) 6. Zhou, W., Lou, C., Li, Z., Lu, L., Yang, H.: Current status of research on optimum sizing of stand-alone hybrid solar and wind power generation systems. Appl. Energy 87(2), 380–389 (2010) (Hong Kong, China Popular Republic) 7. Departement of Energy: Swimming Pool Covers (2009). http://energy.gov/energysaver/ swimming-pool-covers Accessed 22 Oct 2015 8. Trianti-Stourna, E., Spyropoulou, K., Theofylaktos, C., Droutsa, K., Balaras, C.A., Santamouris, M.: Energy conservation strategies for sports centers: part B. Swimming pools. Energy Build. 27(2), 109–122 (1998) 9. Shah, M.M.: Prediction of evaporation from occupied indoor swimming pools. Energy Build. 35, 707–713 (2003) 10. Ngan, M.S., Tan, C.W.: Assessment of economic viability for PV/wind/diesel hybrid energy system in southern Peninsular Malaysia. Renew. Sustain. Energy Rev. 15(9), 4659–4666 (2012) (Johor, Malásia) 11. Katsaprakakis, D.: Comparison of swimming pools alternative passive and active heating systems based on renewable energy sources in Southern Europe. Energy 81(March), 738– 753 (2015). https://doi.org/10.1016/j.energy.2015.10.1 12. Fazelpour, F., Soltani, N., Rosen, M.A.: Feasibility of satisfying electrical energy needs with hybrid systems for a medium-size hotel on Kish Island. Energy 73, 856–865 (2014) (Iran) 13. Milliken, J., Joseck, F., Wang, M., Yuzugullu, E.: The Advanced Energy Initiative. J. Power Sources 172, 121–131 (2007) (San Francisco) 14. Guimarães, B.M.A.: Pools: Associated Energy Consumption and the Application of Renewable Energies. Case Study Conducted in Private Education Establishment. University of Porto. Faculty of Sport (2010) 15. www.weatherbase.com.weather: Climate averages, forecasts, current conditions and normals search

Contribution of Renewable Energies in Existing Building Retrofits M. Badeche1,2 and Y. Bouchahm1,2(&) 1

Faculty of Earth Sciences and Architecture, Larbi Ben Mhidi University Oum El Bouaghi, Oum El Bouaghi, Algeria [email protected] 2 Department of Architecture and Urban Planning, University 3 Constantine, El Khroub, Algerie

Abstract. Current trends apply simulation tools for the design and the operation phases of the building. Simulation approaches are used to quantify building energy consumption, and support the development of possible energy use reduction. The residential sector presents a significant potential of energy reduction, through the integration of renewable energy. This research work aims to contribute towards energy savings by applying the solar collection strategy as an indoor heating system. For this purpose, a simulation model of a building has been implemented in Pleiade-Comfie software. The potential of indoor space heating by glazing balcony, to make use of solar energy is investigated. This approach allowed finding optimal values for design parameter, in terms of energy savings, in a semi arid climate of Algeria. Keywords: Numerical simulation balcony


Retrofitting
Renewable energy
Glazed

1 Introduction In Algeria, the building is the biggest energy consumer, with a contribution of almost 41% of the national end-use energy [1]. The residential sector for over 43% of energy consumption [2], which is increasing, with a rate of 147% [3]. While Space heating represents up 46% of this consumption [4]. To deal with limited energy resources, and to achieve a low carbon emission, sustainability measures such as integrating renewable energy in the building sector must be used. The renewable energy is an efficient management of building energy which plays a vital role on environment improvement. The renewable energy sources employed in buildings are (solar, hydroelectric, wind, biomass, and geothermal power and hybrid systems of these resources.). They require large investments in infrastructure in Algeria [5]; however the use of a strategy of passive solar is more adequate solution. Since Algeria has one of the largest solar resources in the world (exceeding 5 billion GWh/yr) [2]. Energy loads in old buildings are greater than those in recent building. Consequently, sustainable energy consumption is intimately linked with the improvement of their thermal performance [3]. In Algeria, 30% of building stock was built before 1977, © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 55–61, 2021. https://doi.org/10.1007/978-3-030-63846-7_6

56

M. Badeche and Y. Bouchahm

therefore a retrofit strategy of existing buildings become a priority [3]. Energy efficiency retrofits are defined as actions that allow an upgrade of the building’s energy and environmental performance to a higher standard than was originally planned [6]. Energy performance of the building envelope deserves great attention nowadays [7]. According to Ma et al. [8], the main factors that affect the building energy consumption are: envelope, lighting and air conditioning system [9]. In retrofitting procedure, efforts are to be made in envelope reinforcing. The commonly applied solutions correspond to external wall insulation, replacing window glazing or resizing it, and closing balconies. Integration of sunspaces to the building, have been used successfully in retrofitting existing buildings with poor energy performance [10]. Sunspaces consist of attached glass house or glass covering the balconies [11]. Glazed balcony acts as a thermal buffer between inside and outside of the building, and as a solar collector when heat is transmitted to the adjacent spaces through the masonry common wall. It serves also for pre-heating air before reaching adjacent space. Today, Building Performance Simulation (BPS) tools have an important role to play in identifying and determining the best retrofit strategies among a wide range of candidate options [12]. They allow building retrofit strategy to be simulated in a structured way and assessed before it is implemented. The first aim of this paper is to present an approach which leads to an effective and feasible energy retrofitting of residential building by closing balconies. The adopted solution will respond to the issue of energy saving in the built environment by solar energy use as a passive heating strategy. The glazed balcony must be designed to take full advantage of local climate all year round. From the review of various studies, luís sanhudo et al. [13] identified four methods applied on research for retrofitting energy efficiency: statistical approach, artificial neural networks, computational models, and simulation software. In this work, the fourth method is adopted. The work is based on a numerical model of a typical residential apartment. The improvement of the energy efficiency of its envelope is applied, with pleiade-comfie software. 1.1

Local Climate

Every solution developed to reduce the energy consumption of buildings must first consider the climate. This study considered Constantine city located at (36°, 17 N and 7°, 23’ E), 694 m altitude. Having a semi arid climate, which is characterized by very cold and wet winters, and hot summers. The average intensity of solar radiation in this location is significant, as it is about 4230 W/m2/day on a horizontal surface [14]. Such type of climate has contrasting needs depending on the period of the year: heating in winter and cooling in summer, therefore it requires careful design decision. 1.2

Case Study

A residential building (1994s) was selected according to the most existing in Algeria. The base unit is an apartment of 120 m2 area, located in the second floor. Five thermal zones were defined and modeled in the software as shown in Fig. 1.

Contribution of Renewable Energies in Existing Building Retrofits

57

Fig. 1. The internal view of the simulated apartment, with Pleiade-Comfie software.

The base case balcony (2.7  1.5 m2) was assumed to be closed by glazing with a transparency ratio of 50% of the external wall. It included a simple clear glazing. It has no protection from sun. The reflectance values of walls, ceiling and floor were assumed to be 75, 75 and 36% respectively. The external ground surface was assumed to have a 10% reflectance value. Materials characteristics of base case building can be seen in Table 1.

Table 1. Characteristics of base case building materials used in simulations. External opaque walls Thickness Type (mm) 20 Mortar cement 150 Hollow brick 100 Air gap 100 Hollow brick 20 Plaster U-value (W/m2K) 0.52 Solar heat gain – coefficient Components

Slabs Thickness (mm) 13

windows Thickness (mm) Plasterboard 6

200

Slab corps creux Mortar cement

50

3.33 –

Type

5.10 0.90

Type Single glass

58

M. Badeche and Y. Bouchahm

2 Data Input Simulation A large amount of parameters play a role on the heating and cooling loads. The most determining are the following: outdoor condition variations, the building shape, envelop insulation level, ventilation and outdoor air infiltrations, heating and cooling temperature set points, efficiency of the heating and cooling systems, human behavior and equipment energy consuming [15]. The building was modeled in a three dimension environment, and all building cited parameters are set in as data input. Thermostat schedules for heating and cooling were also assigned, assuming a set point temperatures of 19 °C and 27 °C respectively.

3 Simulation Cases According to K. Hilliaho et al. [16], the among key factors affecting the energy engineering design of a glazed balcony are: losses from the building to the balcony and from the balcony to outdoor, and the absorption coefficients of its surfaces. Joe Clarke et al. [17] in another study, showed that the design elements affecting the solar radiation entering the building (e.g. balcony depth, glazing areas covering the external façade, glazing optical properties, etc.) have a significant impact on the heating loads. Thus, analysis was conducted for the following parameters: Number of glazed layers (single and double); Orientation (South, West, North and East) and glazing ratio (50%, 60%, 80% and 100%). Table 2 summarizes the types of glass studied, including their thermal characteristics.

Table 2 Types of studied glazing and their thermal characteristics. Glazing type Type Composition

Thermal characteristics Thermal conductivity U- value (W/m2.°K)

(a) (b)

5.10 2.80

Solar heat gain coefficient (SHGC) 0.90 0.72

2.80

0.40

(c)

Single glazing (6 mm pane) Double glazing (6 mm pane/12 mm air gap/6 mm pane) Double glazing (6 mm pane/12 mm air gap/6 mm pane and low-e film)

4 Results and Discussion The energy simulation will indicate the heating and cooling demands and assess the effect of glazed balcony retrofit from the energy perspective for a whole year.

Contribution of Renewable Energies in Existing Building Retrofits

4.1

59

The Effect of Glazing Ratio on Heating and Cooling Loads

Figure 2 shows the results for all the studied glazing, in the four principal orientations. It was found that the lower ratio of glazing (50%) is more adapted for reducing heating and cooling loads. Indeed, the glazing contributes in the reduction of energy heating in the cold season. Also it reduces solar rays then avoids overheating in the hot season. This is due to its configuration having less glazing area in contact with exterior [11].

Fig. 2. The effect of glazing ratio on heating and cooling loads.

4.2

The Effect of Orientation

The South orientation is the most efficient direction of glazed balconies where both heating and cooling are concerned. This is understandable considering that South orientation receives lower solar energy in the hot season and upper in the cold season. Positioning a 50% of glazing in the south face appears to be most effective, for all year round (Fig. 3). 4.3

The Effect of Glazing Types

The use of double glazing with low e film (type c) leads to better performs (Fig. 4). The impact of the thermal conductivity (U-value) on the annual load savings is smaller compared to that of the Solar heat gain coefficient (SHGC), as concluded by Tibi and Mokhtar [18].

60

M. Badeche and Y. Bouchahm

Fig. 3. The effect of orientation on heating and cooling loads.

Fig. 4. Heating and cooling loads comparison between the three types of glazing, in south orientation with WWR = 50%

5 Conclusion The first aim of this paper is to present an approach which leads to an effective and feasible energy retrofitting of residential building by closing balconies. The study identifies the energy potential of integration of solar energy for heating spaces of old residential building, situated in semi arid climates. Design variables such as glazing ratio, glazing type and space orientation were studied. Analysis indicated that glazed

Contribution of Renewable Energies in Existing Building Retrofits

61

balconies can be an appropriate and effective system all year round if properly designed. The most successful combination having higher energy savings in kilowatt hours can be achieved with a 50% ratio of glazing type (c), having lower SHGC, and facing south.

References 1. Bilan Energétique National de l’année 2017, Ministère de l’Energie et des Mines, Edition (2018) 2. Stambouli, A.B.: Algerian renewable energy assessment: the challenge of sustainability. Energy Policy 39, 4507–4519 (2011) 3. Khelifa, I., et al.: Analysis of strategies to reduce thermal discomfort and natural gas consumption during heating season in Algerian residential dwellings. Int. J. Sustain. Build. Technol. Urban Dev. 45–76 (2020) 4. El Hassar, S.M.K.: Guide pour une construction éco-énergétique en Algérie. Editions universitaires européennes, Saarbrüken, Allemagne (2016) 5. Bencheikh, D., Bederina, M.: Assessing the duality of thermal performance and energy efficiency of residential buildings in hot arid climate of Laghouat, Algeria. Int. J. Energy Environ. Eng. 11, 143–162 (2020) 6. Jaggs, M., Palmer, J.: Energy performance indoor environmental quality retrofit – a European diagnosis and decision making method for building refurbishment. Energy Build. 31, 97–101 (1999) 7. Badeche, M., Bouchahm, Y.: The energy savings potential of windows in the office buildings sector: state of the art. Nat. Technol. J. Vol. A: Fundam. Eng. Sci. 18, 27–32 (2018) 8. Ma, H., et al.: Analysis of typical public building energy consumption in northern China. Energy Build. 136, 139–150 (2017) 9. Kadraoui, H., et al.: Analysis of energy consumption for Algerian building in extreme NorthAfrican climates. Int. J. Sustain. Energy Plan. Manag. 19, 45–58 (2019) 10. Grunewald, S, Rottensteiner. S.: Task 37 - Advanced Housing Renovation with Solar & Conservation - Apartment Buildings in Dornbirn. International Energy Agency (2004) 11. Aelenei, D., et al.: The use of attached-sunspaces in retrofitting design: the case of residential buildings in Portugal. Energy Procedia 48, 1436–1441 (2014) 12. Cascioa, E., et al.: Residential building retrofit through numerical simulation: a case study. Energy Procedia 111, 91–100 (2017). Turin. Italy 8th International Conference on Sustainability in Energy and Buildings 13. Sanhudo, L., et al.: Building information modeling for energy retrofitting – a review. Renew. Sustain. Energy Rev. 89, 249–260 (2018) 14. National office of weather data. https://www.meteo.dz/ 15. Clement, P.: Building energy retrofitting: from energy audit to renovation proposals the case of an office building in France. Master of Science Thesis KTH School of Architecture and the Built Environment SE-100 44 STOCKHOLM (2012) 16. Hilliaho, K., et al.: Energy saving potential of glazed space: sensitivity analysis. Energy Build. 99, 87–97 (2015) 17. Clarke, J., et al.: Study of the Energy Performance of Korean Apartment Buildings with Alternative Balcony Configurations. World Renewable Energy Congress (WRECX) Editor A. Sayigh © 2008 WREC (2008) 18. Tibi, G., Mokhtar, A.: Glass selection for high-rise buildings in the united arab emirates considering orientation and window-to-wall ratio. Energy Procedia 83, 197–206 (2015)

Robust Control of Grid-Interfaced Wind Energy Conversion System Based on Active Disturbance Rejection Control Abdeldjabar Benrabah1(&), Farid Khoucha1, Fayçal Benyamina1, Ali Raza2, and Mohamed Benbouzid3 1

2

Department of Electrical Engineering, Ecole Militaire Polytechnique, Algiers, Algeria [email protected] Department of Electrical Engineering, University of Lahore, Lahore, Pakistan 3 Institut de Recherche Dupuy de Lome, University of Brest, Brest, France

Abstract. Considering control limitations in permanent magnet synchronous generator (PMSG) based-grid connected-wind energy conversion (WEC) systems operating under parameter uncertainties and disturbed conditions, a new control strategy based on active disturbance rejection control (ADRC) is proposed in this paper. The proposed control method uses an extended state observer to estimate the total disturbance including parameters uncertainties, grid disturbances and wind speeds variation then compensate it in the disturbance rejection control loop. In order to illustrate the dynamic performances of the proposed control strategy, various simulation studies are performed under MATLAB/Simulink environment. It is shown that the proposed ADRC strategy exhibits significant improvements in both tracking performance and antidisturbance ability. Keywords: Permanent magnet synchronous generator
Active disturbance rejection control
Grid connected-wind energy conversion systems
Robust control
Renewable energy

1 Introduction The energy production potential of wind energy conversion (WEC) systems relies not only on the speed of wind at the installation area, but also relies on the control strategy used for WEC systems to deal with intermittent wind conditions, parameter uncertainties and disturbances [1, 2]. Recently, variable-speed permanent magnet synchronous generators (PMSGs) are gaining much attention in grid-connected WEC systems as they are presenting many advantages such as reliability, loss reduction, higher power density and optimal efficiency [2]. In the ongoing researches, many control strategies have been proposed for WEC systems with PMSG to cope with parameters uncertainties and disturbances which can affect system operation and may lead to system instability. However, the controller design for WEC systems with PMSG still needs to be improved so as to guarantee safe and suitable integration to the electrical grid. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 62–70, 2021. https://doi.org/10.1007/978-3-030-63846-7_7

Robust Control of Grid-Interfaced Wind Energy Conversion System

63

Generally, proportional integrator (PI) based-controllers are the most adopted control technique for interfacing renewable energy sources to the utility grid as they provide the benefit of simplicity and easy implementation [3]. Nevertheless, taking into account parameters uncertainties and disturbances in grid-tied PMSG-based WEC systems, the control performance of classical PI controllers becomes poorer and insufficient to achieve grid connection requirements. Therefore, many research works have been oriented to develop other controllers in order to be applied in WEC systems. In [4], a sliding mode control system for generator and grid side control of WEC systems is proposed but the problem of chattering phenomenon is not completely resolved. In [5], the authors have proposed an H-infinity based robust control technique applied to PMSG-based WEC system to regulate the active and reactive power but only the machine side converter control is studied while the grid-side converter control is not explored. The authors in [6] proposed an uncertainty and disturbance estimator-based control but the experimental validation is not provided. To deal with the over mentioned control problems under parameter uncertainties and disturbances, another control method named active disturbance rejection control (ADRC) is proposed in this paper [7–10]. The flow of this paper is organized as follows: Sect. 2 provide a detailed description of the PMSG based WEC. Section 3 presents the proposed controller based on ADRC for grid side and machine side converter control. In Sect. 4, the effectiveness and robustness of the ADRC controller is illustrated through simulation results. Finally, conclusions are drawn at the end.

2 Wind Energy Conversion System Description Figure 1 shows the typical configuration of grid connected WEC system. The machine side converter (MSC) control ensures the control of machine torque according to the reference of power. The grid side converter (GSC) control ensures the regulation of the dc-link voltage and the injected power into the electrical grid. The direct-drive PMSG offers higher efficiency and reduces the cost by eliminating the gearbox [11]. The control of the stator currents is greatly affected by variations of the resistances and inductances of the stator windings which may yield an inefficient control of the MSC. This could in turn results in worse tracking performance of the active and reactive power references. Moreover, the cross-coupling terms between active and reactive current control loops may complicate the independent control of the active and reactive power of PMSG [12]. The same stability issues of cross coupling and parameter variations may compromise the current regulation in grid side converter control [13]. On the other hand, grid-connected WEC systems are subject to grid disturbances. Classical proportional-integral (PI) controllers present a major disadvantage in grid connected applications as they are known to be ineffective to deal with low-order harmonics in the synchronous reference frame. Proportional-resonant controllers can deal with this problem but they may not ensure system stability under a large band of harmonic cancellations [14]. The drawbacks mentioned above can be resolved by designing an ADRC controller appropriately. In the ADRC control scheme, parameter uncertainties, grid voltage disturbances and coupling effects are modeled as an aggregate disturbance.

64

A. Benrabah et al.

The proposed control scheme compensates for the aggregate disturbance which ensures convenience of implementation in practical applications.

Fig. 1. Simplified scheme of grid-connected PMSG based WEC system.

3 Proposed Control Strategy 3.1

Brief Overview of ADRC

The main idea of the ADRC is based on to treating the external and internal disturbance and uncertainties as “a total disturbance”, and using an extended state observer (ESO) to estimate and then compensate them in the disturbance rejection loop [7, 8]. ADRC is independent of the model and can explicitly deal with coupled systems, various unknown disturbances, and dynamic uncertainties. In order to reduce the control complexity, the traditional nonlinear ADRC was greatly simplified by Gao into a linear ADRC [10]. The corresponding linear ESO used to estimate the state variables and total disturbance is as follows:


z_ ðtÞ ¼ AzðtÞ þ Buðt  Td Þ þ LðyðtÞ  ^yðtÞÞ ^yðtÞ ¼ CzðtÞ

ð1Þ

where L = [b1 b2]T is the observer gain vector and Td is the time delay. The gain vector of the ESO is chosen so that the eigenvalues of (A – LC) are located at −xo; which gives b1 = 2xo, b2 = x2o. In this way, the observer states z1 and z2 will track y and f respectively [10]. The ADRC canonical model is given by: y_ ¼ f ðt; y; wÞ þ bu

ð2Þ

where f is the total lumped disturbance acting on the system, b is a given constant, d is the internal disturbances including modeling errors and parameters uncertainties, u is the system input and y is the system output. The association of the ESO and the feedback controller forms the ADRC control scheme whose structure is depicted in Fig. 2. The feedback gain is kp given by kp = −xc = 4/Tsettle where Tsettle is the desired settling time and xo is closed-loop bandwidth. The observer bandwidth xobs is selected such that xobs = 3  10 as the ESO dynamic is higher than the process dynamic [10].

Robust Control of Grid-Interfaced Wind Energy Conversion System

65

Fig. 2. First order ADRC control scheme

3.2

MSC Control

The stator windings voltage for PMSG in dq-axes rotating reference frame are expressed by (

usd ¼ Lsd didtsd þ Rs isd  Lsq xisq di usq ¼ Lsq dtsq þ Rs isq þ Lsd xisd þ xwf

ð3Þ

where usd and usq are the stator windings voltages in dq-axes, isd and isq are the stator windings currents in dq-axes, Rs is the resistance of stator windings and wf is the permanent magnet flux. Lsd and Lsq are the dq-stator windings inductances. The regulation of the stator windings currents is achieved by using the ADRC controller as shown in Fig. 3. The reference direct stator current isd is set to zero while the reference quadrature stator current isd is obtained by the optimal torque control maximum power point tracking OTC-MPPT [15]. When isd is set to zero, the electromagnetic torque is controlled by isq as follows 3 Te ¼ pwf isq 2

ð4Þ

Then, isq can be expressed by isq ¼ ð

2 ÞT  3pwf e

ð5Þ

The stator windings currents isd and isq are written in the ADRC canonical form as follow: _isdq ¼ fsdq ðy; d; tÞ þ bisdq 0 usdq

ð6Þ

66

A. Benrabah et al.

The direct stator current is expressed as Lsq disd Rs 1 ¼ isd þ xisq þ usd Lsd dt Lsd Lsd

ð7Þ

And quadrature stator current is expressed as disq Rs Lsd 1 1 ¼ isq  xisq þ usq  w Lsq Lsq f dt Lsq Lsq

ð8Þ

or fsd and fsq are the lumped disturbances affecting isd and isq respectively. usd and usq are the control inputs. bisd and bisq are given parameters of the system. 3.3

GSC Control

The GSC control consists of an outer control loop for dc link voltage control and an inner control loop for grid current regulation as shown in Fig. 3. The grid currents are conditioned and converted to their corresponding dq-values using the information obtained from the phase locked loop (PLL). The d-axis current reference is obtained from the outer dc link voltage control loop while the q-axis current reference is obtained from the desired delivery of reactive power. Accordingly, the active and reactive can be controlled by the d-axis and q-axis grid currents respectively as follows


Pg ¼ 32 vgd igd Qg ¼  32 vgd igq

ð9Þ

Additionally, the power in the DC-link capacitor is given by Pdc ¼ Vdc ðis  ig Þ

ð10Þ

Then, Pdc ¼ CVdc

dVdc dt

ð11Þ

We also have, Pdc ¼ Ps  Pg

ð12Þ

where Ps and Pg are the source and grid powers, respectively. From the above equations, the DC link voltage can be derived as follows 2 dUdc 2Udc 3vgd ¼ is  igd dt C C

ð13Þ

Robust Control of Grid-Interfaced Wind Energy Conversion System

67

By putting X = U2dc, the dc link voltage control can be written in the ADRC canonical form as follows 2 dUdc 2Udc 3vgd ¼ is  ifd dt C C

ð14Þ

For the q-axis current reference, it is calculated as follows igq  ¼ 

2 Qfref 3vgd

ð15Þ

Then, similarly to the MSC control, the grid-side currents are given in the canonical form of ADRC. is

iwind

$&

V

θ PCC

3:0

DEF

0337

Sψ

id ,ref = 0

iq ,ref

Vdc

DEF

Vdc ,ref

GT

$'5&

id id ,ref

$'5&

θ

Vq ,ref

DEF

Vd ,ref

+ vg

3//

vg

3:0

GT iq

3&&

$&

θ

Tem ,ref

L

Vdc

'&

PMSM

Turbine

ig

'&

C

GT

$'5&

Vq ,ref GT DEF

Vd ,ref

id

iq

$'5& $'5&

iq ,ref id ,ref

Fig. 3. MSC control scheme based on ADRC

4 Performance Evaluation Results In order to evaluate the performances of the proposed control, numerical simulations are performed under MATLAB/Simulink environment. The parameters of the PMSG and the grid connected electrical system are given in Table 1. The wind speed profile is depicted in Fig. 4. The wind speed varies between 10 and 12 m/s. The dq-axis stator currents responses in the dq-axis are shown in Fig. 5. The obtained result illustrates the robustness of the ADRC controller in tracking the desired references in spite of transient and gets to the reference current with fast dynamic response. Similarly, it can be observed from the evolution of the d–q-axis grid currents shown in Fig. 6 that the grid currents follow their references. Moreover, from Fig. 7, it is clear that the DC-link voltage is constant and equal to its desired reference, which is 800 V in spite of the wind speed variation. The ADRC controller is then effective to keep the controlled variables variation within an acceptable range. Figure 8 shows the power flow from the turbine to the grid. As can be noted, the active power tracks the maximum power under wind speed variations. The reactive is set to zero to ensure unity power factor. Hence, the voltage and current of the grid are in phase as seen in Fig. 9.

68

A. Benrabah et al.

Finally, in order to evaluate the robustness of the proposed ADRC controller under parameter variations and disturbances, we have performed a change in the grid inductance by an increase of 50% of its nominal value and introduced the following grid distortion levels: 5% fifth harmonic and 3% seventh harmonic. Figure 10 shows the high quality of the current injected to the grid with a THD of 2.84%. Hence, the proposed controller provides a robust performance under parameter variations and disturbances.

Fig. 4. Wind speed profile

Table 1. Grid-connected wind energy system parameters Parameters Symbols Value Number of pole pairs p 48 Stator winding RS, LS 0.06 X, 0.1 mH Viscous friction f 0.01 N.m.s coefficient Moment of inertia J 3500 kg.m2 Magnetic flux wf 1.48 Wb

Fig. 5. dq-axis grid current response using ADRC

Parameters DC link voltage Output filter Grid voltage and frequency Switching frequency Rated power

Symbols Value 800 V Vdc L, R 1 mH, 0.1 X vg, f 380 V, 50 Hz fs 10 kHz P 1.5 MW

Fig. 6. dq-axis grid current response using ADRC

Robust Control of Grid-Interfaced Wind Energy Conversion System

Fig. 7. DC link voltage response using ADRC

Fig. 9. Three phase grid current and grid voltage response under nominal operation

69

Fig. 8. Active and reactive power injected into the grid

Fig. 10. Three phase grid current and grid voltage response under grid inductance variation and grid voltage disturbances

5 Conclusion In this paper, an ADRC control scheme is developed for a WEC system based on PMSG. Machine side and grid power converters are controlled in order to extract the maximum of the power and inject it into the utility grid. Simulation results show that the proposed control provides a good dynamic performance under wind speed variation and high quality of the injected grid current. Moreover, the proposed controller provides a robust performance under parameter variations and grid disturbances.

References 1. Sharma, S., Singh, B.: Control of permanent magnet synchronous generator-based standalone wind energy conversion system. IET Power Electron. 5(8), 1519–1526 (2012) 2. Tripathi, S.M., Tiwari, A.N., Singh, D.: Grid-integrated permanent magnet synchronous generator based wind energy conversion systems: a technology review. Renew. Sustain. Energy Rev. 51, 1288–1305 (2015) 3. Tripathi, S.M., Tiwari, A.N., Singh, D.: Optimum design of proportional-integral controllers in grid-integrated PMSG-based windenergy conversion system. Int. Trans. Elect. Energy Syst. 26(5), 1006–1031 (2016)

70

A. Benrabah et al.

4. Zhang, S., Tseng, K.J., Vilathgamuwa, D.M., et al.: Design of a robust grid interface system for PMSG-based wind turbine generators. IEEE Trans. Ind. Electron. 58(1), 316–328 (2011) 5. Kolar, J.W.: Novel three-phase AC-DC-AC sparse matrix converter. IEEE Trans. Power Electron. 22(5), 1649–1661 (2007) 6. Ren, B., Wang, Y., Zhong, Q.-C.: UDE-based control of variable-speed wind turbine systems. Int. J. Control 90, 121–136 (2017) 7. Han, J.: From PID to active disturbance rejection control. IEEE Trans. Ind. Electron. 56(3), 900–906 (2009) 8. Huang, Y., Xue, W.: Active disturbance rejection control: methodology and theoretical analysis. ISA Trans. 53(4), 963–976 (2014) 9. Zhou, Z., Elghali, S.B., Benbouzid, M., Amirat, Y., Elbouchikhi, E., Feld, G.: Tidal stream turbine control: an active disturbance rejection control approach. Ocean Eng. 202, 107190 (2020) 10. Gao, Z.: Scaling and bandwidth-parameterization based controller tuning. In: Proceedings of the 2003 American Control Conference, pp. 4989–4996 (2003) 11. Li, S.Q., Zhang, K.Z., Li, J., et al.: On the rejection of internal and external disturbances in a wind energy conversion system with direct-driven PMSG. ISA Trans. 61(3), 95–103 (2016) 12. Zhao, Y., Wei, C., Zhang, Z., et al.: A review on position/speed sensorless control for permanent-magnet synchronous machine-based wind energy conversion systems. IEEE J. Emerg. Sel. Top. Power Electron. 1(4), 203–216 (2013) 13. Benrabah, A., Xu, D., Gao, Z.: Uncertainty and disturbance estimator based current control of LCL-Filtered grid-connected inverters. Presented at the 2017 IEEE ITEC Asia-Pacific, Harbin, China (2017) 14. Blaabjerg, F., Teodorescu, R., Liserre, M., Timbus, A.: Overview of control and grid synchronization for distributed power generation systems. IEEE Trans. Ind. Electron. 53(5), 1398–1409 (2006) 15. Kumar, S.S., Jayanthi, K., Kumar, N.S.: Maximum powerpoint tracking for a PMSG based variable speed wind energy conversion system using optimal torque control. In: Proceedings of the 2016 ICACCCT, Ramanathapuram, India, pp. 347–352 (2016)

Wind Speed Forecasting Based on Discrete Wavelet Transform, Moving Average Method and Gated Recurrent Unit K. Zouaidia(&), S. Ghanemi, and M. S. Rais Embedded Systems Laboratory, Badji Mokhtar University, Annaba, Algeria [email protected]

Abstract. Wind power generation is a very sophisticated process that requires a highly accurate wind speed forecast. In this study, a short term wind speed forecasting model was proposed based on the Discrete Wavelet Transform technique, Moving Average method and the Gated Recurrent Unit model (DWT-MA-GRU), the DWT followed by MA were used for Data Denoising process then Max-Min normalization technique was applied to standardize the data and finally the GRU model was used for the wind speed prediction. GRU, LSTM, CNN, DWT-LSTM, DWT-CNN, DWT-GRU, DWT-MA-LSTM, DWT-MA-CNN models were used as benchmark models. The experimental results showed that the proposed model outperformed the other models which were validated by the RMSE (root mean square error), MAE (mean absolute error) and MAPE (mean absolute percentage error) from experimental results of two different datasets from two different cities. Keywords: Gated Recurrent Unit speed
Deep learning


Discrete Wavelet Transform
Wind

1 Introduction The increasing development of industrial factories and the acceleration of economic activities led to a considerable growth in the energy demand. While the world is still suffering from the environmental pollution and global warming caused by the abused utilization of conventional energy resources such as fossil fuels (oil, natural gas and coal). Therefore the world today is in desperate need for alternative energy sources more than any other time and the wind power generation is considered as the most rentable renewable energy in matter of cost and materials. But due to the instable nature of the wind, forecasting its parameters remains a difficult task which requires a powerful model capable of automatically learning features from a sequence within the temporal ordering and the best way to achieve that is the use of Recurrent Neural Networks. Latest researches showed that forecasting the wind speed parameter is the main clue to get an accurate power generation schemes where the deep learning techniques proved its efficiency over traditional models. In our study we will focus on hourly wind speed forecasting using one of the most popular Recurrent Neural Networks known as GRU network. The selection of this model was based on its simplicity © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 71–78, 2021. https://doi.org/10.1007/978-3-030-63846-7_8

72

K. Zouaidia et al.

and efficiency in matter of computation time and its high performance. For the data denoising process we needed a method that have a high resolution in the frequency and also in the time domain so the Discret Wavelet Transform in conjunction with the Moving Average Technique were the best choice to fulfil these two conditions. Using datasets of Adrar and Tindouf city which both have the highest levels of wind speed in Algeria. The rest of the paper is structured as follows: Sect. 2 describes the related works. Section 3 presents the architecture of the proposed model. Section 4 discusses the obtained results and the final section summarizes the study and proposes future works.

2 Related Works In 2018, Yusuf Elmir proposed an Artificial Neural Network (ANN), combined with the Genetic Algorithm (GA) for the weights generation, where the input layer of this model is made up of 4 values: day, month, year, time; alongside an output layer composed of air temperature, relative humidity, atmospheric pressure, mean wind speed of Béchar city the GA-ANN model showed such a good improvements where the mean error was reduced for each predicted variable from 3.8632° for air temperature, 3.0006% for humidity, 11.0101 (mmHg) atmospheric pressure, 2.4065 (m/s) for the wind speed to 2.7918°, 3.0454%, 10.3271 (mmHg), and 2.3599 (m/s) respectively for the GA-ANN compared to the ANN model once used alone. [1]. In the same year, Yiwei Fu et al., proposed a framework for an effective wind power forecasting as well as an LSTM/Two-layer GRU forecasting models with a wind speed correction process using NWP data. The proposed multi-input and multi-hiddenlayer model were fed with the corrected wind speed data at the prediction moment. The experiments results showed that the LSTM and GRU are equivalent when the training data set is reduced but when it is extended the LSTM gave better performance [2]. A year later, Yaping Deng et al., presented a wind power prediction model based on bidirectional GRU alongside an Adam optimization adaptive learning rate algorithm which is used for the GRU weights update. The proposed model used a dataset provided from a wind farm containing 33 wind turbines in Ningxia, China. The results showed an upgraded in the forecasting accuracy by making full use of the information obtained from multiple data sources of numerical weather forecast [3]. Then Yao Liu et al. came to propose the DWT_LSTM model for wind power generation, where the DWT was used for the data preprocessing stage to get approximate and detailed signals then z-score normalization method was applied to standardize the decomposed data. Every sub-signal resulted from the previous decomposition was assigned to a separate LSTM. The proposed model overcame the limitations of the benchmarks models [4]. By the end of the year, Hui Liu et al., proposed a new hybrid multi-step short-term wind speed forecasting model based on SSA-CNNGRU and SVR model where the SSA is used for data pre-processing in order to get one trend component and several detail components; then the trend component of wind speed were predicted using an enhanced CNNGRU, while the detail components were obtained using the SVR model.

Wind Speed Forecasting Based on Discrete Wavelet Transform

73

Authors worked with three sets of 15-min averaged wind speed data from three wind farms in China. The results showed that the proposed method outperformed the other methods [5].

3 The Proposed Methotology The entire process of the DWT-MA-GRU model is shown in Fig. 2 which can be represented as follows: 3.1

Discrete Wavelet Transform (DWT)

The DWT is a mathematical function that is localized in time and frequency; it is also a tool that transforms the original signal with time domain into its frequency sub-bands for analysis and processing [6]. the application of the DWT in removing highfrequency noise proved its effectiveness and robustness. In this paper we make use of the Daubechies family of wavelets specifically db5 wavelet (5 filter coefficient). To show the power of the DWT we applied the db5 up to 3 levels of decomposition. The general process for wavelet-based denoising proposed in this study can be summarized as follows: 1. Estimation of the noise variance r2 from the finest wavelet coefficients d1 using: r ¼ med ðjdb5jÞ=0:6745

ð1Þ

2. Determination of the noise threshold limit as: (Where the N is the length of the signal) k ¼ r  sqrtð2  lnðN ÞÞ

ð2Þ

3. Application of the soft thresholding to remove noise that falls under the estimated threshold limit. 4. Reconstruction of the original signal by performing an inverse wavelet transformation, which results in the denoised signal [6]. After the application of DWT denoising, the noise level at the lower band frequencies remains high. To further denoise the data, Moving Average technique was applied. 3.2

Moving Average Technique

The idea behind using moving average for smoothing data is that observations which are nearby in time tend to be close in value so the use of the average of the points near every observation will grant a reasonable evaluation of the trend at that observation. The average eliminates some of the randomness in the data, leaving a smooth trend component. In this Study the value at time (t) is calculated as the average of the raw observations at and before the time (t) with a window of 9 [7].

74

K. Zouaidia et al.

yðtÞ ¼ mean ðobsðt  8Þ; obsðt  7Þ; obsðt  6Þ; obsðt  5Þ; obsðt  4Þ; obsðt  3Þ; obsðt  2Þ; obsðt  1Þ; obsðtÞÞ

ð3Þ Where: obs is observation at t, y(t) is the result of data smoothing using trailing moving average at t. 3.3

Min-Max Normalization

Rescaling (min-max normalization) is a widely used normalization data technique which sets the minimum value of each feature to 0 and the maximum value to 1 and every other value gets transformed into a decimal between 0 and 1. [8]. Y 0 ¼ ðyðtÞ  minÞ=ðmax  minÞ

3.4

ð4Þ

Gated Recurrent Unit

The GRU is the most appropriate structure for dealing with time sequence problems and the wind speed data studied here presents a typical time series data. This network presents the newer generation of Recurrent Neural networks that came to solve the problem of gradient descent faced with the RNN, the cell as illustrated in Fig. 1 is adopted to construct the GRU Network. It is composed of reset gate used to decide how much precedent information to forget and update gate that uses a hidden state to transfer information and decides what information to eliminate and what new information to add. Once the reset gate is closed, the previous state is ignored and the current state is only determined by the current input, that is to say, all the useless information for the current hidden state is dropout. [9] In this study, the proposed network consists of Single GRU layer, while the mean squared error, Adam algorithm and RELU are adopted as the loss function, optimization function and activation function, respectively.

Fig. 1. The structure of the GRU.

Wind Speed Forecasting Based on Discrete Wavelet Transform

75

4 Dataset Datasets for this research was obtained from Raspisaniye Pogodi Ltd, “Weather for 243 countries of the world” site in St. Petersburg, Russia, since 2004. The company has the license for activity in hydrometeorology and other close areas [10]. Five years of wind speed dataset of Adrar and Tindouf city were used for these experimentations from the period between 01 January 2014 and 01 January 2019. 67% of data were set for training from the year 2014 to 2017 and 33% years for test and validation from 2017 to 2019 where the data was captured every hour. The choice of the sites was based on the Algerian Wind Atlas map [11] where the highest wind speed levels were given in Adrar and Tindouf city showing a great potential for the study and also to confirm the efficiency of the proposed model with testing it on two different datasets that have different characteristics.

Fig. 2. The whole process of the DWT-MA-GRU model.

76

K. Zouaidia et al.

5 Experiments Results In these experiments, three classifications with three models for each were built for Adrar and Tindouf city’s wind speed: (CNN, LSTM and GRU), (DWT-CNN, DWTLSTM and DWT-GRU) and (DWT-MA-CNN, DWT-MA-LSTM and the proposed DWT-MA-GRU model). As can be seen from Tables 1, 2 and Figs. 4, 6 the RMSE, MAE and MAPE achieved highest values when using GRU with DWT-MA. This confirms the effectiveness of the proposed method. The comparison of the models for Adrar and Tindouf city ‘s wind speed forecasting are shown in Figs. 3 and 5 respectivly.

Fig. 3. Comparison of the 9 models for Adrar City wind speed Fig. 4. RMSE Adrar city. Forecast.

Results

for

Wind Speed Forecast for Adrar City: The RMSE results of the models: CNN, LSTM, GRU then DWT-CNN, DWT-LSTM, DWT-GRU then DWT-MA-CNN, DWT-MA-LSTM and the DWT-MA-GRU for Adrar city are as follows: 2.077 (m/s), 1.476 (m/s), 1.402 (m/s), then 0.170 (m/s), 0.046 (m/s), 0.022 (m/s) then 0.095 (m/s), 0.034 (m/s) and 0.015 (m/s) respectively, this indicates that the proposed model gives a better performance in comparison with the other two forecasting models followed by the DWT-MA-LSTM. Wind Speed Forecast for Tindouf City: The RMSE results of the models: CNN, LSTM, GRU then DWT-CNN, DWT-LSTM, DWT-GRU then DWT-MA-CNN, DWTMA-LSTM and the DWT-MA-GRU for Tindouf city are as follows: 2.201 (m/s), 1.561 (m/s), 1.481 (m/s) then 0.181 (m/s), 0.051 (m/s), 0.031 (m/s) then 0.105 (m/s), 0.034 (m/s), 0.020 (m/s) respectively, this results confirms that the proposed model gives a better performance in comparison with all the other forecasting models.

Wind Speed Forecasting Based on Discrete Wavelet Transform

Fig. 5. Comparison of the 9 models for Tindouf City wind Fig. 6. RMSE Tindouf city. speed Forecast.

Results

77

for

The MAE and MAPE results for Adrar and Tindouf city ‘s wind speed forecasting are shown in Tables 1 and 2 respectively.

Table 1. MAE and MAPE results of Adrar City. Models CNN DWT-CNN DWT-MA-CNN LSTM DWT-LSTM DWT-MA-LSTM GRU DWT-GRU DWT- MA-GRU

MAE (m/s) MAPE (m/s) 0.0693 0.0080 0.0071 22.0132 2.8060 2.6763 0.0792 0.0026 0.0022 16.0408 0.6923 0.7192 0.071 0.0022 0.0020 10.9617 0.7610 0.6421

Table 2. MAE and MAPE results of Tindouf City Models CNN DWT-CNN DWT-MA-CNN LSTM DWT-LSTM DWT-MA-LSTM GRU DWT-GRU DWT- MA-GRU

MAE (m/s) MAPE (m/s) 0.0755 0.0080 0.0071 21.0652 2.6100 2.5773 0.0692 0.0026 0.0022 14.8603 0.7723 0.7309 0.0663 0.0024 0.0018 11.0167 0.7501 0.5668

6 Conclusion In this paper, a short term wind speed forecasting model is developed by combining the DWT method, MA technique with the GRU model, in the architecture of the proposed DWT-MA-GRU the DWT and MA were used for denoising the datasets then the GRU model was designed to predict the hourly wind speed, we worked on Adrar and Tindouf city datasets having the highest wind speed levels in Algeria. A benchmark of models were selected for comparison purpose GRU, LSTM, CNN, DWT-GRU, DWTLSTM, DWT-CNN, DWT-MA-CNN, DWT-MA-LSTM where the proposed DWTMA-GRU showed more accurate results with the RMSE, MAE and MAPE being 0.015 (m/s), 0.0018 (m/s) and 0.5668 (m/s) respectively and the fastest computation

78

K. Zouaidia et al.

over the two datasets due to its simple and effective structure followed by the LSTM. In further works we will focus on gathering high accurate data from multiple stations to get better results, keeping up with trying other smoothing data techniques, investigating more models combinations and using different linear and nonlinear models for comparison.

References 1. Elmir, Y.: Weather forecasting using genetic algorithm based artificial neural network in South West of Algeria (Béchar). In: Mustapha, H.: Artificial Intelligence in Renewable Energetic Systems, LNCS in Networks and Systems, vol. 35, pp. 273–280. Springer International Publishing AG (2018) 2. Fu, Y., Hu, W.T., Tang, M., Yu, R., Liu, B.: Multi-step ahead wind power forecasting based on recurrent neural networks. In: Proceedings of IEEE PES Asia-Pacific Power and Energy Engineering Conference, Kota Kinabalu, Malaysia, pp. 217–222 (2018) 3. Deng, Y., Jia, H., Li, P., Tong, X., Qiu, X., Li, F.: Deep learning methodology based on bidirectional gated recurrent unit for wind power prediction. In: Proceedings of 14th IEEE Conference on Industrial Electronics and Applications, Xi’an, China, pp. 592–595 (2019) 4. Liu, Y., Guan, L., Hou, C., Han, H., Liu, Z., Sun, Y., Zheng, M.: Wind power short-term prediction based on LSTM and discrete wavelet transform. Appl. Sci. 9(1108), 1–17 (2019) 5. Liu, H., Mi, X., Li, Y., Duan, Z., Xu, Y.: A smart wind speed deep learning based multi-step forecasting model using singular spectrum analysis, convolutional gated recurrent unit network and support vector regression. Renewable Energy 143, 842–854 (2019) 6. Donoho, D., Johnstone, I.: Ideal spatial adaptation by wavelet shrinkage. Biometrika 81(3), 425–455 (1994) 7. Moving average smoothing for data preparation and time series forecasting in python. https://machinelearningmastery.com/moving-average-smoothing-for-time-series-forecastingpython/. Accessed 20 Apr 2020 8. Normalization. https://www.codecademy.com/articles/normalization. Accessed 03 May 2020 9. Illustrated guide to LSTM’s and GRU’s: a step by step explanation. https:// towardsdatascience.com/illustrated-guide-to-lstms-and-gru-s-a-step-by-step-explanation44e9eb85bf21. Accessed 03 Jun 2020 10. Raspisaniye Pogodi Ltd.: Weather for 243 countries of the world. https://rp5.ru/Weather_in_ the_world. Accessed 07 Apr 2020 11. The wind deposit map in Algeria. https://www.cder.dz/spip.php?article1765. Accessed 24 Apr 2020

DTC-DPC of Induction Machine Connected to Wind Generator Hamidia Fethia(&) and Abbadi Amel LREA Laboratory, Medea University, Medea, Algeria [email protected]

Abstract. Wind energy is currently one of the most sought-after renewable and clean energies, both for isolated sites and as extra for connected networks. It can be a competitive alternative contributing to the reduction of the increasingly galloping demand for electricity. The development and multiplication of the use of renewable energy conversion chains have led industrialists and scientists to invest in improving the technical and economic indices of this conversion and the quality of the energy supplied. This paper proposes one of the best know controlled Induction machine named direct torque control (DTC) based on the flux and torque estimation, and in order to control DC voltage, this paper proposes also one of an interesting technique based on power and voltage estimation to produce a direct current voltage from an alternating network, called direct power control (DPC), this technique is similar to the direct torque control (DTC) of asynchronous machines, it is developed in this paper to control the DC voltage to be used later as input of our inverter. This hybrid system will be supplied by Wind energy; Finally, the simulation results are given to show the effectiveness and feasibility of the hybrid system ‘Wind-DPC-DTC-IM’. To improve the performance of the system, we propose a fuzzy logic type-2 controller in DTC to replace the hysteresis comparators and in DPC to replace PI-DC voltage regulator. Keywords: Wind generator
Direct torque control Induction machine
Fuzzy logic type-2


Direct power control


1 Introduction Wind Energy Conversion Systems have become an alternative to oil-based energy sources around the world [1]. Wind energy is energy that uses the wind. The wind turbine is a device intended to transform the kinetic energy of the wind into mechanical energy then into electrical energy. Wind energy is an indirect form of solar energy, it is due to the displacement of air masses linked to the sunshine of the earth. The warming of certain areas of the planet and the cooling of others creates a difference pressure which causes constantly moving air masses. The use of wind is not new today, it is one of the first natural resources to have been used by man. Wind turbines must be installed in open spaces to improve their performance and for safety reasons they are far from urban areas. They are perfectly suited for an isolated structure, such as a farm, where space is available. The different elements of a wind turbine are designed to maximize © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 79–87, 2021. https://doi.org/10.1007/978-3-030-63846-7_9

80

H. Fethia and A. Amel

energy conversion, to have a good match between the torque/speed characteristics of the turbine and the electric generator. The electrical network signals (especially currents) are often disturbed and they are not perfectly sinusoidal. For this reason, it is necessary to know all the energy exchanges between the network and the different loads in order to compensate for any disturbances. In many cases, the current consumed by the charges no longer has a pure sinusoidal form. Current distortion involves distortion of the voltage also dependent on the impedance of the source. The harmonic disturbances are caused by the introduction on the electrical network of non-linear loads such as equipment incorporating power electronics (inverters, static converters, light dimmers, welding stations). More generally, all materials incorporating rectifiers and cutting electronics deform currents and create voltage fluctuations on the distribution network. It is the concentration of many polluters in harmonics that generates a lot of disturbances on the network. The fast development of power electronics in the last forty years has made its use very abundant in the current industrial world, especially in the conversion and storage of electrical energy. In most applications (such as, for example, drives with DC or AC motors, battery chargers, DC power supply systems, etc.), the electrical energy distributed by the transmission network energy is converted to another form to facilitate its exploitation. This conversion is often done through electronic interfaces which are, in most cases, rectifiers diodes. This type of interface behaves against the supply network as a non-linear load, and the conversion cannot be done, therefore, without a significant degradation of the power quality, which is mainly manifested at the waveforms of the currents absorbed. The presence of harmonics in the supply network could, on the one hand, adversely affect the electrical equipment connected to it (application of overvoltages, overheating and damage to the devices) and, on the other hand, lead to improper operation of the equipment. Energy (excessive losses in transmission lines). Several rectifier control strategies have been proposed, such as direct power control (DPC), numerous works have already dealt with the study of DPC and different solutions have been suggested to control the DC voltage as sliding mode control, DPCSVM, etc., in this work, we propose DPC based on Fuzzy Logic type-2. In general, the most used electric actuators in the majority of industrial applications are built around the asynchronous squirrel cage machine. The latter is distinguished in particular by its robustness, its reliability, its low cost and it does not require regular maintenance. However, its dynamic behavior is often very complex, since its modeling results in a system of nonlinear equations, strongly coupled and multivariable. In addition, some of its state variables, such as flux are not measurable. These constraints require more advanced control algorithms to control the torque and flux of these machines in real time. Several control strategies have been proposed in the literature to achieve this objective. In the mid-1980s, a strategy for controlling the asynchronous machine, known as Direct Torque Control or DTC (Direct Torque Control), appeared to compete with conventional commands. Its principle is based on a direct determination of the control pulses applied to the switches of the voltage inverter, in order to maintain the electromagnetic torque and the stator flux inside two bands with predefined hysteresis, Such an application of this technique allows provide a decoupling between torque and flux control without the need to use pulse width modulation or coordinate

DTC-DPC of Induction Machine Connected to Wind Generator

81

transformation, however these two regulators causes a high ripples in torque and flux, to resolve this problem, this work proposes also a two fuzzy controllers to replace the hysteresis controllers. In modern high-performance AC drives, the Field Oriented Control (FOC) and the Direct Torque Control (DTC) have been investigated rigorously [2]. This performance is produced by the well-known fast dynamic response of DTC and DPC algorithms, so the objective of this paper is to combine the advantage of the DTC with DPC in same time by using wind renewable energy.

2 Proposed Hybride System 2.1

Hybrid System Principle

In this section, we propose hybrid technique (DPC-DTC), wich contain direct power controlled rectifier and direct torque controlled inverter, this system is supplied by wind energy. 2.2

Wind Turbine

Basically, the wind turbine is a device which transforms the kinetic energy of the wind into mechanical energy; in the general case; the wind turbine can be presented by a genetic model with its main parts, the rotor and the generator system. The modeling of the turbine thus consists of modeling the torque developed by the blades of the turbine. the power extracted from the wind can be expressed as follows 1 Px ¼ qSVx3 2

ð1Þ

With S ¼ pR2 . The mechanical power recovered by the wind turbine can be written: 1 PT ¼ qSVx3 Cp ðkÞ 2 k¼

RX Vx

1 TT ¼ qR3 Vx2 Cr ðkÞ 2

ð2Þ ð3Þ ð4Þ

The wind torque depends to three variables: the wind speed, the rotation speed of the turbine shaft and wedge angle b [3] (Fig. 2). 2.3

Direct Power Control DPC

DPC method provide robust performance and makes power quality improvement. It minimize the transient behavior of the system and reduces the complexity of control

82

H. Fethia and A. Amel

system and distortion whereas vector control method is complex control which require coordinate transformation, current controller, power controller and large tuning work which deteriorates the system performance and increases transient [4]. The main idea of direct power control (DPC) initially proposed by Ohnishi (1991) and developed later by Noguchi and Takahachi in 1998, is similar to the direct torque control (DTC) of asynchronous machines. Instead of the stator flux and the torque, the instantaneous active (p) and reactive (q) powers are chosen as two quantities to be controlled (Fig. 1). The active and reactive power are estimated by:
 dia dib dic ^p ¼ L ia þ ib þ ic þ vdc ðSa ia þ Sb ib þ Sc ic Þ dt dt dt 
  1 dia dic ^q ¼ pffiffiffi 3L ic  ia  vdc ½Sa ðib  ic Þ þ Sb ðic  ia Þ þ Sc ðia  ib Þ dt dt 3

ð5Þ ð6Þ

Sa, Sb, Sc, represent the state of the three phase legs 0 meaning that the phase is connected to the negative and 1 meaning that the phase is connected to the positive leg. The vector is calculated from the estimated active and reactive power; it can be written as: 

^ea ^eb

 ¼

 1 ia 2 2 i ia þ ib b

   ib ^ p
ia ^ q

ð7Þ

The vector PWM control plane (in a and b) is divided into twelve sectors so that each control vector divides each region into two equal parts. The DPC based on the selecting a control vector from a switching table. The latter used the errors of the instantaneous active and reactive powers, provided by the two-level hysteresis regulators, as well as the angular position of the estimated voltage vector which be calculated from the relationship cited below:
 ^he ¼ arctg eb ea

ð8Þ

The reference of the active power is obtained by regulating the DC voltage, using a PI regulator. To ensure a unit factor of the reactive power, the control at zero is performed. 2.4

Direct Torque Control DTC

The direct torque control is considered as one of the best alternative solution compared to classical one as direct field control; DTC is also usually implemented in control of induction machine [5]. In the DTC, the stator flux (estimation) follows from the stator voltage by integrating [6]:

DTC-DPC of Induction Machine Connected to Wind Generator

83

Fig. 1. Schematic diagram of hybrid system

Fig. 2. Turbie power characteristic (b = 0)

Z /s ¼

t

ðVs  Ris Þ dt

ð9Þ

0

Let’s us replace the estimate of the stator voltage Vs with the true value and write it as: j4p j2p 2 Vs ðSa ; Sb ; Sc Þ ¼ Udc ðSa þ Sb e 3 þ Sc e 3 Þ 3

ð10Þ

Sa, Sb, Sc, represent the state of the three phase legs 0 meaning that the phase is connected to the negative and 1 meaning that the phase is connected to the positive leg. The stator current space vector is calculated from measured currents ia, ib, ic: j4p j2p 2 is ¼ ðia þ ib e 3 þ ic e 3 Þ 3

ð11Þ

The component a and b of vector us can be obtained: Z t ðVsa  Risa Þ dt /sa ¼ 0 Z t   /sb ¼ Vsb  Risb dt 0

ð12Þ

84

H. Fethia and A. Amel

Torque relationship cited below:  3  Tem ¼ p /sa isb  /sb isa 2

ð13Þ

the voltage plane is divided into six sectors so that each voltage vector divides each region into two equal parts. [7]. Both actual torque and actual flux are fed to the comparators where they are compared, to a torque and flux reference value, Torque and flux status signals are calculated using a two level hysteresis control method [8]. 2.5

Induction Motor

The transformation of PARK brings back to the equation stator in reference frame related to the rotor [9, 10]. h

2 Msr dIsd Msr Msr 1 dt ¼ rLs ðRs þ Lr Tr ÞIsd þ xs rLs Isq þ Lr Tr urd þ Lr xr urd h dIrq Msr2 Msr Msr 1 dt ¼ rLs xs rLs Isd  ðRs þ Lr Tr ÞIsq  Lr xr urd þ Lr Tr urq durd Msr 1 dt ¼ Tr Isd  Tr urd þ ðxs  xr Þ urd durq Msr 1 dt ¼ Tr Isq  Tr urq  ðxs  xr Þ urd 2 P Msr dx F P dt ¼ Lr J ðIsq urd  Isd urq Þ  J x  J Tr

þ Vsd þ Vsq

i i ð14Þ

With ðr ¼ 1 

2.6

Msr2 Þ Ls Lr

Direct Torque Control DTC and Direct Power Control Based on Fuzzy Logic Type-2

The input and the output present respectively torque or flux error (eTem = Tem* − Tem, e/ = /s* − /s) and Sce or Sph. So, in this section, we have two fuzzy regulators replacing the traditional hysteresis comparators of flux and torque as presented in Fig. 1. This controller has Gaussian seven T2 fuzzy sets (both for input and output) as presented in Fig. 3b and the reduction type used is centroid

Fig. 3. Membership function of input of the fuzzy controllers

DTC-DPC of Induction Machine Connected to Wind Generator

85

DC voltage fuzzy controller used in DPC control has five T2 fuzzy sets intervals (both for input and output) as shown in (Fig. 3a): Negative Grand (NG), Negative Petit (NP), Zero (Z), Positive Petit (PP), Positive Grand (PG), described by Gaussian membership functions uniformly distributed on the discourse universe [−1, +1]. The input and the output present respectively error of the DC Voltage (eU = Udc* − Udc) and the output is presented by DC current reference as shown Fig. 1. [11]. Table 1. Parameters of PMSG Power Stator resistance Rotor resistance Inertia Friction Stator inductance Rotor inductance Mutual Inductance Poles

3.5 Kw 4.85 O 3.805 O 0.031 kg.m2 0.001136 0.274 H 0.274 H 0.258 H 2

Table 2. Induction motor parameters Power Poles Stator resistance Stator inductance Magnet flux linkage Rated speed

6 Kw 5 0.425 X 8.4 mH 0433 Wb 153 rad=s

3 Simulation Results In this section, All Simulations were carried out for an electric machine in MATLABSimulink environment, using the induction motor parameters can be found in Table 2. The nominal parameters of the generator used in the conversion chain are indicated in Table 1. The time is given in seconds on the horizontal axis of all figures with wind speed 12 m/s. Figure 4 presents the performance of dynamic response between the actual and reference of the DC link voltages, the time needed to reach the steady stated is nearly equal to 0.25 s; presents the active and reactive power (in kW), and the zoom of the Voltage and current V and I are close to the shape sinusoidal; the same figure illustrates the stator flux in the complex plane, its trajectory is almost circular and starts at point (0,0) and rotates counterclockwise in a circle of radius 1.1 Wb fixed by the reference, the figure visualizes the speed curve of the induction machine which reaches its value 126 rad/sec of the steady state with a load torque applied at 1 s. The curve indicates the robustness of the PI controller used in the pursuit of the reference value with an acceptable overshoot time. According to same figure, the torque follows the value of the reference reflecting the effectiveness of the control technique used. The Fig. 5 shows the effectiveness of the proposed fuzzy controller used to replace hysteresis comparators, it has seen clearly the minimization of the high torque and flux ripples, the Fig. 5 presents the power (P and Q) in kW, the current and voltage responses, by using fuzzy logic type-2 controller in DPC to replace PI DC voltage, the THD of supplied current has reduced from 5.82 to 4.8% (Fig. 6).

86

H. Fethia and A. Amel

Fig. 4. DTC-DPC-Wind system reponses

Fig. 5. DTC based on FLC type-2 in open loop

Fig. 6. DPC based on FLC type-2

DTC-DPC of Induction Machine Connected to Wind Generator

87

4 Conclusion In this paper the control structures of DPC-DTC system supplied by wind turbine have been presented. The control algorithm of induction motor is based on Direct Torque Control. To control DC voltage the Direct Power Control method has been applied. DPC as DTC method provide robust performance and makes torque and power quality improvement. It minimize the transient behavior of the system and reduces the complexity of control system with no need current controller, coordinate transformation, power and flux controller. To improve the performance od DTC and DPC, a fuzzy logic controllers have been proposed to replace traditionl controllers. The presented simulation results confirmed that the Fuzzy logic type-2 offers a very satisfied results compared to classical one; and the considered hybrid system with both techniques (DTC-DPC) supplied by wind turbine has good performances and good control properties.

References 1. Cipriano, D., Rengifo, J., Aller, J.M.: Transient stability evaluation of high penetration of DFIG controlled by DTC and DPC into power systems. In: IEEE Third Ecuador technical Chapter Meeting, pp. 282–291 (2018) 2. Dubey, M., Saxena, R., Sharma, S.: Direct torque and power controlled PMSM drive. In: Proceedings of the IEEE 7th Power India International Conference (PIICON) (2016) 3. Hamidia, F., Abbadi, A., Benbouabdllah, O., Chiba, Y.: Direct torque controlled doubly fed induction motor supplied by WG and based on ANN. In: Proceedings of the 3rd International Conference on Artificial Intelligence in Renewable Energetic System, ICAIRES (2019) 4. Savarkar, V., Dyanamina, G., Singh, S.B.: MATLAB simulation of DPC-SVM of DFIG based on wind energy system, In: Proceedings of ICAEEC-2019, IIIT Allahabad, India, 31st May–1st June (2019) 5. Cirrincione, D.M., Pucci, M., Vitale, G.: Power Converters and AC Electrical Drives with Linear Neural Networks. CRC Press, Taylor & Francis Group (2012) 6. Melkebeek, J.A.: Electrical machines and drives: fundamentals and advanced modelling 7. Hamidia, F., Abbadi, A., Boucherit, M.S.: PV/battery water pumping system based on firefly optimization algorithm. In: 2rd International Conference on Artificial Intelligence in Renewable Energetic System, IC-AIRES (2018) 8. ABB Drives - Technical Guide Book, No. 1, Direct Torque Control, pp. 07–32 (2014) 9. Hamidia, F., Abbadi, A., Boucherit, M.S.: Maximum power point tracking control of photovoltaic generation based on fuzzy logic. In: Proceedings of the 1st International Conference on Artificial Intelligence in Renewable Energetic System, IC-AIRES (2017) 10. Hamidia, F., Abbadi, A., Boucherit, M.S.: Wind water pumping system based on ANN. In: Proceedings of the 6th International Conference on Control Engineering & Information Technology (CEIT 2018), Environmental Science (2018) 11. Hamidia, F., Abbadi, A., Tlemçani, A.: Improved pumping system supplied by double photovoltaic panel. Rev. Roum. Sci. Techn.–Électrotechn. et Énerg. 64(1), 87–93 (2019)

Detecting Partial Shading in Grid-Connected PV Station Using Random Forest Classifier Abderrezzaq Ziane1(&), Rachid Dabou1, Noredine Sahouane1, Ammar Necaibia1, Mohammed Mostefaoui2, Ahmed Bouraiou1, and Abdeljalil Slimani1 1

2

Unité de Recherche en Energie Renouvelables en Milieu Saharien, URERMS, Centre de Développement des Energies Renouvelables, CDER, 01000 Adrar, Algeria [email protected] École Supérieure en Informatique de Sidi Bel Abbès, Sidi Bel Abbes, Algeria

Abstract. The data-driven fault detection techniques particularly artificial intelligent ones have many advantages over model-based methods is that not much information about system parameters is needed. In this work, a datadriven method based on machine learning random forest technique was proposed to instantaneous detecting and diagnosing a partial shading fault in a gridconnected PV system in real-time, a PV system installed in the desert area of Adrar, Algeria was used as a case study. The feasibility of the tree-based ensemble method (random forest) in detecting and diagnosing a partial shading fault in a grid-connected PV system was assured with high performance, the error was recorded with less than 1.3%. Keywords: Fault detection
Shading
PV system
Grid-connected
Machine learning
Random forest

1 Introduction Nowadays, there is a global trend to the utilization of renewable energy especially photovoltaics due to several geopolitical, economical, and environmental reasons [1]. The PV station behavior can be influenced by numerous factors and such as different PV technology [2], different tilt angle of PV array [3], accumulation of dust on the photovoltaic panels [4–6], partial shading on PV module [7, 8] and variation of meteorological conditions [9, 10]. Those parameters effects can be classified into two categories; system losses and system faults. The development of monitoring and diagnostic tools is necessary to ensure the healthy performance of the PV system and detecting the fault in real-time to be handled as soon as possible. Several studies[11–13] attempts to develop their techniques to detect and diagnose the system fault based on analytical modeling and comparing the simulation results to the actual measurements. The data-driven fault detection particularly artificial intelligent techniques have many advantages over model-based methods is that not much information about system parameters is needed [14]. Several researchers attempted to use machine learning for © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 88–95, 2021. https://doi.org/10.1007/978-3-030-63846-7_10

Detecting Partial Shading in Grid-Connected PV Station

89

fault detection in photovoltaic systems, Garoudja et al. [15] used a model based on a probabilistic neural network (PNN) classifier to detect and diagnose DC side anomalies. A local PV shading detection based on AC power and regional irradiance data and SVM was developed by Bognár et al. [16]. In this work, we proposed a different approach the studies mentioned above, using a data-driven method and machine learning classification namely the ensemble treebased random forest alongside an experimental study of shading effect and the healthy performance of the 7 kWp photovoltaic station connected to the grid installed in the desert region of Adrar, Algeria. The ultimate goal is online real-time fault detection and diagnosing partial shading.

2 Materials and Methods 2.1

Experimental Setup

The experimental measurement had been carried out on a grid-connected PV station located in Renewable Energy Research Unit in the Saharan Region (URER.MS), in Adrar, Southern Algeria. The photovoltaic station of 7 kWp has 28 solar modules Mono-Crystalline (BJP250SA with a power of 250 W), and it is apart from a larger PV system described in earlier work [17]. The solar modules are mounted on metal frames supported by fixed concrete pillars at a 27° angle facing the south. The system schematic is presented in Fig. 1.

Fig. 1. Schematic of grid-tied PV station

90

A. Ziane et al.

The SMA Sunny WebBox system measures the solar irradiation, PV module temperature, DC current, DC voltage periodically. The measured data is stored in an SD card with 5 min intervals. The training data were collected during a week-long experience with shading effect (faulty performance) and without shading (healthy performance) as shown in Fig. 2.

Fig. 2. Shading experience in grid-tied PV station

2.2

Classification Method

The forests of decision trees (random forest classifier) were first proposed by Ho [18– 20] and were formally proposed by Leo Breiman [21, 22]. They are part of ensemble tree techniques. The basis of the calculation is based on learning by a decision tree. Breiman’s proposal aims to correct several known drawbacks of the initial decision tree method, such as the sensitivity of single trees to the order of predictors, by calculating a set of partially independent trees as shown in Fig. 3. A quick presentation of the proposal can be expressed as follows: Step 1 – Create a new learning sets by a double sampling process: the new observations identical to that of the original data (a technique known as bootstrap). Step 2 – On each sample, this algorithm will construct and train the decision tree, limiting its growth by cross-validation. Step 3 – Storing the predictions of the variable of interest for each original observation. Step 4 – Predicting the random forest is then a simple majority vote (Ensemble learning).

Detecting Partial Shading in Grid-Connected PV Station

91

Fig. 3. Random forest algorithm

The main downside of this method is that you lose the visual aspect of single decision trees. The implementation of a random forest classifier included in the scikit-learn module of python programming language was used for all developmental and experimental work [23]. The work was carried out on a personal computer (Intel Core i5 2.50 GHz with 12 GB of RAM).

3 Resultats and Discussion A six days’ experience was performed in May 2018 during and which the system was healthy for four days and faulty (one PV panel shaded) for two days, the system recorded four parameters irradiance, module temperature, DC current, DC voltage, those data were considered as the training data and shading parameter (1 for shading and 0 for healthy) is the target data for supervised learning, which in our case is the random forest classifier. The training data irradiance, module temperature, DC current, DC voltage, and shading are represented in Fig. 4. The visual examination of the record data cannot be sufficient to determine the presence of shading instantaneously for every point.

92

A. Ziane et al.

Fig. 4. Training data for the healthy and shaded system

The DC voltage of the shaded system is lower than of healthy one overall, but not each value of the shaded voltage lower than the healthy one. Therefore, we have to train a random forest model with all measured data. The trained data was divided randomly into sub-dataset one for training (80%) and the other one (20%) for scoring the model to tune the model parameters. The score of the model was recorded 0.9698 for these random forest classifier parameters: number of trees = 100, maximum depth of the decision trees = 16, number of random splits per node = 2 and minimum number of samples per leaf node = 1.

Detecting Partial Shading in Grid-Connected PV Station

93

After the training and the scoring of the random forest model, we applied the final trained model on another dataset containing Irradiance, module temperature, DC current, and DC voltage for one day in the same month, with a healthy system from sunrise to 1:00 pm then a faulty system with one shaded PV panel from 1:05 pm to sunset. The decision of the model is compared to the actual case as shown in Fig. 5.

Fig. 5. Test data for the trained model

94

A. Ziane et al.

The obtained results of the detection of shading using the trained random forest model show that the model is accurate 98% with an only error of 1.29%, the model failed to determine that the system is healthy for lower values of irradiance. The final python module based on the trained model was used to detect shading in real-time alongside another monitoring software developed by our team. To solve the problem of fault detection in low irradiance cases, we proposed training different models for each irradiance range using tree-based ensemble methods.

4 Conclusion In this work, a tree ensemble method based on the random forest classifier technique was proposed to instantaneous detecting and diagnosing a partial shading fault in a grid-connected PV system in real-time, a PV system installed in a desert area in Adrar, Algeria was used as a case study. The feasibility of the proposed model in detecting and diagnosing a partial shading fault in a grid-connected PV system was assured with high performance, the error was recorded with less than 1.3%. Nevertheless, the model can be improved in the future by proposed training different models for each irradiance range using tree-based ensemble methods.

References 1. Bouraiou, A., et al.: Status of renewable energy potential and utilization in Algeria. J. Cleaner Prod. 246, 119011 (2020) 2. Edalati, S., Ameri, M., Iranmanesh, M.: Comparative performance investigation of monoand poly-crystalline silicon photovoltaic modules for use in grid-connected photovoltaic systems in dry climates. Appl. Energy 160, 255–265 (2015) 3. Saber, E.M., Lee, S.E., Manthapuri, S., Yi, W., Deb, C.: PV (photovoltaics) performance evaluation and simulation-based energy yield prediction for tropical buildings. Energy 71, 588–595 (2014) 4. Abderrezzaq, Z., Mohammed, M., Ammar, N., Nordine, S., Rachid, D., Ahmed, B.: Impact of dust accumulation on PV panel performance in the Saharan region. In: Proceedings of the 18th International Conference on Sciences and Techniques of Automatic Control and Computer Engineering (STA 2017), pp. 471–475 (2017) 5. Mostefaoui, M., Ziane, A., Bouraiou, A., Khelifi, S.: Effect of sand dust accumulation on photovoltaic performance in the Saharan environment: southern Algeria (Adrar). Environ. Sci. Pollut. Res. 26(1), 259–268 (2019) 6. Mostefaoui, M., et al.: Importance cleaning of PV modules for grid-connected PV systems in a desert environment. In: Proceedings of the 4th International Conference on Optimization and Applications (ICOA 2018), pp. 1–6 (2018) 7. Dabou, R., et al.: Impact of partial shading and PV array power on the performance of grid connected PV station. In: Proceedings of the 18th International Conference on Sciences and Techniques of Automatic Control and Computer Engineering (STA 2017), pp. 476–481 (2017) 8. Silvestre, S., Chouder, A.: Effects of shadowing on photovoltaic module performance. Prog. Photovoltaics Res. Appl. 16(2), 141–149 (2008)

Detecting Partial Shading in Grid-Connected PV Station

95

9. Dabou, R., et al.: Monitoring and performance analysis of grid connected photovoltaic under different climatic conditions in south Algeria. Energy Convers. Manage. 130, 200–206 (2016) 10. Abderrezzaq, Z., et al.: Performance analysis of a grid connected photovoltaic station in the region of Adrar. In: Proceedings of the 5th International Conference on Electrical Engineering - Boumerdes (ICEE-B 2017), vol. 2017–Janua, pp. 1–6 (2017) 11. Silvestre, S., Chouder, A., Karatepe, E.: Automatic fault detection in grid connected PV systems. Sol. Energy 94, 119–127 (2013) 12. Chouder, A., Silvestre, S.: Automatic supervision and fault detection of PV systems based on power losses analysis. Energy Convers. Manag. 51(10), 1929–1937 (2010) 13. Chaibi, Y., Malvoni, M., Chouder, A., Boussetta, M., Salhi, M.: Simple and efficient approach to detect and diagnose electrical faults and partial shading in photovoltaic systems. Energy Convers. Manage. 196, 330–343 (2019) 14. Ziane, A., Necaibia, A., Mostfaoui, M., Bouraiou, A., Sahouane, N., Dabou, R.: A fuzzy logic MPPT for three-phase grid-connected PV inverter. In: Proceedings of the 20th International Middle East Power Systems Conference, MEPCON 2018, pp. 383–388 (2019) 15. Garoudja, E., Chouder, A., Kara, K., Silvestre, S.: An enhanced machine learning based approach for failures detection and diagnosis of PV systems. Energy Convers. Manage. 151, 496–513 (2017) 16. Bognár, A., Loonen, R.C.G.M., Valckenborg, R.M.E., Hensen, J.L.M.: An unsupervised method for identifying local PV shading based on AC power and regional irradiance data. Sol. Energy 174(2018), 1068–1077 (2018) 17. Sahouane, N., et al.: Energy and economic efficiency performance assessment of a 28 kWp photovoltaic grid-connected system under desertic weather conditions in Algerian Sahara. Renewable Energy 143, 1318–1330 (2019) 18. Ho, T.K.: Random decision forests. In Proceedings of 3rd International Conference on Document Analysis and Recognition, vol. 1, pp. 278–282 (1995) 19. Ho, T.K.: Random decision forests perceptron training. In: AT&T Bell Laboratories (1995) 20. Ho, T.K.: The random subspace method for constructing decision forests. IEEE Trans. Pattern Anal. Mach. Intell. 20(8), 832–844 (1998) 21. Breiman, L.: Random forests. Mach. Learn. 45(1), 5–32 (2001) 22. Breiman, L.: Bagging predictors. Mach. Learn. 24(2), 123–140 (1996) 23. Varoquaux, G., et al.: Scikit-learn: machine learning in python. J. Mach. Learn. Res. 12(1), 2825–2830 (2011)

Network Reconfiguration Management in Intelligent Distribution System Taking into Account PV Production Variation Using Grey Wolf Optimizer Mustafa Mosbah1(&), Rabie Zine2, Mustapha Hatti3, Samir Hamid-Oudjana4, and Salem Arif1 1

4

LACoSERE, Electrical Engineering Department, University of Amar Telidji, Laghouat, Algeria [email protected] 2 School of Science and Engineering, Al Akhawayn University in Ifrane, Ifrane, Morocco 3 Unité-de Développement des Equipements Solaires, Bou Ismaïl, Tipaza, Algeria Unité de Recherche Appliquée en Energies Renouvelables, Ghardaia, Algeria

Abstract. The management of modern distribution networks or smart grids needs the integration of new technologies and advanced software. This paper proposes the Grey Wolf Optimizer or GWO technique to determine the optimal configuration of the network in real-time in the presence of photovoltaic sources. This technique is able to establish the state of the looping switches during each hour and this according to the variation of the power load and the photovoltaic power produced. The goal function chosen is to minimize the active losses under the imposed constraints. This study was applied on a practical network, namely the Algerian distribution network under MATLAB code. The combined results showed the efficiency and robustness of the proposed technique. Keywords: Reconfiguration management
Grey wolf optimizer
Photovoltaic source

1 Introduction With the increase in the insertion of intermittent sources (photovoltaic, wind power) in electrical distribution system, and the high variation in load during the day, the currents flowing through the lines have changed significantly [1]. For this reason, it is necessary to integrate new technologies (hardware and software) in order to operate the grid with the best performance [2]. The Algerian distribution networks have undergone major changes to fat of starting the operation of modernization of networks, in order to have an automated network or smart grid [3]. Such as the installation of intelligent protection systems, fiber optic links between the different source substations, smart switches installation, fault detectors, photovoltaic sources, the smart meters installation, SCADA implementation and automatic voltage regulators [4]. The network operator first © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 96–104, 2021. https://doi.org/10.1007/978-3-030-63846-7_11

Network Reconfiguration Management in Intelligent Distribution

97

concern is to minimize losses on distribution lines as much as possible [5]. For this, the most interesting task in the dynamic management system (DMS) is the Network Reconfiguration Management (NRM) [6]. NRM consists of reconfiguring the network topology in order to improve the efficiency of the electrical distribution power system by changing the state of the loop switches (closing or opening the switch) [7]. It is important to note that the distribution power networks are constructed with a loop structure; however, they are operated with a radial structure [8]. The challenge is to have structures in real-time with minimum power losses by considering the photovoltaic production variations and load variations, while observing the technical, topological and security constraints [8]. With the development of technology the dynamic reconfiguration of the distribution network has become an easy task to implement [9]. The literature review has shown that static reconfiguration is largely discussed [10–14]. However, the dynamic reconfiguration has fewer papers. To this purpose, it is important to think about the problem of the Dynamic DistributionNetwork Reconfiguration (DD-NR) [15–23]. This work presents the study of the problem of searching the optimal distribution network reconfiguration, which gives the minimum of active losses, for each time instant. This reconfiguration changes according to the photovoltaic production and the load, because the losses of the lines vary according to the currents transited through them. The objective is to determine the best configuration of the network for each instant, by modifying the states of the loop switches with respect to the physical and operational constraints using the GWO technique. This technique offers a random permutation of the lines between the different loop points, and depending on the considered constraints the best configuration will be selected. The proposed technique is applied on Algerian network 116-bus.

2 Problem Formulation 2.1

Fitness Function

The currents flowing through the distribution lines are consistently high. For this reason, network operators have continuously aimed to minimize active loss, mathematically it is given by [22]: 2 Rpq ¼ PTloss ðtÞ ¼ Ipq

S2pq

P2pq þ Q2pq

Vp

Vp2

R ¼ 2 pq

Rpq

ð1Þ

where Vp , Ipq , Rpq , Ppq and Qpq , are voltage, current, resistance, active and reactive power of the line between bus p and q respectively.

98

M. Mosbah et al.

2.2

System Constraints

Among the constraints to be respected in the optimization process is the balance at all times between the power generated and the power consumed, which is presented by the equation: SG þ SPV ¼ SD þ SL

ð2Þ

where SG , SPV , SD and SL are apparent power of centralized generator, apparent power of photovoltaic source, apparent power of load and apparent power of losses respectively. The equations below represent the technical and security limits of the various structures of the network.

0

PG1min  PG1  PG1max

ð3Þ

Sk  Skmax for k ¼ 1. . .. . .:NB

ð4Þ

0:95  Vi  1:05 for i ¼ 1. . .. . .N

ð5Þ

NPV X i¼1

PPVi  0:3 

Nbus X

PDi for i ¼ 1. . .. . .NPV

ð6Þ

i¼1

In addition to the constraints mentioned above, there are two conditions related to the topology of the distribution network: The first requires that the obtained configuration does not contain closed loops. The second requires that the resulting configuration does not contain isolated nodes. When looking for an optimal configuration it is necessary to apply graph theory to ensure topological constraints [23].

3 Grey Wolf Optimizer Method In 2014, S. Mirjalili et al., proposed a ranked optimization algorithm with population based metahistoricals (Grey Wolf Optimizer) [24]. The main inspiration of this algorithm is the social direction of the Grey Wolf hunting technique (see Fig. 1). The process is started by a random set of candidate solutions (Wolfs). During each iteration, the three best wolves are considered as a, b and d which guide the wolf x towards a promising area of a search space, The hunting behaviour of Grey Wolfs: (A) pursuing, approaching and following a prey (B/D) chasing, harassing and encircling a stationary prey (E) attacking.

Network Reconfiguration Management in Intelligent Distribution

99

Fig. 1 Grey wolf optimizer strategy

4 Analyzes Results In this study, the GWO-based metaheristic method was used to determine the optimal configuration in each time interval by taking into account the intermittency of the load that is imposed by the consumers, and the intermittency of the power produced by PV source is imposed by the meteorological conditions. Before beginning the application on the Algerian network, a test was carried on the IEEE 69 bus network to make sure that the code developed under Matlab is suitable, through comparisons with works in the literature. The initial configuration of the 69 network is identified by the opening of the following switches: N° 69, 70, 72, 73 [25]. The 116 bus network represents a real power system feeding an urban area with a voltage of 30 kV to 10 kV, this network is distributed in nine feeders. Four PV sources have been considered in this study, which are injected at different points, such as node 109, 106, 66, and 62. The total power generated by all the PV is 9980 kW, which is less than 10 MW according to the grid code specifications of the Algerian electricity company. The initial configuration of the network 69 is identified by opening switches No. 116 to No. 124 [23]. In order to evaluate the dynamic characteristics of the Algerian grid it is essential to have instantaneous data of the load and the power generated by the PV sources. For this purpose, recordings were made on 12/07/2017 to determine the daily load curve as well as the PV production curves (Figs. 3 and 4). The date chosen represents the day with the maximum power reached during the year 2017. It is important to note that the PV source and the 116bus network are located on the same locality.

100

M. Mosbah et al.

The calculation process carried out in this study to determine the best configuration in each hour follows the following steps: calculate the active losses for the initial configuration. Randomly select a network configuration by the GWO method. Check the topological constraints by applying graph theory. If these constraints are satisfied, and calculate the power flow by Newton Raphson method, then check the constraints of the voltage and thermal limits of the lines. If not, penalize the objective function to eliminate the unfeasible solutions. Metaheuristic methods need a good choice of parameters, which requires several attempts to reach the best values. In this study, the best values of the GWO method parameters used population size is 120, number of generations is 1000.

Fig. 2 The daily load curve

Fig. 3 PV production curve

It is important to indicate that in the simulation of the IEEE network it is considered a moment of time and in the absence of PV. In other hand, the configuration is determined as static, because it is the same process repeated 24 h. This simulation is made initially to check the developed code. After the convergence of the GWO method, the results are shown in the tables and figures below. The application of the proposed method on the 69 network shows the efficiency and reliability of the GWO technique compared to the work of the literature (see Table 1). After this test phase, a validation on the Algerian network shows the success of the GWO method towards the determination of the dynamic configuration with minimum active losses (see Table 2). The dynamic optimization proves the improvement of the voltage profile throughout the day (see Fig. 4). The same applies to the improvement of the losses in each hour (see Fig. 5). Also a clear improvement of the grid parameters during PV operation or during the presence of radiation. Table 2 shows the dynamically varied active losses weighting the day, confirming the need to optimize the configuration in an instantaneous way.

Network Reconfiguration Management in Intelligent Distribution

101

Table 1. Comparisons results with works in bibliography Optimization methods GAM in [26] PCGAM in [27] FGAM in [28] MHB-MO in [29] BBOM in [30] GAM in [22] GWO method

Switches to be opened 9/28/33/34/36 no reported by paper 12/55/61/69/70 7/9/14/32/37 14/70/69/58/61 14/70/69/58/61 14/55/61/69/70

Real losses (kW) 140.600 100.950 99.620 139.510 99.580 99.580 99.580

Fig. 4 Minimum voltage for every hour

Fig. 5 Minimum total losses for every hour

According to Figs. 4 and 5, the improvement of the voltage profile and active losses is greatest between 08:00 and 18:00, as this is the period of operation of the PV sources. Other periods may be less affected by load variation. According to the results obtained, the losses are minimized at a rate of 15%, and consequently the voltage profile is improved. Another observation that has been made with regard to the

102

M. Mosbah et al.

improvement of losses and voltages is that the higher the PV generation, the better the grid parameters (voltage loss) (or even the hour at 1 p.m.). The total losses of the day decreased from 5206.41 kW to 4629.96 kW. Where has a minimization rate of 11.07%. This reduction does not cost anything from an investment point of view, all that is required is to open or close the loop switches. The task of dynamic reconfiguration of the network has become indispensable in the face of technological modernization and development. It is important to point out that the Algerian distribution networks have been equipped with new technologies that facilitate the integration of the DDNR task.

Table 2. DDNR determined by GWO method in presence PV sources Hours Switches to be opened

Before Optimization

After Optimization

3:00

103 75 79 89

19 121 68

Vmin (pu) Losses (kW) Vmin (pu) Losses (kW) 60 107 0.9814 153.7 0.9841 141.1

4:00 5:00 6:00 7:00 8:00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:00 20:00 21:00 22:00 23:00 0:00 1:00 2:00

103 103 99 103 103 88 88 88 88 88 88 88 103 103 103 99 103 103 103 103 103 103 103

19 19 19 19 29 29 29 29 29 29 29 29 29 19 19 19 19 19 19 19 19 19 19

60 60 60 60 10 69 69 10 70 70 70 69 13 10 98 60 60 60 60 60 60 60 60

75 75 75 75 75 75 75 75 75 75 75 75 75 75 75 75 75 75 75 75 75 75 75

79 79 79 79 79 79 79 79 79 79 79 79 79 79 79 79 79 79 79 79 79 79 79

89 89 105 105 105 105 105 105 105 105 105 105 105 105 105 105 89 89 89 89 89 89 89

121 121 121 121 121 121 121 121 121 121 121 121 121 121 121 121 121 121 121 121 121 121 121

68 68 68 83 83 122 43 43 43 43 122 122 122 83 83 68 68 68 68 68 68 68 68

107 107 107 107 107 124 67 67 67 67 124 124 124 107 107 107 107 107 107 107 107 107 107

0.9817 0.9805 0.9818 0.9842 0.9856 0.9849 0.9833 0.9829 0.9820 0.9799 0.9774 0.9778 0.9785 0.9790 0.9794 0.9754 0.9734 0.9734 0.9737 0.9746 0.9761 0.9778 0.9793

148.9 168.7 151.3 139.9 113.3 120.5 154.8 176.4 203.4 235.2 274.1 256.7 238.8 244.6 260.5 283.3 313.7 313.7 306.8 286.4 254.1 217.9 189.8

0.9844 0.9834 0.9843 0.9858 0.9897 0.9914 0.9909 0.9891 0.9913 0.9903 0.9875 0.9875 0.9871 0.9846 0.9813 0.9785 0.9773 0.9773 0.9776 0.9783 0.9796 0.9811 0.9824

136.6 154.8 139 129.6 104.1 103.7 124.4 137.7 156.2 186.4 229.1 219.8 212.7 224.8 241.6 260.7 287.9 287.9 281.5 262.9 233.2 199.9 174.2

5 Conclusion In this paper, a metaheuristic method based on GWO has been proposed to solve the NMSD problem in the presence of PV sources with Matlab code. NMSD takes into account the instantaneous power of the PV loads and sources. A test was done on the IEEE 69 network to compare the results with the literature, followed by a validation on

Network Reconfiguration Management in Intelligent Distribution

103

the real Algerian network. The proposed method shows its robustness and reliability from a technical and economical viewpoint. The future work of this study is to propose to integrate other types of renewable sources and to apply more recent optimization techniques under industry codes.

References 1. Shefaei, A., Vahid-Pakdel, M., Mohammadi-ivatloo, B.J.C.: Application of a hybrid evolutionary algorithm on reactive power compensation problem of distribution network. Comput. Electr. Eng. 72, 125–136 (2018) 2. Dixit, M., Kundu, P., Jariwala, H.R.J.C.: Optimal integration of shunt capacitor banks in distribution networks for assessment of techno-economic asset. Comput. Electr. Eng. 71, 331–345 (2018) 3. Merlin, A.: Search for a minimal-loss operating spanning tree configuration for an urban power distribution system. Proc. of 5th PSCC 1, 1–18 (1975) 4. Shirmohammadi, D., Hong, H.W.: Reconfiguration of electric distribution networks for resistive line losses reduction. IEEE Trans. Power Delivery 4(2), 1492–1498 (1989) 5. de Assis, L.S., et al.: Switch allocation problems in power distribution systems. IEEE Trans. Power Syst. 30(1), 246–253 (2015) 6. Kumar, R., Maaß, H., Hagenmeyer, V.J.C.: Comparison of lossless compression schemes for high rate electrical grid time series for smart grid monitoring and analysis. Comput. Electr. Eng. 71, 465–476 (2018) 7. Ding, F., Loparo, K.A.: Hierarchical decentralized network reconfiguration for smart distribution systems—Part I: Problem formulation and algorithm development. IEEE Trans. Power Syst. 30(2), 734–743 (2015) 8. Zidan, A., El-Saadany, E.F.: Distribution system reconfiguration for energy loss reduction considering the variability of load and local renewable generation. Energy 59, 698–707 (2013) 9. Chao-Shun, C., Chia-Hung, L., Hui-Jen, C., Chung-Sheng, L., Ming-Yang, H., Chia-Wen, H.: Optimal placement of line switches for distribution automation systems using immune algorithm. IEEE Trans Power Syst 21(3), 1209–1217 (2006) 10. Mosbah, M., Arif, S., Mohammedi, R.D., Oudjana, S.H.: A genetic algorithm method for optimal distribution reconfiguration considering photovoltaic based dg source in smart grid. In: Hatti, M. (eds) Renewable Energy for Smart and Sustainable Cities. ICAIRES 2018. Lecture Notes in Networks and Systems, vol 62, pp. 162–170. Springer, Cham (2019) 11. Mohammedi, R.D., Zine, R., Mosbah, M., Arif, S.: Optimum network reconfiguration using Grey Wolf Optimizer. 2428–2435 (2018) 12. Mosbah, R., et all.: Optimum-distribution network reconfiguration in presence DG unit using BBO algorithm. J. Electr. Syst. 14(4), 180–189 (2018) 13. Mosbah, M., Arif, S., Mohammedi, R.D., Oudjana, S.H.: A genetic algorithm method for optimal distribution reconfiguration considering photovoltaic based DG source in smart grid. In: Hatti, M. (eds) Renewable Energy for Smart and Sustainable Cities. ICAIRES 2018. Lecture Notes in Networks and Systems, vol 62, pp. 62–170. Springer, Cham (2019) 14. Mosbah, M., Arif, S., Mohammedi, R.D., Zine, R.: Optimal Reconfiguration of an Algerian Distribution Network in Presence of a Wind Turbine Using Genetic Algorithm. In: Hatti, M. (eds.) Artificial Intelligence in Renewable Energetic Systems. ICAIRES 2017. Lecture Notes in Networks and Systems, vol 35, 392–400. Springer, Cham (2018)

104

M. Mosbah et al.

15. Masteri, K., Venkatesh, B.: Real-time smart distribution system reconfiguration using complementarity. Electr. Power Syst. Res. 134, 97–104 (2016) 16. Souza, S.S., Romero, R., Pereira, J., Saraiva, J.T.: Artificial immune algorithm applied to distribution system reconfiguration with variable demand. Int. J. Electr. Power Energy Syst. 82, 561–568 (2016) 17. Wen, J., Tan, Y., Jiang, L., Lei, K.: Dynamic reconfiguration of distribution networks considering the real-time topology variation. IET Gener. Transm.Distrib. 12(7), 1509–1517 (2018) 18. Kovački, N.V., Vidović, P.M., Sarić, A.T.: Scalable algorithm for the dynamic reconfiguration of the distribution network using the Lagrange relaxation approach. Int. J. Electr. Power Energy Syst. 94, 188–202 (2018) 19. Hamida, I.B., Salah, S.B., Msahli, F., Mimouni, M.F.: Optimal network reconfiguration and renewable DG integration considering time sequence variation in load and DGs. Renewable Energy 121, 66–80 (2018) 20. Bineeta, M., Debapriya, D.: Multi-objective dynamic and static reconfiguration with optimized allocation of PV-DG and battery energy storage system. Renew. Sustain. Energy Rev. 124, 109777 (2020) 21. Bernardon, D.P., Mello, A.P.C., Pfitscher, L.L., Canha, L.N., Abaide, A.R., Ferreira, A.A. B.: Real-time reconfiguration of distribution network with distributed generation. Elec Power Syst Res 107, 59–67 (2014) 22. Hamid-Oudjana, S., Mosbah, M., Zine, R., Arif, S.: Optimum Dynamic Network Reconfiguration in Smart Grid Considering Photovoltaic Source. In: Hatti, M. (eds.) Smart Energy Empowerment in Smart and Resilient Cities. ICAIRES 2019. Lecture Notes in Networks and Systems, vol 102, pp. 557–565 Springer, Cham (2020) 23. Mosbah, M. et al.: Optimum dynamic distribution network reconfiguration using minimum spanning tree algorithm. In: 5th International Conference on Electrical Engineering, IEEE Proceeding, Boumerdes, 28–31 Oct (2017) 24. Mirjalili, S., Mirjalili, S.M., Lewis, A.: “Grey Wolf Optimizer” Advances in Engineering Software, pp. 46–61 (2014) 25. Baran, M.E., Wu, F.F.: Optimal capacitor placement on radial distribution systems. IEEE Trans. Power Delivery 4(1), 725–734 (1989) 26. Hong, Y.-Y., Ho, S.-Y.: Determination of network configuration considering multiobjective in distribution systems using genetic algorithms. IEEE Trans. Power Syst. 20(2), 1062–1069 (2005) 27. Qin, Y., Wang, J., Gui, W.: Particle clonal genetic algorithm using sequence coding for solving distribution network reconfiguration. In: The 9th International Conference for Young Computer Scientists ICYCS 2008, pp. 1807–1812 (2008) 28. Liu, L., Chen, X.: Distribution network reconfiguration based on fuzzy genetic algorithm, pp. 66–69 (2000) 29. Niknam, T.: An efficient multi-objective HBMO algorithm for distribution feeder reconfiguratio. Expert Syst. Appl. 38(3), 2878–2887 (2011) 30. Kouzou, A., Mohammedi, R.D., Hellal, A.: An efficient biogeography-based optimization algorithm for smart radial distribution power system reconfiguration. In: 2015 First Workshop on Smart Grid and Renewable Energy (SGRE), pp 1–7 (2015)

An Enhanced MPPT Method Combining Fractional-Order and Fuzzy Logic PID Controller for a Photovoltaic-Wire Feeder System (PV-WFS) N. Hamouda1, B. Babes1(&), A. Boutaghane1, S. Kahla1, and B. Talbi2 1

2

Research Center, Industrial Technologies (CRTI), Algiers, Algeria [email protected] Department of Electronics, University of Mohamed El Bachir El Ibrahimi, Bordj Bou Arréridj, Algeria [email protected]

Abstract. The use of photovoltaic (PV) module as a power source for wire feeder systems (WFSs) of arc welding machines is one of the promising domains in the solar PV applications. This paper provides a new kind of welding WFS and investigates the PV penetrated power systems. The considered system consists of a PV module, a DC-DC buck converter, and PMDC motor. The power of the PV-WFS can be widely enhanced by using a Fractional-order Fuzzy PID (FO-Fuzzy-PID) controller based P&O MPPT algorithm. In this work, a FO-Fuzzy-PID controller is also proposed for PMDC motor driven WFS. This will lead consequently to optimize the mechanical motor speed of the WFS. The dynamic response of the PV-WFS relies upon the parameters of these FO-Fuzzy-PID controllers, which are optimized by using Particle Swarm Optimization (PSO) algorithm. Simulation results found are satisfactory and prove the stability, accuracy and dynamic response of the synthesized optimized wire feeder regulating system and the proposed intelligent MPPT algorithm. Keywords: Solar photovoltaic (PV) module
Wire feeder system (WFS)
Arc welding machines
DC-DC buck converter
MPPT control
FO-Fuzzy PID controller
Particle Swarm Optimization (PSO) algorithm

1 Introduction There are many types of welding power supplies utilized for a variety of welding processes in isolated area [1]. The diesel generators are utilized for gas metal arc welding (GMAW) process, since they are simple to install, but these kinds of power supplies incorporate a diesel fueled engine attached to an electrical generator create numerous problems such as fuel cost, maintenance, noise, and short lifetime. The PVWFS presents other solution to replace the used diesel generator, because it generates electricity without damaging the environment, and they are mostly utilized due to high distribution costs/non-availability of grid-power [1]. According to [2], frequent © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 105–113, 2021. https://doi.org/10.1007/978-3-030-63846-7_12

106

N. Hamouda et al.

maintenance and repairs of diesel engine-driven welding machines are regularly (2–4) times more than PV-WFSs. Therefore, the use of PV as power source for welding machines can be considered as one of the most promising areas of solar PV application. The PV-WFS requires only a PV module with a DC-DC buck converter and a storage bank. Moreover, the proposed system works in isolated areas, which needs an efficient and reliable production system to extract the maximum power from the solar energy. The PV module coupled directly to the DC-DC switching converter can meet these requirements. In the WFS-side, a DC constant voltage power supply based on a DC-DC buck converter is adopted for a PMDC servo motor. The DC-DC buck converter responds to the requirements of the modern WFSs due to high power handling ability and a better use of the power switch. Furthermore, the problem of identifying the maximum power point (MPP) of solar PV module can be solved by using an intelligent Perturb and Observe (P&O) MPPT algorithm combining fractional-order PID and fuzzy controller. The main contribution is a maiden application has been made to tune all the possible parameters of FO-Fuzzy-PID based MPPT controller of PV module and FO-Fuzzy-PID based speed controller of PMDC motor, simultaneously with PSO algorithm to handle the uncertainties caused by the WFS and the PV generator.

2 System Description and Modeling The PV-WFS is depicted in Fig. 1. It composed of four main elements: PV module, DC-DC buck converter for MPPT, and storage bank connected to the WFS.

Fig. 1. Description of the solar photovoltaic powered wire feeder system.

2.1

Solar PV Generator Model

Figure 2 illustrates the equivalent circuit of the solar cell which comprises a singlediode (D), parallel resistor (Rp), and serial resistor (Rs). By applying kerchief’s current law, the output current of the solar cell is written by [3] IPV ¼ Iph  Is ½expð

qðVp þ Rs :Ic ÞÞ VPV þ Rs :IPV  1  KTc A Rp

ð1Þ

An Enhanced MPPT Method Combining Fractional-Order

107

Fig. 2. Solar cell model with a single-diode and two resistors.

where, Iph is the photocurrent (A), Id is the diode current (A), IPV is the output current (A), VPV is the output voltage, Is is the reverse saturation current (A), q is the charge of the electron (1.6  10−19 C), K is the Boltzmann’s constant (1.38  10 − 23 J/K), A is the ideality factor of the diode, Tc is the actual cell temperature (°C). 2.2

Group Wire Feeder System and GMAW Process Model

The graphic illustration of WFS for GMAW process application is depicted in Fig. 3. The wire feed servo motor is in itself a feedback controlled system which is capable of delivering wire of wire spool to the weld process at a controlled wire feed rate Vf.

Fig. 3. Physical representation of WFS for GMAW process application.

108

N. Hamouda et al.

The dynamic equation for the electrical circuit of GMAW process, is Voc ¼ L1

dIW þ Rs IW þ Varc dt

ð2Þ

where Voc is the open circuit voltage, Iw is the welding current, R1 is the resistance of arc welder power supply, and L1 is the inductance of arc welder power supply. The dynamic equation of arc voltage Varc is expressed as [4] Varc ¼ ka la þ kp IW þ Vc

ð3Þ

where ka, kp, Vc are parameters of arc characteristics, and larc is the arc length. The dynamic equation of arc length larc, is dla ¼ Vm  Vf dt

ð4Þ

where Vm represents the wire melting rate may be expressed as V m ¼ km I W

ð5Þ

where km indicates the coefficient of wire melting rate. The dynamic equation of the power source Voc, is 2.3

PMDC Motor Driven Wire Feeder System Model

The equivalent circuit of the wire feed servo motor is illustrated in Fig. 4.

Fig. 4. Equivalent circuit of the wire feed servo motor.

According to the Kirchhoff’s voltage equation around the armature circuit and the motion equation of the PMDC motor, the corresponding mathematical model of PMDC motor is as follows [5] dIa þ k1 x dt

ð6Þ

dx þ TLoad dt

ð7Þ

Va ¼ Ra Ia þ La Tem ¼ B x þ J

An Enhanced MPPT Method Combining Fractional-Order

109

where, Ia and Va are the phase current and voltage of the PMDC motor, Ra is the winding resistance, La is the winding inductance, J is the rotor inertia, B is the motor friction coefficient, k1 is the back EMF constant. x is the rotor speed.

3 Fractional-Order Fuzzy Logic Pid (FO-Fuzzy-Pid) Controller FO-Fuzzy-PID structure utilized in this research paper has fractional-order Fuzzy PI and Fractional-order Fuzzy PD controllers [6]. On this basis, we realized a FO-FuzzyPID controller for solar PV-WFS system and its structure is illustrated in Fig. 5

Fig. 5. Fractional-order fuzzy logic PID control block diagram.

In this figure Db is the fractional derivative and I−a is the fractional integrator orders. The control action U(t) is a nonlinear function of error E, fractional change of error DE, and fractional integral of error IE with the following model:

  db d a eðtÞ KV UðtÞ ¼ f KP eðtÞ þ KD eðtÞ þ KI dt dt

ð9Þ

4 The Optimization Problem This section presents a PSO algorithm for the optimal parameter calibration of FOFuzzy-PID controller. The PSO calibration process consists of finding the optimal FOFuzzy-PID controllers parameters that present the best possible performance for the regulation of a solar PV-WFS system. Therefore, the FO-Fuzzy-PID controller parameters represent the dimensions of each candidate solution for the optimization problem. To evaluate the performance of FO-Fuzzy-PID controller under each parameter configuration, the Integral of Time multiply Absolute Error (ITAE) criterion has been adopted based on the instantaneous error of PV module output current control loop e1(t) and instantaneous error of motor speed control loop e2(t) to find a better solution in a minimum computation time and accuracy. The quality of each candidate solution is evaluated according to the following objective function.

110

N. Hamouda et al.

Ztsim J ¼ J1 þ kJ2 ¼

Ztsim tje1 jdt þ k

0

tje2 jdt

ð10Þ

0

where k is weighting factor and tsim is the time of simulation. The performance of this objective function is evaluated according to PSO algorithm depicted by the flowchart of Fig. 6.

Fig. 6. Schematic diagram of the PSO Algorithm.

5 Control Design of the Proposed MPPT Algorithm In this section, we consider an improved P&O algorithm with adaptive increment step [7]. Basic principle of this strategy is increment step variation to converge faster towards maximal power point (MPP) while reducing oscillations around. Indeed, in order to rapidly converge, increment step ‘H’ is reduced or adapted from a region to

An Enhanced MPPT Method Combining Fractional-Order

111

another: H = 0.01 in ‘‘F’’ region and H = 0.001 in ‘‘G’’ region. The flow chart for the proposed P&O MPPT algorithm is explained in Fig. 7.

Fig. 7. Flowchart of the improved P&O MPPT algorithm.

6 Simulation Results and Discussions We have investigated three test cases in order to show the tracking capability of the proposed FO-Fuzzy-PID controller-based MPPT algorithm at different irradiance levels conditions. 6.1

Performance Test with Uniform and Constant Solar Irradiance

The results presented in these figures (Fig. 8(a–b)) illustrate the rapidity and robust effective control of the response provided by the proposed FO-fuzzy-PID controllers.

Fig. 8. FO-Fuzzy-PID and traditional PID based: (a) MPPT and (b) speed control of solar PV-WFS.

112

6.2

N. Hamouda et al.

Performance Test Under Ramp-Change in Solar Irradiance

From analysis in Fig. 9 (a–b), the proposed FO-Fuzzy-PID method has superior power and speed tracking performances than other traditional PID method.

Fig. 9. FO-Fuzzy-PID and traditional PID based: (a) MPPT and (b) speed control of solar PV-WFS.

6.3

Performance Test Under Step-Change in Solar Irradiance

These figures show that the proposed FO-Fuzzy-PID controller has the highest tracking performance than its competitor.

Fig. 10. FO-Fuzzy-PID and traditional PID based MPPT and speed control of solar PV-WFS.

7 Conclusion In this work, an advanced synthetic study of a standalone solar PV-WFS system is introduced. It includes: solar PV generator modeling, an improved P&O algorithm with an adaptive step increment, a detailed method to the modeling of WFS and PMDC motor, and a thorough PSO algorithm for calculating the optimal parameters of the FOFuzzy-PI controllers used for MPPT algorithm and wire feed speed control of solar PV module and PMDC servo motor of WFS, respectively. This work has yielded some improvement simulation results: (i) the response time presented by the proposed FOFuzzy-PID based MPPT control is almost less than that given by the traditional PID based P&O control. (ii) the control test results validate the FO-Fuzzy-PI speed

An Enhanced MPPT Method Combining Fractional-Order

113

controller robustness and tracking effectiveness for the PMDC motor drive. Finally, this article with its amount of information, its references and its synthetic aspect will be helpful for researchers and PhD students, who required a simple and effective way to model, control and simulate solar PV-WFS driven by PMDC motor.

References 1. Boussiala, B., Nezli, L., Mahmoudi, M.O., Deboucha, A.: Novel welding machine based on small PMSG wind turbine. J. Renew. Sustain. Energy 10(2018), 053304 (2018). https://doi. org/10.1063/1.5042609 2. Errouhaa, M., Derouicha, A., Nahid-Mobarakeh, B., Motahhir, S., El Ghzizal, A.: Improvement control of photovoltaic based water pumping system without energy storage. Sol. Energy 190(2019), 319–328 (2019) 3. El Khazane, J., Tissir, E.H.: Achievement of MPPT by finite time convergence sliding mode control for photovoltaic pumping system. Sol. Energy 166(2018), 13–20 (2018) 4. Hamouda, N., Babes, B., Boutaghane, A., Kahla, S., Mezaache, M.: Optimal tuning of PIkDl controller for PMDC motor speed control using ant colony optimization algorithm for enhancing robustness of WFSs. In: 020 1st International Conference on Communications, Control Systems and Signal Processing (CCSSP), EL Oued, Algeria, pp. 364–369 (2020) 5. Hamouda, N., Babes, B., Hamouda, C., Kahla, S., Ellinger, T., Petzoldt, J.: Optimal tuning of fractional order proportional-integral-derivative controller for wire feeder system using ant colony optimization. Proceedings Journal Européen des Systèmes Automatisés 53(2), 157– 166 (2020). https://doi.org/10.18280/jesa.530201 6. Al-Dhaifallah, M., Kanagaraj, N., Nisar, K.S.: Fuzzy Fractional-order pid controller for fractional model of pneumatic pressure system. Hindawi Math. Probl. Eng.2018, Article ID 5478781, 9 pages (2018). https://doi.org/10.1155/2018/5478781 7. Dhaker, A., André, M., Gérard, C., Benoit, R.: Real time supervision for a hybrid renewable power system emulator. Simul. Model. Pract. Theory 42(2014), 53–72 (2014)

MPPT Charge Regulator and Monitor for Photovoltaic/Battery System Based on Microcontroller H. Assem1,2,3(&), F. Bouchafaa2, T. Azib3, N. Allam2, and N. Belhaouas1 1

2

3

Centre de Développement Des Energies Renouvelables, CDER, Algiers, Algeria [email protected] Université Des Sciences et de La Technologie Houari Boumediene, USTHB, Algiers, Algeria [email protected] Ecole Supérieure Des Techniques Aéronautiques et de Construction Automobile, ESTACA, Paris, France [email protected]

Abstract. Many different type of systems have been designed to adapt photovoltaic (PV) generators to their charges. They ranged from basic and low efficient linear controllers, to more sophisticated power electronic based converters. Accordingly, depending on the system requires, further attention should be devoted to the development of the converter design that connects the PV system to the battery (BAT) or load, in conjunction with adding new and more advanced control strategies to the system. This paper outlines the process of designing a new electronic concept of a PV regulator for autonomous applications, which uses a buck converter to ensure high efficiency over a wide operating range. The power converter is driven by a very precise algorithm, adapted for implementation in a low-cost microcontroller (PIC184550), that has been designed to allow PV generator to track and function in their MPP including a control strategy that ensures an appropriate charging/discharging process for improved power management and prolonged life cycle of BAT. Besides, a software tool has been developed to supervise the various phases of the charging/discharging and the entire energy in the PV stand-alone system including the availability of continual telemetric monitoring of both PV and BAT units. The system has been developed and experimentally evaluated to investigate the behaviors of each unit of the isolated PV/BAT system. Keywords: Microcontroller
Maximum power point (MPP)
DC-DC converter
Battery charge control
Isolated photovoltaic systems

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 114–124, 2021. https://doi.org/10.1007/978-3-030-63846-7_13

MPPT Charge Regulator and Monitor

115

1 Introduction An important advantage of renewable energy sources (RES) is their ability to make sustainable electricity available in regions not covered by the conventional utility grid. PV Systems combined with other sources such as batteries have been for years been an excellent alternative energy source for isolated areas [1, 2]. Off-grid and micro-grid systems using RES have acquired a strategic competitive advantages in the latest five years, and this transition is predicted to be accelerated in the future [1]. Over half of the population in developing countries live in remote locations [2], where solar PV power could be a key for rural electrification. Consequently, autonomous PV systems are a promising solution for isolated regions [2, 3]. Furthermore, the cost reduction of PV panels (by 81%) and BATs in the last few years [4] has driven to an increasing interest in “living off-grid” or “leaving grid” [5]. Basically, stand-alone PV systems consist of a PV generator, a regulator, a BAT system and DC or AC loads with inverter. The regulator control the PV generator, the load, and the BAT currents to avoid overdischarge/overcharge of the BAT. BAT charge controllers are commonly classified according to the strategies used for regulating the load power from the PV to the BAT systems [6, 7]. Maximum power point tracking (MPPT) and Pulse width modulation (PWM) and are the most popular utilized regulators in the PV stand-alone industries [8]. A PWM BAT charge controllers could be categorized into shunts and series groups. PWM technique controls the duty cycle, which is the ratio of pulse length to period duration of the waveform, of the current from the PV array to the BAT system [9]. The duty cycle is related to the state of charge (SOC) of the BAT. These kinds of approaches, which are applicable to lead–acid BATs, can allow rapid recharging of the BAT and improve the life-cycle efficiency of low maintenance BATs [10]. However, between the cost of the monitoring system and the precision of the supervised data, adequate supervising of PV systems using PWM methods is a challenging task. Alternatively, the MPPT method generally uses a power electronic DC/DC converter which functions the PV generator at its MPP across the P(V) curve. Since the power of MPPT regulator is provided by DC currents and voltages, there is no difficulty in measuring them. For this reason, most of the PV off-grid systems rely on a BAT system being direct coupled to the to the PV generator via a converter using a simple power switching device. This method is regarded as the most reliable and least expensive. A large number of MPPT techniques are found in the literature [11, 12]. After the abovementioned methods are validated, they can be implemented in a multitude manners utilizing sophisticated or less complex data acquisition tools (DAQ), as well as transducers for effective and real parameter measurements. In addition, DAQ system could be replaced by microcontroller and graphical interface user, for example: a Nano USB Arduino microcontroller using SD memory card for data storage is developed in [13], a Macro Arduino with two micro SD memory cards is used in [14], or an Arduino UNO board connected to the PC in [15]. The transducers and sensors can be used with electronic circuits designed to obtain the required parameters. The duty cycle can also be achieved through a specified microcontroller. In the literature, different algorithms implemented on variety of microcontrollers to generate the duty cycle waveform, can be found [16, 17].

116

H. Assem et al.

This article proposes a regulator capable of operating in the MPP of PV generator under the variation of meteorological conditions regardless their impacts on the characteristics of the PV panels. The PV generator and the BAT system have been conceived to supply power to DC loads. Figure 1 shows a global schematic of the proposed stand-alone PV system with the control and instrumentation tool designed to regulate and monitor the system. Several sensors and transducers are utilized to measure the required parameters, in accordance with the recommended practices detailed in [18], to define the efficiency of the off-grid PV-systems. The in-plane irradiance, PV panel and BAT temperatures are measured. Moreover, the PV generator and BAT currents/voltages are, also, measured simultaneously by a DC hall-effect calibrated transducers to supervise the PV stand-alone system with MPPT charge regulator. A process for data acquisition system has been designed with a microcontroller (PIC184550) used to control and monitor PV/BAT systems. In addition, a graphical user interface was used to enhance the monitoring systems. In this way, the data acquisition system could be governed and controlled by the graphical interface, allowing real-time interface. The interface has been built in Delphi program and it is capable of recording various system behaviors such as the power produced by the PV and stocked in the BAT system where Excel has been embedded for real time data plotting and analysis of PV autonomous systems. i

Load

PLoad

PPV

System

LEM Module Buck converter iPV A

VPV

DC V

A

PBAT

V

VBAT

α

Battery

TPV

TBAT

Microcontroller

Ir Reconfiguration

Control - Acquisition

Load

LEM Module iBAT

DC

PV Generator

VLoad

iPV ref

iBAT ref SOC

MPPT Ir

T

PV

V

PV

BMS-BAT i

PV

T

BAT

V

BAT

i

BAT

Fig. 1. The proposed stand-alone photovoltaic system with regulator and monitor devices.

MPPT Charge Regulator and Monitor

117

2 Paprameters Measurement 2.1

Voltage and Current Measurement

The main components used for designing the measurement of electrical parameters are LEM type current and voltage transducers: LA 55-P and LV 25-P respectively, which are both Hall Effect based transducers. The LA 55-P has a current measuring range of up to 50A, while the LV 25-P is capable of measuring a voltage of up to 500 V according to their datasheets. The electrical conditioning circuits developed to measure DC currents and voltages using these transducers are shown in (Fig. 2).

-a-

-b-

Fig. 2. Electrical conditioning circuit of: (a) the current and (b) the voltage transducers.

The measured currents and voltages provided by their respective transducers in the form of a DC signal are amplified with an operational amplifier (LM324). Precision potentiometers and resistors have been selected to transform the measured current and voltage into a voltage in the range of the ADC of the microcontroller in the data acquisition board. Moreover, Zener diodes are added to the input of the microcontroller to ensure overshoot protection. The developed measuring circuit of the LEM module is shown in (Fig. 3).

Fig. 3. Developed circuit of the LEM module.

118

2.2

H. Assem et al.

Temperature and Irradiation Measurement

Surface-mounted DS18B20-PT100 sensor were used for measuring temperature of the PV panel. The latter is located on the rear side of the PV module as shown in (Fig. 4-a). In-plane irradiance is measured by means of a calibrated cell with the same technology of the used PV modules (Fig. 4-b).

Fig. 4. Measurement of: (a) Temperature and (b) irraditon.

3 DC-DC Converter Design and Control 3.1

Design

Many topologies may be appropriate for the specified purpose but probably the most straightforward one is the buck topology: it consists of a small number of components and a single magnetic device. In order to maximize the efficiency of the converter, the buck topology is used with a rectification circuits (Fig. 5), in which the optocoupler for galvanic isolation IR2117 is added to the circuit.

Fig. 5. Electrical and develped circuit of the DC-DC buck converter.

3.2

MPPT Control Method

Over the years, research has concentrated on different MPPT techniques to get the maximum power out of the PV generator. Some of them, the incremental conductance (IncCond) and the perturbation and observation (P&O) methods have attracted more

MPPT Charge Regulator and Monitor

119

interest. The MPPT algorithms previously implemented by the authors are presented in [19]. The IncCond algorithm was selected because of the benefits it offers [20]: • High effectiveness in case of suddenly changing meteorological conditions. • There is no oscillations in the MPP. • It is not required to have the manufacturer’s features and the characteristics of the PV module. • This technique allows to define with high accuracy when the MPP was attained. 3.3

Battery Control Method

Other functions of the regulator and monitor device is to control the BAT and provide a proper charging and discharging process. For this reason, the limits for the BAT voltage are defined in order to prevent over charging and deep discharging. When the BAT voltage is below 11 V the BAT is disconnected from load (BAT current is switched off), and BAT starts its charging mode (negative current). By this stage, the BAT current charge is given through the MPPT program. Then, when the BAT voltage attains 13.7 V, the BAT is relinked to the load. Similarly, when the BAT voltage attains 15 V, the BAT be disconnected the PV generator in order to prevent the BAT overcharging. The voltage control in the regulator is enabled, subsequently, a reference voltage of 13.7 V is defined for the BAT. Furthermore, if the BAT temperature attains 50°C the PV module will be detached. An additional function of the regulator is the ability for disconnecting the PV generator for the period of the night. This function prevents unnecessarily power consumption at this period.

4 Control Board/Softoware 4.1

Control Baord

In (Fig. 6) we can see the electrical circuit of the whole control board. In order to reduce the complexity and cost of the prototype, the selected microcontroller (PIC184550) has analog and digital inputs, thus avoiding the use of external components. The Analog/Digital Converter (ADC) embedded in the PIC184550 has 10-bit resolution and a number ports to read the external analog level (Currents and voltages of PV generator and BAT). A straightforward but precise algorithm was developed in this paper that allowing to minimize the programming code to be implemented in a low-cost microcontroller PIC18F4550. The simplified algorithm implemented is summarized in the flowchart of the (Fig. 7). 4.2

Graphical Interface

In order to monitor the regulator, a graphical interface was developed in Delphi program (Fig. 8). The communication between the acquisition card (microcontroller) and the PC is done through a USB connection. The system monitors the principal electrical and

120

H. Assem et al.

Fig. 6. Electrical an develped circuit of the control board.

Fig. 7. Main flowchart of the PIC18F4550 microcontroller programming.

meteorological parameters. In effect, the acquisition system with its graphical user interface allows to visualize and monitor the input and output currents and voltages (and therefore the power and efficiency) as well as the temperature of the PV generator and the BAT system. All measured parameters data are transmitted every 5 min and can be saved to a text file with a specified sampling rate.

MPPT Charge Regulator and Monitor

121

Fig. 8. Graphical user Interface.

5 Results and Discussion The developed MPPT charge regulator and monitor has been tested below real conditions exploiting a test bench installed in Algiers (Algeria) in Algiers (Algeria). The measures were carried out utilizing a prototype and variable load consumption (rheostat) with controlled switching in order to obtain diverse charge profiles (Fig. 9).

Precision multimeter

Power Supply

Real-time monitoring

PV panel

Rheostat

Conditioning circuit

Fig. 9. The components of the experimental test.

Battery

122

H. Assem et al.

The nominated MPPT charge control and monitor has been conceived for a PV system with the specification bellow: • PV panel of 110 Watts. • Lead-acid Battery 12 V/100Ah. • Three of different power loads (82 W, 92 W, 112 W). Derived system performance parameters are determined by using the monitoring data registered over the daily reporting period. A graphical user interface provides historical data and daily system behaviors as well as various performance indicators in real time. The data recovered from the experimental tests were saved in a text file and then plotted on MATLAB in order to have a better resolution of the figures representing the results (Fig. 10 and Fig. 11). 16

14

14 12

Voltage (V), Current (A)

Voltage (V), Current (A)

Vpv Ipv

-a-

12 10 8 6 4

10 8 6 4 2 0 -2

-b

-4

2

-6 0

0

2

4

6

8

10

12

14

Time (hours)

16

18

20

22

24

-8

0

2

Vbatt Ibatt 4

6

8

10

12

14

16

18

20

22

24

Time (hours)

Fig. 10. Experimental results of PV and BAT voltage and current measured.

Figure 10-a shows the PV voltage and current measured during a day with good solar insolation. The PV generator supplies the load and charges the BAT. At a given time during the experimental test (2:58 PM), a drop in current is observable due to the passage of clouds above the test location. At this point in time, the PV module is incapable to provide enough power to the electrical load. Figure 10-b shows the voltage and current of the BAT during the blue sky day. The BAT current (ibatt) is assumed as negative in charge mode and positive during the BAT discharging mode. As can be seen in (Fig. 10-b) at 7:43 AM, the BAT changes from discharge to charge mode when the power provided from the PV generator is in excess. Similarly, the BAT switches from charge to discharge mode at 6:08 PM to compensate the drop in current from the PV generator. During the experimental tests, the voltage at buck converter was maintained at 13.7 V. Regulating the voltage at 13.7 V guaranteed the electrical load would be supplied by 13 V. Figure 11 shows the variation of PV power (PPV), BAT (Pbatt) and load power (Pload). PV and BAT work in complementarity to ensure that they provide the required power to the load.

MPPT Charge Regulator and Monitor

123

120 Ppv Pload Pbatt

110 100 90

Power (Watts)

80 70 60 50 40 30 20 10 0 -10 -20

0

2

4

6

8

10

12

14

16

18

20

22

24

Time (hours)

Fig. 11. Power curves obtained during the experimental test.

6 Conclusion The design process of MPPT charge control and monitor was proposed which consists of three blocks. The first block uses buck topology converter analog circuitry to drive the MOSFET and to make the system operates at the MPP of the PV panel and controls BAT charge/discharge. The second block is digital (based on PIC18F4550) and is responsible for modification of the current references of the first block. The third block complements the second block by instrumentation tools and graphical interface that monitors the behaviors of the whole system. The experimentation tests reveal that the MPP of the PV generator has been attained. The use of a simple algorithm minimizes the code programing software and permits the possibility using of low-cost microcontroller. The updating of MPP is only done when the conditions at the PV generator or at the charge power changes. The function of the MPPT can therefore be ensured at all times. As maximalized power is the power effectively provided to the charge, temperature changes or other specific features of the PV generator are considered. Besides, the designed charge regulator improves the process of BAT charging/discharging. Indeed, the BAT charge regulator stage avoids the overcharge and deep discharge of the BAT. As a result, the system guarantees a safe operation and a longer life of the BAT, which increases the lifetime of the whole system.

References 1. International Energy Agency. Energy Access Outlook 2017: From Poverty to Prosperity, Vol. 94. International Energy Agency: Paris, France (2017) 2. Ghaib, K., Ben-Fares, F.-Z.: A design methodology of stand-alone photovoltaic power systems for rural electrification. Energy Convers. Manag. 148, 1127–1141 (2017) 3. Alnejaili, T., Drid, S., Mehdi, D., Chrifi-Alaoui, L., Belarbi, R., Hamdouni, A.: Dynamic control and advanced load management of a stand-alone hybrid renewable power system for remote housing. Energy Convers. Manag. 105, 377–392 (2015) 4. International Renewable Energy Agency. Renewable Power Generation Costs in 2017; International Renewable Energy Agency: Abu Dhabi, UAE (2018)

124

H. Assem et al.

5. Khalilpour, R., Vassallo, A.: Leaving the grid: An ambition or a real choice? Energy Policy 82, 207–221 (2015) 6. Muñoz, F.J., Torres, M., Muñoz, V., Fuentes, M.: Monitoring array output current and voltage in stand alone photovoltaics systems with pulse width modulated charge regulators. J. Sol. Energy Eng. 135(2), 021008 (2013) 7. Munoz, F.J., Jiménez, G., Fuentes, M., Aguilar, J.D.: Power gain and daily improvement factor in stand-alone photovoltaic systemswith maximum power point tracking charge regulators. case of study: south of, spain. J. Sol. Energy Eng. 135, 41011 (2013) 8. Phocos. Comparing PWM & MPPT Charge Controllers. Phocos, Ulm, Germany, 2015 9. IEEE Instrumentation and Measurement Society: IEEE Standard for Transitions, Pulses, and Related Waveforms, vol. 2011. IEEE Instrumentation and Measurement Society, New York, NY, USA (2011). ISBN 9780738167060 10. Chen, L.R.: A design of an optimal battery pulse charge system by frequency-varied technique. IEEE Trans. Ind. Electron. 54, 398–405 (2007) 11. Tse, K.K., Ho, M.T., Chung Henry, S.H., Hui, S.Y.R.: A novel maximum power point tracker for PV panels using switching frequency modulation. IEEE Trans. Power Electron. 17(6), 980–985 (2002) 12. Krein, P.T., Tumbull, R.J., Reppa, R., Kimball, J.: Dynamic maximum power point tracker for photovoltaic applications. In: Proceedings of the IEEE Power Electronics Specialists Conference (1996) 13. Fanourakis, S., Wang, K., McCarthy, P., Jiao, L.: Low-cost data acquisition systems for photovoltaic system monitoring and usage statistics. In: IOP Conference Series: Earth and Environmental Science vol. 93, p. 12048. IOP Publishing, Bristol, UK (2017) 14. Mahzan, N.N., Omar, A.M., Rimon, L., Noor, S.Z.M., Rosselan, M.Z.: Design and development of an arduino based data logger for photovoltaic monitoring system. Int. J. Simul. Syst. Sci. Technol. 17, 11–15 (2017) 15. El Hammoumi, A., Motahhir, S., Chalh, A., El Ghzizal, A., Derouich, A.: Low-cost virtual instrumentation of PV panel characteristics using Excel and Arduino in comparison with traditional instrumentation. Renew. Wind. Water Sol. 5, 3 (2018) 16. Bauer, J.: Microchip Various Solutions for Calculating a Pulse and Duty Cycle, pp. 1–22. Microchip Technology Inc., Chandler, AZ, USA (2012) 17. Atmel Corporation. Atmel AVR135: Using Timer Capture to Measure PWM Duty Cycle Table of Contents. Atmel Corporation, San Jose, CA, USA, pp. 1–20 (2016) 18. IEEE: IEEE recommended practice for testing the performance of stand-alone photovoltaic systems. IEEE Std. 1526–2003, 1–18 (2004) 19. Assem, H., Bouchafa, F., Bouzidi, B., Hadj Arab, A.: Fuzzy logic controller in optimizing of power management in stand-alone photovoltaic system. In: Renewable Energy Journal SIENR 2014, Ghardaïa, Vol. 18, N°41–48 (2014) 20. Lokanadham, M., Vijaya Bhaskar, K.: Incremental conductance based maximum power point tracking (MPPT) for photovoltaic system [Paper]. Int. J. Engin. Res. Appl. (IJERA) (2012)

Experimental Validation of a Prototype for Fault Detection and Classification of a Photovoltaic System Using dSPACE A. Hamied1, A. Rabhi2, N. Rouibah3, and A. Mellit1(&) 1

2

Renewable Energy Laboratory, University of Jijel, 18000 Jijel, Algeria [email protected] Modeling, Information and Systems Laboratory, University of Picardie Jules Verne, 3 Rue Saint Leu, 80039 Amiens Cedex 1, France 3 Electric and Industrial Systems Laboratory, Faculty of Electronics and Informatics, USTHB, Algiers, Algeria

Abstract. In this paper, an experimental validation of a low-cost prototype for fault detection and classification of a photovoltaic (PV) system is presented. The prototype can detect, classify faults and displaying the results on the website via Internet of Things (IoT). The experiments result have been carried out at MIS Laboratory of Picardie Jules Verne University, (France). The code is developed under Matlab/Simulink environment and then implemented into a dSPACE1104 for experimental verification. The obtained results of the developed prototype offer similar results as the one obtained using a dSPACE-1104. This confirm the accuracy of the implemented fault detection and classification algorithm (the investigated fault are: short-circuit, open-circuit, shading effect, and dust accumulations on PV modules). Keywords: Photovoltaic system


Monitoring
IoT
Fault detection

1 Introduction With reference to the International Energy Agency (IEA) cumulative installed capacity for photovoltaic (PV) at the end of 2019 reached at least 627 GW [1]. Millions of PV plants were installed worldwide, so in order to avoid losses due to some faults, the status of the PV systems should be checked and monitored online. Some equipment is recently commercialized, however, most available equipment are mainly used to protect and isolate the system from faults [2]. They are not able to identify the type of the faults, even to predict the nature or the origin of the faults [3]. Recording data from PV systems is the first step to detect any faults, as discussed in [4], the data is recorded and sent to the website via Internet of Things (IoT) based on ESP 8266. Despite the low-cost of the system, this model has some limits in the measurement accuracy, in addition using temperature (LM335) sensor has a negative impact in the accuracy of the measurement, as it was used only for ambient temperature. Furthermore, this prototype needs human intervention to detect faults that occur

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 125–131, 2021. https://doi.org/10.1007/978-3-030-63846-7_14

126

A. Hamied et al.

in the system. In [5] the authors presented a very simple algorithm that is integrated into this prototype [4] to detect the status of the PV system. In the literature, there are some attempts about remote sensing, monitoring, fault detection and diagnosis of PV plants on IoT. For example, in [6], a real-time remote monitoring of PV system was developed based on a low-cost ZigBee wireless network, the system allows to control the power. In [7], the authors presented another application of remote sensing of solar power generation system using a ZigBee network. According to the authors the designed remote monitoring system can help users by decision-making reference to the safe operation and daily maintenance as well as management of PV power generation. In [8] the authors presented a process for fault detection of solar module using IoT technique-based Wi-Fi module ESP 8266, the advantage of this process is the possibility of detecting faults without any human involvement. A new effective methodology based IoT technique for: facilities fault detection, preventive maintenance and real-time monitoring has been described in [9]. In [10] the authors presented an online monitoring and fault detection for solar module in remote areas using IoT technique based XBeeS2 module. A wireless lowcost solution based on long-range (LoRa) technology for monitoring PV power plants has been proposed in [11]. A monitoring system (IoT-DAS) for grid-connected PV systems is designed in [12], it can identify non-ideal operating conditions of the system. A new technique for wireless fault diagnosis based independent component analysis (ICA) for PV systems is proposed in [13]. The main objective of this work is to assess the capability and the reliability of the designed prototype in [5] to detect and classify faults in another location, by making a comparison with a system based on a dSPACE-1104. The designed prototype was used to monitor electrical and climate data of a PV system (such as PV array current, PV array voltage, module temperature and solar irradiance) in real time. Sensing an electronic circuit based on the Arduino Mega 2560 was developed. IoT technology is used to transmit the monitored data on the internet using an ODBC (Open Access Databases) database MATLAB window, as well as a website to store and display the monitored data in real time, are developed. A dSPACE-1104 with the test facility available at MIS Laboratory of Picardie Jules Verne University, (France), are used for real time verification and comparison. The rest of the paper is organized as follows: Sect. 2 presents the description of the whole system, including test facility and the designed prototype. Results and discussion are reported in Sect. 3. The last section reports some concluding remarks.

2 System Description The present study is carried out at the MIS laboratory in France. The considered system consists of two PV modules (The PV module characteristic is shown in Table 1), voltage regulator DC-DC according to MPPT and a resistance load of 22 O. Figure 1 shows the block diagram of the investigated system. The developed fault detection and classification code in [5], is implement into Matlab/Simulink for real time verification based on dSPACE 11-04.

Experimental Validation of a Prototype for Fault Detection

127

Table 1. Electrical characteristic of A-300 Solar module. Specifications Open-circuit voltage (Voc) Short-circuit current (Isc) Maximum power voltage (Vmp) Maximum power current (Imp) Rated power (Pmax) Temperature Coefficients of (Voc) Temperature Coefficients Power

Values 23.45 V 5.9 A 19.6 V 5.54 A 108.5 W -1.9 mV/°C -0.38%/°C

Fig. 1. Block diagram of the overall system.

For measurement purpose, current sensor (ACS712), voltage sensor (DC < 25 V), solar radiation sensor (Spektron 320), and cell temperature sensor (Pt100) are used. Figure 2 depicts the developed electrical circuit for measuring voltage (Voc) and current (Isc) by using a simple switch K1 and K2 [5]. DC

A Nc

PV

PV

V

K1

DC

K2

No

Fig. 2. Electrical circuit for measuring Voc and Isc using two relies K1 and K2 [5].

L

128

A. Hamied et al.

The Isc and the Voc are used to identify the fault, in order to know the experimental value in real time a control switch K1 and K2 are used. For more detail on the functioning principal, see [5]. In order to compare the collected results on the website by the designed prototype and the implemented code into Matlab/Simulink using dSPACE, both experiments are run at the same time. Figure 3a shows the implemented system based dSPACE and Fig. 3b shows the experimental prototype.

Fig. 3. a) Test facility at MIS Lab, Amiens University. b) Photo of the designed prototype.

3 Results and Discussion 3.1

Simulation Results

Figure 4a and Fig. 4b show the simulation results obtained by Matlab/Simulink. These results have been obtained for the following averaged weather conditions: solar radiation level = 70 W/m2, cell temperature = 17 ° C (the evolution of solar irradiance and cell temperature are shown in Fig. 4c and Fig. 4d). 3.2

Experimental Results

Experimental tests have been conducted on December 17, 2019 (overcast sky, there were some raindrops). The parameters such as output current, output voltage, solar irradiance and cell temperature) are recorded using dSPACE-1104. Measured parameters are sent and displayed immediately on the web-site based on the IoT technique in order to check any abnormal on the evolution of the PV output power. Figure 5a and Fig. 5b illustrate the current changes in a faulty state and can be compared to Fig. 4a and Fig. 4b. These figures show voltage changes in the same condition and can be compared to Fig. 4. In ‘zone 1’ of both Fig. 4 and Fig. 5, the system is working in normal conditions, although the current is low, as well as the voltage. In this case, the system does not send any notice of fault to the user. So the system works correctly without any fault (normal

Experimental Validation of a Prototype for Fault Detection

129

(zone 6)

(zone 5)

(zone 4)

(zone 3)

(zone 2)

(zone 1)

Current (A)

(a)

Time (s)

(zone 6)

(zone 5)

(zone 4)

(zone 3)

(zone 2)

(zone 1)

Voltage (V)

(b)

Time (c)

(d)

Fig. 4. Simulated data: a) PV output Current evolution, b) PV output Voltage evolution, measured data: c) Solar irradiation and d) Cell temperature.

130

A. Hamied et al.

Current (A)

(a)

(zone 5)

(zone 6)

(zone 3)

(zone 2)

(zone 1)

(zone 4)

Voltage(V)

(b)

Time (s)

Fig. 5 The effect of faults on the current and voltage curves.

operation). In ‘zone 2’ of both Fig. 4 and Fig. 5, an open circuit for one module is occurred, a decrease in the current value is observed. By comparing the region ‘zone 2’ of Fig. 4a and Fig. 4b with those in Fig. 5a and Fig. 5b it can be concluded that there is a similarity in current and voltage variation. Also, it can be observed, after that in ‘zone 3’, the system returns to work correctly without any fault (normal operation). In ‘zone 4 ‘of figures (Fig. 4 and Fig. 5) the observed current and voltage are decreased, the fault detected concern the dust accumulation on both PV modules, and immediately after that a part of the module is covered in order to simulate the absence of the sun. In the region ‘zone 5’ of Fig. 4 and ‘zone 6’ of Fig. 5, the fault detected is shading effect. From the above results it can be concluded that the compared systems (dSPACE and the prototype) operate accurately and the prototype is more appropriate for real time applications.

4 Conclusion and Future Works In this paper, a prototype for fault detection and classification of a PV system is validated and experimentally compared with a real time system based on a dSAPCE. Compared experimental results (by the designed prototype and the implemented system based on dSAPCE) showed a very close similarity. These results confirm the ability of the designed prototype to detect and classify fault in another location (MIS Laboratory

Experimental Validation of a Prototype for Fault Detection

131

of Picardie Jules Verne University, France), under a specific climatic condition. The detected and classified faults are: short-circuit, open-circuit, shading effect and dust accumulations on PV modules. As further contributions, advanced fault diagnosis algorithms based on artificial intelligence (AI) techniques [3] will be integrated to examine the scalability of the designed prototype in real time applications, the reliability issue will be also considered. Acknowledgments. Dr. A. Mellit would like to thank the DGRSDT, Algeria for the financing of the socioeconomic project, code 07/2019.

References 1. IEA: snapshot of global photovoltaic markets, Accessed April 2020 2. Daliento, S., et al.: Monitoring, diagnosis, and power forecasting for photovoltaic fields: a review. Int. J. Photoenergy 1356851, 13 (2017) 3. Mellit, A.: Recent applications of artificial intelligence in fault diagnosis of photovoltaic systems. In: A Practical Guide for Advanced Methods in Solar Photovoltaic Systems, Advanced Structured Materials 128, pp. 257–271 (2020) 4. Hamied, A., Mellit, A., Zoulid, M., Birouk, R.: IoT-based experimental prototype for monitoring of photovoltaic arrays. In: IEEE, International Conference on Applied Smart Systems (ICASS), pp. 1–5 (2018) 5. Hamied, A., Boubidi, A., Rouibah, N., Chine, W., Mellit, A.: IoT-based smart photovoltaic arrays for remote sensing and fault identification. In: Hatti, M. (eds) Smart Energy Empowerment in Smart and Resilient Cities. ICAIRES 2019. Lecture Notes in Networks and Systems, vol 102, pp. 478–486. Springer, Cham (2020) 6. Xiaoli, X., Daoe, Q.: Remote monitoring and control of photovoltaic system using wireless sensor network. In: International Conference on Electric Information and Control Engineering, IEEE, pp. 633–638 (2011) 7. Xu, X.L., Wang, H.: Construction of solar PV power generation remote monitoring system in the architecture of internet of things. In: Advanced Materials Research, pp. 178–182 (2012) 8. Hariprabhu, M., Sundararaju, K.: IoT based fault detection in solar panel using arduino UNO with Wi-Fi Module ESP 8266. Int. J. Recent Technol. Eng. (IJRTE), pp. 2277–3878 (2019) 9. Adhya, S., Saha, D., Das, A., Jana, J., Saha, H.: An IoT based smart solar photovoltaic remote monitoring and control unit. In: 2nd International Conference on Control, Instrumentation, Energy & Communication (CIEC) IEEE, pp. 432–436 (2016) 10. Karthik, S., Mahalakshmi, M., Mahitha, R., Meena, S.: IoT based solar panel fault monitoring and control. Int. J. Inf. Comput. Sci. 238–243 (2019) 11. Paredes-Parra, J.M., García-Sánchez, A.J., Mateo-Aroca, A., Molina-García, Á.: An alternative Internet-of-Things solution based on LoRa for PV power plants: data monitoring and management. Energies 12, 881 (2019) 12. Dupont, I.M., Carvalho, P.C., Jucá, S.C., Neto, J.S.: Novel methodology for detecting nonideal operating conditions for grid-connected photovoltaic plants using Internet of Things architecture. Energy Convers. Manage. 200, 112078 (2019) 13. Qureshi, F.A., Uddin, Z., Satti, M.B., Ali, M.: ICA‐based solar photovoltaic fault diagnosis. Int. Trans. Electr. Energy Syst. e12456 (2020)

Heuristic Optimization, Modeling and Control of Energetic Systems

Optimization Assembly Line Balancing Variables Using Genetic Algorithm Based on Desirability Function Approach Samah A. Aufy(&) and Allaeldin H. Kassam Production Engineering and Metallurgy Department, University of Technology, Baghdad, Iraq {70221,70150}@uotechnology.edu.iq Abstract. Mixed Model-Assembly Line Balancing (MM-AL) recently become more important problems in production management area by producing different types of models of the same product on the assembly line. Thus, any improvement in the performance of the assembly line may lead to improvement in both profit and time. For the purpose, it is almost important to improve practical solution procedures that result high– quality design decisions. Due to the nature MM-AL problem that can classified as NP- hard problem resulting in it being mostly difficult to yield an optimal solution in adequate time. Heuristic algorithm usually employed to solve these types of problems. This paper presented a novel multi-objective genetic algorithm as a meta–heuristic algorithm classification. It based on heuristic treated initial population with recursive heuristic algorithm to approach optimal related of objectives of cycle time, efficiency, smoothness index, variation, and idle time, simultaneously optimized the stated objectives was arrived at employing desirability function approach for reliable effectiveness of efficiency of the design solution. Finally, a comparison study with the proposed approach indicate that population size (50) outperforms population size (500) on 23 replication with regard to both quality of solution and increasing convergence velocity. Keywords: Heuristic algorithm
Multi-objective genetic algorithm model assembly line
Desirability function approach


Mixed-

1 Introduction Assembly Line Balancing Problem (ALBP) was introduced for the first time by Henry Ford in early 1900 considers one of the most debated problems in IE [1, 2]. The objective of solving an ALBP is to assign the needed tasks for producing a product into a set of workstations, which arranged a long material handling system [3]. According to the competitive market characterized with the raising and varied consumer demand, the ALBP become lead that is more complicated their manufacturing to address additional issues, such as producing different models on the same line [4]. Under the term ALBP various models classified into two categories with respect to their layout [2]. Firstly, straight assembly line balancing (SALB), deals with assigning tasks only when predecessors already assigned to straight workstation. While, U-shaped (UALB), as assigning the task whose predecessors and successors are already assigned to © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 135–154, 2021. https://doi.org/10.1007/978-3-030-63846-7_15

136

S. A. Aufy and A. H. Kassam

workstation with respect to the efficiency of solution due to more options of assignable tasks than straight assembly line, referring to [5] for more details. Heuristic algorithm are used for solving ALBP and considered generally problem specific, their application are limited and employed to find good solutions in reasonable amount of time [6]. Since the ALBP characterized as the NP–hard class of combinatorial optimization problems, numerous research efforts turned their attention towards of meta– heuristic algorithm for the solution [6, 7]. A meta–heuristic provides a general algorithm framework that implemented to various optimization problems. Evolutionary algorithm as Genetic algorithm (GA) is one of the most common of meta–heuristic algorithm based on the principle of natural evolution [8, 9]. GA is an effective intelligent random search technique due to its ability and flexibility to move from one solution to other [7].

2 Relevant Literature Zhang and Gen (2011) proposed a novel concept for solving mixed-model assembly line balancing problem related objectives as minimize cycle time, variation of workload and total cost using a generalized Pareto –based scale –independent fitness function genetic algorithm and it validated efficiently in solving problem. [4]. Hwang et al. (2008) proposed multi–objective genetic algorithm using priority – based coding method for solving U–shaped assembly line balancing problem. The proposed approach was provide its ability and support the planner to find numerous feasible solutions and also the coding method can be utilized for distribution scheduling problems [10]. Sabuncuoglu et al. (2000) developed a new genetic algorithm with special chromosome structure for solving the deterministic in the single model assembly line balancing problem, [7]. Tarimoradi et al. (2015) proposed a hybrid multi–objective GA to solve straight and U–shaped assembly line balancing problems in fuzzy circumstance. Furthermore, a one fifth-success rule method was adopted for selection and mutation operator of genetic algorithm performance. The intended result was controlled by convergence and diversity of genetic algorithm, which was enhanced by numerical example for satisfactorily performance. [2]. Chong (2008) adopted GA based on heuristics–treated initial population and compared with randomly generated initial population. He concludes that the approach improved the solution for large–size problems [8]. Chen J.C. et al. (2019) proposed a novel mathematical programming model to find the optimal allocation of tasks, workers, machines, and workstations. It distinguished by two phases of heuristic approach; firstly to optimize initial population, then is devoted to mechanism of GA. The results point out that the proposed twophases of GA had improved the performance compared with traditional GA [11]. Aufy and Kassam (2020) [12] proposed a new methodology for balancing a mixed– model assembly line using worker– task assigned to workstation heuristic (W-TAWH) model developed to address both straight and U-shaped problems. This model generates performance evaluation criteria according to number of the suitable worker and tasks assigned to suitable workstation. For optimizing, these criteria simultaneously integrated into a single score using desirability function approach. The purpose of this paper is to develop a genetic algorithm to solve heuristic treated initial population to approach optimal design solution related of multi-objective problem. According to the

Optimization Assembly Line Balancing Variables

137

reviewing literature, this paper makes one of the first attempts to show how a heuristic algorithm, desirability function approach, and genetic algorithm are hybrid to capture optimal design solution concerning the quality of solution and convergence velocity.

3 The Details of the Mathematical Model There are several algorithms in solving combinatorial optimization problems, whilst most of these algorithms are restricted to solve the following three basic characteristics, which are the focus of this research paper: 1. Finding feasible initial solution to allocate proper worker and assembly tasks to suitable workstation subjected to some constraints such as: precedence relationships meet among tasks, only one worker assign to workstation, a task assign to only one workstation, processing time differs among workers according to their accumulated work experience and capabilities. 2. Solving MM-ALB problem for both straight and U-shape models. 3. Optimizing proposed design solution about multi-objective criteria such as minimizing cycle time (CT), maximizing efficiency (E), minimizing smoothness index (SI), minimizing idle time (Id), and minimizing variation (V). Generally, the outline structure of the model is compromised of two phases as depicted in Fig. 1, phase one devoted to show recursive heuristic assignment (RHA) algorithm, while the second one is adaptive genetic algorithm (AGA) based on desirability function approach.

Fig. 1. Outline of the mathematical model

3.1

Phase 1: Recursive Heuristic Assignment (RHA) Algorithm

The algorithm developed to attain a satisfactory feasible solution with respect to both straight and U-shaped models. It generates performance evaluation criteria according to required data represented by the number of worker, layout, worker sequence, and

138

S. A. Aufy and A. H. Kassam

workstation number. In general, the framework of the developed algorithm comprises of two main parts i.e. Task Assigned-Heuristic Recursive (TA-HR) algorithm, as a searching process based on maximal equality of the total execution time for all workstations, by achieving maximal equality in partition sequence vector (SV) data among workstations. Under this consideration, low variation time among workstations occurred. While, the second part was Worker Assigned-Heuristic Recursive (WA-HR) algorithm dedicated for assigning workers to given workstations. 3.1.1 Part 1: Task Assigned – Heuristic Recursive (TA – HR) Algorithm This algorithm required to handle the imposed precedence relationships constraint among tasks. In literature, a list of heuristic priority rules is used for ranking the set of tasks in SV according to their priority function and precedence relationships among them. In this study, the maximum immediate number of predecessor tasks and the maximum total number of predecessor tasks is assigned for SAL. While the maximum immediate number of predecessor or successor tasks and the maximum total number of predecessor or successor tasks is focused for UAL were used. Stepwise procedure is summarizing below with support of Fig. 2: Step 1: Step 2:

Step 3:

Segmentation the SV into A & B parts by dividing the number of given workstations equally by 2. Calculate workstation ratio (WR), that exhibit the ratio of the number of workstation allocated to each part under the impost condition, that say. WR is less than or equal to 1. Calculate time ratio (TR), shows the data ratio assigned to each A & B part (workstation), and the idea is based on dividing the SV into two parts called sub-vector. At this point, each sub-vector can be depicted by the left position (PL), and right position (PR), as shown in Eq. (1). TR ¼

Step 4:

Step 5: Step 6:

Xi

APTj = j¼PL

XPr j¼i þ 1

APTj

ð1Þ

Checking condition that says ðTR  WR), if yes, a new position (i + 1) must be added, while if no go to step (6), that ensures the amount of allotted time for the sub-vectors with less variation. The last position (i) must be deleted from the sub-vector (A), to ensure not violating TR  WR condition. Repeat steps (1–5) until the given workstations become 1, in another word each sub-vector refers to the workstation that has been filled with number of tasks.

3.1.2 Part 2: Worker Assigned-Heuristic Recursive (WA-HRA) Algorithm The algorithm developed to minimize cycle time associated with processing time for worker assembly line balancing problem. Workers assigned to given workstations have been summarized in the following procedure and can be shown in Fig. 3.

Optimization Assembly Line Balancing Variables

Fig. 2. Flowchart of TA-HR algorithm

139

140

S. A. Aufy and A. H. Kassam

Step 1:

Compute workstation time (Ts(w)), using Eq. (2). It shows the total time needed to finish the assigned tasks to the workstation. TsðwÞ ¼

Step 2: Step 3: Step 4:

Xn X i¼1

i2s

TTki Asi

ð2Þ

Repeat step (1) for all given workers until all workers have been assigned according to minimum Ts(w) to a workstation. The above steps repeated for all workstations. Finally, the minimum cycle time of assembly line is determined using Eq. (3) CT ¼ maxðTsðwÞÞ

3.2

for k ¼ 1; . . .; W

for 8s 2 S

ð3Þ

Phase 2: Adaptive Genetic Algorithm (AGA)

Genetic algorithm is probabilistic search method that utilize search techniques introduced by Darwinm’s evolutionary theory involved the principles and mechanisms of natural selection and the survival of the more suitable. Genetic algorithm based on random search to reach the global optimal solution [13]. Recently, the research effort towards exploiting the characteristics of the ALBP to further improve the existing genetic algorithm structures, herein AGA is proposed to capture the optimal design solution. Major characteristics in the design solution involved are layout shape of assembly line, sequence vector, sequence of worker, and number of workstations in achieving optimum values subjected to limitations and constraints imposed on the objective functions. The main structure of the AGA depicted in Fig. 4 and further detailed illustrated as following: 3.2.1 Initial Population In general, the initial population has a significant effect on solution quality in GA, whilst it must be has more diversity in producing the better performance of GA. For this purpose, a heuristic treated initial population with RHA algorithm adopted as each chromosome represent the proposed design solution generated.

Optimization Assembly Line Balancing Variables

141

Fig. 3. Flowchart of WA-HR algorithm

3.2.2 Chromosome Encoding and Decoding The chromosome designed according to required data to generate feasible solution by involving the RHA represented by four genes; the position label refers to the information of shape layout, sequence vector, worker sequence, and workstation number. Encoding is a transformation process from original information to a chromosome, whereas decoding is the opposite way of transformation. The integer permutation method used for encoding chromosome. Each gene has a specified range of levels listed in Table 1. The letters from A to D is the code of variable generated in constant sequence, for instance Fig. 5 denote the chromosome problem representation in [A B C D], while numbers in genes will be created randomly, for instance [1 2 3 4]. These chromosomes generated to be feasible chromosomes. Therefore, their weakness will improve when they have infeasible initial solution. Hence, the repair mechanism to overcome the mentioned conditions represented repair procedure. 3.2.2.1 Repair Procedure In order to ensure imposed constraints to find feasible solution represented in form chromosome, the following procedure required to perform: Step1: Step2: Step 3:

Select Chromosome X. Select Gene. Decision, if Gene A = 2, check Condition(1) that says, if Gene B value with range (3-4) go to step 4, otherwise replacement gene A with value = 2.

142

Step4: Step 5:

S. A. Aufy and A. H. Kassam

Decision, Condition (2) that says, if the values in Genes C and D equal go to step 5, otherwise select the value in Gene D. Check all population size, if yes go to end, otherwise go to step 2. End

3.2.3 Evaluation of Fitness Value In general, Fitness evaluation is a function to identify the performance of the solution (chromosome). Hence, each individual of the population is evaluated using objective function and fitness function. Formulation of the objective function can identified by the individual desirability value (di), where di(h) is the value of desirability function for performance measure (i) in solution (h) with range 0  di  1. The corresponding performance measures depends on whether these measures (ih ) to be one–sided in cases

Fig. 4. The Main structure of the AGA

Optimization Assembly Line Balancing Variables

143

Table 1. Data set of each gene and their levels Assignment variable Type Layout Straight, U-shaped Sequence vector Priority Rules for Straight Priority Rules for U-Shaped Worker sequence / Workstation number /

Level [1,2] [1,2] [1,2] [4,5] [4,5]

Code A B C D

Fig. 5. Encoding and decoding the chromosome

belong to smaller-the-better CTh, SIh, Idh, and Vh, Eq. (4) or belong to larger–thebetter Eh, Eq. (5). After that, these individual desirability values integrated to overall desirability function (D(h)) as defined by Eq. (6) [14, 15]. 8 > >

> : di  di 0 8 > >

> : di  di 1 DðhÞ ¼

A Y

ih \dmin i dmin  ih  dmax i i ih [ dmax i ih \dmin i dmin  ih  dmax i i

ð5Þ

ih [ dmax i !PA1

wi di;h

ð4Þ

i¼1

wi

ð6Þ

i¼1

Whereas the fitness function is generated according to the overall desirability function (D(h)) based penalty function (B(h)), for objective in desirability function type that belong to smaller the better estimated by Eq. (7) [14], while that belong to the type larger the better by Eq. (8).

144

S. A. Aufy and A. H. Kassam

(

   ih dimax wi ; ih [ dimax 0:01  d max  dimin PðdiÞh ¼ i 0:01 ; ih  dimax   (  ih dimin wi ; ih \dimin 0:01  dmax  dimin PðdiÞh ¼ i 0:01 ; ih  dimin BðhÞ

 Y 2  PA  A 1 w i i¼1 ¼ ð i¼1 Pðdi Þh Þ  0:01

ð7Þ

ð8Þ

ð9Þ

Hence, the fitness function (the overall desirability function based on penalty function) found as: Df ðhÞ ¼ DðhÞ  BðhÞ

ð10Þ

max Where indicate the individual penalty function dmin initial bounds for each i , di performance measure, overall penalty function.

3.2.4 Selection In the selection process, individual chromosomes selected for generating offspring with maximum fitness values. The chosen chromosomes called parents; these parents can create offspring using genetic operators by Roulette wheel method; the selection of parents is in accordance with probability distribution of the revised fitness values. 3.2.5 Generating Offspring Reproduction is the crossover of two chromosomes to produce a new offspring that has genes from both parents. A technique for permutation-based chromosomes was used to ensure that, when applied on two permutation–based chromosomes, the chromosomes of the resulting offspring are also valid permutations [2]. 3.2.5.1 Crossover Procedure The critical attribute of the genetic algorithm is that contain some sort of reproduction procedure. The simple version of this operator inherits individual as they are. Generating new offspring created by Position–Based Crossover (POS) [16]. Firstly, two chromosomes with unequal fitness value randomly are selected, each one have random value less than or equal to the crossover rate (Pc). The basic concept of this type is randomly selected of genes from the first chromosome to enjoin on the identical genes for the second parent, in addition repair procedure is also done to meet the feasibility of new offspring [4, 16].

Optimization Assembly Line Balancing Variables

145

3.2.5.2 Mutation Procedure A new offspring is generated by random mutation. Whereas random number is less than or equal to mutation rate (Pm) to ensure that gene will be muted [4, 16]. 3.2.5.3 Stopping Criteria Stopping criteria is the number of generations possible to control the time of the optimization assignment variables. The stopping criteria must be a set of generations across all tests been considered sufficiently large. It will allow the full optimization performed in short time.

4 Implementation Mechanism The developed mechanism is experimental tested using theoretical data to account for the optimization associated with designing assembly line. Hence, the effectiveness, reality, and validity of the developed approach can highlight. Therefore two-product (A&B) mixed assembly line with data given in Tables 2 and 3 is examined. In addition, the precedence diagram given by Fig. 6. Both products (A&B) require 21 tasks to accomplish. Any individual of five workers (Wi) with different capabilities of processing time can perform any task. Table 4 rank tasks based on the ordered by four different heuristic rules as described in Sect. 3.1.1, each row represents the ranking of 21 tasks in form a sequence vector (SV). AGA stepwise procedure is summarized as following:

Fig. 6. Combined precedence graph

146

S. A. Aufy and A. H. Kassam Table 2. Data set of the product (A) Task No. Task time (per W1 W2 W3 1 3 1.5 1 2 2 2 1 3 6 6 4 4 3 2.5 1.7 5 7 5 2 6 1 1.9 1 7 3 4.5 3 8 4 3.5 3 9 3 2 1 10 0.5 0.3 0.2 11 2 1.5 1 12 0.7 0.4 0.3 13 4 1.9 2 14 2 2 0.9 15 2.5 1 1.8 16 1 2 1 17 8 7 4 18 2 3 2 19 1 1 0.9 20 2 2 1 21 5 3 2.5

unit W4 2 1.3 5 1.9 3 1.3 2.5 2 1.6 0.2 0.2 0.3 1.3 1 2.2 1.2 4.7 1 0.5 1.2 1.9

Table 3. Data set of the product (B)

time) W5 2.6 1.7 6.5 3 4.3 2.5 5.1 4 2 0.2 1 0.35 2 1.5 3 1.3 8 3 1.2 1.8 3.8

Task No. Task time (per W1 W2 W3 1 1 1.7 1 2 1 0.4 0.5 3 5 3 3 4 2 1.5 0.8 5 2 2.2 2.5 6 3 1.3 1 7 5 1.9 1 8 3 2.1 0.5 9 2 2 1.5 10 0.5 0.5 0.3 11 1 0.9 0.5 12 0.3 0.4 0.2 13 1 2 0.5 14 1 0.4 0.6 15 2.5 3 0.7 16 2 0.4 0.5 17 5 3.4 2.5 18 3 1 0.5 19 1 0.6 0.1 20 1 0.4 0.5 21 2 2.6 1

unit W4 0.4 0.5 3 1.1 2.4 1.1 2.3 2.3 1.4 0.4 1.6 0.3 1.7 0.8 0.8 0.6 3.1 2 0.7 0.6 2.3

time) W5 1 1 2.5 1.5 3.1 1.1 2.1 2.3 2.5 0.7 1.7 0.55 2.5 1.2 1.5 1.4 3.7 1.5 0.6 0.9 2.5

17 19 15 10

21 16 16 11

Table 4. Ranking of tasks according to four priority rules Rules Straight 1 2 U– 3 shaped 4

4.1

Tasks 1 2 3 1 2 3 1 2 3 1 2 19

4 4 4 18

20 20 21 20

5 5 19 3

6 6 18 4

7 7 5 5

8 8 17 6

9 9 7 7

10 10 6 14

11 11 8 21

12 12 9 17

13 13 10 13

15 14 11 8

18 15 12 9

16 18 13 16

19 17 14 12

14 21 20 15

Initial Population

Randomly generate 100 chromosomes, each one having 4 genes [gene1 (layout), gene2 (SV), gene3 (worker sequence), gene4 (workstation number)], for example 10 chromosomes generated randomally which represent intial population as shows in Table 5.

Optimization Assembly Line Balancing Variables

147

Table 5. Initial population generated randomly 1 2 3 4 5 6 7 8 9 10

4.2

1 1 2 1 2 1 2 1 2 2

1 1 3 2 4 2 4 1 3 4

5 4 5 3 5 5 4 3 4 3

5 4 5 3 5 5 4 3 4 3

Application of RHA

After specified initial population, each chromosome basically encoded in such away that involve data according to imposed constraints of the developed heuristic algorithm. At first, RHA is run and the obtained results can be clearified as in Table 6. These results correspond to the assigned workers, assigned tasks, and performance measures, and how switch tasks and workers contributed in improving performance measures among chromosoms. That is each chromosome will be represent feasible solution. Table 6. Assigned tasks and workers into workstations for each individual chromosome Ch. no. Assigned W3 W4 1 1 4 2 20 3 5 0 0 0 0 2 W3 W4 1 20 2 5 3 6 4 7 0 0 0 0 0 0 3 W3 W4 1 3 2 4 0 21 0 19 0 18 0 0

worker W2 W5 6 11 7 12 8 13 9 15 10 18 W2 W1 8 18 9 16 10 17 11 21 12 19 13 14 15 15 W2 W5 5 6 17 8 7 9 0 10 0 11 0 12

Performance measurement W1 CT E SI Id V 16 24 0.34 158.24 1798.1 1.154 17 21 19 14 / CT E SI Id V 34 0.31 192 1317.2 0.9730

W1 CT E SI 13 29 0.30 271.7 15 16 14 20 0

Id V 1727.3 1.719

(continued)

148

S. A. Aufy and A. H. Kassam Table 6. (continued) Ch. no. Assigned W3 W4 4 W3 W4 1 5 2 6 3 7 4 8 20 9 0 10 0 11 0 12 5 W3 W4 1 20 2 3 19 4 18 5 0 0 0 0 0 0 6 W3 W4 1 4 2 20 3 5 0 0 0 0 7 W3 W4 1 3 2 4 19 5 18 6 20 0 0 0 0 0 0 0 8 W3 W4 1 5 2 6 3 7 4 8 20 9 0 10 0 11 0 12

worker W2 W5 W2 / 13 14 15 18 19 16 17 21 W2 W5 6 17 7 13 14 0 21 0 0 0 0 0 0 0 W2 W5 6 11 7 12 8 13 9 14 10 15 W2 W1 7 13 14 8 21 9 17 16 0 10 0 11 0 12 0 15 W2 / 13 15 18 16 17 21 19 14

Performance measurement W1 CT E SI Id V / CT E SI Id V 39 0.36 174.67 821 1.102

W1 8 9 16 10 11 12 15 W1 18 19 16 17 21 /

CT E SI Id V 25 0.37 178.96 1794.8 1.5

CT E SI 30 0.38 166.6

Id V 1333.4 1.39

/

CT E SI 39 0.38 164.4

Id 855.1

CT E SI Id V 25 0.37 178.96 1794.8 1.5

V 1.102

(continued)

Optimization Assembly Line Balancing Variables

149

Table 6. (continued) Ch. no. Assigned W3 W4 9 W3 W4 1 20 2 5 3 6 4 7 0 0 0 0 0 0 10 W3 W2 1 3 2 4 19 5 18 6 20 7 0 14 0 21 0 0 0 0

4.3

worker Performance measurement W2 W5 W1 CT E SI Id W2 W1 / CT E SI Id 8 18 34 0.32 179 1357 9 16 10 17 11 21 12 19 13 14 15 0 W1 / / CT E SI Id 17 43 0.37 258.27 828.5 13 8 9 16 10 11 12 15

V V 0.973

V 1.50

Evaluation of Fitness Values

The obtained results of objective function represented individual desirability function (di) was computed using Eq. (4) and (5), where overall desirability function (D(h)) using Eq. (6). All these results reported in Table 7, so can be noticed that the chromosoms number 1, 3, 7, 8, and 10 are signes the objective function with zero value. It means that these chromosomses will be extracted therefore the panelty function was developed in order to increase number of design solutions. Whereas Table 8 shown the results of panelty function (which include individual panelty function (P(di)h) for each measure, computed by Eq. (7 and 8), where overall panelty (B(h)) represented in Eq. (9)) and fitness function for each chromosme (Df(h) computed using Eq. (10)), its clearified that the panelty value was computed only for chromosome with D(h) value equal to zero. This procedure will enable chromosome to be used in the next generation with new fitness function.

150

S. A. Aufy and A. H. Kassam Table 7. Desirability function results for each individual chromosome Chromosome no. Individual desirability function d1 d2 d3 d4 d5 1 1 0.870 1 0 0.945 2 0.861 0.659 0.931 0.867 1 3 0.940 1 0 0.591 0 4 0.732 0.944 0.969 1 0.962 5 0.989 0.973 0.960 0.320 0.782 6 0.989 0.973 0.960 0.320 0.782 7 0.926 0 0.984 0.861 0.848 8 0.732 0 0.988 0.992 0.962 9 0.861 0.757 0.960 0.852 1 10 0 0.973 0.652 0.998 0.782

D(h) 0 0.855 0 0.916 0.746 0.746 0 0 0.882 0

Table 8. Penalty function results for each individual chromosome Chromosome no. Individual penalty function Fitness Function DfðhÞ Pðd1Þh Pðd2Þh Pðd3Þh Pðd4Þh Pðd5Þh BðhÞ 1 2 3 4 5 6 7 8 9 10

4.4

0.01 0 0.01 0 0 0 0.01 0.01 0 0.01

0.01 0 0.01 0 0 0 0.01 0.01 0 0.01

0.01 0 0.01 0 0 0 0.01 0.01 0 0.01

0.01 0 0.01 0 0 0 0.01 0.01 0 0.01

0.01 0 0.01 0 0 0 0.01 0.01 0 0.01

1 0 1 0 0 0 1 1 0 1

 10−10 0.00009 0.855  10−10 0.00009 0.916 0.746 0.746  10−10 0.00009  10−10 0.00009 0.882  10−10 0.00009

Selection, Crossover, Mutation Procedure

Based on the individuals fitness values, two parents are choosen for genetic replication. Individuals with higher score fitness function will have higher chance to select. After that, create offspring operators (crossover & mutation) applied on the selected parents. Position – based crossover (POS) structure was adapted as a method of producing offspring. Crossover procedure for generating offspring, based on selecting from random position of parents with unequal fitness value, variables copied to produce offspring. Finally, repair procedure is adapted to setting layout with corresponding priority rule in addition satisfy the assumption of RHA. It randomly decides whether each gene should mutate according to mutation probability. Then, repair procedure is required to satisfy the imposed constraints.

Optimization Assembly Line Balancing Variables

151

5 Evaluation of AGA Performance In order to validate the proposed approach, the performance of AGA evaluated through study of the effect of population size on the AGA convergence velocity in finding optimum solution, where population size taken values 50 and 500, while crossover rate taken 0.6, mutation rate taken 0.001, and max generation is 300 and kept constant. Tables 9 and 10 shows the experimental results for AGA, note that columns denotes the number of replication, max fitness function, max generation, best chromosomes represent the optimal design solution for balancing MM-AL problem, and performance measures of the 30 replication. A heuristic treated initial population is consider to perform better if the AGA produced optimal solution with lesser population size, if both examined population sizes produce close results, the performance is determined by cycle time, efficiency, smoothness index, idle time, and variation. From tables can notice that the obtained results shows there is not much difference in targeted improvement for both tested population sizes, CT results values in range [12–27.5], E varying from [52%–72%], SI taken values in range [397.27–864.72], Id taken values in Table 9. Numerical results over evolutionary genetic algorithm, pop_ size = 50, pm = 0.001, pc = 0.6 Rep.

Fitness function

1

0.9541185858131

2

Max generation

Best chromosome

Performance measures CT

E

SI

Id

V

70

1

2

3

3

25.5

0.55

450.38

641.12

2.41

0.8763906860452

80

2

4

4

4

15

0.67

590.72

629.76

2.78

3

0.9468226387796

46

1

4

4

4

17

0.60

444.14

764.61

2.03

4

0.9077064691148

125

1

2

4

4

17

0.61

397.27

731.85

1.94

5

0.9243979639310

97

2

4

5

5

12.5

0.67

709.32

785.66

3.07

6

0.9024797701610

68

2

4

4

4

15

0.67

590.72

629.76

2.78

7

0.8912395748116

100

1

2

5

5

13

0.63

542.91

876.18

2.16

8

0.9476764301084

102

1

2

4

4

17

0.58

540.42

788.89

2.43

9

0.9592615223362

78

2

4

4

4

14.9

0.65

540.24

662.97

2.73

10

0.9583191093181

98

1

2

4

4

17

0.58

540.42

788.89

2.43

11

0.9432182139228

85

1

1

5

5

12

0.68

535.36

766

2.30

12

0.9504772745954

79

2

4

5

5

12.9

0.63

808.63

868.46

3.73

13

0.9347863292426

112

2

3

3

3

27.5

0.52

483.72

682.12

2.74

14

0.9420553005796

69

2

3

4

4

16.5

0.60

609.16

760.43

2.83

15

0.9565372325870

84

2

3

4

4

17

0.58

630.21

792.84

2.82

16

0.7979957036090

72

1

2

3

3

19.5

0.70

400.57

426.33

2.20

17

0.9473730374743

107

2

4

3

3

21.5

0.64

554.12

514.38

3.00

18

0.9290447100333

79

1

1

3

3

19.5

0.70

400.57

426.33

2.20

19

0.8873114796170

116

2

4

4

4

15

0.67

590.72

629.76

2.78

20

0.9390600710193

99

2

4

3

3

25

0.55

478.20

640.12

2.65

21

0.9187324511078

67

2

4

5

5

12.5

0.67

709.32

785.66

3.07

22

0.9378771872494

82

2

4

5

5

12.9

0.63

808.63

868.46

3.73

23

0.9214502092958

74

1

1

4

4

17

0.61

397.27

731.85

1.94

24

0.8884101513689

115

1

2

5

5

13

0.63

542.91

876.18

2.16

25

0.9595140845831

95

2

3

5

5

13.3

0.60

864.72

941.95

3.79

26

0.9312761307394

77

2

3

4

4

17

0.58

630.21

792.84

2.82

27

0.9524669077322

73

1

1

4

4

17

0.61

397.27

731.85

1.94

28

0.9513265422285

103

2

3

3

3

20

0.72

571.84

389.28

2.80

29

0.9208037174851

86

1

2

4

4

17

0.60

444.14

764.61

2.03

30

0.9165497411697

69

2

4

4

4

15

0.67

590.72

629.76

2.78

152

S. A. Aufy and A. H. Kassam Table 10. Numerical results over evolutionary genetic algorithm, Pop_ size = 500, Pm = 0.001, Pc = 0.6 Rep.

Fitness function

Max generation

Best chromosome

Performance measures CT

E

SI

Id

1

0.8838396528578

85

1

2

3

3

19.5

0.70

400.57

426.33

2.20

2

0.9605042309150

77

1

2

4

4

17

0.58

540.42

788.89

2.43

3

0.8784630628200

95

2

3

3

3

27.5

0.52

483.72

682.12

2.74

4

0.9009877772662

150

2

4

4

4

15

0.67

590.72

629.76

2.78

5

0.9314516802816

116

2

4

3

3

25

0.55

478.20

640.12

2.65

6

0.9148851771054

77

1

1

4

4

17

0.62

397.27

731.85

1.94

7

0.9503624688830

80

1

2

4

4

17

0.58

540.42

788.89

2.43

8

0.9423675140988

130

2

4

4

4

14.9

0.65

540.24

662.97

2.73

9

0.8867210350112

97

1

1

3

3

19.5

0.70

400.57

426.33

2.20

10

0.9231772064512

103

2

4

3

3

25

0.56

478.20

640.12

2.65

11

0.8684261124267

94

2

3

5

5

13.3

0.61

864.72

941.95

3.79

12

0.9467955580038

82

2

4

3

3

25

0.56

478.20

640.12

2.65

13

0.9258232526872

125

2

3

3

3

27.5

0.52

483.72

682.12

2.74

14

0.9430296672724

76

2

4

3

3

21.5

0.64

554.12

514.38

3.00

15

0.9155589955248

69

2

3

5

5

13.3

0.60

864.72

941.95

3.79

16

0.9443795907346

124

2

4

3

3

21.5

0.64

554.12

514.38

3.00

17

0.9233522393553

110

1

2

4

4

17

0.58

540.42

788.89

2.43

18

0.9343654863244

90

2

4

4

4

15

0.67

590.72

629.76

2.78

19

0.9036260730160

133

2

4

4

4

15

0.67

590.72

629.76

2.78

20

0.9371860584696

79

2

4

3

3

21.5

0.64

554.12

514.38

3.00

21

0.9289849895313

73

1

2

4

4

17

0.60

444.14

764.61

2.03

22

0.9288361099856

97

1

1

5

5

12

0.68

535.36

766

2.30

23

0.9309472861735

81

2

4

3

3

21.5

0.64

554.12

514.38

3.00

24

0.9520115960067

92

2

4

5

5

12.9

0.63

808.63

868.46

3.73

25

0.9346380957120

145

1

2

5

5

13

0.63

542.91

876.18

2.16

26

0.9229115015590

89

1

2

3

3

25.5

0.5547

450.38

641.12

2.41

27

0.9440287048832

152

2

3

4

4

17

0.59

630.21

792.84

2.82

28

0.8001867843047

87

1

2

3

3

19.5

0.72

400.57

426.33

2.20

29

0.9141269123685

71

2

4

4

4

16.50

0.60

609.16

760.43

2.83

30

0.9386019392131

94

2

4

5

5

12.9

0.63

808.63

868.46

3.73

V

range [426.33–941.95], and V taken values in range [1.94 – 3.79]. From a total of 30 replications, the population size 50 of the heuristic treated initial population shows its superiority over the other population size (500) in capturing the optimal design solution associated with lowest max generation for 23 replications (76.6%), all algorithms coded by MATLAB 2018 SOFTWARE. The reason is the search space is small and the AGA can quickly find the optimum solution. Herein, the benefit of using a heuristic treated initial population is apparent in increasing convergence velocity of global optimum and so kept the solution efficiency.

6 Conclusions In this paper, adaptive GA has proposed. Heuristic treated initial population to approach optimal design solution related of multi-objectives is inflict. Heuristic treated initial population addressed by using recursive and consecutive heuristic algorithms for solving the MM-AL problem with objective: to minimize cycle time, smoothness

Optimization Assembly Line Balancing Variables

153

index, idle time, variation, and maximum efficiency of the proposed design solution. The fitness function is formulated as the overall desirability function (D(h)) based penalty function (B(h)). Two different methods for computing fitness function were studied namely smaller-the-better and larger – the better to correspond type of objective respectively. The influence of this novelty in GA were examined over theoretical data. The results obtained showed that the use novelty exhibited superiority for small population size (with regard to both the quality of the solution obtained and increase convergence velocity. This work is limited for straight and U-shaped ALBP, however, it is represented a start point for future studies, such as more heuristic or meta-heuristic may be developed to solve the problem such as simulated annealing approach, and secondly extended the proposed approach to solve other complex of ALBP such as parallel and two sided models.

References 1. Lolli, F., Balugani, E., Gamberini, R., Rimini, B.: Stochastic assembly line balancing with learning effect. IFAC 50(1), 5706–5711 (2017) 2. Alavidoost, T.M., Tarimoradi, M., Fazel Zarandi, M.H.: Fuzzy adaptive GA for multiobjective assembly line balancing. Appl. Soft Comput. 34, 655–677 (2015) 3. Venkatesh, J., Balaji, D.: Application of simple genetic algorithm to U-shaped assembly line balancing problem of type II. The international Federation of Automatic Control (2014) 4. Wenqiang, Z., Mitsuo, G.: An efficient multiobjective genetic algorithm for mixed-model assembly line balancing problem considering demand ratio-based cycle time. J. Intell. Manuf. 22, 367–378 (2011) 5. Alavidoost, M., Babazadeh, H., Sayyari, S.: An interactive fuzzy programming approach for bi – objective straight and U – shaped assembly line balancing problem. Appl. Soften Comput. 40, 221–235 (2016) 6. McMullen, P.R., Tarasewich, P.: Using ANT techniques to solve the assembly line balancing problem. IIE Trans. 35, 605–617 (2003) 7. Sabuncuoglu, I., Erel, E., Tangyer, M.: Assembly line balancing using genetic algorithm. J. Intell. Manuf. 11, 295–310 (2000) 8. Chong, K.E., Omar, M.K., Abu Bakar, N.: Solving assembly line balancing problem using genetic algorithm with heuristics-treated initial population. WCE, London. U.K (2008) 9. Alavidoost, M.H., Zarandi, M.H., Tarimoradi, M., Nemati, X.: Modified genetic algorithm for simple straight and U-shaped assembly line balancing with fuzzy processing time. J. Intell. Manuf. 28, 313–336 (2017) 10. Katayama, H., Gen, M.: U-shaped assembly line balancing problem with genetic algorithm. Int. J. Prod. Res. 46, 4637–4649 (2008) 11. Chen, J.C., Chen, Y.-Y., Chen, T.-L., Kuo, Y.-H.: Applying two-phase adaptive genetic algorithm to solve multi-model assembly line balancing problems in TFT-LCD module process. J. Manuf. Syst. 52(Part A), 86–99 (2019) 12. Aufy, S.A., Kassam, A.H.: A consecutive heuristic algorithm for balancing a mixed-model assembly line type II using a (W-TAWH) model developed for straight and U-shaped layouts. In: 3rd International Conference on Engineering Science, Material Science and Engineering, vol. 671 (2020) 13. Zacharia, P.T., Nearchou, A.: Multi-objective fuzzy assembly line balancing using genetic algorithms. J. Intell. Manuf. 23, 615–627 (2012). https://doi.org/10.1007/s10845-010-0400-9

154

S. A. Aufy and A. H. Kassam

14. Pasandideh, S.H.R., Niaki, S.T.A.: Optimizing muti-response statistical problems using genetic algorithm and simulation. In: Conference of Computer Society of Iran (2005) 15. Liuyang, Z., Yizhong, M., Linhan, O., Feng, W.: Application of the modified genetic algorithm to multi-response robust design based on entropy weight and the desirability function. In: IEEE, pp. 164–68 (2013) 16. Gen, M., Cheng, R.: Genetic algorithms & engineering optimization. John Wiley & Sons, New York (2000)

Design of Optimal Decentralized Controller Using Overlapping Decomposition for Smart Building System Mohamed Z. Doghmane1(&), Madjid Kidouche1, S. Eladj3, and B. Belahcene2 1

3

LAA Laboratory, Department of Automation FHC, University M’hamed Bougara, Boumerdes, Algeria [email protected] 2 Abou Bekr Belkaid University of Tlemcen, Chetouane, Algeria [email protected] LABOPHYT Laboratory of Physics of the Earth FHC, University M’hamed Bougara, Boumerdes, Algeria [email protected]

Abstract. Many industrial systems are known to have complex structure with large dimension variables. For such type of complexities, it is generally preferable to evade the design of centralized controller because of dimensionality augmentation in the step of implementation. Many research studies have been focused on designing decentralized controller for large scale systems. The aim of this paper is not just designing high dimension decentralized controller but also increase the robustness and improve systems’ performance, the optimality of these systems has been considered and discussed in the frame work of mathematical development of inclusion-contraction principle and overlapping decomposition. Furthermore, the proposed control strategy has been applied to a smart building system in order to minimize the damage caused by earthquake; the obtained results allow us to conclude that the proposed control strategy can be so useful for constructing smart cities. Keywords: Optimal decentralized control
Smart building system
Overlapping decomposition
Overlapping decomposition
Smart cities

1 Introduction The mechanical systems behavior is in many cases very complex in a way the mathematical model that describes its dynamic can be very dimensional [1]. Thus, stability analysis and controller design of such type of systems become more difficult; because these steps will require much more computational efforts and the control tasks with required performance cannot achieved easily. The latter imposes the development of new strategies by beneficiating from the mathematical structure of the system in order to decompose it into smaller and low dimension subsystems easy to deal with; the subsolutions of each subsystem can be then joined together with respect to the interaction constraints to construct a solution of the original system [2, 3]. The main objective of © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 155–167, 2021. https://doi.org/10.1007/978-3-030-63846-7_16

156

M. Z. Doghmane et al.

this manuscript is to design a decentralized optimal controller for smart building system by using the overlapping decomposition strategy and extension-contraction principle; this work is development of (L.Bakule and J.Rodellar 1995) study by improving the performance of responses using optimization technique given by [4, 5].

2 System Description Engineering is a large domain that gathers many disciplines, which are all unified in a principle of application of science for practical reasons, it can be said the applied science and engineering are almost equivalent. Civil engineering is one of the applied sciences in which people have constructed many important things such buildings, dams, canals, roads, bridges. It is the scientific vision of the construction who improved the science of modern civil engineering [6]. The combination of practical knowledge of materials and construction with mathematics and science has accelerated the development of building toward smart cities [1].Consider the mechanical second order building system shown in Fig. 1; the building is composed of six floors (Fig. 1.b). The mathematical model that describes the system shown in Fig. 1 is given as:

Fig. 1. a) Schema highlighting the overlapping structure of building system, b) Figure of Real Building System.

8 < M€q þ Dq_ þ Sq ¼ Bu y ¼ Cq : v ¼ V q_

ð1Þ

Design of Optimal Decentralized Controller

157

M66 : is the mass matrix, it symmetric and definite positive, D66 : is the damping coefficients matrix, S66 : is the stiffness coefficients matrix, q61 : is the displacement vector, represents the degree of freedom of the system, B63 : is the input matrix, represents locations of actuators in the floors of the smart building, u31 : is the input signal, it is sinusoidal signal for this model. Equation (1) is designated to represent the response of smart building system to continuous earthquake disturbances; Fig. 2 demonstrated a real example of response failure of building system without actuator to earthquake [6, 7]. Equation (1) can rewritten in the following state space form.

Fig. 2. Example of failure response of a building system to earthquake disturbances [3].

⎧ ⎡ M 11 ⎪⎢ ⎪ ⎢ M 21 ⎪ ⎢⎣ M 31 ⎪ ⎪ ⎪ ⎨ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩

M 12 M 22 M 32

M 13 ⎤ ⎡ q1 ⎤ ⎡ D11 M 23 ⎥⎥ ⎢⎢ q2 ⎥⎥ + ⎢⎢ D21 M 33 ⎥⎦ ⎢⎣ q3 ⎥⎦ ⎢⎣ D31

D12 D22 D32

D13 ⎤ ⎡ q1 ⎤ ⎡ S11 D23 ⎥⎥ ⎢⎢ q2 ⎥⎥ + ⎢⎢ S 21 D33 ⎥⎦ ⎢⎣ q3 ⎥⎦ ⎢⎣ S31

⎡ y1 ⎤ ⎡C11 0 ⎢ ⎥ ⎢ ⎢ y2 ⎥ = ⎢ 0 C22 ⎢⎣ y3 ⎥⎦ ⎢⎣ 0 0

S12 S 22 S32

S13 ⎤ ⎡ q1 ⎤ ⎡ B11 S 23 ⎥⎥ ⎢⎢ q2 ⎥⎥ = ⎢⎢ 0 S33 ⎥⎦ ⎢⎣ q3 ⎥⎦ ⎢⎣ 0

0 B22 0

0 ⎤ ⎡ u1 ⎤ 0 ⎥⎥ ⎢⎢u2 ⎥⎥ B33 ⎥⎦ ⎢⎣u3 ⎥⎦

0 ⎤ ⎡ q1 ⎤ 0 ⎥⎥ ⎢⎢ q2 ⎥⎥ C33 ⎥⎦ ⎢⎣ q3 ⎥⎦

0 ⎤ ⎡ q1 ⎤ ⎡ v1 ⎤ ⎡V11 0 ⎢v ⎥ = ⎢ 0 V 0 ⎥⎥ ⎢⎢ q2 ⎥⎥ 22 ⎢ 2⎥ ⎢ ⎢⎣ v3 ⎥⎦ ⎢⎣ 0 0 V33 ⎥⎦ ⎢⎣ q3 ⎥⎦

ð2Þ The dashed lines in Eq. (2) defines the subsystems, for building system composed of six floors we have two subsystems: S1 (floors 1, 2, 3, and 4) and S2 (floors 3, 4, 5, and 6), the commons floors are 3 and 4, and the shared information is ðM22 ; D22 ; S22 ; B22 ; C22 ; V22 Þ.

158

M. Z. Doghmane et al.

3 Expansion–Contraction Principle The original system given by Eq. (1) is named the overlapping system; it can be transformed into expanded system described by the following equation 8 < Me €qe þ De q_ e þ Se qe ¼ Be ue Se : ð3Þ y e ¼ C e qe : ve ¼ Ve q_ e To achieve this objective, transformation matrices are proposed between the overlapping and the expanded systems 8 8 qe ¼ Tq q ¼ T I qe > > > > < < I ue ¼ U u u ¼ Uue ð4Þ or y ¼ Gy y ¼ GI ye > > e > > : : I ve ¼ Hv v ¼ H ve Where T I T ¼ In , UU I ¼ Im ; GI G ¼ Ip , and H T H ¼ Ir , T I ; U I ; GI ; H I : are the pseudo-inverse of T; U; G; H respectively. One can say that the system represented by Eq. (3) is an expansion of the system given by (1) (reversely (1) is contraction of the system in (3)) if transformation T; U; G and H can be found and satisfies conditions in Eq. (4) for any initial states ðqe ð0Þ; q_ e ð0ÞÞ and for any input ue ðtÞ 2 Rm , t  0 [8, 9]. 8 8 < qe ð0Þ ¼ Tqð0Þ < qe ðtÞ ¼ TqðtÞ q_ e ð0Þ ¼ T q_ ð0Þ ) q_ e ðtÞ ¼ T q_ ðtÞ : : uðtÞ ¼ Uue ðtÞ ve ðtÞ ¼ HvðtÞ

ð5Þ

Theoretically, there exist two main methods to the necessary and sufficient conditions for expansion principle: A. First Method This method necessitates working on the matrices of the second order system directly for the original system and expanded system as well. It means the use of the matrices M and Me is mandatory. B. Method Two In this method, a first order equivalent system should be obtained from the second order system, thus, it requires working with the inverse matrices M 1 and Me1 . Consider the original system given by Eq. (1) and its expansion given by Eq. (3); T where the state vectors x; xe are defined as: x ¼ ðqT ; q_ T Þ [2, 9], these equations can be rewritten as shown in Eqs. (6) and (7).  Sx :

x_ ¼ Ax x þ Bx u yx ¼ C x x

ð6Þ

Design of Optimal Decentralized Controller

 Sex :

x_ e ¼ Aex xe þ Bex ue yex ¼ Cex xe

159

ð7Þ

 Ax ; Bx ; and Cx , Aex ; Bex ;8and Cex are defined as follows:  8 Where 088 I8 I6 066 > > > > A ¼ > > > Ax ¼ M 1 S M 1 D > < < ex  Me1 Se Me1 De   and 084 063 > > Bex ¼ > Bx ¼ M 1 B > > > Me1 Be > > : : Cx ¼ diagðC; V Þ Cex ¼ diagðCe ; Ve Þ By defining the transformation matrices T; U; G that satisfy Eq. (5) for the original system given by Eq. (6); we can find the transformation matrices of the expanded system (7) as: Td ¼ diagðT; TÞ; Cd ¼ diagðG; HÞ, this equation means that: 

x e ð 0 Þ ¼ Td x ð 0 Þ ) uðtÞ ¼ Uue ðtÞ



xe ðtÞ ¼ Td xðtÞ yex ðtÞ ¼ Cd xðtÞ

ð8Þ

C. Theorem One The system Sex given by Eq. (7) is an expansion of the system S given by Eq. (6) or equally S is the contraction of Se , if and only of there exists full rank transformation matrices T; U; G and H such that 8 1 M S T ¼ TM 1 S > > > e1 e > < Me De T ¼ TM 1 D ð9Þ Me1 Be ¼ TM 1 BU > > > GC ¼ C T e > : HV ¼ Ve T Equation (9) is obtained by converting the systems (1) and (3) into state space model and it be can be rewritten as follows: 8 1 Me ¼ TM 1 T I þ Mcq > > > > S ¼ TST I þ Scq > > < e De ¼ TDT I þ Dcq ð10Þ Be ¼ TBU þ Bcq > > > > > C ¼ GCT I þ Gc > : e Ve ¼ HVT I þ Vc

160

M. Z. Doghmane et al.

The matrices Me ; Se ; De ; Be ; Ce and Ve are given in the following form 2 3 2 0 M13 0 M11 M12 D11 D12 6 M21 M22 7 6 0 M D D 0 23 7 21 22 , De ¼ 6 Me ¼ 6 4 M21 4 D21 0 M22 M23 5 0 D22 M31 0 M32 M33 D 0 D32 2 3 2 31 0 S11 S12 0 S13 B11 0 6 S21 S22 0 S23 7 6 0 B22 B22 7 Se ¼ 6 Be ¼ 6 4 S21 0 S22 S23 5, 4 0 B22 B22 0 0 0 S 0 S32 S33 2 31 3 2 3 0 0 0 0 0 C11 V11 0 6 0 C22 6 0 0 7 0 7 7, and Ve ¼ 6 0 V22 0 7 Ce ¼ 6 4 0 5 4 0 C22 0 0 V22 0 5 0 0 0 0 C33 0 0 0 V33

3 D13 D23 7 7, D23 5 D33 3 0 0 7 7, 0 5 B33

Mqc ; Sqc ; Dqc ; Bqc ; Cqc and Vqc are complementary matrices calculated in a way to respond to the necessary and sufficient conditions of extension-contraction principle given by theorem two [8]. D. Theorem Two [6] If theorem one is satisfied, we can say that system (7) is an expansion of the system (6) if condition given by Eq. (11) is verified. 8 Mqc T ¼ 0 > > >K T ¼ 0 > > qc > < Dqc T ¼ 0 Bqc ¼ 0 > > > > C T ¼0 > > : qc Vqc T ¼ 0

ð11Þ

One of the appropriate choices of the complementary matrices is given by Eq. (12), this form of complementary matrices will guarantee the verification of condition in Eq. (11). 2

0 60 6 ½
qc ¼ 4 0 0

1 2 ½
12 1 2 ½
22  12 ½
22  12 ½
32

 12 ½
12  12 ½
22 1 2 ½
22 1 2 ½
32

3 0 07 7 05 0

ð12Þ

4 Contraction Principle of Controllers In order to discuss the contractibility of controller, consider the control law, given by Eq. (13), applied to building system.

Design of Optimal Decentralized Controller

u ¼ Fy þ Lv þ w

161

ð13Þ

Consider, also, the control law given by Eq. (14), applied to expanded system in (7). ue ¼ Fe ye þ Le ve þ we

ð14Þ

w and we denote external inputs which represent the signal of earthquake in this study [10, 11]. A. Theorem Three The control law (14) is contractible to the control law (13) if and only if 

FC ¼ UFe GC LV ¼ ULe HV

ð15Þ

B. Theorem Four If Eq. (2) is an extension of Eq. (1) and if Eq. (2) is stable (respectively asymptotically stable) then Eq. (1) is stable (respectively asymptotically stable) [5].

5 Decentralized Control Design A. Problematic Description Consider the original system described by Eq. (6), the objective of control design is to find the gain matrix for the output feedback control law u ¼ Kyx that minimizes the following cost function: Zþ 1 J¼

 T  x Qx þ uT Ru dt

ð16Þ

1

So that the closed loop system given by Eq. (17)  Sc ¼

x_ ¼ ðA þ Bx KCx Þx yx ¼ Cx x

ð17Þ

Will be asymptotically stable [13]. B. Proposed Solution The expanded system of building system can be seen as combination of two subsystems, they can be decoupled into two separate low dimension systems each preserve the

162

M. Z. Doghmane et al.

effect of interactivity between them. The two subsystems’ models are given by Eq. (18. a) and Eq. (18.b) respectively. 8 < M1 €qe1 þ D1 q_ e1 þ S1 qe1 ¼ B1 ue1 ð18:aÞ y ¼ C1 qe1 : e1 ve1 ¼ V1 q_ e1 8 < M2 €qe2 þ D2 q_ e2 þ S2 qe2 ¼ B2 ue2 ð18:bÞ y ¼ C2 qe2 : e2 ve2 ¼ V2 q_ e2 The state space representation of the decoupled subsystems is given by Eq. (19.a) and Eq. (19.b) respectively.  S1 :  S2 :

x_ e1 ¼ Aex1 xe1 þ Bex1 ue1 y1 ¼ Cex1 xe1

ð19:aÞ

x_ e2 ¼ Aex2 xe2 þ Bex2 ue2 y2 ¼ Cex2 xe2

ð19:bÞ

Where the matrices Aex1 ; Bex1 ; and Cex1 , Aex2 ; Bex2 ; and Cex2 are defined as follows: 8  8   0ð2 þ 2Þð2 þ 2Þ Ið2 þ 2Þð2 þ 2Þ > 0ð2 þ 2Þð2 þ 2Þ Ið2 þ 2Þð2 þ 2Þ > > > Aex1 ¼ > > 1 1 1 1 > > Aex2 ¼ <  M1 S1  M1 D1 <  M2 S2  M2 D2 ; 0 0 þ 2Þ þ 2Þ > Bex1 ¼ ð2 þ 2Þð2 > Bex2 ¼ ð2 þ 2Þð2 1 1 > > > > M B M B 1 2 > > 1 2 : : Cex1 ¼ diagðC1 ; V1 Þ Cex2 ¼ diagðC2 ; V2 Þ The optimal output feedback of each systems is ui ¼ Ki yi ; i ¼ 1; 2, and the performance index for each subsystem is defined as: Zþ 1 Ji ¼



 xTi Qi xi þ uTi Ri ui dt; i ¼ 1; 2

ð20Þ

1

By respecting the demonstrated theorems, the necessary and sufficient conditions for each subsystem are obtained as given in Eq. (21). 8 T T T > < /i Pi þ Pi /i þ Qi þ Cxi Ki Ri Ki Cxi ¼ 0 1 1 T T ð21Þ Ki ¼ Ri Bxi Pi Li Cxi Cxi Li CxiT > : T /i Li þ Li /i þ X0i ¼ 0

Design of Optimal Decentralized Controller

163

The matrix is /i ¼ Ai þ Bxi Ki Cxi ; and the initial state vector X0i ¼ x0i xT0i ; in this study we took the initial state as an identity vector x0i ¼ I. The optimal cost is the trace of the resultant matrix as given in the following equation: Ji ¼

1  tracðPi X0i Þ 2

The value of the optimal cost corresponds to the optimal control law as ui ¼ Ki yi [11].  i  i i i K13 K14 K11 K12 : The form of the optimal gain matrix is Ki ¼ i i i i K21 K22 K23 K24 The output feedback control law for expanded system in (18) is given by the following equation: 2

1 K11 6 K1 21 Ki ¼ 6 4 0 0

1 K12 1 K22 0 0

0 0 2 K11 2 K21

0 0 2 K12 2 K22

1 K13 1 K23 0 0

1 K14 1 K24 0 0

0 0 2 K13 2 K23

3 0 0 7 7 2 5 K14 2 K24

The contracted form of the controller can now be obtained by applying the contraction principle to the expanded controller, the contracted controller is expressed by Eq. (22). 2

1 K11 1 K ¼ 4 K21 0

1 K12 1 2 K22 þ K11 2 K21

0 2 K12 2 K22

1 K13 1 K21 0

1 K14 1 2 K24 þ K13 2 K23

3 0 2 5 K14 2 K24

ð22Þ

In order to implement the controller in Eq. (22) for smart building system, it is necessary to write it in the following form: u ¼ Fy þ Lv þ w w is the external earthquake input signal to the six-floor building system [12]. The controller of the system described by Eq. (1) is given by the following equations K

F , L with F

K111 1 K 21 0

K121 K K112 2 K 21 1 22

0 K122 , L K 222

K131 1 K 23 0

K141 K K132 2 K 23 1 24

0 K142 K 242

Finally, the implementation of the designed controller in the original building system gives us the following closed loop form for the smart building system: 8 < M€q þ ðD þ BLVÞq_ þ ðK þ BFCÞq ¼ Bw y ¼ Cq : v ¼ V q_

ð23Þ

164

M. Z. Doghmane et al.

6 Results Discussion All results presented in this paper are obtained under Matlab environment, where different controllers have been simulated starting from centralized feedback controller without optimality constraints, to decentralized output feedback controller with optimization algorithms, for the six floors of the smart building system. Figure 3 shows the results of the designed feedback centralized controller for floor 2, 4, and 6 without any optimization condition, the response of the smart building system has been compared to open loop system which conventional building without any actuator. It is clear that the smart building response to earthquake is better than the open loop response (Fig. 3 a and b) this may minimize the effect of disasters of earthquake, however in floor 6 even with actuator the response of the smart building system in this floor is considerable and it may create some damage. For this reason, it is necessary to improve the robustness and performance of smart building for the worst case where we have quick disturbances with large amplitudes.

Fig. 3. Non-optimal centralized control system; a) Floor 2, b) Floor 4, c) Floor 6

As an improvement of the obtained results from the centralized controller, we proposed the use of decentralized controller in order to increase the performance of the system to the structured and unstructured external disturbances as shown in Fig. 4. It is noticeable that in the floor 4 which is the common floor between, the two subsystems (Fig. 4 b and d), the response of smart building system in closed loop form is very considerable with high amplitudes; this is due to the effect of interconnection terms between subsystem 1 (floors 1, 2, 3, and 4) and subsystem 2 (floors 3, 4, 5, and 6). The actuator in this study are placed in floors 2, 4, and 6 respectively. The interaction terms have been considered in the following results, in which an optimization algorithm has been integrated with overlapping decomposition strategy in order to improve the robustness of controller and performance of the smart building

Design of Optimal Decentralized Controller

165

Fig. 4. Non-optimal decentralized control; a) subsystem 1 Floor 2, b) subsystem 1 Floor 4, c) subsystem 2 Floor 2, d) subsystem 2 Floor 4

systems especially for the actuator of the fourth floor [10]. Figure 5 shows the results of centralized controller with an optimization function however Fig. 6 shows the results of decentralized controller with the optimization technique.

Fig. 5. Response of smart building system with centralized controller and optimization algorithm: a) Floor 2, b) Floor 6

We can notice from Fig. 5 that even with integrating the optimization algorithm the responses of the floors 2 and 6 is still considerable, this might damages the structure of the smart building, the cost function value found for this controller is equal to Jt ¼ 7:31  103 , it represents the optimal total energy consumed to generate dynamic counter force that should be applied by the actuators in order to absorb the oscillations smoothly without creating damages to the building. Figure 6 shows the responses of the floors 2, 4, and 6 with decentralized controller designed by using overlapping decomposition and optimization algorithm. Figure 6 demonstrates the effectiveness of the designed controller based on overlapping decomposition strategy, indeed, the response of the common floor has been minimized (Fig. 6b and d), this means that the performance of the smart building

166

M. Z. Doghmane et al.

Fig. 6. Response of smart building system with decentralized controller and optimization algorithm: a) Floor 2 of subsystem 1, b) Floor 4 of subsystem 1, c) Floor 2 of subsystem 2, d) Floor 4 of subsystem 2

system has been improved, the cost functions of subsystems 1 and 2 are respectively J1 ¼ 2:47  103 and J2 ¼ 1:41  103 , these values indicates that the overlapping decomposition strategy has not just improved the robustness of the controller but also optimized the value of the energy that should be applied to protect the building in the presence of harmful earthquake disturbances.

7 Conclusion A new decomposition strategy has been proposed in this study, the mathematical development of the overlapping technique is detailed with a focus on inclusioncontraction principle. The development of decentralized controller is based on the necessary and sufficient conditions of the principle summarized in some theories. In order to introduce the optimization aspect of the proposed strategy, the application of inclusion-contraction for the cost function has been intensively discussed. The contractibility of output feedback controller has been proved for smart building system composed of six floors and with two degrees of freedom. A comparison between centralized and decentralized output feedback controllers allowed us to prove the effectiveness of the decomposition strategy to improve the performance of smart building system and to increase the robustness of its controllers connected directly to the actuators. Furthermore, the optimality of the control law has been also examined; the cost function for smart building system represents the optimal energy that should be generated in order to distribute counter forces that mitigate smoothly the oscillations without damaging the building. it is concluded, through the obtained results, that the proposed decomposition techniques does not just increase the performance of smart building system but also minimize the necessary energy consumed by the actuators to ensure that. Thus, it is highly recommended to use the proposed technique to design building of next smart cities with minimized and renewable energies.

Design of Optimal Decentralized Controller

167

References 1. Bakule, L., Rodellar, J.: Decentralized control and overlapping decomposition of mechanical systems–Part 1. System decomposition. Int. J. Control 61(3), 559–570 (1995). https://doi. org/10.1080/00207179508921918 2. Bakule, L., Lunze, J.: Decentralized design of feedback control for large-scale systems. Prilohacasopisu Kybernetika, Supplement to the journal Kyberneitika, ACADEMIA24 (sup.2), pp. 3–96 (1988) 3. Stipanovic, D.M., Inalhan, G., Teo, R., Tomlin, J.-T.: Decentralized overlapping control of a formation of unmanned aerial vehicles. Automatica 40, 1285–1296 (2004). https://doi.org/ 10.1016/j.automatica.2004.02.017 4. Doghmane, M.Z.: Conception de commande décentralisée des systèmes complexes en utilisant les stratégies de décomposition et optimisation par BMI. Doctoral thesis, University M’hamed Bougara of Boumerdes, Algeria (2019) 5. Doghmane, M.Z.: Optimal decentralized control design with overlapping structure. Magister thesis, University M’hamed Bougara of Boumerdes, Algeria (2011) 6. Reitherman, R., Anagnos, T., Meluch, W.: Building bridges between civil engineers and science. J. Clerk Maxwell, Museums. Consortium of Universities for Research in Earthquake. Engineering. 1301 S. 46th Street, Building 420, Richmond, CA 94804, USA (2008) 7. Doghmane, M.Z., Kidouche, M., Habbi, H., Bacetti, A., Bellahcene, B.: A new decomposition strategy approach applied for a multi-stage printing system control optimization. In: 4th International Conference on Electrical Engineering (ICEE), Boumerdes, Algeria (2015). https://doi.org/10.1109/intee.2015.7416751 8. Iftar, A.: Decentralized estimation and control with overlapping input, state, and output decomposition. Automatica 29(2), 511–516 (1993). https://doi.org/10.1016/0005-1098(93) 90148-M 9. Ikeda, M., Siljack, D.D.: Overlapping decompositions, expansions and contractions of dynamic systems. Large Scale Syst. 1(1), 29–38 (1980). North-Holland Publishing 10. Francisco, P.-Q., Rodellar, J., Rosell, J.M.: Sequential design of multi-overlapping controllers for longitudinal multi-overlapping systems. Appl. Math. Comput. 217(3), 1170–1183 (2010). https://doi.org/10.1016/j.amc.2010.01.130 11. Doghmane, M.Z., Kidouche, M.: Decentralized controller Robustness improvement using longitudinal overlapping decomposition—Application to web winding system. Elektronika ir Elektronika 24(5), 10–18 (2018). https://doi.org/10.5755/j01.eie.24.5.21837 12. Ifar, A.: Overlapping decompositions and controller design for neutral distributed-time-delay systems. In: IEEE 14th International Conference on Control and Automation (ICCA), June 12–15, 2018. Anchorage, Alaska, USA (2018). https://doi.org/10.1109/icca.2018.8444323 13. Doghmane, M.Z., Kidouche, M., Habbi, H., Lamraoui, W.: A new decomposition strategy approach applied for web winding system control optimization. In: IEEE 3rd International Conference on Control, Engineering & Information Technology (CEIT), Tlemcen, Algeria (2018). https://doi.org/10.1109/ceit.2015.7233047 14. Riane, R., Kidouche, M., Doghmane, M.Z., Illoul, R.: Modeling of torsional vibrations dynamic in drill-string by using PI-observer. In: Bououden, S., Chadli, M., Ziani, S., Zelinka, I. (eds.) Proceedings of the 4th International Conference on Electrical Engineering and Control Applications, ICEECA 2019. Lecture Notes in Electrical Engineering, vol. 682. Springer, Singapore (2021). https://doi.org/10.1007/978-981-15-6403-1_12

Energy Management Strategy and Optimization of a MicroGrid System Based on Echo State Networks I. Abadlia(&), L. Hassaine, F. Abdoune, and A. Beddar Centre de Développement des Énergies Renouvelables, CDER. BP, 62 Route de l’Observatoire Bouzareah, 16340 Algies, Algeria {i.abadlia,l.hassaine,f.abdoune,a.beddar}@cder.dz

Abstract. This paper discusses the investigation of the Echo State Networks (ESNs) used to optimize a MicroGrid (MG) system operation as well as, an application of a performant Energy Management System (EMS). MG is the grid connected-multisource which composed of an Hybrid Renewable Energy Source (HRES) with an hydrogen fuel transformation system (H2FTS). HRES includes a Photovoltaic Generator (PVG) and a Wind Power Generation System (WPGS). In this work, the ESN is applauded for the control and the optimization of the MG power flow, where the main objectives are, power flow control improvement, minimization in possible the loss in power and increasing the H2 storage efficient. Simulation tests prove that, the proposed EMS unit enables to convert the total power for controlling the Hydrogen Tank Level (H2TL), optimizing the injected energy and get involved the H2FTS to increase system efficiency with a contentious of service. Keywords: Microgrid
Multisource system management
Optimization


Echo state networks
Energy

1 Introduction With the integration of multisource, in opposition to transport grids, the design of the traditional grids distribution has led to less sophisticated control and management structures and weak in terms of automation. To cope, it is necessary and important to take technical and logistical procedures. Many such mix systems comprising of renewable energies and storage systems for MG application with EMS have been discussed in the literature. In MG, compounding of RESs with energy storage systems (ESSs) is representing an extremely efficient solution. Energy management is very important in multisource energy applications, in which to present a preferment algorithm of management, it recommended to considered many conditions and system constraints. Generally, the storage/generation systems control, the power flow disturbance and the equipment costs are the major task of the EMS. Intelligent algorithms offered always the big success in EMS applications that permitted to overcome the nonlinearity problems and present a good solution to perform energy systems, defeat the problem of modelling and the complexity of the conventional algorithms. In © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 168–177, 2021. https://doi.org/10.1007/978-3-030-63846-7_17

Energy Management Strategy and Optimization

169

addition, among the potential advantages of MGs are an increase in the quality of service, reliability and security, which proves that it will be intelligently integrated into the planning of exploitation system. As well, the topic of new architectures for electrical grids is the subject of a various published documents. In [1], the authors proposed the EMS of a PV source associated to an ESS which composed of a fuel cell system and batteries. Fuzzy Logic controller is used to ensure a good management of the power flow, with maximizing the production of H2 and controls the charge/discharge mode operation of the battery. In these applications, as well as in the hybrid power supply system introduced previously, the main challenge is to properly control the demand sharing among the main and the secondary power sources in order to comply with the availability of the different energy sources considered and with the variability of the demand. Also, the control of hybrid power sources includes H2FTS for many applications are developed. In [2], the nonlinear MPC method based EMS is proposed to control the air flow and ensure the adequate oxygen supply for the fuel cell stack. In this paper, the EMS based on ESNs in an hybrid RES compounded the H2FTS for the MG application is proposed. In which, we adopt the ESN as a control unit to manage and optimize the power flow. The main contributions of the work are to investigate the ESN performance for achieving a good EMS with best characteristics of control and optimization procedures; maximizing the use of the RES; benefit the use of H2FTS as a support of store/generation with the maximization of the H2 storage.

2 Microgrid System Description Figure 1 shows the proposed the MG system which represent a grid-connected a hybrid RES (PVG+WPGS) and a H2FTS composed of alkaline water electrolyzer, H2 tank and, proton exchange membrane fuel cell (PEM-FC). Generated renewable energy is injected to the grid where the surplus is used for water electrolysis to produce H2 by

Fig. 1. MicroGrid topology.

170

I. Abadlia et al.

using the alkaline electrolyzer. The generated FC power works as a support to supplement the balanced grid powers when the generated renewable energy is deficient during a period of low solar radiation, wind speed decreasing or in the case of high demands. H2FTS is used as a source or storage system. ESN control unit is used to elaborate the EMS for controlling and optimizing the power flow via activation of intelligent switches.

3 Proposed EMS In this section, the creation steps of the proposed EMS approach based on ESN intelligent algorithm are elaborated. 3.1

Intelligent Applications For EMS in Microgrid Systems

A list of recent studies reported in the literature that use artificial intelligence for EMS in MG applications is summarized in the Table 1 for rappelling some shames and selecting EMS application objectives. Table 1. Summary of modern researches for EMS in MG applications based artificial intelligence. Ref.

System

Meisam Hemmati et al. [3]

MG controllable sources (PV Multi-cross learning-based arrays and a WT) and nonchaotic differential evolution controllable sources (DEG algorithm and MT)

EMS

Sanjari et al. [4]

MG including a FC with Hyper-Spherical Search combined heat and power and algorithm a battery-based ESS

X Fang et al. [5] Multi-energy system (RES, FC, ESS, combining heat generators) in MG on-grid

Neuro-dynamic algorithm

S Leonori et al. [6]

Grid-connected Microgrids

Fuzzy logic with genetic algorithms

Loau Tawfak Al Bahrani and Jagdish Chandra Patra [7]

Smart Power Grid on-grid application

Orthogonal PSO algorithm: solving economic dispatch problem under various constraints

Remarks The algorithm is presented to solve the optimization problem of MG operation. Results obtained are compared with several other recently published optimization methods confirm the validity of the developed approach The results in using real load demand data, compared with other optimization algorithms show effectiveness to find the optimal dispatch strategy EMS applauded to maximize the social welfare of all participants and to balance the energy supply and demand for power. Simulations are given to prove the effectiveness Different strategies for the synthesis of a Fuzzy based EMS by using the genetic algorithms to reduce the EMS rule base system complexity. Where, the rule base system is reduced The algorithm is tested with three practical power systems and it has shown evidence of superior performance compared to several existing techniques in providing reliable, consistent and optimal to be statistically significant against

(continued)

Energy Management Strategy and Optimization

171

Table 1. (continued) Ref.

System

Khalil Gholami and Ehsan Dehnavi [8]

renewable generation (WT, A modified PSO PV, and a combined heat and power plants) in a micro-grid

P.K. Ray and A Mohanty [9]

Wind/PV/FC based MGs

Firefly-swarm

Saraiva et al. [10]

Smart grids with

ANNs

3.2

EMS

Remarks PSO is modified for the optimal power sharing among several RESs within MG with the aim of cost minimization without uncertainty. The algorithm is verified and compared with some different Meta-heuristic algorithms, where it is presents encouraging results A frequency regulation in MG using firefly and PSO based hybrid optimization technique is proposed to tune parameters the PID controller to minimize the frequency deviation in MG system. As well as, EMS in MG is studied. Proposed technique is validated in real-time digital simulation ANNs used as a classification tool of nonlinear loads in a simulated smart grid environment

Modes Operation of the ESN-EMS

Five mode operations characterize the proposed EMS P unit functioning. In the natural state, grid ensure the demands side located loads ( PL) where, Pg ¼

N X

PL

ð1Þ

1

Where, N indicates energy demands. N P In case of Pg \ PL and grid cannot ensure the demands, power side generation 1

PGN composed of PV (PPV), generated wind turbine power (PWT) and the generated power from fuel cell (PFC) intervening and calibrating the rest of the demands which, PGN ¼ ðPWT þ PPV þ PFC Þ

ð2Þ

In the proposed management system, H2TL is defined in max of contents by 90% of the total volume for the security designing. Furthermore, total generated RES power is injected to the grid where, the excess power (Pexcess) is used for water electrolyze by using the electrolyzer to produce H2. In addition, the proposed EMS capable to control the hydrogen tank level that able to ensure efficient functions for the modes operation as presented following. Mode 1: If power grid satisfy demands with existing of excess in renewable power and H2TL < 90%, excess power is used for water electrolysis to produce H2.

172

I. Abadlia et al.

Mode 2: In this mode there are three cases of operation which explained as following: Case 1: the power grid satisfies demands with no exciting of excess power. It is result of no production of the H2 and electrolyser meet in repos state. Case 2: the power grid satisfies the demands with exciting of the excess power and H2TL > 90%, there is no production of H2 and electrolyser is meeting in repos state. Case 3: the power grid cannot satisfy the demands with no exciting of any excess power. It is the result of no production of H2 and electrolyser is meeting in repos state, which grid is forced to put-out some loads for ensuring the balance of powers. Mode 3: The power grid cannot satisfy demands with exciting of excess power and H2TL > 90%, all excess renewable power is injected to the grid with no production of H2 and so; electrolyser is meeting in repos state. Mode 4: If the power grid cannot satisfy demands with exciting of the excess power and H2TL < 90%, in this case it recommended the testing of the total power side generation and to compare it with the renewable generation where, PRE \PGN the FC is activated to produce energy with no possibility of H2 production. Mode 5: the same conditions of the mode 4, but PRE [ PGN . In this mode, excess renewable power is injected to the grid with an existing of a possibility to produce the H2. Figure 2 shows the flowchart of the EMS with five modes of operation.

Fig. 2. Flowchart of EMS modes operation.

Energy Management Strategy and Optimization

3.3

173

Proposed EMS Based ESN Topology

ESN which was introduced by Jaeger [11] is a recurrent neural network (RNN) with a sparsely connected hidden layer. ESN is a popular technique today that helps to train efficiently RNNs, which showed excellent performance, it has a structure composed of an input layer, a recurrent dynamical reservoir, and an output layer. In our study, we investigate the use of this recent approach for the both management of the electrical power in the microgrid system which, this purpose has not been explored yet previously. EMS unit is capable to control the H2TL, the powers flow system and to optimize the error in power between the delivering and generation powers gridmultisource system and the consumed energy by the loads, adopting the mean square error (MSE) as an objective function given by: MSE¼Pg ðPWT þ PPV þ PFC Þ

ð3Þ

So, the goal of the EMS control unit is multiple. With the optimization of the objective function, intelligent switches are activates by the unit to operate powers flow. Proposed EMS is composed of three ESN cells where, their roles explain as following. • ESN Cell 1: Control of generated power: in this cell, inputs are generated power (PPV, PWT and PFC) and the output is the switch (SRE) used to controlling the renewable generated power. • ESN Cell 2: Control of grid power: in this cell, Pexcess, Pg and PL are inputs and a switch (Sg) is used to control and ensure the grid connection on/off. • ESN Cell 3: Control of H2FTS (generation FC and production H2 using electrolyzer) and optimize the MSE: this cell is very important, which it capable to activate switches of electrolyze (SElec) operation and for the fuel generator (SFC) with the process of the MSE optimization (Fig. 3).

Fig. 3. Architecture of the proposed EMS based on ESN.

174

I. Abadlia et al.

4 Results Table 2 shows the main characteristics of proposed MG sources. Table 2 Parameters of the simulated MG sources model. Fuel cell generator Type Nexa Ballard 310–0027 Rating 10000 W Number of cells 40 Rated current 45 Rated voltage 28 – – PMSG Wind Turbine Rated power 5000 W Rated voltage 220 V Rated frequency 50 Hz Pole-pairs 2 Stator resistance 2.9 X Rotor resistance 1.52 X Stator inductance 223 mH Rotor inductance 229 mH Moment of inertia 0.0048 kg.m2

Electrolyzer Electrolyte Electrolyte section Distance between electrodes Temperature of the electrolyzer Cathode transfer coefficient Anode transfer coefficient PV System Type Rating Series modules Parallel modules Short-circuit current Open-circuit voltage Maxi power/cell – –

Alkaline 300 °Cm2 3.010−4 52 °C 0.5 0.3 ELR-615 5000 W 25 4 3.28 A 21.6 V 50 W – –

MG model is simulated and the performances have been evaluated. Figure 4 (a) show the generated PV with the wind renewable powers and Fig. 4 (b) illustrate a typical profile of the demand which, Figs. 4 (a) and 4 (b) represent a Data-bases for the initial ESN control unit. Figure 5 (a) shows the generated fuel power as a support to reinforcing the renewable generation and Fig. 5 (b) represents the excess power used for water electrolysis to produce H2 by using the electrolyzer. It proved that the validation of the applauded EMS operation modes and justify the H2TL presented by Fig. 6. MSE is shown by Fig. 7 (a). It is observed that, a very small values in power compared with all system powers. This was proved the minimizing of the power loss. Switches states of the system management represented by Fig. 7 (b), which proves the manager of the power flow and the activation of different switches of the MG control power system respecting the EMS operation modes.

Energy Management Strategy and Optimization

Fig. 4. (a): (PV and WT) generated powers, (b): demands side.

Fig. 5. (a): FC generated power, (b): electrolyzer used power.

Fig. 6. H2 tank level.

175

176

I. Abadlia et al.

Fig. 7. (a): MSE of the optimization problem, (b): Switches state of the Management system (0: closed; 1: opened).

5 Conclusion An efficient EMS unit control and optimization based-ESNs was developed and applauded for a MG-multisource system application. Under the proposed scheme of power flow management, the MG power plant is able to effectively operate under both normal conditions. The scheme was to provide dynamic support to the grid during abnormal conditions, thus making it compatible with the modern grids.

References 1. Abadlia, I., Bahi, T., Bouzeria, H.: Energy management strategy based on fuzzy logic for compound RES/ESS used in stand-alone application. Int. J. Hydrogen Energy 41(38), 16705–16717 (2016) 2. Gruber, J.K., Bordons, C., Oliva, A.: Nonlinear MPC for the airflow in a PEM fuel cell using a Volterra series model. Control Eng. Pract. 20(2), 205–217 (2012) 3. Hemmati, M., Amjady, N., Ehsan, M.: System modeling and optimization for islanded micro-grid using multi-cross learning-based chaotic differential evolution algorithm. Int. J. Electr. Power Energy Syst. 56, 349–360 (2014) 4. Sanjari, M.J., Karami, H., Yatim, A.H., Gharehpetian, G.B.: Application of Hyper-Spherical Search algorithm for optimal energy resources dispatch in residential microgrids. Appl. Soft Comput. 37, 15–23 (2015) 5. Fang, X., He, X., Huang, J.: A strategy to optimize the multi-energy system in microgrid based on neurodynamic algorithm. Appl. Soft Comput. 75, 588–595 (2019) 6. Leonori, S., Paschero, M., Mascioli, F.M.F., Rizzi, A.: Optimization strategies for Microgrid energy management systems by Genetic Algorithms. Appl. Soft Comput. 86, 105903 (2020) 7. Al Bahrani, L.T., Patra, J.C.: Orthogonal PSO algorithm for economic dispatch of thermal generating units under various power constraints in smart power grid. Appl. Soft Comput. 58, 401–426 (2017)

Energy Management Strategy and Optimization

177

8. Gholami, K., Dehnavi, E.: A modified particle swarm optimization algorithm for scheduling renewable generation in a micro-grid under load uncertainty. Appl. Soft Comput. 78, 496– 514 (2019) 9. Paital, S.R., Ray, P.K., Mohanty, A.: Firefly-swarm optimized fuzzy adaptive PSS in power system for transient stability enhancement. In: 2017 Progress in Electromagnetics Research Symposium-Fall (PIERS-FALL), Nov 2017, pp. 1969–1976. IEEE (2017) 10. Saraiva, F.D.O., Bernardes, W.M., Asada, E.N.: A framework for classification of non-linear loads in smart grids using Artificial Neural Networks and Multi-Agent Systems. Neurocomputing 170, 328–338 (2015) 11. Jaeger, H.: Echo state network. Scholarpedia 2(9), 2330 (2007)

Control and Supervision of Water Tower in Smart City L. Miloudi(&), A. Djenadi, and A. Hadj Youb Electrical Systems Engineering Department, Faculty of Technology, Applied Automatic Laboratory, University M’Hamed Bougara Boumerdès, Boumerdes, Algeria [email protected]

Abstract. This study is dedicated to automation and supervision of a water tower. The automation of electric hydraulic pumps group installations of water tower was carried out by a programmable logic controller (PLC) S7-1200, and the supervision by a system of control and data acquisition SCADA (Supervisory Control And Data Acquisition). This is a system of teleprocessing on large scale allowing to treat in real time a great number of data and to control technical installations remotely. The interface for the supervision was created using the TIA PORTAL V13 which is the last software of engineering developed by SIEMENS and best adapted to the material used. Keywords: Water tower
Electric hydraulic pumps group EHPG Programmable logic controller (PLC) S7-1200
SCADA




1 Introduction In smart city two principal factors must be optimally managed which are electric energy and water, this study is based on practical and rapid technique for solving an industrial problem of water distribution, when the theoretical data are not available and system modeling is difficult. Many works was developed by researchers for modeling and control various tanks in particular the application of the theoretical methods. A. Thamallah et al. [1] proposed predictive control of four tanks inter-connected after modeling the process by Takagi Sugeno method, and optimization the controller actions by particle swarm optimization algorithm. Abdullah Bas_çi et al. [2] performed control level position of two coupled tank water, the method developed is based on an adaptive fuzzy control system divided on adaptive model identifier and controller. The parameters of model system are obtained using recursive last square algorithm. Amruta S. Jondhale et al. [3] investigated a review of level control tank system based on PID controller (proportional, integral and derivative actions). It is a very widespread and practical tool in industry, this study shows that some PLC are equipped with PID controller by using Ziegler-Nichols and Cohen-coon methods. Zi-Jiang Yang et al. [4] developed a robust nonlinear control for tracking water level of three tank system. This theoretical method based on back stepping design performed good numerical results. Yun-Hyung Lee [5] published a interesting study for control level of single water. Three linear sub models were obtained according the water level then a fuzzy—PID © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 178–183, 2021. https://doi.org/10.1007/978-3-030-63846-7_18

Control and Supervision of Water Tower in Smart City

179

controller was implemented, the results of simulation show that the fuzzy—PID controller can effectively stabilize and control the system studied. Yuli Wei et al. [6] carried out the control of a double tank based on neural network controller with back propagation algorithm. The results obtained of the system are definitely better than those of a system controlled by a traditional derived integral proportional controller. An experimental study was presented by Nuno Brito et al. [7], using an on–off regulator to control the water tank level, this controller is supervised by a LabVIEW software. Tanks in general are among the most favorable systems to be controlled and supervised by various software. Nomenclature Cap NB Cap NH PID PLC SCADA TIA PORTAL C–E RV

Low level sensor High level sensor Derived integral proportional controller Programmable logic controller Supervisory Control And Data Acquisition Software Water tower Water storage

2 Process Description and Equipments 2.1

Process

The water tower is fed by a storage tank with capacity of 1500 m3 through of electric hydraulic pumps group (EHPG), which counts three pumps, as represented in Fig. 1. The aim of the study is to put the pumps under a PLC programmable logic controller which allows the opening and the closing pumps automatically by connecting them to levels sensors of storage tank and water tower. 2.2

Equipments

For realization this project, the following equipment is necessary: Piezometric level sensor, analogical sensor, analogical pressure sensor, chlorine sensor and flow meter. – high level sensor is noted: cap NH – low level sensor is noted: cap NB Programmable logic controller (PLC) proposed in this project is S7-1200 (SIMATIC S7 of SIEMENS). The PLC is used for small to medium sized automation applications. It has a compact and modular architecture with: 14 digital inputs,10 digital outputs, two integrated analogical inputs, 6 fast counters and 4 integrated pulses integrated with the possibility of extensions to 8 input-outputs modules. 100 KB working memory with a PROFINET interface for HMI (Human Machine Interface). The basic functional modeling system is shown on Fig. 2.

180

L. Miloudi et al. Water storage 5000m3

EHPG 3pumps

Water tower 1500m3

Fig. 1. Basic process diagram

PLC

Fig. 2. Functional modeling system

3 Automation The variables table of the PLC must be created in the first. The variables table allows us to define the list of variables that will be used during the programming and producing the specifications. It enables us to indicate all inputs and outputs of our system, the feedbacks, the different flow, pressure, and water level transmitters, the defects, the memories, the starting and the stopping of the pumps. During programming, we can access PLC variables via their names, through software “TIA portal V13” of SIEMENS.

Control and Supervision of Water Tower in Smart City

181

The control of the pumps for discharge is done by cascade permutation. Each time when the system gives authorization signal the comparator compiles with the counter which will gives an optimal solution. The following Table 1 explains the three control cases of pumping in permutation with start to work = 1 and stop work = 0. Table 1. Pumps permutation Pump control 1st case 2nd case 3rd case

Pump 1 Pump 2 Pump 3 1 1 0 0 1 1 1 0 1

After creating our project and to finish the configuration, the program must be simulated and to detect any error using the ‘on line’ command in the menu bar. After compiling the software, then it checks the correct operation of this program. The following window is displayed with information on the compilation of program, what it shows that the program has been compiled successfully and without any error as represented on Fig. 3.

Fig. 3. Display messages compilation

4 Suprvision The Human Machine Interface (HMI) will allow us to control and supervise the station remotely, to connection with PLC as shown in Fig. 4. A system control and data acquisition (SCADA) is a large scale remote management system, making it possible to treat on real time a great number of telemetries measurements and to control technical installations. It is an industrial technology in the field of instrumentation, which the implementations can be regarded as working

182

L. Miloudi et al.

Fig. 4. PLC-HMI connection

Fig. 5. The station view

frameworks, HMI are used as well for visualization of figures as represented in Fig. 5 and for control of machines. They make it possible to collect input/output data from PLC to present its as to make its understandable and exploitable by the operator.

Control and Supervision of Water Tower in Smart City

183

5 Conclusion Main goal of this study is to give a program which monitors the water level in the water tower and the tank principal, to control the pumping group. This program makes it possible to give an optimal solution to the pumping operation which is the permutation pumps and control level for a long period to use the pumps. The PLC S7-1200 was programmed with TIA Portal software. As well as the supervision of faults that the system may present and display them as alarms on the supervision screen which to allow remote management of the station. This technique allows engineers and managers to make a good management, a good control of distribution and storage of drinking water, and the maintenance of the equipment station.

References 1. Thamallah, A., Sakly, A., M’Sahli, F.: A new constrained PSO for fuzzy predictive control of Quadruple-Tank Process. Meas. J. 136, 93–104 (2019). https://doi.org/10.1016/j. measurement.2018.12.050 2. Bas_çi, A., Derdiyok, A.: Implementation of an adaptive fuzzy compensator for coupled tank liquid level control system. Meas. J. 91, 12–18 (2016). https://doi.org/10.1016/j.measurement. 2016.05.026 3. Jondhale, A.S., Gaikwad, V.J., Jondhale, S.R.: Level control of tank system using PID controller—A review. IJSRD—Int. J. Sci. Res. Dev. 3(10) (2015). ISSN (online): 2321–0613 4. Yang, Z.-J., Sugiura, H.: System in the presence of mismatched uncertainties robust nonlinear control of a three-tank. In: IFAC (International Federation of Automatic Control Papers), pp. 4088–4093 (2017). www.sciencedirect.com, https://doi.org/10.1016/j.ifacol.2017.08.793 5. Lee, Y.-H.: Level control of single water tank systems using Fuzzy-PID technique. J. Korean Soc. Mar. Eng. 38(5), 550–556 (20140. ISSN 2234-7925 Print http://dx.doi.org/10.5916/ jkosme.2014.38.5.550 ISSN 2234-8352 Online https://doi.org/10.5916/jkosme.2014.38.5.550 6. Wei, Y.: Study of double-holding water tank liquid level control base on neural networks PID control. In: Advances in Intelligent Systems Research (AISR), Vol. 151, p. 320, 2018 International Conference on Computer Modeling, Simulation and Algorithm (CMSA 2018) 7. Brito, N., Ribeiro, P., Soares, F.: A remote system for water tank level monitoring and control —A collaborative case-study. In: 3rd IEEE International Conference on E-Learning in Industrial Electronics (ICELIE) Porto, Portugal, 3–5 Nov 2009. https://doi.org/10.1109/icelie. 2009.5413218. Source: IEEE Xplore

Preventive Maintenance Optimization of Top Drives in Smart Rotary Drilling Systems Idir Kessai1(&), Samir Benammar1, Mohamed Z. Doghmane2, and Sadek Khelifa1 1

2

Laboratoire Energétique- Mécanique & Ingénieries (LEMI), Université M’Hamed Bougara de Boumerdes, 35000, Boumerdes, Algeria [email protected], [email protected] Laboratory of Applied Automatic, Faculty of Hydrocarbons and Chemistry, M’hamed Bougara University Boumerdes, Boumerdes, Algeria [email protected]

Abstract. The increased competition in petroleum industry has pushed many companies to revise the preventive maintenance schedules in a way that allows them to use effectively the drilling equipments, and improve the availability of backup equipments while ensuring a continuous production for the hydrocarbons markets. This paper treated the optimization of equipments availability as an objective function, wherein, the manipulated parameters are failure rate, maintenance costs and production loss costs. The optimization algorithm has been transformed into Visual Basic Application (VBA), which provides optimum results in reasonable time. Moreover, the proposed approach has been applied to CANRIG 8050-AC 712 Top Drive used in smart rotary drilling systems; it is very expensive equipment and suffers a lot of failures due torsional vibrations. Furthermore, the effectiveness of the approach has been demonstrated through an application to field rotary drilling system belongs to the national drilling company ENAFOR. Keywords: Preventive maintenance
Equipment availability
Objective function optimization
VBA
Rotary drilling systems
Top drive

1 Introduction Oil and gas are considered as one of the most important energies of this century, they are used in many countries all over the world. This energy forms the basis of the world economy, so that industrialized countries as well as developing countries cannot progress without exploring this energy. Indeed, our dependence on this fossil energy is very strong, for that and to meet the needs of the market, oil companies around the world have decided to extract as much oil as possible by drilling wells using different methods [1]. Drilling of wells involves creating a connection between the surface and the deposit containing hydrocarbons by making holes in several phases [2]. This connection is carried out using a drilling rig fitted with a derrick, capable of supporting the weight caused by the drillstring and drill bit screwed at its end [3]. In the past, the rotational drilling process was achieved by the rotary table but in recent years, and with © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 184–196, 2021. https://doi.org/10.1007/978-3-030-63846-7_19

Preventive Maintenance Optimization of Top Drives

185

the increase in demand for oil, this device was not able to meet the growing requirements due to its low rate [4]. Therefore, companies have prompted a new drilling technology which is based on the introduction of a new equipment to speed up the drilling process; this equipment is called the Top Drive [1, 4]. The Top Drive is developed to improve the performance of a drilling rig, to save time, money and lower the burden on operators [5]. However, in Algeria this equipment has often problem of unavailability due to lack of a preventive maintenance schedule that has been recommended by the manufacturer, this creates an acceleration of Top Drive aging [2]. To remedy this problem, in this study an action schedule procedure has been carried out on a CANRIG 8050AC-712 Top Drive. Then, the most penalizing elements in terms of availability have been selected and formulated as an objective function, and optimal solution was calculated based on the VBA algorithm. The manuscript is organized as follows: in section two, a general description of the Top Drive and its parts has been given in order to construct an idea about the type of more penalizing elements ([6, 7]). The third section focuses on CANRIG 8050AC-712 Top Drive as a case study. In the fourth section, the optimization algorithm of preventive maintenance has been detailed with all calculations and objective function implementation for CANRIG Top Drive [8]. The obtained results have been demonstrated and discussed in section five; they allowed us to end this paper with conclusive remarks and recommendations for the national drilling company.

2 Top Drive Types The Top Drive is a rotating device, held at the derrick by means of a slide and a movable muffle, it constitutes the connection between the movable block and the drill pipe and it makes a part of the lifting device of a drilling installation [2]. It is integrated into the structure of mast via a guide rail system and can be moved up and down the mast between the crown bearing and the working platform; it is carried out using a control console. The Top Drive can perform the functions of the hook, injection head, rotary table, Kelly, keys…etc. [6]. This allows the drillers to go faster during the drilling process, save time, reduce work effort, eliminate dangerous situations and improve control of unforeseen events [1]. Generally the Top Drive encompasses three main functions: the rotation of the drill string [9], the circulation of drilling fluid, and maneuvering operations (ascent and descent).There are many types of Top drives like BENTEC, VARCO, TESCO, and CANRIG (Fig. 1). The Top Drive system can be chosen according to the user’s needs, i.e. dimensions, maximum loading capacity of the hook, Torque and speed needed [10], AC/DC electric motor…etc. [6].

186

I. Kessai et al.

Fig. 1. Types of top drive used in petroleum industry

3 Canrig 8050ac-712 Top Drive CANRIG is one of the world’s leading suppliers of drilling systems in the oil and gas industry [1]. The company manufactures markets and maintains a full range of superior, mobile and stationary drive systems, critically designed for most land and offshore platforms. This includes superior motors in all sizes and configurations to meet all drilling applications [11]. CANRIG superior drives are designed and manufactured according to API standards including the standard for safety and reliability [12]. The CANRIG Top Drive Type 8050AC-712 is powered by an AC asynchronous electric motor; it is used to transform electric energy into mechanical energy to ensure the rotation of the drill string [4]. These types of Top Drive are equipped by intelligent accessories in order to improve the efficiency of drilling [2]. Among these accessories we can find the following: A. Advanced Diagnostic System The advanced diagnostic system (ADS) monitors the strategic points of the upper unit and its associated control and power system which allow intelligent evaluation of defaults and report the possible causes. B. Anti-collision System CANRIG developed an anti-collision system (TAC) which triggers an alarm when contact between certain components is on point to happen [13]. If the alarm is activated and the problem is not corrected, the equipment will be stopped by an automatic shutdown device.

Preventive Maintenance Optimization of Top Drives

187

C. Equipment Condition Monitoring System Using sophisticated sensors, specially designed software and the collection of data, CANRIG can offer an even higher level of reliability and higher performance by using equipment conditions monitoring system (ECM).

4 Preventive Maintenance Optimization Algorithm All activities in the oil drilling companies have undergone considerable changes since 1980. These changes have only been possible through the development of new application, organization and management, which have made it possible to develop new work forms. Nowadays, most companies have become digitalized, which requires to have, in our time, the computation tools to manage them. Currently we do not choose the software applications only for their ability to perform an IT function but also for their roles in the company’s production chain [1]. In this section, we will present an algorithm of optimization used for preventive maintenance operation of the Top drive, the algorithm is developed in VBA environment with graphical interface to simplify its use for drillers ([8, 14]). All the following mathematical equations have been integrated in the algorithm code [15]. A. The Objective Funtions In order to estimate the number of interventions during the time interval ½0; T , and according to the Poisson distribution taking the random variable h as the number of failures in that interval [16], we present the probability of having h failures by PðT; hÞ ¼

ah a e h!

ð1Þ

a: Mathematical expectation of the number of failures in the interval ½0; T , it is calculated as follows ZT

a ¼ kðtÞdt

ð2Þ

0

With kðtÞ: Instant failure rate. For a material mainly composed of m components whose failure rates progress linearly, we have ki ðtÞ ¼ ki þ ki t; i ¼ 1;


; m

ð3Þ

188

I. Kessai et al.

ki : is the initial failure rate of the ith hardware component, and ki : is the coefficient of proportionality [16]. The overall failure rate, considering the serial components from the reliability point of view is determined by the following expression kð t Þ ¼ k0 þ

Xm 1

ð4Þ

ki t

P With k0 ¼ m 1 ki . It is considered that after each planned repair at the subset level ensemble i its rate of default takes the initial value ki , the mathematical expectation of the number of failures in the interval ½0; T  is written as follows Xm k i T 2

a ¼ k0 T þ

1

ð5Þ

2

B. Optimal Cost Estimation We consider the frequency and structure of planned maintenance interventions over a time interval ½0; T . The planned periods are designated by D1 ; D2 ; D3 . . .Dm1 for equipment of which m components require systematic maintenance beforehand planned one at times t1 ; t2 ; t3 ; . . .:tm1 [13], we can formulate this by the following relationships t2 ¼ t1 þ iD1;t t3 ¼ t1 þ iD1 þ jD2 ; tm ¼ t1 þ iD1 þ jD2 þ


þ pDm1

ð6Þ

Let us denote by n1 ; n2 ; n3 ; . . .:nm the quantities of periods of time whose durations are respectively D1 ; D2 ; D3 ; . . .; Dm spread over the period ½0; T  such that n1 ¼

D2 D3 D4 T ; n2 ¼ ; n3 ¼ n ; nm ¼ Dm1 D1 D2 D2

ð7Þ

The relationships between the quantity of time period of duration D1 ; D2 ; D3 . . . and Dn and the quantity of interventions planned over the period ½0; T  are given by the following expressions np1 ¼ ðn1  1Þ

Ym 2

ni; np2 ¼ ðn2  1Þ

Ym 3

ni ; np3 ¼ ðn3  1Þ

Ym 3

ni ; npm ¼ ðnm  1Þ ð8Þ

Taking into account the expressions (6), (7) and (8), relation (5) can be written in the following form a ¼ ðt1 ; n1 ; n2 ;


nm Þ ¼ nm

Xnm 1 Xn2 1 Xn1 1 Z p¼0

j¼0

i¼0

D1

½k0 þ k1 t1 þ k2 ðiD1 þ t1 Þ þ

0

k3 ðiD1 þ jD2 þ t1 Þ þ


þ km ðiD1 þ jD2 þ


þ t1 Þdt1

ð9Þ

Preventive Maintenance Optimization of Top Drives

189

The timing of the planned interventions ðiÞ corresponds to drops in failure rate corresponding to their initial values ki and the failure rate is reduced up to the value k0 only when carrying out the general overhaul [16]. The structure and frequency of the planning of preventive interventions are determined using as an optimization criterion, the expenses of planned repairs and not planned during the calculation period of duration T [8]. By taking into account the relation (9), the function of the costs to be optimized (objective function) has the following form:  2 2 T2 S ¼ CA k0 T þ CA1 2n;nk21nTk


nm þ CA2 2nnkn2xT


nm þ CA3 2nk23


n þ


þ m CAm

km T 2 þ C1 ðn1  1Þn2 n3


nm þ C2 ðn2  1Þn3


nm þ :C3 ðn3  1Þn4


nm Þ þ


þ 2nm

C m ð nm  1Þ

ð10Þ

S : is the summation maintenance cost. CAi : is the cost of an unplanned repair of the ith component [13]. Ci : is the cost of a planned repair of the ith component. So, in this study the optimization of maintenance based on equipment reliability, consists in optimizing the objective function; that is to say determine the combination ðn1 ; n2 ; n3 ; . . .; nm Þ which gives the following minimum maintenance cost min S ¼ CA k0 T þ

Xm i¼1

" CAi

2

k T2 Qim j¼1

nj

þ C i ð ni  1Þ

Ym j¼i þ 1

# nj

ð11Þ

Therefore, this equation was integrated into the calculation code algorithm in order to give it a mechanical aspect (i.e. making the program more user-friendly).

5 Results and Discussion Avoiding failures of critical data is always a top priority [17], when a few minutes of service disruption can compromise the value of a company on the market, the reliability of the elements and their availability become importantly crucial [16]. In ENAFOR (Algerian Drilling company), we noticed that equipment that causes a lot of problems in a drilling site and that often falls broken down is the Top Drive system, which causes a lot of losses for the company since it is a strategic piece of equipment [4]. In order to master these failures and break the cycle of breakdowns, we decided to conduct a study based on the ABC method (PARETO analysis), to select the most more penalizing in terms of availability, then we will perform preventive maintenance according to a predetermined schedule to increase performance ([8, 18]). To develop a plan for preventive maintenance actions for essential components constituting an equipment from their direct costs (maintenance costs) and indirect (production loss costs) by knowing their survival models (Weibull), we developed a calculation code in Visual Basic

190

I. Kessai et al.

Application (VBA) which is based essentially on the objective mathematical functions given by Eqs. (10) and (11). A. Pareto Method This method necessarily assumes that you have a history of a previous period or forecast. To apply this method, we have followed the chronological steps: 1) Definition of the objective of the study and its limits (materials, causes of breakdowns…). 2) Choose of the classification criteria. 3) Building a Pareto graph. 4) Determination of the ABC zones (in general the curve has three paces). 5) Interpretation of the curve. In the diagram of Fig. (2), there are three zones • Zone A: 20% of breakdowns are responsible of 80% of the costs. • Zone B: the 30% additional breakdowns cost only additional 15%. • Zone C: the remaining 50% of failures concern only 5% of the total cost. The graph in Fig. 2 illustrates that 80% of the cumulative costs are consumed by only 20% of failure; this will allow us to classify failures according to their order of importance based on their cost. B. Essential Subsets Equipment

Failures %

Based on the available technical data, we have considered certain elements as being essential for the operation of the Top Drive. The subsets on which the study was carried out are: Washpipe, Upper Kelly valve, Torque Boost [10], Stabing bell, Hpu (Hp Pump), Link Tilt, Back up wrench, Guide gear case, Rotary manifold, Gage-bearing [19].

Number of elements % Fig. 2. General PARETO diagram of the top drive failures [17]

Preventive Maintenance Optimization of Top Drives

191

C. Elements with High Unavailability Rate On the basis of the availability criterion, the data collected in the history files of a top drives set have been analyzed by using the PARETO method [19]; the obtained results are summarized in Table 1. Table 1. History of the elements with high unavailability rate for top drive equipments Order Equipment 1 2 3 4 5 6 7 8 9 10

Back up wrench Torque Boost Rotary Manifold Hpu (Hp Pump) Guide gear case Link Tilt Gage bearing Washpipe Upper Kelly Valve Stabing Bell

Number of interventions (n) 3 3 2 2 3 2 2 6 2 2

Technical time for reparation (T) 90 43 45 30 19 21 21 4 6 1

nT 270 129 90 60 57 42 42 24 12 2

Table 2 shows for this case that the penalizing elements in terms of availability are also the most penalizing elements in terms of maintainability (level maintenance is greater than or equal to 3), which shows that these elements can be repaired at workshop level, the results found are listed in the Table 2.

Table 2. Ranking of the elements based on their importance Order 1 2 3 4 5 6 7 8 9 10

Element Elements % Days of stop Cumulated Back up wrench 10 90 90 Rotary manifold 20 45 135 Torque Boost 30 43 178 Hpu (Hp pump) 40 30 208 Guide gear case 50 22 230 Link Tilt 60 21 251 Gage bearing 70 21 272 Upper Kelly Valve 80 6 278 Washpipe 90 4 282 Stabing Bell 100 1 283

Stop % 31.80 47.70 62.89 73.49 81.26 88.68 96.10 98.22 99.63 100

192

I. Kessai et al.

D. Graphic Representation From the data in Table 2, the Pareto curve has been drawn as given in Fig. 4; this curve represents the unavailability rate as a function of the element rate. It contains three zones determined by the change in the shape of the curve ([20, 21]).

Fig. 3. Pareto curve of the Top drive data: Unavailability rate as a function of the element rate

From Fig. 3, we can clearly see that 40% of the considered elements are responsible for 73.49% of the overall unavailability of CARNIG Top drives. We note that our study is extended to the second zone to draw the elements concerned by the study. In the graph of the figure, we see that the elements that are in the area A and B are those on which priority should be given. The elements that causes the unavailability of the CARNIG Top Drive with a rate of 73.5% are:S1 : Back up wrench, S2 : Rotary Manifold, S3 : Torque Boost, S4 : Hpu [19]. Thus, in order to improve the performance and optimize (minimize) this rate, we proposed a preventive planning maintenance for these equipments as described in the next subsection. E. Development of the Calculation Algorithm in VBA The code is created as a programming language in Visual Basic for Application. It should be noted that this code is based on an objective function given by (11). This function is used to manipulate technical and economic parameters of the given subassemblies, with: S: Maintenance summary cost, CAi : Cost of an unplanned repair, Ci : Cost of a planned repair, nj : Quantity of time periods. The introduction of data in the VBA algorithm is done through a dialog box shown in Fig. 4.

Preventive Maintenance Optimization of Top Drives

193

Fig. 4. Data input for program execution

The objective function is then written in the form of a loop going from zero to 1000. Figure 5 shows brief instructions used in our paper, in order to find the optimal cost Smin [7].

Fig. 5. Source code of the optimization algorithm

Fig. 6. Element unavailability rates

194

I. Kessai et al.

In order to use this algorithm, we have to: • Select the most penalizing elements in terms of availability from Fig. 6; • Linearize failure rates to get technical parameters Ki & Bi ; • Calculate failure costs CAi and maintenance costs Ci (on the basis of the documentation provided by ENAFOR); • Fill in the dialog box with the necessary data; • Run the VBA algorithm. After running the program, we will obtain optimal results for our study: a) The cost necessary Smin to carry out maintenance interventions is 363843,293 $. b) Intervention periods are: T1 = 180 D; T2 = T3 = T4 = 500 D. F. Proposed Preventive Maintainance Planning The proposed preventive action schedule is illustrated in Fig. (7)

0

S1

S1

180 D

360 D

S1+S2+S3+S4 500 D

S1

S1

680 D

860 D

S1+S2+S3+S4 1000 D

Fig. 7. Structure of the proposed intervention schedule

This result constitutes the first planning of systematic optimal preventive actions to offer top drives after several years of applying a practically corrective strategy, including default expenditure greater than $ 670000 (for a defined number of top drives). The implementation of this new strategy requires a preliminary investment which will allow us to lower the financial effects caused by production losses; and will allow the company to earn up to 46% of the expenses of the previous strategy. To implement this planning, the preparation department must carry out modes of the concerned sub-assemblies ðS1 ; S2 ; S3 and S4 Þ. These operating modes (preparation) must contain the chronological order of operations, the necessary tools and the time allocated to the operation. This information is necessary for the scheduling department to plan the proposed schedule over time. This planning is essential for spare parts supply managers. This schedule requires the support of drilling companies.

6 Conclusion In this study, we were able to distinguish the elements responsible for the unavailability of the Top drive in the field, the method used to find these elements is the Pareto technique. The main objective was mainly focused on the availability of sub-assemblies making up the Top Drive, we found that the elements penalizing in terms of availability are also penalizing in terms of maintainability. Therefore, we can say that the mastery of these elements allows us to control the availability and maintainability of the equipment and decrease the technical repair time. The objective function provided the possibility to apply a maintenance policy based on preventive actions planned with

Preventive Maintenance Optimization of Top Drives

195

optimal cost, this approach requires that the elements concerned by the study follow the Weibull model and their failure rates are linearized. Moreover, it is automatically calculated the optimal cost necessary to carry out preventive actions as well as their corresponding periods (T1, T2, T3 and T4). The results of this study can be considered as a technical and industrial manual that will be used in the future as maintenance reference. It is highly recommended to implement the proposed preventive schedule, which can recover up to 70% of rig availability and save up to 46% of costs. Acknowlegments. This study has been sponsored by dgrsdt (direction generale de la recherche scientifique et du developpement technologique) algiers- algeria.

References 1. Richarson, A.S., Kuttel, B.: Canrig Drilling Technology Ltd, 2000. Drilling method and apparatus. U.S. Patent 6,050,348 (1997) 2. Boone, S.G., Ellis, B., Gillan, C.J., Kuttel, B.: Canrig Drilling Technology Ltd, 2014. Automated directional drilling apparatus and methods. U.S. Patent 8,672,055 (2008) 3. De Mul, A.A., Roodenburg, J.: ITREC BV, 2018. A drilling rig with a top drive system operable in a drilling mode and a tripping mode. U.S. Patent Application 15/737,068 (2015) 4. Scott, G.B.: Drilling scorecard, US Patent US8510081B2. Nabors Drilling Technologies USA Inc (2009) 5. Doghmane, M.Z., Kidouche, M.: Decentralized controller Robustness improvement using longitudinal overlapping decomposition- Application to web winding system. Elektronika ir Elektronika 24(5), 10–18 (2018). https://doi.org/10.5755/j01.eie.24.5.21837 6. Heidecke, K., Rials, J.R., Fisher, R., Olstad, D.M.: Weatherford Technology Holdings LLC, 2016. Method of using a top drive system. U.S. Patent 9,528,326 (2012) 7. Doghmane, M.Z.: “Optimal decentralized control design with overlapping structure”, Magister Thesis. University M’hamed Bougara of Boumerdes, Algeria (2011) 8. Javanmard, H., Koraeizadeh, A.: Optimizing the preventive maintenance scheduling by genetic algorithm based on cost and reliability in National Iranian Drilling Company. J. Ind. Eng. Int. 12(4), 509–516 (2016). https://doi.org/10.1007/s40092-016-0155-9 9. Riane, R., Kidouche, M., Doghmane, M.Z., Illoul, R.: Modeling of torsional vibrations dynamic in drill-string by using PI-observer. In: Bououden, S., Chadli, M., Ziani, S., Zelinka, I. (eds.) Proceedings of the 4th International Conference on Electrical Engineering and Control Applications, ICEECA 2019. Lecture Notes in Electrical Engineering, vol. 682. Springer, Singapore (2021). https://doi.org/10.1007/978-981-15-6403-1_12 10. Riane, R., Kidouche, M., Illoul, R., Doghmane, M.Z.: Unknown resistive torque estimation of a rotary drilling system based on kalman filter. IETE J. Res. 27 Feb 2020. https://doi.org/ 10.1080/03772063.2020.1724834 11. John, P., Schroeder, E.M.: Industrial plant equipment, process and maintenance optimization. US Patent US9626634B2. ABB Schweiz AG (2012) 12. Wan, J., Tang, S., Li, D., et al.: A manufacturing big data solution for active preventive maintenance. IEEE Trans. Ind. Inform. 13(4), 2039–2047 (2017). https://doi.org/10.1109/ TII.2017.2670505 13. Zhicheng, Z., Yisha, X., Bo, Z.: Multi-component maintenance optimization: a stochastic programming approach. under 2nd round revision at INFORMS J. Comput. arXiv:1907. 00902v2 (2020)

196

I. Kessai et al.

14. Doghmane, M. Z., Kidouche, M., Habbi, H., et al.: A new decomposition strategy approach applied for a multi-stage printing system control optimization. In: 4thInternation Conference on Electrical Engineering (ICEE), Boumerdes, Algeria (2015) 15. David, E., Johnson, S., Gray, K., Uppuluri, B.: Maintenance optimization system through predictive analysis and usage intensity. US Patent US20190147413A1. GE Energy Power Conversion Technology Ltd (2017) 16. Zhang, C., Qian, Y., Dui, H., Wang, S., Shi, J.: Component failure recognition and maintenance optimization for offshore heave compensation systems based on importance measures. J. Loss Prevent. Process Ind. 63, 103996 (2020). https://doi.org/10.1016/j.jlp. 2019.103996 17. Kessai, I., Benammar, S., Doghmane, M.Z., Tee, K.F.: Drill bit deformations in rotary drilling systems under large-amplitude stick-slip vibrations. Appl. Sci. 10, 6523 (2020) 18. Fadeji, J.A., Okuwa, M.O., Mgbennena, C.O., Ezekiel, K.C.: The Pareto principle and a hazard model as tools for appropriate scheduled maintenance in a manufacturing firm. African J. Sci. Technol. Innov. Develop. 8(2), 173–177 (2016). https://doi.org/10.1080/ 20421338.2016.1147203 19. Faisal, Y., Tommy, V.: Top drive back-up wrench with thread compensation. US Patent US20190203546A1. Nabors Drilling Technologies USA Inc (2017) 20. Noghin, V.D.: Edgeworth-pareto principle. In: Reduction of the Pareto Set”. Studies in Systems, Decision and Control. vol 126. Springer, Cham (2018). https://doi. org/10.1007/978–3-319-67873-3_1 21. Beata, N.: Preventive services of residential buildings according to the pareto principle. In: IOP Conference Series: Materials Science and Engineering. vol. 471(11), p. 112034 (2019)

Stick-Slip Vibrations Control Strategy Design for Smart Rotary Drilling Systems Mohamed Z. Doghmane1(&), Abdelmoumen Bacetti2, and Madjid Kidouche1 1

Laboratory of Applied Automatic (LAA), Faculty of Hydrocarbons and Chemistry, M’hamed Bougara University, Boumerdes, Algeria [email protected] 2 Industrial Products Systems and Innovation Laboratory (IPSIL), National Polytechnic School of Oran (ENPO), Oran, Algeria

Abstract. The objective of this paper is to design new strategy using three controllers for rotary drilling systems located in Algerian oil field. Torsional vibration is one of the most challenges facing petroleum drilling technology, it creates the stick- slip phenomenon that decreases Borehole quality, increases drilling cost through additional operations, and reduces penetration rate. The controllers are designed to minimize the vibrations so that maximize penetration rate and decreases drilling operation costs through avoiding unnecessary operations such as side track. The simulations results were so promising; thus a 3D prototype of rotary drilling systems has been realized in laboratory to validate the proposed approaches. Moreover, a graphical interface has been created to facilitate the use of this strategy by drilling field supervisors. Keywords: Control strategy
Rotary drilling system Stick-Slip phenomenon
Side track
3D prototype


Torsional vibrations


1 Introduction Petroleum industry has played a major role in the Algerian economy for the last few years, wherein many researchers have focused on problematic of production costs optimization regarding the increased need of this source of energy. Since drilling system is one of the most expensive parts of petroleum production industry, therefore, a deep study of new optimization control strategy is very important for efficient production capacities increase. Rotary drilling system is designed by combining mechanical and electrical subsystems for the purpose of drilling a well, the mechanical part of the system can be simply seen as a beam rotating at constant speed and in its limits drills with drill bit [1]. The stick-slip phenomenon appears under certain circumstances; sudden change of rock hardness stops the drill bit and creates the stick phase, then the drill bit is released and starts again with speed higher than the drill string surface speed and it creates slip phase [2]. Consequently, these two phases generate torsional vibrations along the drill string. It has been noticed in the drilling field that the torsional vibrations have many effects on drill string equipment, the drilling system damage is subject to high couple created, and it also imposes very © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 197–209, 2021. https://doi.org/10.1007/978-3-030-63846-7_20

198

M. Z. Doghmane et al.

expensive extra operations to deal with the low quality of the borehole. Moreover, the performance of drilling process is decreased in term of capacity due to postponing axial advancement [3]. Many researchers have interested in decreasing the ‘Stick-slip’ phenomenon effects using different types of controllers ([4–7]). However, most of these works are based on simplified mathematical model of the rotary drilling system and no experimental validations have been realized [8]; we can find also that proportional integral (PI) controller in used in [9], H infinity controller in ([10, 11]), and sliding mode controller in [12]. On the other side, few researchers have taken the generalized nonlinear model of the system in consideration to deal with this phenomenon, wherefore; no control strategy is applied in rotary drilling system of oil field industry. For these reasons, this paper takes the objective of designing strategy using PI, PID, and Hinfinity controllers for the generalized nonlinear model of the rotary drilling system. In order to experimentally validate this strategy, a prototype of rotary drilling system has been designed in the laboratory.

2 Mathematical Model of Rotary Drilling System The mathematical model proposed in this paper is the generalized model of n elements; the special cases of this model converges to models used in [8], and in [11]. In this study we consider that the drill-string behaves as a torsional pendulum, i.e. the drill pipes are represented by torsional spring, the drill collar behaves as a rigid body and the top drive rotates at constant speed (Fig. 1). The inertial masses Jp and Jb , locally damped by dp and db , are connected one to each other by a linear spring with torsional stiffness k and torsional damping l [13]. The equations of motion can be represented as follows
   Jp €hp þ dp h_ p  h_ b þ k hp  hb þ lp h_ p ¼ uT

   : Jb €hb  db h_ p  h_ b  k hp  hb þ lh_ b ¼ Tob h_ b 8
> > > > < x_ i ¼ dJi xi1  ðdi þ diJþ 1 þ lÞ xi þ diJþ 1 xi þ 1 þ KJ i xi þ 4  KiJþ 1 xi þ 5 ; i ¼ 2; 3 i i i i i > > > > > :

x_ 4 ¼ dJ44 x3  ðd4 þJl4 þ lb Þ x4 þ KJ44 x8 ; i ¼ 4 x_ i ¼ u  x1 ; i ¼ 5 x_ i ¼ xiNs 1  xiNs ; i ¼ 6


8

ð10Þ

200

M. Z. Doghmane et al.

3 Controllers Design for General Model A state space model of the drill-string torsional behavior can be derived from (11) as 

X_ ¼ AX þ Bu þ Ed Y ¼ CX

ð11Þ

 T  T Where X ¼ h_ 1 h_ 2


h_ 2Ns is the state vector, Y ¼ h_ 1 h_ Ns is the output
 _ vector, u ¼ h1 is the top drive input velocity, ðd ¼ TobÞ is the known/measured input torque-on-bit. C and E are known matrices with appropriate dimensions [15]. 

E ¼ ð 0 0 0 1=J4 0 0 0 0 ÞT C ¼ ð 1 0 0 1 0 0 0 0 ÞT

ð12Þ

The error is added in order to achieve some desired robustness performance. 3.1

Conventional PI, PID Controllers

Conventional PI (or PID) control is an implementation of error-based feedback control; its fundamental theory is very simple. Suppose that yr ðtÞ is the set of reference input, and yðtÞ is the real output of the system. Then the error is eðtÞ ¼ yðtÞ  yr ðtÞ, and the classical PID control input is defined as follows Z uðtÞ ¼ a0

t

ðyðsÞ  yr ðsÞÞds  a1 ðyðtÞ  yr ðtÞÞ  a2

t0

d ð yð t Þ  yr ð t Þ Þ dt

ð13Þ

Where a0 ; a1 ; and a2 are the design parameters, they are called integral, proportional, and derivative gains respectively [16]. PI controller Eq. (17) can be reduced into Z uðtÞ ¼ a0

t

ðyðsÞ  yr ðsÞÞds  a1 ðyðtÞ  yr ðtÞÞ

ð14Þ

t0

3.2

Improved Ziegler-Nichols Tuning

The classical tuning rules have been widely used in industrial systems, however, the obtained control system lacks of robustness especially for system with fast dynamic [17]. For this reason, an improved Ziegler-Nichols tuning process has been chosen, the process is obtained by fitting the model P ðsÞ ¼

a0 esTd 1 þ sT

ð15Þ

The process steady state gain K is found from the steady state value of the step response [16].

Stick-Slip Vibrations Control Strategy Design

3.3

201

H Infinity Controller

The mathematical model of the system given in Eq. (10) is written as 

x_ ðtÞ ¼ Ax þ Bu yðtÞ ¼ Cx

ð16Þ

A more general state space representation of the system is given as follows 8 < x_ ðtÞ ¼ AxðtÞ þ B1 xðtÞ þ B2 W ðtÞ zðtÞ ¼ C1 xðtÞ þ D11 xðtÞ : yðtÞ ¼ C2 xðtÞ þ D21 xðtÞ

ð17Þ

Where the matrix form of Eq. (21) is given as 3 2 x_ ðtÞ A 4 zðt Þ 5 ¼ 4 C 1 C2 yð t Þ 2

B1 D11 D21

3 2 3 B2 xð t Þ 0 5
4 w ðt Þ 5 ¼ P 0 uð t Þ

ð18Þ

The solution of this H1 problem in Eq. (22) based on Riccati equations is implemented and the following conditions are verified • • • •

A; B2 is stabilisable and C2 ; A is detectable, D12 and D21 have full rank, A  jwI, B2 ; C1 ; D12 has full column rank for all wR hence,D12 is tall, A  jwI, B1 ; C2 ; D21 has full column rank for all wR hence, D21 is wide.

The expended model is produced by accounting for the weighting functions W1 ; W2 ; and W3 . To reach best control robustness, the outputs were chosen to be transfer weight functions as:z1 ¼ W1  e; z2 ¼ W2  u; z3 ¼ W3  y [18]. The cost function of mixed sensibility is given for 2

Ty1u1

3 W1 S ¼ 4 W2 R 5; S ¼ ðI þ GK Þ1 ; R ¼ K ðI þ GK Þ1 ; T ¼ GK ðI þ GK Þ1 ð19Þ W3 T

Where S and T are called sensibility and complementary sensitivity functions respectively. The transfer function from W to z1 is the weighted sensitivity function W1S , which characterizes the performance objective of good tracking; the transfer function from W to z2 is the complementary sensitivity function T, whose minimization ensures low control gains at high frequencies, and the transfer function from W to z3 is Ks , which measures the control effort. It is also used to impose the constraints on the control input [18].

202

M. Z. Doghmane et al.

(a)

(b)

Fig. 1. Rotary drilling system: a) Image of real system, b) Detailed schematic diagram

4 Results and Discussion 4.1

Simulation Results

Parameters used in this study are summarized in Table 1. Table 1. Parameters used in rotary drilling system mathematical model Parameters Dbit kb ODs IDs Lsp Nsp Ysp Gsp Psp RHOsp ODbha IDbha Lbhap Nbhap

Definition bit diameter Boussaada’s simplified model coefficient outer diameter of the string pipe inner diameter of the string pipe length of one string pipe number of string pipes Young modulus of string pipes Shear modulus of string pipes Poisson’s ratio of string pipes density of string pipes material outer diameter of the BHA inner diameter of the BHA length of one BHA pipe number of BHA pipes

Value 0.2159 m 0.3 0.127 m 0.1086 m 9.14 m 383 200*1e9 79.3*1e9 0.3 7850 0.1651 m 0.0714 m 9.14 m 22 (continued)

Stick-Slip Vibrations Control Strategy Design

203

Table 1. (continued) Parameters Ybhap Gbhap Pbhap RHObhap Nes Nebha

Definition Young modulus of BHA pipes Shear modulus of BHA pipes Poisson’s ratio of BHA pipes density of BHA pipes material number of elements of the drill string number of elements of the BHA

Value 200*1e9 79.3*1e9 0.3 7850 10 4

The parameters of the generalized model have been calculated using basic equations ([19], [20]) as follows Number of elements Ne ¼ Nes þ Nebha

ð20Þ

Les ¼ Nsp  Lsp =Nes

ð21Þ

Lebha ¼ Nbhap  Lbhap =Nebha

ð22Þ

    Depth ¼ Nsp  Lsp þ Nbhap  Lbhap

ð23Þ

  Jes ¼ p  RHOsp  Les  OD4s  ID4s =32

ð24Þ

  Jebha ¼ p  RHObhap  Lebha  OD4bha  ID4bha =32

ð25Þ

   Jeff ¼ 4  Jes = p2 þ Jebha

ð26Þ

  Kes ¼ Gsp  p  OD4bha  ID4bha =ð32  Les Þ

ð27Þ

  Kebha ¼ Gbhap  p  OD4bha  ID4bha =ð32  Lebha Þ

ð28Þ

Length of one string element

Length of one BHA element

Drilling depth

Moment of Inertia For string elements

For BHA

Stiffness coefficients vector

204

M. Z. Doghmane et al.

Mass   Mes ¼ RHOsp  Les  p  OD2s  ID2s =4   Mebha ¼ RHObhap  Lebha  p  OD2bha  ID2bha =4

ð29Þ ð30Þ

Axial Spring constant   Ses ¼ Ysp  p  OD2s  ID2s =ð4  Les Þ

ð31Þ

  Sebha ¼ Ybhap  p  OD2bha  ID2bha =ð4  Lebha Þ

ð32Þ

Dynamic Viscosity coefficient ETAes ¼ ðSes  Jes =ðMes  Kes ÞÞ1=2

ð33Þ

ETAebha ¼ ðSebha  Jebha =ðMebha  Kebha ÞÞ1=2

ð34Þ

Damping coefficients vector   des ¼ 30  Les  D2bit  OD2s  ETAes = D2bit  OD2s

ð35Þ

  debha ¼ 30  Lebha  D2bit  OD2bha  ETAebha = D2bit  OD2bha

ð36Þ

Torque on bit
  
1=2 


     1= x2Ns þ O20 Tob ¼ð2  kb  xNs Þ= kb2 þ x2Ns þ þ ðp  O0 Þ= x2Ns þ O2O  ððxNs Þ=O1  1Þ

ð37Þ Simulation results of the designed controller are show in Fig. 2

Open-Loop/ Closed loop Responses to Zero Initial Condition 300

u Open loop Responses PI controller PID controller H infinity controller

Drill bit Velocity (rpm)

250

200

150

100

50

0 0

50

100

150

200

250

300

350

400

Time (s)

Fig. 2. Open loop, closed loop responses for PI, PID, and Hinfinity controllers – linearized model

Stick-Slip Vibrations Control Strategy Design

4.2

205

Experimental Results of the Drilling System Prototype

A graphical user interface has been developed (Fig. 3) and implemented with DSpace interface so that it can facilitate the controllers manipulations of prototype system.

Fig. 3. Graphical user interface for the developed control strategy

The dimensions of rotary drilling shown in Fig. 4. A have been reduced by keeping the proportionality between them (Fig. 4.B).

(a)

(b)

Fig. 4. Rotary drilling system; a) Image of realized prototype, b) Detailed schematic diagram

206

M. Z. Doghmane et al.

The torque on bit of the prototype has been given in Fig. 5; it is stabilized at 8 T after 160 s. Torque on bit response 9

8

7

Torque on Bit

6

5

4

3

2

1

0 0

20

40

60

80

100

120

140

160

180

200

Time (s)

Fig. 5. Torque on bit response of the nonlinear model of the system

The open loop responses for five elements model are shown in Fig. 6 where it is noticed that this responses are stabilized after 160 s also, the torsional vibrations have affected the velocity responses in the interval [0 160 s]. It is also noticed that element 3 has the highest velocity response value (Fig. 6), where it has the most remarkable vibration effect, it can be interpreted as the most risky point of drill string damage. Open-Loop Velocities

Angular Velocity (RPM)

150

100

Set Point Top Drive Element 1 Element 2 Element 3 Drill Bit

50

0

0

20

40

60

80

100

120

140

160

180

200

Time (s)

Fig. 6. Five elements responses of the nonlinear model of the system

In order to generalize the designed algorithm, many input references have been used (Figs. 7, 8, and 9); these signals can represent any type of input signals in oil field drilling system. As it is shown in Figs. 7, 8, and 9 the controller have tracked different types of input signals, where it is noticed that PI and PID controllers give the same tracking performances. However, Hinfinity controller offered a better tracking performance with better minimization of torsional vibration and time response, it can be considered as the best solution in the status of stuck or slip phases, however it can

Stick-Slip Vibrations Control Strategy Design

207

reduce the axial advancement, thus a switch between PI (unfavorably derivative term) and Hinfinity controllers will increase the efficiency of control tasks; PI controller in normal situation without any vibrations, as soon as there is stuck of drill bit automatic switch to Hinfinity controller is done, when the drill bit speed return near to the drill string surface speed an automatic switch back to PI controller is done and so on. Open-Loop/ Closed loop Responses to Zero Initial Condition 350

u Open loop Responses PI controller PID controller H infinity controller

300

Drill Bit Velocity (rpm)

250 200 150 100 50 0 -50

-100

0

50

100

150

200

250

300

350

400

Time (s)

Fig. 7. Response of the nonlinear system with second type of top drive velocity input

Open-Loop/ Closed loop Responses to Zero Initial Condition 350 u Open loop Responses PI controller PID controller H infinity controller

300

Drill Bit Velocity (rpm)

250 200 150 100 50 0 -50

-100 0

50

100

150

200

250

300

350

400

Time (s)

Fig. 8. Result of nonlinear system with third type of top drive velocity input

Open-Loop/ Closed loop Responses to Zero Initial Condition

350

u Open loop Responses PI controller PID controller H infinity controller

300

Drill Bit Velocity (rpm)

250 200 150 100 50 0 -50

-100 0

50

100

150

200

250

300

350

Time (s)

Fig. 9. Result of the nonlinear system with fourth type of top drive velocity input

400

208

M. Z. Doghmane et al.

5 Conclusion This study dealt with the design of control strategy for the purpose of controlling torsional vibrations in rotary drilling system, the strategy is based on PI, PID, and Hinfinity controllers’ responses; it combines these controllers to have desired maximum system performance with minimum drill string vibrations. The promising simulation results guided us to design, experimentally, a prototype of rotary drilling system, it gave the opportunity to validate the proposed algorithms, thus propose the implement of the developed strategy to an operating Algerian drilling system so that extra drilling cost can be avoided.

References 1. Saldivar, B.: Analysis, modeling and control of an oil well drilling system. Cotutelle Ph.D. Thesis. Centro de Investigación y de Estudios Avanzados del Instituto Politécnico Nacional, Mexico. Institut de Recherche en Communications et Cybernétique de Nantes, France (2013) 2. Lear, W., Dareing, D.: Effect of drill string vibrations on MWD pressure pulse signals. J. Energy Res. Technol. 112(2), 84–89 (1990) 3. Saldivar, B., Mondié, S.: Drilling vibration reduction via attractive ellipsoid method. J. Franklin Institute Elsevier 350(3), 485–502 (2013) 4. Zhu, X., Tang, L., Yang, Q.: A literature review of approaches for stick-slip vibration suppression in oil well drill string. Adv. Mech. Eng. 6, 967952 (2014) 5. Karkoub, M., Abdel-Magid, Y.L., Balachandran, B.: Drill string torsional vibration suppression using GA optimized controllers. J. Canad. Petroleum Technol. 48(12), 32–38 (2009) 6. Puebla, H., Alvarez-Ramirez, J.: Suppression of Stick-Slip in Drill strings: A Control Approach Based on Modeling Error Compensation. J. Sound Vib. 310(4–5), 881–901 (2008) 7. Richard, T., Germay, C., Detournay, E.: A simplified model to explore the root cause of stick slip vibrations in drilling systems with drag bits. J. Sound Vib. 305, 432–456 (2007) 8. Fubin, S., Linxiu, S., Lin, L., Qizhi, Z.: Adaptive PID control of rotary drilling system with stick slip oscillation. In: The 2nd International Conference on Signal Processing Systems, ICSPS 2010. pp. 289–292 (2010) 9. Serrarens, A.F.A., van de Molengraft, M.J.G., Kok, J.J., van den Steen, L.: H∞ Control for Supressing Stick-Slip in Oil Well Drillstrings. IEEE Control Syst. Magaz. 18(2), 19–30 (1998) 10. Hernández-Suárez, R., Puebla, H., Aguilar-López, R., Hernández-Martínez, E.: An integral high-order sliding mode control approach for stick-slip suppression in oil drill strings. Petrol. Sci. Technol. 27(8), 788–800 (2009) 11. Liao, C., Balachandran, B., Karkoub, M., Abdel-Magid, Y.L.: Drill-string dynamics: reduced-order models and experimental studies. J. Vib. Acoust. 133(4), 8 (2011). 041008 12. Kessai, I., Benammar, S., Doghmane, M.Z., Tee, K.F.: Drill bit deformations in rotary drilling systems under large-amplitude stick-slip vibrations. Appl. Sci. 10, 6523 (2020) 13. Li, Z., Gaob, H., Agarwalc, R., Kaynakd, O.: H∞ control of switched delayed systems with average dwell time. Int. J. Control 86(12), 2146–2158 (2013) 14. Germay, C., Van De Wouw, N., Nijmeijer, H., Sepulchre, R.: Nonlinear drilling dynamics analysis. SIAM J. Appl. Dynamical Syst. 8(2), 527–553 (2005)

Stick-Slip Vibrations Control Strategy Design

209

15. Navarro-López, E.: An alternative characterization of bit-sticking phenomena in a multidegree-of-freedom controlled drill string. Nonlinear Anal. Real World Appl. 10(5), 3162– 3174 (2009) 16. Riane, R., Kidouche, M., Illoul, R., Doghmane M. Z.: Unknown Resistive torque estimation of a rotary drilling system based on kalman filter. IETE J. Res. (2020) https://doi.org/10. 1080/03772063.2020.1724834 17. Doghmane, M. Z., Kidouche, M., Habbi, H., et al.: A new decomposition strategy approach applied for a multi-stage printing system control optimization. In: 4th International Conference on Electrical Engineering (ICEE), Boumerdes, Algeria, (2015) 18. Mendil, C., Kidouche, M., Doghmane, M. Z.: Automatic control of a heat exchanger in a nuclear power station: The classical and the fuzzy methods. In: IEEE International Conference on Advanced Electrical Engineering (ICAEE), 19–21 Nov. 2019, AlgiersAlgeria (2019) https://doi.org/10.1109/icaee47123.2019.9014661 19. Doghmane, M.Z., Kidouche, M.: Decentralized controller Robustness improvement using longitudinal overlapping decomposition- Application to web winding system. Elektronika ir Elektronika 24(5), 10–18 (2018). https://doi.org/10.5755/j01.eie.24.5.21837 20. Riane, R., Kidouche, M., Doghmane, M.Z., Illoul, R.: Modeling of torsional vibrations dynamic in drill-string by using PI-observer. In: Bououden, S., Chadli, M., Ziani, S., Zelinka, I. (eds.) Proceedings of the 4th International Conference on Electrical Engineering and Control Applications. ICEECA 2019. Lecture Notes in Electrical Engineering, vol. 682. Springer, Singapore (2021). https://doi.org/10.1007/978-981-15-6403-1_12

Comparison Study Between Improved JAYA and Particle Swarm Optimization PSO Algorithms for Parameter Extraction of Photovoltaic Module Based on Experimental Test Y. Rehouma1(&), Z. Tir2, A. Gacem2, F. Rehouma2, M. A. Hamida1, and A. Gougui1 1 LAGE Laboratory, University of Ouargla, 30000 Ouargla, Algeria {rehouma.youssef,hamida.assaad}@univ-ouargla.dz, [email protected] 2 LEVRES Laboratory, University of El Oued, 39000, El Oued, Algeria {tir-zoheir,gacem-abdelmalek, rehouma-ferhat}@univ-eloued.dz

Abstract. To enhance of the photovoltaic systems, use of an exact model for photovoltaic module is necessary. Parameter extraction of the equivalent circuit of photovoltaic module represents a challenge that was surmounted by means of metaheuristic optimization algorithms. One of the most used algorithms is the particle swarm optimization (PSO) algorithm which attracts an important attention due to its rapid response and structure simplicity. This paper employs recent optimization techniques called JAYA to compare its performance with well-known PSO algorithm to identify the exact parameters for different photovoltaic models (DDM and DDMM). Based on real data, the performance of used algorithms is verified. The simulation results illustrate the better performance of improved JAYA than the classical PSO algorithm. Keywords: Improved JAYA algorithm identification
Solar modules


PSO algorithm
Parameter

1 Introduction In the face of the energy challenges, the exploit of solar energy is considered as better solution to classical energy generation sources. In fact, such kind of energy is a clean energy resource [1]. The exploitation of solar energy through the photovoltaic (PV) technology has attracted the interest of academic researchers [2]. Actually, the maximum extraction of solar energy via PV modules is the required task. The PV solar modules are mathematically presented under nonlinear model. In the literature, the most proposed PV models are provided by using single and double diode models [3]. The first one is usually employed due to its guarantees of a good compromise between simplicity and precision. However, the second one gives more precision although it requires computation time. To obtain an accurate model, the precise estimation of PV © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 210–221, 2021. https://doi.org/10.1007/978-3-030-63846-7_21

Comparison Study Between Improved JAYA and PSO

211

model parameters is very important task [1]. In this context, the different approaches have been proposed to extract of PV panels models parameters. Among the proposed methods, the numerical approach which is based on nonlinear optimization algorithms whatever the climatic conditions. These optimization techniques are classified under two principal categories: iterative and metaheuristic optimization techniques. To avoid the drawbacks of conventional (iterative) technique, metaheuristic optimization methods have shown great interest in PV modules parameter extraction issues [1]. Among the optimization algorithms are usually used; the Pattern Search (PS) [4], the Genetic Algorithm (GA) [5], the Bird Mating (BM) [6], the Artificial Bee Swarm Optimization (ABSO) [7], the Simulated Annealing (SA) [8], the PSO algorithm [9], and the improved JAYA [10]. In this work, an improved JAYAO algorithm is compared with PSO for parameters extraction using different type of PV modules on the basis of two equivalent circuits. The experimental results demonstrate that IJAYA provides good performance compared to the PSO algorithm. The rest of the paper is structured as follows: In the next Section, the problem is stated. Optimization techniques are recalled in Sect. 3. In Sect. 4, the results of simulation tests and experimental verification are presented. Finally, conclusion is done in Sect. 5.

2 Problem Formulation Among the frequently used models for PV modules we can find the single SDM and double diode models DDM [1]. 2.1

Single Diode Model

The SDM model of PV cell is: I ¼ Iph  Id  Ip

ð1Þ

From (1), the SDM is rewritten as:
V þ Rs:I  V þ Rs*I I ¼ Iph  Is1 eð n1Vth Þ  1  Rp

ð2Þ

Where, Rs and Rp represents the series and parallel resistances. The PV cell current and voltage are represented by I and V, respectively.

Fig. 1. Equivalent-circuit of SDM PV cell

212

Y. Rehouma et al.

The SDM equivalent circuit is shown by Fig. 1. 2.2

Double Diode Model

The DDM is given as: I ¼ Iph  Id1  Id2  Ip

ð3Þ

Where
Id1 ¼ Is1 e
Id2 ¼ Is2 e Ip ¼

 V þ RsI n1:Vth

! 1

 V þ RsI n2:Vth

ð4Þ !

1

V þ Rs*I Rp

ð5Þ ð6Þ

with Iph: photo-generated current, depending on solar, Is1and Is1: the dark diffusion and saturation currents, Ip: parallel resistor current (A), T: PV cell temperature (C), q: electron charge (C),

Fig. 2. Circuit equivalent of DDM for a cell

k: Boltzmann constant, Vth ¼ KT q : junction thermal voltage. 2.3

PV Module Model

The double diode PV module model is composed by numerous connected cells. The current-voltage relationship is formulated by:

Comparison Study Between Improved JAYA and PSO


V þ Rs:IM  V þ Rs*I*M V þ Rs:IM ð Þ ð Þ n1MVth n2MVth  1  Is2 e 1  I ¼ Iph  Is1 e Rp*M


213

ð7Þ

From this relationship, it can be seen that seven parameters should be identified. 2.4

Objective Function

For parameters estimation of PV cells, it is essential to define an objective function. Then, the role of the optimization algorithm is minimizing the error between the simulated and measured currents. For that the following function is used to realize this task: vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi u N u X ðI  IsÞ2 RMSE ¼ t½  N k¼1

ð8Þ

Where, N: represents the real data size, I and Is: represent measured and identified currents, respectively.

3 Optimization Methods 3.1

PSO Method

PSO is one of the most known optimization algorithm based on the animal swarms social behaviors [11]. A PSO technique performs search using particles population. During the search process, the searching velocity is updated Vjk ðt þ 1Þ to be used to determine the following position Xjk ðt þ 1Þ. The current position Xjk ðtÞ and velocity of each particle Vjk ðtÞ can be modified by the following equations:     Vjk ðt þ 1Þ ¼ w:Vjk ðtÞ þ C1 :r1 Xpbest;jk  Xjk ðtÞ þ C2 :r2 Xgbest;k  Xjk ðtÞ

ð9Þ

Xjk ðt þ 1Þ ¼ Xjk ðtÞ þ Vjk ðt þ 1Þ

ð10Þ

Where,

214

Y. Rehouma et al. Table 1. Pseudo-Code Of PSO

r1 and r2 : represent random numbers, C1 and C2 : cognitive and social acceleration constant respectively, w: Inertia weight factor. The pseudo-code of PSO is set in Table 1. 3.2

Improved JAYA Optimization Algorithm

IJAYA is a recent technique proposed by R. Venkata Rao in (For more details See) [12]. For ith iteration, the design variables are m numbers (i.e.j ¼ 1; 2; . . .; m) and n number of candidate solutions which gives the population size k ¼ 1; 2; . . .; n. Amongst entire candidate solutions, the best candidate obtains the best value of f ðxÞ (i.e. say f ðxÞbest ) and the worst candidate obtains the worst value of f ðxÞ (i.e. say f ðxÞworst ). If Xj;k;i is the value of the jth variable for the kth member of a set of possible solution during the ith iteration, then this value is modified as per the following Eq. (11):     0 Xj;k;i ¼ Xj;k;i þ :r1;j;i Xj;best;i  Xj;k;i  r2;j;i Xj;worst;i  Xj;k;i

ð11Þ

Where, Xj;best;i is the value of the variable j for the best candidate and Xj;worst;i is 0 the value of the variable j for the worst member of a set of possible solution. Xj;k;i is the

Comparison Study Between Improved JAYA and PSO

215

updated value of Xj;k;i . Finally, after iteration, all the accepted function values are used in the following iteration [12]. The IJAYA presents some merits compared to the old one, such that the selfadaptive weight employed to avoid the worst solution. Moreover, the learning strategy is modified in order to maintain good population diversity; this strategy is based on chaotic method. The final solution using IJAYA can be obtained by using the following two equations:

Table 2. Pseudo-Code of the IJAYA [10]

Zm ¼ 4:Zm ð1  Zm Þ

ð12Þ

Xbest;j ¼ Xbest;j þ ranð2:Zm  1Þ

ð13Þ

0

216

Y. Rehouma et al.

The pseudo-code of IJAYA is given in Table 2.

Fig. 3. Experimental prototype.

4 Results

Table 3. Parameter search range for three diode model and PV module double diode. Parameter

Double Lower Iph(A) 0 Is1,Is2(uA) 0 Rs(Ω) 0 Rp(Ω) 0 n1, n2 1

diode Upper 1 1 0.5 100 2

Module PV Lower Upper 0 1.5 0 50 0 2.5 0 1500 1 2.5

Table 4. The estimated parameters cells of each algorithm Parameter Algorithm PSO Iph(A) 0.76104 Is1(uA) 0.20327 Rs(Ω) 0.03275 Rp(Ω) 74.69531 n1 1.62019 Is2(uA) 0.55074 n2 1.55958 RMSE 0.0052

IJAYA 0.76008 0.17885 0.03690 61.489214 1.71104 0.22692 1.45402 0.0048

Comparison Study Between Improved JAYA and PSO

217

Fig. 4. Extraction Results of IJAYA and PSO Algorithms: (a) convergence for double diode model cell. (b) Comparison with two algorithm simulated IAE_V curves, Individual Absolute Error (IAE).

Fig. 5. Extraction Results of IJAYA Algorithm for Comparison between measured and simulated (a) Current-Voltage (b) Power-Voltage

Fig. 6. Identification Results of PSO Algorithm for Comparison between measured and simulated (a) Current-Voltage (b) Power-Voltage.

218

Y. Rehouma et al.

Fig. 7. Optimal estimated parameters cells of each algorithm with limit.

The presented two optimization algorithms are used in this section for PV cell parameters estimation considered two cases. The DDM based on the experimental data [10] is firstly verified. Then, a solar panel analyzer type VA 200 is used to generate 63 points to illustrate the current voltage evolution as illustrated in Fig. 3, a thermal sensor is also installed firmly on the back surface of PV panel to measure the temperature of cell and a pyranometer is used to measure the solar radiation. The parameters search ranges are given in Table 3 Table 5. The estimated parameters Module double diode of each algorithm Parameter Algorithm PSO IJAYA Iph(A) 0.65510 0.65948 Is1(uA) 10.71418 42.09201 Rs(Ω) 0.042046 0.03169 Rsh(Ω) 1618.90935 1044.88941 N1 2.35029 2.41093 Is2(uA) 15.20219 19.87926 N2 2.07097 2.24268 RMSE 0.0072 0.0067

Case 1: DDM cell The estimated parameters obtained by IJAYA and PSO are shown in Table 4. On this table, one can see that that IJAYA is a value of 0.0048 while PSO is 0.0052.

Comparison Study Between Improved JAYA and PSO

219

Fig. 8. Identification Results of IJAYA Algorithm (module Double diode IJAYA): Experimental and Estimated I_V curve, and IAE_curves

Fig. 9. Identification Results of PSO Algorithm (module two diode PSO): Experimental and Estimated I_V curve, IAE_curves.

220

Y. Rehouma et al.

Fig. 10. Convergence of two employed optimization techniques for DDM.

The I-V and P-V characteristics curve resulted from the extracted seven parameters cells by PSO and IJAYA algorithms along with the measured points have been illustrated in Fig. 5 and Fig. 6. Case 2:DDM module The same identification procedure was adopted under solar radiation of E = 431w/m2 and a temperature of T = 24C° with identification parameters listed in Table 3. Table 5 presents the extracted PV cell parameters the convergence error obtained by the two employed algorithms. On this table, one can see that that IJAYA is a value of 0.0067 while PSO is 0.0072. From Fig. 4(a)-10, it becomes clear that the IJAYA converges quickly than the PSO, especially in the two-diode model and the two-cell model.

5 Conclusion This paper presents an enhanced method of parameter extraction for PV panel based on an improved JAYA optimization algorithm. Two equivalent circuits are employed for this porpose. The implementation of IJAYA algorithm is very simple compared to other competitive algorithms. To prove that, the IJAYA is compared with PSO algorithm. Experimental results demonstrate that IJAYA provides good performance in terms of reliability and accuracy.

References 1. Merchaoui, M., Sakly, A., Mimouni, M.F.: Particle swarm optimisation with adaptive mutation strategy for photovoltaic solar cell/module parameter extraction. Energy Convers. Manag. 175, 151–163 (2018) 2. Mekhilef, S., Saidur, R., Safari, A.: A review on solar energy use in industries. Renew. Sust. Energ. Rev. 15, 1777–1790 (2011)

Comparison Study Between Improved JAYA and PSO

221

3. Ma, J., Bi, Z., Ting, T.O., Hao, S., Hao, W.: Comparative performance on photovoltaic model parameter identification via bio-inspired algorithms. Sol. Energy 132, 606–616 (2016) 4. AlHajri, M.F., El-Naggar, K.M., AlRashidi, M.R., Al-Othman, A.K.: Optimal extraction of solar cell parameters using pattern search. Renew. Energy 44, 238–245 (2012) 5. Ismail, M.S., Moghavvemi, M., Mahlia, T.M.I.: Characterization of PV panel and global optimization of its model parameters using genetic algorithm. Energy Convers. Manag. 73, 10–25 (2013) 6. Askarzadeh, A., Rezazadeh, A.: Extraction of maximum power point in solar cells using bird mating optimizer-based parameters identification approach. Sol. Energy 90, 123–133 (2013) 7. Askarzadeh, A., Rezazadeh, A.: Artificial bee swarm optimization algorithm for parameters identification of solar cell models. Appl. Energy 102, 943–949 (2013) 8. El-Naggar, K.M., AlRashidi, M.R., AlHajri, M.F., Al-Othman, A.K.: Simulated annealing algorithm for photovoltaic parameters identification. Sol. Energy 86, 266–274 (2012) 9. Macabebe, E.Q.B., Sheppard, C.J., van Dyk, E.E.: Parameter extraction from I–V characteristics of PV devices. Sol. Energy 85, 12–18 (2011) 10. Yu, K., Liang, J.J., Qu, B.Y., Chen, X., Wang, H.: Parameters identification of photovoltaic models using an improved JAYA optimization algorithm. Energy Convers. Manag. 150, 742–753 (2017) 11. Kim, D.H., Abraham, A., Hirota, K.: Hybrid genetic: particle swarm optimization algorithm. In: Abraham, A., Grosan, C., Ishibuchi, H., (eds.) Hybrid Evolutionary Algorithms, pp. 147– 170. Springer Berlin Heidelberg (2007) 12. Rao, R.V., Rai, D.P., Balic, J.: Surface grinding process optimization using jaya algorithm. In: Computational Intelligence in Data Mining. vol. 2, pp. 487–495. New Delhi (2016)

Performance Improvement of IFOC Based on Particle Swarm Optimization Algorithm in Induction Motor Control S. E. Rezgui(&), H. Benalla, and B. Nemmouchi Electrotechnic Department, Faculty of Technology Sciences, Frères Mentouri Constantin 1 University, Constantine, Algeria [email protected] Abstract. In this paper, a PSO algorithm is used to tune simultaneously the parameters of all proportional-integral (PI) controllers included in inner current control loops and outer speed loop for indirect rotor field-oriented control (IFOC) of induction motor (IM) wich is persistently used in renewable energy. The approach consists of tuning the PI parameters that meet high dynamics and tracking behavior using the Simulink model which takes in consideration all the IFOC dynamics. The performances are evaluated using the integral time squared error (ITSE) plus the overshoot term due to its better results compared to other tested usual performance indices. The presented technique has shown superior results compared to conventional calculated PI method. Keywords: Particle swarm optimization
IFOC
Induction motor
PI tuning

1 Introduction In renewable energies, a great number of control methods follow certain laws and design hypothesis which gives for each their own degree of complexity and accurateness. The classical or conventional methods require a mathematical model to control the system (like; Bode method, linear quadratic regulator LQR, Kessler, etc…), but as is well known, the model of the induction motor is nonlinear and strongly coupled, which leads the designer of the controller to make simplifying assumptions in order to reduce the computational effort, but this maneuver sometimes results in a loss of the high performance of the system response. Also, although empirical methods such as of ZieglerNichols seem good choices, but this is preferable only in cases where the model of the process is not fully known and the system allows bringing it to its limits of stability. The combination of vector control integrated in variable speed drives and the asynchronous machine, offers an ability to meet the specific needs in several renewable energies applications (wind, solar, or hybrid); as in [1], the paper dealt with a squirrel cage induction generator connected to the grid through a back-to-back converter driven by vector control. Or as in [2] where the authors have described the system and control structures for vector controlled induction generators used for variable speed wind energy conversion (WEC) systems. Or in solar based indirect vector controlled induction motor drive for water pumping systems as in [3, 4]. Some researches dealt with a control of an isolated microgrid fed from wind and solar based hybrid energy sources as in [5]. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 222–232, 2021. https://doi.org/10.1007/978-3-030-63846-7_22

Performance Improvement of IFOC

223

A very promising compromise has occurred in recent decades, which is the introduction of the meta-heuristic optimization techniques in the field of control. These are algorithms for finding optimal solutions in complex searching spaces. Among them, those who imitate human biological behavior to approach or have an optimal solution in control such as; the genetic algorithm which is based on the theory of evolution and natural selection (Darwinian/Mendelian) resulting from the best individual (survivor) [6, 7], or the mechanism of processing of emotions in the brain, which is essentially based on a selection of action based on sensory inputs and emotional signals, the regulators proposed in [8] are called intelligent controllers based on emotional brain learning, where they have been used in flux and speed control and tested experimentally via an FPGA card. There has also been a growing interest in engineering applications in using the collective behaviors of colonies of social insects, as well as other animal societies. They use different type of strategies to search for and obtain nutrients so that maximizes the proportion between the energy obtained and the time spent foraging. Somme popular methods are Ant (ACO) and Bees (BCO) colonies algorithms, and bacteria foraging algorithm [9–11]. More specifically, particle swarm optimization (PSO) which was firstly introduced by [12], has aroused more attention thanks to its simplicity and flexibility. Through contributions of individuals in a population of particles (fish school or bird flocks), it uses mainly two iterative equations of the particles; one for the positions and the other for the velocities. There are number of recent publications found in IM drives. In [13], a PSO algorithm based output feedback control to determine the controller gains for vector control scheme is proposed, the method was stable and showed good transients behavior. The authors of [14] introduced a PSO approach for the determination of the PI controller gains in a field oriented control (FOC) chain. But only one speed controller was used. The authors of [15] proposed a novel method to eliminate harmonics in a solar powered CHMLI (cascaded H-bridge multilevel inverter). The Newton Raphson (NR) and PSO based selective harmonic elimination algorithm were developed for nineteen- level inverter. The results depicted that the proposed method can eliminate the deadliest lower order harmonics, and thereafter, dramatically decrease the THD (total harmonic distortion) of the output voltage of a CHMLI with induction motor drive. In this work the PSO algorithm is designed to achieve an extensive PI parameters optimization in the search space. The approach of the proposed method consists in the fact that all the parameters of the PI controllers are optimized using the Simulink model, which takes in consideration all the IFOC dynamics including the model of the switches and the nonlinear parts. After manipulation with multiple familiar performance indices (ITAE, ISE, IAE, ITSE), we have opted for the integral time squared error (ITSE) plus the term of overshoot because of its most better results. The paper is organized as follows: in Sect. 2, model of the IM and the fundamentals of the IFOC are presented. We outline the design of control strategy with the tuning scheme based on the PSO technique in Sect. 3. The results of simulations and discussions for the tracking trajectory and the performance indices of both methods (PI, and PSO) are given in Sect. 4. Finally, in Sect. 5, some end remarks are presented in the conclusion.

224

S. E. Rezgui et al.

2 IM Model and IFOC Fundamentals The aim of the indirect field oriented vector control is to equalize the behavior and simplicity obtainable by the separately excited DC machine control, where the decoupling between the quantities controlling the flux and the torque is intrinsically achieved. 2.1

Induction Motor Model

The equations translating the dynamic behavior of the electric and electromagnetic modes of induction machine are expressed in the two-phase rotating Park reference (d, q) in terms of stator currents isd, isq, rotor flux wrd, wrd, and stator voltages vsd, vsq as follows: 3 2  1  1r isd rTs rTr 6 isq 7 6 xs 6 6 7 ¼ s4 Lm wrd 5 6 4 Tr wrq 0 2

xs  1r rTr 0

 rT1 s

Lm Tr

1r  rT r Lm 1r rLm xr  T1r xsl

1r rLm ðxs  1r rTr Lm

xsl  T1r

xsl Þ

3

2 3 isd 7 6 1 i 7 sq 7 7 7 þ 6 4 wrd 5 þ rLs 5 wrq

2

1 60 6 40 0

3 0   17 7 vsd : 5 vsq 0 0

ð1Þ   Where, xsl is the slip frequency, r ¼ 1  L2m =Ls Lr is the total leakage factor. xs is the synchronous rotating angular speed. xr is the electrical angular speed of the rotor. Tr ¼ Lr =Rr , Ts ¼ Ls =Rs are the rotor and stator time constants respectively. The electromagnetic torque Te is expressed as: Te ¼ p

 3 Lm  wrd isq  wrq isd 2 Lr

ð2Þ

Where; p is pole pair number. The motional equation of IM is described as: JX_ ¼ Te  TL  f v X

ð3Þ

Where J is the moment of inertia, TL is the load torque, fv is the viscous friction coefficient, and X=xr/p is the mechanical rotor speed, and ð
Þ is the derivative. 2.2

Indirect Rotor Field Oriented Vector Control Fundamentals

The implementation of the indirect rotor field oriented vector control is based on the orientation of the rotating coordinate system of axes d, q, such that the d axis coincides with the direction of wrd. This leads in a decoupling of the variables so that the torque and the flux can be separately controlled by quadrature-axis current isq, and stator

Performance Improvement of IFOC

225

direct-axis current isd respectively (as in separately excited DC machine). Thus, the electromagnetic torque, the synchronous angular speed, and the position can be expressed as: Te ¼ Kt ¼ pð3=2ÞðLm=LrÞwrd ¼ Kt isq :

ð4Þ

xs ¼ pX þ Lm Rr isq =Lr wrd :

ð5Þ

Z

Z

hs ¼ xs :dt ¼ ðxr þ xsl Þdt:

ð6Þ

All others IM parameters are shown in Table 1.

Fig. 1. Scheme of indirect rotor field oriented vector control using PSO algorithm.

The Fig. 1 shows the block diagram of IFOC with tuning PI controllers using PSO. esd and esq are the d and q axis decoupling terms respectively. Note that an IP structure has been used in the speed regulation loop. It has been proven that an IP regulator is equivalent to a PI regulator with a first-order filter at the input, which considerably reduces overshoots [16]. Finally the overall system of equation becomes:   8 disd 1 1r 1r 1 > ¼   > rTs rTr isd þ xs isq þ rTr Lm wrd þ rLs vsd < dt disq 1 1r 1r 1 dt ¼ xs isd  rTs  rTr isq  rLm xr wrd þ rLs vsq > > : dwrd ¼ Lm i  1 w dt

Tr sd

Tr

rd

ð7Þ

226

S. E. Rezgui et al.

3 Im Model PSO Technique Description The PSO technique utilizes concepts borrowed from the field of social psychology. It uses a swarm of particles that communicate with each other to determine the optimal behaviour to solve a problem. Thus, the contribution to the solution of a member of a group is due to the development of a cognitive coherence acquired by personal experience and the influence of collective knowledge of the social group. 3.1

Basic Concepts of PSO

At each step (k), each particle moves in a way so it reaches a better local solution (evaluated by a fitness), it remembers the position where it achieved the best value ðkÞ compared to previous searches. This is called the individual best position (PiD ). In addition, the group keeps track of the position where the bestvalue of the whole swarm was achieved, and what it’s called global best position

PðgkÞ . The position and ðkÞ

velocity of the ith particle in the kth iteration are denoted by vector of D-dimension XiD ðkÞ

and ViD respectively. And PSO algorithm is performed on the basis of the following two iterative equations [17]: ðk þ 1Þ

ViD

    ðkÞ ðkÞ ðkÞ ðkÞ ðkÞ ¼ wViD þ C1 r1 PiD  XiD þ C2 r2 PgD  XiD ð k þ 1Þ

XiD

ðkÞ

ð k þ 1Þ

¼ XiD þ ViD

ð8Þ ð9Þ

Where; w denotes the inertia weight, it’s used to provide a balance between global and local search, thus requiring less iteration on average to find a sufficiently optimal solution, it’s set according to (10), and decreases linearly from about 0.9 to 0.4 [18]. C1 and C2 are respectively cognitive and social parameters. The coefficients r1, r2 are random numbers belonging 0 and 1. w ¼ wmax  Iter:ðwmax  wmin Þ=Itermax

ð10Þ

Where Itermax is the maximum number of iterations, and Iter is the current iteration. In the right hand side of (8), the second segment refers to the cognitive part, and it ðkÞ ðkÞ represents the distance between the particle XiD and its best located solution PiD . While the third segment represents the social component, which reflects the distance ðkÞ between the same particle and the global best solution PgD . Once the velocity vector is updated, based on the individual history and collective experience of the swarm, the particle moves to a new position through (9). This procedure continues until the best solution is reached or the algorithm meets user-defined stopping criteria.

Performance Improvement of IFOC

227

The basic algorithm for PSO optimization is given by the following steps: ð0Þ

ð0Þ

• Step 1: Initialize: XiD ; ViD (set random values at k = 0). ðkÞ

• Step 2: Evaluate the fitness f ðXiD Þ. Update best local position and best global position from (11) and (12) respectively: ð k þ 1Þ

PiD

  8 < PðiDkÞ iff ðXðiDk þ 1Þ Þ  f PðiDkÞ   ¼ : Xðk þ 1Þ iff ðXðk þ 1Þ Þ\f PðkÞ iD iD iD   ðkÞ ðk þ 1Þ PgD ¼ arg min f PiD i;D

ð k þ 1Þ

ð11Þ

ð12Þ

ð k þ 1Þ

and ViD using (8) and (9). • Step 3: Update XiD • Step 4: If user-defined stopping criteria are met, stop. Otherwise, set k +1 and go to Step 2. 3.2

Setting the Controllers Parameters in IFOC

To apply the PSO method for searching optimum controllers’ parameters, we have used the three IFOC’s compensators; one in the outer speed loop (IP), and two PI in the inner current loop. Hence, the dimension of the searching space is six since we have three couple of KP and Ki parameters. In order to emphasize the effectiveness of the PI-PSO controller, we have compared the results with those obtained in simulated IFOC with conventional PI parameters derived by the standard calculations’ form. The calculations gave: Kpw = 3, Kiw = 13 for the speed IP parameters, and Kpi = 14.512, Kii = 1731.6 for the currents PI parameters. In fact, we have used these results of calculus to determine the searching space for each parameter. In Table 1 are summarized the optimum values that we have obtained after several tries. Table 1. Parameters used in PSO Kpw, Kiw Kpi, Kii for isd Kpi, Kii for isq Swarm size [wmin, wmax] C1, C2 Itermax, D

[0 15], [0 30] [0 30], [0 3000] [0 30], [0 3000] 50 [0.4 0.96] 2, 2 50, 6

228

S. E. Rezgui et al.

The performance index (or fitness) JP used in this study is the integral of time multiplied by squared error (ITSE) in addition to system overshoot according to the following equation: Z

JP ¼ a:ITSE þ b:overshoot ¼ a t:e2 ðtÞdt þ b:overshoot

ð13Þ

Where e(t) is the error between the reference and desired speed such that: eðtÞ ¼ X  ðtÞ  XðtÞ ¼ N  ðtÞ  NðtÞ: This choice is justified because when we compare (13) with several existing and widely used indices in controller tuning via optimization, which are ITAE and ISE and IAE in (14), with addition to settling time. We found that (13) gives much better results. ISE ¼ R e2 dt IAE ¼ R jejdt ITAE ¼ R tjejdt

ð14Þ

4 Testing Results and Discussions The performances of the presented method are simulated in Matlab/Simulink. Real parameters are used in the simulation, they were obtained by identification procedure in the laboratory of three phase Y connected squirrel cage induction motor, 1 kW, 2880 rpm, 220/380 V, 4/2.3 A, 50 Hz, [19]. The identified parameters are depicted in Table 2. Table 2. Parameters of IM Parameters Number of pole pair (p) Stator resistance (Rs) Rotor resistance (Rr) Stator cyclic inductance (Ls) Rotor cyclic inductance (Lr) Mutual inductance (Lm) Moment of inertia (J) Friction coefficient (fv)

4.1

Value 1 6.58 [Ω] 5.81 [Ω] 0.7490 [H] 0.7490 [H] 0.7209 [H] 0.00207 [kg.m2] 0.173[N.m/(rad/s)]

Performance Tests

In Fig. 2 and Fig. 3 are depicted the swarm of the speed controllers (Kpw, Kiw) at the beginning and the end of generations respectively, where the red dot represents the global best position.

Performance Improvement of IFOC

229

Fig. 3. Swarm position at the end of generations.

Fig. 2. Swarm position at the beginning of generations

The optimal values found by the PSO swarm are; Kpw = 0.5779, Kiw = 42.4032, and Kpi = 83.4951, Kii = 2146.8. Note that all the curves in green represent the PSO results, and those of PI are in red. Figures 4 and 5 show speed tracking curves between reference N* and actual rotor speed N for both PI and optimized PI with PSO. Initially, a step command with 1400 rpm is applied without load, and at t = 0.8 s the machine is fully loaded with TL = 3.2 N.m. Thereafter, a second step (2800 rpm) is applied at t = 1.3 s. In Fig. 5 a zoom-in shows clearly the time response and the disturbance rejection. 3000 N-PI

N-PI

N* N-PSO

N* N-PSO

2000

N vs N* :[rpm]

1400 [rpm]

1000

0

0

0.5

t:[s]

1

1.5

2

Fig. 4. Speed tracking test with PI and PIPSO

t :[s] 0,2

0,4

0,6

0,8

1

Fig. 5. Zoom-in of the speed response.

In Fig. 4, we can see that the speed reaches and follows the reference value without overshoot in the two step references (1400, 2800 rpm), with a faster dynamic response than that obtained with the PI. We can also see (Fig. 5) that the speed drop due to the

230

S. E. Rezgui et al.

application of the load torque and takes a considerably shorter time to return to the stability compared to the PI, and therefore, the rejection of the disturbance has been improved significantly.

15

Te :[N.m]

10 5 0 0

0.5

1 t:[s]

1.5

2

Fig. 6. Torque response in PI and PI-PSO.

The torque on Fig. 6 presents more ripples when compared with PI-PSO method. In Fig. 7, each method is evaluated when a sudden change with negative speed reference is demanded (from 1800 rpm to -1800 rpm). The IM is loaded at rated torque at t = 0.6 s, and set at no load at t = 1 s, after that it’s loaded again at t = 1.8 s. It is noted that the response of the system is the same as the first test in terms of rapidity and disturbance rejection. The Fig. 8 shows the torque curve, both regulators realize high dynamic performance in response to changes in torque demand, but as it can be seen, the torque ripple in the PI-PSO is less than in conventional PI technique. In addition, the torque has a rather rapid dynamic response compared to the PI technique. 2000

20 10 Te :[N.m]

speed :[rpm]

1000

Te-PI Te-PSO

0

0

-10

-1000

-20 -2000 0

0.5

t:[s]

1

1.5

2

Fig. 7. Speed response with PI and PI-PSO.

0

0.5

t:[s]

1

1.5

2

Fig. 8. Electromagnetic torque response in PI and PI-PSO.

Performance Improvement of IFOC

231

5 Conclusion The performance of IFOC with classic PI was tested and compared with the PSO tuned PI controllers. The results have shown that the presented method improves the system transient response and reduces the effect of disturbances as it decreases the time response of the system. The simulation results reveal the effectiveness of the PI-PSO approach to achieve optimal solutions and contribution in the torque ripples attenuation.

References 1. Lethwala, P.D.: Modeling and vector control of grid-connected doubly fed induction generator. In: Innovations in Power and Advanced Computing Technologies (i-PACT), Vellore, India, pp. 1–6 (2019) 2. Domínguez-García, J.L., Gomis-Bellmunt, O., Trilla-Romero, L., Junyent-Ferré, A.: Indirect vector control of a squirrel cage induction generator wind turbine. Comput. Math. Appl. 64 (2), 102–114 (2012) 3. Kumari, R., Dahiya, R.: Speed control of solar water pumping with indirect vector control technique. In: 2nd International Conference on Inventive Systems and Control (ICISC), Coimbatore, pp. 1401–1406 (2018) 4. Shukla, S., Singh, B.: Single-stage PV array fed speed sensorless vector control of induction motor drive for water pumping. IEEE Trans. Ind. Appl. 54(4), 3575–3585 (2018) 5. Kumar Tiwari, S., Singh, B., Goel, P.K.: Design and control of microgrid fed by renewable energy generating sources. IEEE Trans. Ind. Appl. 54(3), 2041–2050 (2018) 6. Trentin, A., Zanchetta, P., Gerada, C., Clare, J., Wheeler, P.W.: Optimized commissioning method for enhanced vector control of high-power induction motor drives. IEEE Trans. Ind. Electron. 56(5), 1708–1717 (2009) 7. Betin, F., Moghadasian, M., Lanfranchi, V., Capolino, G.-A.: Fault-tolerant control of sixphase induction machines using combined fuzzy logic and genetic algorithms. In: IEEE Workshop, Electrical Machines Design Control and Diagnosis, pp. 138–147 (2013) 8. Markadeh, G.R.A., Daryabeigi, E., Lucas, C., Rahman, M.A.: Speed and flux control of induction motors using emotional intelligent controller. IEEE Trans. Ind. Appl. 47(3), 1126– 1135 (2011) 9. Rajasekara, N., Mohana Sundaram, K.: Feedback controller design for variable voltage variable speed induction motor drive via Ant Colony Optimization. Appl. Soft Comput. 12 (8), 2132–2136 (2012) 10. Sharma, F.B., Kapoor, S.R.: Induction motor parameter estimation using disrupted black hole bee colony algorithm. Int. J. Metaheurist. 6(1/2), 85–106 (2017) 11. Souza Santos, V., Felipe, P.V., Sarduy, J.G.: Bacterial foraging algorithm application for induction motor field efficiency estimation under unbalanced voltages. Measurement 46(7), 2232–2237 (2013) 12. Kennedy, J., Eberhart, R.C.: Particle swarm optimization. In: Proceeding of the IEEE International Conference on Neural Network, pp. 1942–1948 (1995) 13. Reddy, S.R.P., Loganathan, U.: Particle swarm optimization based output-feedback control of vector-controlled induction motor drives considering core loss resistance. In: IEEE International Conference on Power Electronics, Smart Grid and Renewable Energy (PESGRE2020), pp. 1–6 (2020)

232

S. E. Rezgui et al.

14. Marouane, R., Malika, Z.: Particle swarm optimization for tuning PI controller in FOC chain of induction motors. In: 4th International Conference on Optimization and Applications (ICOA), pp. 1–5 (2018) 15. Sudha Letha, S., Thakur, T., Kumar, J.: Harmonic elimination of a photo-voltaic based cascaded H-bridge multilevel inverter using PSO (particle swarm optimization) for induction motor drive. Energy 107, 335–346 (2016) 16. Low, K.S., Deng, Y.Z., Chan, C.Y.: Discrete-time sliding mode control of a brushless DC drive. In: IEEE Conference on Power Electronics and Drive Systems, vol. 1, pp. 286–290 (1997) 17. Shi, Y., Eberhart, R.C.: A modified particle swarm optimizer. In: Proceedings of the IEEE International Conference on Evolutionary Computation, pp. 69–73 (1998) 18. Gaing, Z.-L.: A particle swarm optimization approach for optimum design of PID controller in AVR system. IEEE Trans. Energy Convers. 19(2), 384–391 (2004) 19. Rezgui, S.E., Mehdi, A., Legrioui, S., Meddouce, H., Boulahia, A.M., Benalla, H.: IRFOC vs DTC performance comparison analysis. In: 3rd International Conference on Electric Power and Energy Conversion System, pp. 1–6, November 2013

Analyze and Fault Diagnosis of Double Star Induction Motor Using Wavelet Transformation Guermit Hossine(&) and Kouzi Katia LSMF Laboratory Electrical Engineering Department, Amar Telidji University Ghardaia, Street BP G37, Laghouat (03000), Algeria {ho.guermit,k.kouzi}@lagh-univ.dz

Abstract. This paper investigates the influence of broken bars faults in double star induction motor (DSIM) by analysing the motor stator current using two different methods of diagnosis. The first one is the Fast Fourier Transformation; this method gives good knowledge of how many frequencies exist but without any information where its frequencies are localized in time. So the analysis method is not satisfied, hence the difficulty of obtaining localized information over time. So we cannot study signals whose frequency varies over time (Nonstationary signals). To overcome this problem we applied the wavelet transformation, this method offers a very fine analysis of one-dimensional and twodimensional signals through a continuous wavelet transform and discrete wavelet decomposition. These two techniques are performed in Matlab environment, the results obtained show the effectiveness of these techniques in the detection of electrical rotor defects even the detection of non-stationary. Keywords: Double star induction motor (DSIM)
Modified winding function theory (WFA)
Motor current signature analysis (MCSA)
Fast fourier transform (FFT)
Discrete transformation wavelet (TDW)

1 Introduction Asynchronous Double Star Induction Motor are widely used in the industry such as aviation, electric traction, power plants, and industrial production plants. Ensuring their continuity of operation requires the implementation of preventive and corrective maintenance programs. Indeed, the reliability and the safety of their operation allow in part to ensure the safety of the people, the quality of the service and the profitability of the installations [1–3]. Linear transformations have always been played in signal processing, among them, as well as in the old Fourier transformation (1822). This transformation allows to explore the frequency composition of the signal. Very early in the history of signal processing, it turned out that the decomposition obtained by Fourier was not always more. In the 1940 s, Gabor discovered the first form of time-frequency representation. His technique consists in cutting the signal into different fixed length ranges or windows. Each segment of the signal limited by a window is studied separately from the others by the Fourier analysis [4, 5]. All of these localized transforms form the Gabor © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 233–242, 2021. https://doi.org/10.1007/978-3-030-63846-7_23

234

G. Hossine and K. Katia

transform of the signal. The major disadvantage of this method is that the length of the window being fixed, it is not possible to simultaneously analyze phenomena whose time scales are different. Another technique of analysis which does not privilege any particular scale but which generalizes on all the scales the local analysis of the frequencies obtained by the method of Gabor becomes more than necessary. In 1982, J. Morlet opened the path leading to the solution by constructing wavelet analysis, based on a concept somewhat different from that of frequency. This procedure, developed by Stéphane Mallat and systematized by Ingrid Daubechies, is called multiresolution and suggests a different interpretation of wavelet analysis. Wavelets are therefore one of the newest tools in signal processing and are only a few decades old. They allow us to perform robust analysis and lead to multiple applications. In contrast to the short-term Fourier transform, the wavelet transform uses the notion of time-scale involving dynamic length analysis windows [5, 6]. For that, the goal of this paper is to study the effect of a broken rotor bars in double star induction motor by modeling the double star induction motor for diagnosis purpose in Sect. 2. And in Sect. 3 we deals with Fast Fourier Transformation (FFT) and Discrete Transformation Wavelet (TDW) to analyze the stator current for detection any faults. And presenting some simulation results are implemented in Sect. 4. Conclusion is presented in Sect. 5.

2 Model of the Dual Star Asynchronous Motor with Taking into Account Space Harmonics In order to develop a novel model to simulate and analyze the behavior of the double star induction motor when a fault occurs, in a first time, we present the emerging faults between coils caused by mechanical vibration of the heads of the stator coils. Firstly, we present the system of equations of a healthy DSIM [7–9]: The first equation of the dual three-phase induction can be written in vector –matrix form as follows: ½vs  ¼ ½Rs :½is  þ

d ½u  þ ½vn  dt s

Where: ½vs  ¼ ½vsa1 vsa2 vsb1 vsb2 vsc1 vsc2  ½is  ¼ ½isa1 isa2 isa3 isb1 isb2 isb3  ½vn  ¼ ½vn1a1 vn1a1 vn1a1 vn1a1 vn1a1 vn1a1 

ð1Þ

2 6 6 6 ½R s  ¼ 6 6 6 4

Rsa1 0 0 0 0 0

0 Rsa2 0 0 0 0

Analyze and Fault Diagnosis of Double Star Induction Motor 235 3 3 2 0 0 0 0 Rsa1 0 0 0 0 0 6 0 Rsa2 0 0 0 0 7 0 0 0 0 7 7 7 6 6 0 0 Rsb1 0 0 0 7 Rsb1 0 0 0 7 7 7 6 ½L s  ¼ 6 7 0 Rsb2 0 0 7 7 6 0 0 0 Rsb2 0 0 7 5 4 0 0 0 0 Rsc1 0 5 0 0 Rsc1 0 0 0 0 0 0 Rsc2 0 0 0 Rsc2

½us  ¼ ½Ls :½is  þ ½Msr :½ir  3 ½Rr  Re       6 d ½ur  ½ vr  : 7 ½ir  6 7 : ¼ ½0 ¼ 4 þ ve : 5 ie dt ue Re . . . . . .: nb Re

ð2Þ

2

The mechanical equation is expressed as: dx p ¼ ðTe  Tr  f xÞ dt J

ð3Þ

The mutual inductances are calculated using winding function approach take into account all the space harmonics in the motor as follow: Msqij ¼

      1 4l0 rl Nc2 X 1 2p 2p 2p 2p 2p 2p  q  i  q  i cos h j þ f ð  1 Þ a þ f ð  1 Þ a cos h j 1 1 1 1 s 2 2 2 2 s g0 p h¼1 h2 Ns p 3 Ns p 3

ð4Þ

3 Analysing Stator Current of the Dsim Using Fast Fourier Transformation (FFT) and Discrete Transformation Wavelet (TDW) 3.1

Fast Fourier Transformation (FFT)

Fourier analysis is a major basis of physics and mathematics. It is inseparable from signal processing for two main reasons. The first is the universality of the concept of frequency on which it is based. The second is due to the very structure of Fourier analysis, which lends itself easily to common transformations such as linear filtering by translating them in a particularly simple way. In 1807, the Fourier transform was born, it consists of breaking down any periodic signal into a sum of sinusoidal signals of different amplitudes and phase shifts [10, 11]. Xð f Þ ¼

þZ1 1

xðtÞej2p

ð5Þ

236

G. Hossine and K. Katia

The Discrete Fourier Transform, generally denoted TFD, is a finite sequence of N terms xð0Þ; xð1Þ; xð2Þ. . .. . .. . .. . .::xðN  1Þ and is calculated by the relation [12]: 1 XN1 2pnk xðnÞej N n¼0 N with k ¼ 0; . . .. . .. . .. . .:; N  1 X ðK Þ ¼

ð6Þ

The forms of sine waves of currents and voltages are affected by the presence of faults. Previously, the current of DSIM in default presents of the peak’s co-incident with the occurrence of short circuits between coils. These deformations affect the harmonic spectrum of the currents, voltages, power, etc. In order to visualize the impact of the defect on the frequency spectrum, we have applied the discrete Fourier transform (FFT) to the currents and we compared the spectral when the DSIM operates without fault to the one in presence of faults. 3.2

Discrete Transformation Wavelet (TDW)

Due to the limitations of the Fourier, in the early 1980 s the collaboration of Physical mathematicians introduced the wavelet transform which breaks down the signal both in time and frequency and introduces a window whose size varies with frequency. The term wavelet designates a function w 2 L2 ðRÞ which oscillates over a finite length interval, thus of zero integral. Beyond, the function decreases very quickly towards zero [13]. From the unique function w, a family of wavelets is constructed by translation and expansion:   1 tb wða;bÞ ðtÞ ¼ pffiffiffi w a a a 2 Rþ ; b 2 Rþ

ð7Þ

The choice of the wavelet used for the time-frequency decomposition is the most important point. This has an influence on the resolution in time and frequency of the result. We cannot modify in this way the characteristics of the wavelet transform (the low frequencies have a good frequency resolution but a bad temporal resolution, the high frequencies have a good temporal resolution and a bad frequency resolution), but we can however increase the total frequency resolution or total time resolution. This is directly proportional to the width of the wavelet used in the real space and in the Fourier space.

Analyze and Fault Diagnosis of Double Star Induction Motor

237

When one analyses a signal f ð xÞ with these wavelets, one transforms it into a function of two variables (the time of scale of analysis of signal) that one can call W ða; bÞ : W ða; bÞ ¼ f ; wða;bÞ

ð8Þ

1 Z W ða; bÞ ¼ pffiffiffi f ð xÞwða;bÞ ð xÞdx a

ð9Þ

Which one can also note:

This transformation is in theory infinitely redundant since the wavelet is translated continuously. However, there are methods to reduce this redundancy: one of these methods consists in using the discrete wavelet transform. The factor p1ffiffia normalizes wða;bÞ in order to preserve the energy of the analyzing pattern:  2   wða;bÞ  ¼ 2

Z

 2   wða;bÞ ð xÞ dx ¼ 1

ð10Þ

The discrete wavelet transform is derived from the continuous version, unlike the latter, the Discrete Wavelet Transform DWT uses a scale factor and a discretized translation. A dyadic discrete wavelet transform is called any wavelet basis working with a scale factor u = 2. It is clear that the discrete wavelet transform is practical in implementation on any digital system (PC, DSP, and CARD to lP …).

4 Simulation Results and Discussion The application of discrete wavelet transform (TDW) to fault diagnosis of Double Star Induction Motor is based on electrical signals such as stator currents, stator voltage, velocity, or signals vibration of the machine … etc. Our choice is focused on the stator current sine signatures represent a very rich source of information concerning the defects that are often manifested in the asynchronous machine, for this purpose the majority of the diagnostic work is based on the analysis of the current stator either in its transient part or in its permanent part. The figures represented in Fig. 1 show the phase of stator current with decomposition (details and approximation) at level 12 in healthy case. The approach is focused on the analysis of high level detail (and approximation) signals resulting from wavelet decomposition, whose associated frequency bands are

238

G. Hossine and K. Katia

Fig. 1. Concept of Discrete Wavelets Transformation (DWT).

Fig. 2. TDW of stator current at level 12 with Db12 in healthy Case (a: Original Stator current signal, b: Approximation signal at level 12, c: Details signal at level 12, d: Details signal at level 11, e: Details signal at level 10, f: Details signal at level 9).

Analyze and Fault Diagnosis of Double Star Induction Motor

239

Fig. 3. TDW of stator current at level 12 with Db12 in one Broken Bar Case (a: Original Stator current signal, b: Approximation signal at level 12, c: Details signal at level 12, d: Details signal at level 11, e: Details signal at level 10, f: Details signal at level 9).

included from 0 to the supply frequency, the figures above show, respectively, the simulation results of stator current signal as original signal and the detail signals d9, d10, d11 and d12 obtained by wavelet decomposition using db44 at level 12, of the double star induction machine in healthy case (see Fig. 2), one broke bar case (see Fig. 3), two broken bars case (see Fig. 4) and three broken bars case (see Fig. 5). the Fig. 2 clearly shows the absence of any frequency oscillations in the detail signals which must contain the frequency components characteristic of 2gf fault in non-stationary conditions. On the other hand, in Fig. 3, Fig. 4 and Fig. 5 we can see how the high level signals D9, D10, D11, D12 and A12 increase with the evolution of the left band frequency, there are

240

G. Hossine and K. Katia

Fig. 4. TDW of stator current at level 12 with Db12 in Tow Broken Bars Case (a: Original Stator current signal, b: Approximation signal at level 12, c: Details signal at level 12, d: Details signal at level 11, e: Details signal at level 10, f: Details signal at level 9).

increments in the amplitude, there is still appearance of oscillations in D9 and D10. This characteristic configuration of the high level signals D9, D10, D11, D12 and A12 in the starting current takes into account a reliable diagnosis of the breaking bars defects.

Analyze and Fault Diagnosis of Double Star Induction Motor

241

Fig. 5. TDW of stator current at level 12 with Db12 in Three Broken Bars Case (a: Original Stator current signal, b: Approximation signal at level 12, c: Details signal at level 12, d: Details signal at level 11, e: Details signal at level 10, f: Details signal at level 9).

5 Conclusion The application of the discrete wavelet transform led to very significant results in terms of defects, the direct decomposition of the multilevel stator current gave a real image on the various rotor defects of the cage asynchronous machine. The detection of nonstationary generated in the stator current during bar breaks is achieved by multi-level decomposition Wavelet applications have covered almost every aspect of the fault diagnosis.

242

G. Hossine and K. Katia

References 1. Andriamalala, R.N., Razik, H., Baghli, L., Sargos, F.M.: Eccentricity fault diagnosis of a dual-stator winding induction machine drive considering the slotting effects. IEEE Trans. Ind. Electron. 55(12), 4238–4251 (2008) 2. Razik, H., et al.: A model of double star induction motors under rotor bar defect for diagnosis purpose. In: Proceedings of the IEEE International Conference on Industrial Technology. vol. 2005, pp. 197–202 (2005) 3. Razik, H.: Modelling of double star induction motor for diagnosis purpose. In: IEMDC 2003 - IEEE International Electric Machines and Drives Conference. vol. 2, pp. 907–912 (2003) 4. Siddiqui, K. M., Giri, V.K.: Broken rotor bar fault detection in induction motors using wavelet transform. In: 2012 International Conference on Computing, Electronics and Electrical Technologies, ICCEET 2012, pp. 1–6 (2012) 5. Abbaszadeh, J., Milimonfared, M. H., Toliyat, H.A.: Broken bar detection induction motor via wavelet transformation. In: IECON Proceedings (Industrial Electronics Conference), vol. 3, pp. 95–99 (2001) 6. Lau, E.C.C., Ngan, H.W.: Detection of motor bearing outer raceway defect by wavelet packet transformed motor current signature analysis. IEEE Trans. Instrum. Meas. 59(10), 2683–2690 (2010) 7. Maouche, Y., El, M., Oumaamar, K., Khezzar, A., Lyon, U.: Analysis of stator current of dual-three phase induction motor drive under broken bar fault condition H. RAZIK, IEEE (2018) 8. Guermit, H., Kouzi, K.: Investigate the Performance of an Optimized Synergetic Control Approach of Dual Star Induction Motor Fed by Photovoltaic Generator with Fuzzy MPPT, pp. 297–310. Springer, Cham (2019) 9. Hossine, G., Katia, K.: Improvement of vector control of Dual Star Induction drive using synergetic approach. In: 2017 14th International Multi-Conference on Systems, Signals & Devices (SSD), pp. 643–648 (2017) 10. Krishna, M.S.R., Ravi, K.S.: Fault diagnosis of induction motor using Motor Current Signature Analysis. In: Proceedings of IEEE International Conference Circuit, Power Computer Technology ICCPCT 2013, pp. 180–186 (2013) 11. Nallamekala, K.K., Member, S., Sivakumar, K., Kechida, R.: A fault-tolerant dual threelevel inverter configuration for multipole induction motor drive with reduced torque ripple. IEEE Trans. Ind. Electron. 63(3), 1450–1457 (2016) 12. Talhaoui, H., Menacer, A., Kessal, A., Kechida, R.: Fast Fourier and discrete wavelet transforms applied to sensorless vector control induction motor for rotor bar faults diagnosis. ISA Trans. 53(5), 1639–1649 (2014) 13. Das, S., Purkait, P., Chakravorti, S.: Separating induction motor current signature for stator winding faults from that due to supply voltage unbalances. In: 2012 1st International Conference on Power and Energy in NERIST, ICPEN 2012 – Proceedings (2012)

Modeling of Hydrocarbons Rotary Drilling Systems Under Torsional Vibrations: A Survey Chafiaa Mendil1(&), Madjid Kidouche1, and Mohamed Z. Doghmane1,2 1

Laboratory of Applied Automatic (LAA), FHC, University M’Hamed Bougara, Boumerdes, Algeria [email protected] 2 DOE, Exploration Division, SONATRACH, Boumerdes, Hassi Messaoud, Algeria [email protected]

Abstract. Modeling of vibrations dynamic during drilling hydrocarbons wells is one of the main challenge facing drillers. Indeed, in order to design controllers that can eliminate or at least minimize such phenomenon, it is mandatory to know with some degrees of certain how the bit faces these vibrations. In literature, there are many models that have been designed to represent the drill string behaviour under torsional vibrations since they are the most harmful ones. It has been proven that stick-slip phase generated by the torsional vibrations is the generator of other types of phenomena (i.e. Bit bounce and whirling). The main objective of this study is to review all the models and compare them in order to set the advantages and drawback of each of them. Moreover, this comparison has allowed us to determine which model to use for designing robust controllers for mitigating the torsional vibrations, thus, diminish all of its effects and the other type of vibrations. The obtained results have supported and demonstrated the conclusive comparative study. Keywords: Modeling
Torsional vibrations
Stick-Slip phenomenon Comparative study
Smart rotary drilling system




1 Introduction Fossil hydrocarbons are the main constituents of oil and gas; they come from the decomposition of organic matter accumulated over millions years ago. New sediments will continue to accumulate, causing the source rock to great depths, generally between 2500 and 5000 m, under the action of the high temperatures and pressure which reign there thermal cracking will allow them to be decomposed and put back into liquid or gas forms. Underground fluids are very precious and have great value and since they are found at great depths it requires drilling to explore and exploit them under the best technical, economic and safety conditions. Petroleum exploration begins with the identification of the indices due to the follow-up of geological and geophysical studies which collaborate on this meticulous investigation carried out on a sedimentary basin. It is essential to control the quality of registered seismic data in order to identify the © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 243–251, 2021. https://doi.org/10.1007/978-3-030-63846-7_24

244

C. Mendil et al.

most promising areas. Before considering the exploitation drilling [1], it is a question of evaluating the profitability of the deposit to minimize the possible traps by highlighting their classifications according to their probability of existence and their forecast volume. Oil drilling calls all operations and the manipulated material to reach porous and permeable underground rocks to ensure the extraction of liquid hydrocarbons using a hole that connects the target reservoir to the surface. Only drilling can confirm the assumptions made. The nature of the fluids contained in the rock [2], and the quantity of oil sufficient to economically justify this exploitation must be justified since the cost of exploration drilling is very expensive [3]. Indeed, different techniques are used to dig wells: percussion drilling, bucket drilling and manual construction and rotary drilling [4], knowing that the latter is the most used system in the petroleum industry. It consists in putting a tool in rotation by applying a force oriented towards the direction of advancement from a drilling mast, using bit which is used to destroy the rock [5]. Because in the literature there many mathematical model of these system ([6, 7]), this study focuses on analyzing the different models and comparing them in terms of reliability, complexity and responses in order to use them for designing and implementing robust controllers to mitigate vibrations, thus, improving the performance of drilling systems ([8, 9]).

2 Toolstring Modeling Approaches In the literature several models have been proposed in order to minimize torsional vibrations, among these models two models are chosen, the form of the first model is widely used by literature, as he used at 3 degrees of freedom [10], 4 degrees of freedom [11], n degrees of freedom [9], is the second model used by NOV [12], aims to know which model expresses well the phenomenon of torsion, will give us the simulation obtained in the following section The two models, they have the same degree of freedom, but they have a different structure for the 1st model the input is the speed of the Top drive is the input of the 2nd model is the torque delivered by the motor, in the model 1, the damping coefficient and the stiffness coefficient only exist after disc 1, but for the second model, it exists before and after disc 1. A. The first model Figure 1 represents the 1st model, disk 1 represents the Tope drive and disk 2 represents the bit, the description of the other parameters is given in Table 1.

Modeling of Hydrocarbons Rotary Drilling Systems

245

Fig. 1. The 1st chain model composed of 2 elements

The model with three chain elements is given by the following equations:

 8 < €h1 ¼ 1 d1 ðh_ td  h_ 1 þ k1 ðhtd  h1 Þ  d2 h_ 1  h_ 2  k2 ðh1  h2 Þ  lh_ 1 Þ j1
 : €h2 ¼ 1 d2 ðh_ 1  h_ 2 þ k2 ðh1  h2 Þ  lh_ 2  Tob ðh2 ÞÞ j2

ð1Þ

The different variables are specified in the following table: Table 1. Liste of Symbols Parameters htd hi ki di l ji

Description Unity The angle of the top drive [rad] The joint angle i [rad] Rigidity coefficient of the rope section i [N.m/rad] Internal damping coefficient of the rope section i [N.m.S/rad] wall friction coefficient for rope sections [N.m] The inertia of the rope section i [kg.m2]

To use standard notation for states and inputs in (1), the following variable changes are applied: h_ 1 , x1 , h_ 2 , x2 ; htd  h1 , x3 , h1  h2 , x4 ; h_ td = u. The equations of state of the complete system are given by Eqs. (2-3): 8 x_ 1 ¼ j11 ðx1 ðd1  d2  lÞ þ d2 x2 þ k1 x3  k2 x4 þ d1 uÞ > > > < x_ 2 ¼ j12 ðd2 x1 þ x2 ðd2  lÞ þ k2 x4  Tob ðx2 ÞÞ ð2Þ _ _ > > > x_ 3 ¼ htd  h1 ¼ u  x1 : x_ 4 ¼ h_ 1  h_ 2 ¼ x1  x2

246

C. Mendil et al.

The mathematical Eq. (2) can be rewritten in the state space form as given by Eq. (3). 

Where C ¼ ½ 0

0

1

0

0

x_ ¼ Ax þ Bu Y ¼ Cx

ð3Þ

0 .

B. The second model Figure 2 illustrates a two-degree torsion model, composed of two discs, the first represents the Top drive and the second is the bit. Jb is the equivalent of the moment of inertia of collars with the drill bit, Jr represents the inertia of the Top Drive,C1 is the equivalent viscous damping coefficient of BHA, and C2 is the viscous damping coefficient at the rotary table. Tob is a nonlinear function called the torque on bit [13], and Tm is the torque supplied by the motor to the system.

Fig. 2. The 1st chain model composed of 2 elements

The model with three chain elements is given by the following equations:
 8 < €h1 ¼ 1 kðh1  h2 Þ  C2 h_ 1 þ Tm jr
 : €h2 ¼ 1 kðh1  h2 Þ  C1 h_ 2  Tob jb

ð4Þ

Modeling of Hydrocarbons Rotary Drilling Systems

247

To use standard notation for states and inputs in (4), the following variable changes are applied: h_ 1 , x1 , h_ 2 , x2 , h1  h2 , x3 ; Tm ¼ u þ C2 uref ; The equations of state of the complete system are given by Eqs. (5):   8 < x_ 1 ¼ j1r C2 x1  kx3 þ u þ C2 uref 1 _ : x2 ¼ jb ½C1 x2 þ kx3  Tob ðx2 Þ x_ 3 ¼ x1  x2

ð5Þ

According to models 1 and 2, we can choose parameters: d1 + d2 + l = C2 , K ¼ K1 et d2 + l = C1 , With:K1 = k2 ¼ K = 481.29; N.m/rad,jr ¼ j1 ¼ 999.35 kg. m2,jb ¼ j2 = 127.27, d1 = 51.38 N.m.s/rad,d2 = 39.79 N.m.s/rad, l ¼ 10N:m,C1 = 49.79 N.m.s/rad, C2 = 101.17 N.m.s/rad,Qref = 10. As Tob represents the nonlinear of the drill bit and given by (Table 2): 0

1

x3 pX0 x3 C B Tob ¼ ln Nr @qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi þ 2 2A 2 x 3 þ X0 x23 þ X0

ð6Þ

Table 2. Parameters used in Rock-bit model Parameter ln N r X0 P

Description Friction coefficient The force vector The contact radius vector Chain transition speed The initial friction parameter

Value 40 Nm 9.81 Wob N 0.1 m 1 1.5

3 Results Discussion A. Weight variation on bit We will vary Wob from 10 N to 120 N, with an input U fixed at a value of 100 rpm for the 1st model, and U= 10 * 10 3 N for the second model. we note on Fig. 3 and Fig. 4 that the increase of the weight on the drill, the oscillation is increasing and the phenomenon of sliding of the tool is clearly visible, so that we can see in the 1st model that the speed of the bit is very oscillating ([5, 14]), note also that the speed of this bit is canceled during the time intervals “blocked phase”, on the other hand for the 2nd model this phenomenon does not show itself well as it is stabilized at 30 s.

248

C. Mendil et al.

(a)

(b)

Fig. 3. The first model with a) Wob = 10 N, b) Wob = 120 N

(a)

(b)

Fig. 4. The second model with a) Wob = 10 N b) Wob = 120 N

B. Input variation In the 1st model we will vary U from 10 to 100 rpm, and for the second model, from 100 N to 10.10 ^ 3 N, with the weight on bit fixed at a value of 140 N. we note in Fig. 5 and Fig. 6 that the input increase, the oscillation increases and the tool sliding phenomenon is shown, so we can see in Fig. 5.a that the bit speed has no oscillation ([15, 16]), after increasing the input to 100 rpm the oscillations are increased, for the 2nd model is illustrated in Fig. 6, we also note that in this figure that the increase in entry the oscillation increases.

Modeling of Hydrocarbons Rotary Drilling Systems

(a)

249

(b)

Fig. 5. The first model with a) u = 10 rpm, b) u = 100 rpm

(a)

(b)

Fig. 6. The second model with a) u = 100 N, b) u = 10 K.N

4 Conclusion The main cause of the stick-slip movement is attributed to the speed of the effective friction torque acting on the drill string. To suppress the stick-slip vibration several control techniques can be used like the sliding mode control. It can be very effective to avoid oscillations and suppressing stick-slip hazardous phenomenon in a reasonable time. However, the performance of the controller ([17, 18]) depends directly to the reliability of the model considered, based on this study it has been demonstrated that the use of three element mode can guarantee certain fidelity to the dynamic of rotary drilling under torsional dynamics. In addition to that, it guaranteed the global stability of the systems under unstructured perturbations caused by the torsional vibrations. Therefore, it is highly important to recommend this model for contributions in control of rotary drilling systems.

250

C. Mendil et al.

Acknowledgment. This study was sponsored by DGRSDT (Direction Générale de la Recherche Scientifique et du Développement Technologique) Algiers- Algeria.

References 1. Navarro-L´opez, E.M., Cortés, D.: Avoiding harmful oscillations in a drillstring through dynamical analysis. J. Sound Vib. 307(1-2), 152–171 (2007). https://doi.org/10.1016/j.jsv. 2007.06.037 2. Kidouche, M., Riane R.: On the design of proportional integral observer for a rotary drilling system. In: the 8th CHAOS Conference Proceedings, Henri Poincaré Institute, Paris, France, May 26–29, (2015) 3. Kessai, I., Benammar, S., Doghmane, M.Z., Tee, K.F.: Drill Bit Deformations in Rotary Drilling Systems under Large-Amplitude Stick-Slip Vibrations. Appl. Sci 10(18), 6523 (2020). https://doi.org/10.3390/app10186523 4. Riane, R., Kidouche, M., Illoul, R., Doghmane, M.Z.: Unknown resistive torque estimation of a rotary drilling system based on kalman filter. IETE J. Res. (2020). https://doi.org/10. 1080/03772063.2020.172483416 5. Vaziri, V., Kapitaniak, M., Wiercigroch, M.: Suppression of drill-string stick–slip vibration by sliding mode control: numerical and experimental studies. Eur J. Appl Math. 29(5), 805– 825 (2018). https://doi.org/10.1017/S0956792518000232 6. Vromeni, T., Dai, C.-H., Van de Wouw, N., Oomen, T., Astrid, P., Nijmeijer, H.: Robust output-feedback control to eliminate stick-slip oscillations in drill-string systems. IFACPapersonline 48(6), 266–271 (2015). https://doi.org/10.1016/j.ifacol.2015.08.042 7. Farg, A.: Nonlinear control in petroleum drilling systems: contribution to the elimination of the phenomenon of Stick-Slip. Doctoral Thesis, University Of Paris (2006) 8. Yuan, T.Z., Tai, F.L., Jun, Y.: An intelligent sliding mode control scheme for stabilized platform of rotary steering drilling tool. 46, pp. 246–247 (2013). https://doi.org/10.4028/ www.scientific.net/AMM.246-247.934 9. Liu, Y.: Suppressing stick-slip oscillations in underactuated multibody drill-strings with parametric uncertainties using sliding-mode control. IET Control Theory Appl. 9(1), 91–102 (2015). https://doi.org/10.1049/iet-cta.2014.0329 10. Navarro-L´opez, E.: An alternative characterization of bit-sticking phenomena in a multidegree-of-freedom controlled drillstring. Nonlinear Anal. Real World Appl. 10(5), 3162– 3174 (2009). https://doi.org/10.1016/j.nonrwa.2008.10.025 11. Navarro-Lopez, E.M., Licéaga-Castro, E.: Non-desired transitions and sliding mode control of a multi-DOF mechanical system with stick-slip oscillations. Elsevier Chaos, Solitons Fractals 41(4), 2035–2044 (2009). https://doi.org/10.1016/j.chaos.2008.08.008 12. Morten, K., Johannessen, T.M.: Stick-Slip Prevention of Drill Strings Using Nonlinear Model Reduction and Nonlinear Model Predictive Control. Norwegian University of Science and Technology Department of Engineering Cybernetics, Master thesis (2010) 13. Samir, A., Daniel, G.: A nonsmooth approach for the modelling of a mechanical rotary drilling system with friction. Evol. Eq. Control Theory (2020) 14. Riane, R., Kidouche, M., Doghmane, M.Z., Illoul, R.: Modeling of torsional vibrations dynamic in drill-string by using PI-observer. In: Bououden, S., Chadli, M., Ziani, S., Zelinka, I. (eds.) Proceedings of the 4th International Conference on Electrical Engineering and Control Applications, ICEECA 2019. Lecture Notes in Electrical Engineering, vol. 682. Springer, Singapore (2021). https://doi.org/10.1007/978-981-15-6403-1_12

Modeling of Hydrocarbons Rotary Drilling Systems

251

15. Doghmane, M.Z.: Optimal decentralized control design with overlapping structure. Magister Thesis. University M’hamed Bougara of Boumerdes, Algeria (2011) 16. Doghmane, M. Z., Kidouche, M., and Habbi, H., Bacetti, A., Bellahecene, B., “A new decomposition strategy approach applied for a multi-stage printing system control optimization. In: 4th Internation Conference on Electrical Engineering (ICEE), Boumerdes, Algeria (2015) https://doi.org/10.1109/intee.2015.7416751 17. Doghmane, M. Z., Kidouche, M.: Decentralized controller Robustness improvement using longitudinal overlapping decomposition- Application to web winding system. Elektronika ir Elektronik, 24(5), 10–18 (2018) https://doi.org/10.5755/j01.eie.24.5.2183715o 18. Mendil, C., Kidouche, M., Doghmane, M. Z.: Automatic control of a heat exchanger in a nuclear power station: The classical and the fuzzy methods. In: IEEE International Conference on Advanced Electrical Engineering (ICAEE), Algiers-Algeria (2019)https://doi. org/10.1109/icaee47123.2019.9014661

Optimal Placement of Distributed Generation Based PV Source in Electrical Power System for LVSI Improvement Using GA Algorithm Samir. H. Oudjana1, Rabie Zine2, Mustafa Mosbah3(&), Abdelouahab Khattara4, and Salem. Arif3 1

4

Unité-de Recherche Appliquée en Energies Renouvelables, URAER, Ghardaia, Alegria 2 School of Science and Engineering, Al Akhawayn University in Ifrane, Ifrane, Morocco 3 LACoSERE, Electrical Engineering Department, University of Amar Telidji, Laghouat, Algeria [email protected] Ghardaia University, Energy Systems Modeling Laboratory, Mohamed Kheider University, Biskra, Algeria

Abstract. The modernization of the electric system became one of the first mirrors. This work presents the study of the optimal placement of Distributed Generation Based PV Sources (DG-PV) into the electrical power system. This integration is to determine the optimal placement of DG with different power factor values. For this purpose the Genetic Algorithm (GA) method was applied. The objective function chosen is the improvement of the Line Voltage Stability Index (LVSI), considering the network constraints. The IEEE 14 bus network was selected to be used for the simulations using MATLAB. The results were combined and proved the effectiveness of the GA method. Keywords: Distributed generation
Optimal placement Improvement
Power system
Genetic algorithm


Voltage Stability

1 Introduction Today the world’s need for electricity has grown rapidly and increasing. The increase in the electrical load has been created by the growth of the population, the evolution of economic and industrial activity, household appliances, comfort and the modernization of life [1]. When introducing other power plants (Distributed-generation or DG) into the load bus can have an effect on the electric power transmission. The size of the DG and their position can play an indispensable role in the electricity network. The position and location of the DG in an inappropriate location can lead to no desired impacts and effects on the transmissions network [2]. The effective method to avoid this effect is to integrate DG in an optimal way. In the literature [3], several types of different metaheuristic techniques have been used to find the optimal location and size of DG, example, the article [3] presented a technique based on an OPF (Optimal Power Flow) considering the DG unit, the positions are predetermined on bus 7, bus 10 and bus 30, © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 252–259, 2021. https://doi.org/10.1007/978-3-030-63846-7_25

Optimal Placement of Distributed Generation

253

the minimization of fuel cost and power losses are based on the dual lens function. Reference [4] the continuation of power flow technique is used and the voltage stability, to determine the voltage-vulnerable location, then decide which bus should be targeted (candidate node), then the position of DG. A study planned in [5] to indicate the influence of DG on fuel cost and gas emissions. In this work, the optimization technique used is the interior-point method, the DG site is selected before. Researchers from [6] had formulated the Distribution-Optimal-Power-Flow (D-OPF) taking into account DG units, and the objective function to minimize total real loss, using QPM (Quadratic Programming Method). Ridha Djamel Mohammedi et al. proposed a metaheuristic method based on NSGA-II applied in an electrical distribution network and modulated the DG as a negative charge [7]. The DP-OPF (Distributed and ParallelOPF) algorithm has provided to solve the OPF in presence of DG units based renewable energy in the transmission network [8]. The researchers in [9] have also shown the technique of optimizing particle swarms to solve the problems of site and size of multi-type DG units in distribution power system. A multiobjective technique such as the minimization of the number of DGs and active losses has been provided, in order to deduce the optimal size and adequate site of DGs, using the NLPT (NonLinear Programming Technique) [10]. In another search presented the nonlinear mixedinteger programming approach or (MINLP) to find the optimal site and the number of DGs in the hybrid electricity market [11]. Reference [12] has developed an application of Multi-Objective Particle Swarm Optimization (MOPSO) in order to find the optimal position and size of the DG and the SC (Shunt Capacitor) taking into account the charge uncertainty in the distribution power system. Multiobjective optimization has three objective functions: voltage profile improvement, voltage stability maximization and reduction of active power loss. Another study [13] proposed a single-objective index based technique to find the best location of many wind turbines in the electric power transmission using the genetic algorithm method. In the article [14] presented a meta-heuristic method based genetic algorithm for the optimal size and site of the photovoltaic unit based on DG. The objective minimized in this study is the active losses caused at the level of the transmission lines. In this work, the genetic algorithm was used to find the optimal integration of the PV source based on the distributed generation, while minimizing the line voltage stability index applied on the IEEE 14 bus network under the MATLAB code.

2 Formulation of the Problem 2.1

Fitness Function

To optimize the integration of distributed power plants, it is necessary to maximize the voltage stability in power system, taking into account the operating constraints of the network. To assure that the operating point of the power system is far from the voltage collapse point, the voltage stability index must be improved. Among the effective indices that have been proposed in the literature is the Line Voltage Stability Index (LVSI) [15]. The lower the value of this index, the more stable the network is from a voltage viewpoint. The LVSI of the line between node i and node j is given by following equation:

254

S. H. Oudjana et al.

4Xij Qj LVSIk ¼ 
 2 Vi sin hij  dij

ð1Þ

where Xij , Qj , Vi , hij , and dij are the line reactance, the reactive power, the voltage magnetude, the line impedance angle between bus i and j and voltage phase angle difference between bus i and bus j, respectively. The objective function, subject to set of equality and inequality constraints that should be satisfied while achieving the minimization of active power loss. 2.2

Operating Constrations

These constraints include active, reactive power balance equations. These equations in transmission lines with presence of photovoltaic unit based on DG can be expressed as follows: SG þ SDG ¼ SD þ SL

ð2Þ

where SG , SDG , SD and SL are apparent power of centralized and decentralized generator, apparent power of load and apparent power of losses respectively. The equations below represent the technical and security limits of the various structures of the network.

0

0:95  Vi  1:05 for i ¼ 1. . .. . .N

ð3Þ

Sk  Skmax for k ¼ 1. . .. . .NB

ð4Þ

PGimin  PGi  PGimax for i ¼ 1. . .NG

ð5Þ

N DG X i¼1

2.3

PDGi  0:3 

N bus X

PDi for i ¼ 1. . .. . .NDG

ð6Þ

i¼1

Solution Feasibility

In the search for the best locations of DGs in the power system, probably stumbled on infeasible solutions, in this case the objective function should be penalized to eliminate these infeasible solutions. This function is given as follows [14]: ! NL
NB
X X 2 2 4Xij Qj VLi  Vlim þ ks  Sli  Slim ð7Þ Fp ¼ Min 
2 þ kv  Li li Vi sin hij  dij i¼1 i¼1 where kv and ks are penalty factors, in this study.

Optimal Placement of Distributed Generation

255

3 Applied Approach The genetic algorithm technique is larged is largely defined in the litturature [14]. To provide an optimal location for the DG sources, state variables and controls variables must be chosen correctly. These vectors are given as follows:   vT ¼ PG1 ; VL1 . . .VLNL ; d2 . . .dN; QG1 . . .QGNG ; Sl1 . . .SlNl

ð8Þ

vT ¼ ½LDG1 ; LDG2 . . .. . .. . .. . .LDGn 

ð9Þ

To identify the best locations and sizes of DGs, the GA method was used. It is noted that, Note that the calculation of power flow has been carried out by the MATPOWER software., and we added a new control variable (DG placement). Figure 1 shows the procedure for finding the best locations for DGs. The Fig. 2 present the flow chart for the research of the best placement of the DGs in the network.

Fig. 1. The procedure for searching the best placement of DG in the power system

4 Simulations and Results The study proposed in this paper was carried out on the ieee 14 bus network. The latter is largely defined in the literature [16]. The objective of this simulation is to study the problem of searching for the best places in the network which gives better LVSI value and consequently better voltage stability. For this reason, several scenarios are considered such as, proposal of different DGs with different power factor values (PF = 1 or 0.95, 0.9, 0.85 and 0.8). The constraints of voltage limits, line limits, the penetration level of sources DG, generation limits of centralized and decentralized generators are

256

S. H. Oudjana et al.

Fig. 2. Flow chart for the research of the best placement of the DGs in the network

all taken into account in the simulations. The number of DG sources is fixed beforehand (05 DG of 10 MW from each). All load bus are considered condidate nodes. After several runs of the program, the parameter values of the GA method are fixed at 100 populations and 50 generations. Simulation results for different scenarios are shown in Table 1. The voltage plan of the system bus for the various scenarios are shown by the Fig. 3. The MVA powers of the grid lines in the different scenarios studied are shown in Fig. 4. The results presented in Table 1 show that the lower the value of the power

Optimal Placement of Distributed Generation

257

Table 1. Simulation results for different scenarios Parameters Pg1 in MW Pg2 in MW Pg3 in MW Pg4 in MW Pg5 in MW Qg1 in MVAr Qg2 in MVAr Qg3 in MVAr Qg4 in MVAr Qg5 in MVAr Power factor Placement (bus N°) Size from each DG in MW Size from each DG in MVAr Losses in MW LVSI in pu

Cas01 177.63 39.96 0.00 0.00 0.00 0.00 19.95 20.85 4.10 15.47 1 4, 11, 12, 13, 14 10

Cas02 177.57 39.94 0.00 0.00 0.00 0.00 16.65 19.58 −6.00 12.10 0.95 10, 11, 12, 13, 14 10

Cas03 177.58 39.92 0.00 0.00 0.00 0.00 12.44 18.08 −6.00 9.38 0.9 10, 11, 12, 13, 14 10

Cas04 177.60 39.92 0.00 0.00 0.00 0.00 8.50 16.30 −6.00 8.96 0.85 4, 11, 12, 13, 14 10

Cas05 177.63 39.92 0.00 0.00 0.00 0.00 4.93 14.97 −6.00 7.07 0.8 4, 11, 12, 13, 14 10

0

3.28

4.84

6.19

7.5

8.586

8.519

8.504

8.527

8.561

0.14127

0.12943

0.094778

0.08215

0.077146

factor, the more the LVSI parameter improves and by consequence the voltage stability of the network increases. According to Table 1, it is to be noted that, the more reactive power of the DGs increases, the more stable the network is from a voltage viewpoint.for total active losses are practically the same for the different scenarios. The best placement for DGs are usually the same, since the location is depending on the size of the integrated DG. According to Fig. 3 the voltage profile improves as the power factor value decreases. According to Fig. 4, the transmitted powers in the lines change slightly. As a conclusion, the best value of the power factor that can be selected is 0.8. This senarios represents the best value of LVSI.

258

S. H. Oudjana et al. Cas 01

Cas 02

Cas 03

Cas 04

Cas 05

1.12 1.1

Voltage (pu)

1.08 1.06 1.04 1.02 1 0.98 0.96

1

2

3

4

5

6

7

8

9

10

11

12

13

14

Bus N°

Fig. 3. The voltages in each bus from different scenarios Cas 01

Cas 02

Cas 03

Cas 04

Cas 05

140 120 Power (MVA)

100 80 60 40 20 0

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

Line N°

Fig. 4. Apparent power in each transmission line from different scenarios

5 Conclusion In this paper, a study on the optimal integration of DGs based PV source in the power system was carried out. The influence of the power factor value on the placement of the DGs and on the LVSI parameter was demonstrated. The objective is to search for the best placement of DGs considering different power factor values, in order to improve the LVSI using the GA method. The results obtained show that the best power factor value is 0.8.

References 1. Pham, T., Energiewende and competition in Germany: diagnosing market power in wholesale electricity market. Econ. Policy Energ. Environ. (2015) 2. Mosbah, M., Khattara, A., Becherif, M., Arif, S.: Optimal PV location choice considering static and dynamic constraints. Int. J. Emerg. Electr. Power Syst. 18 (2017) 3. Canard, J.F.: Impact de la génération d’energie dispersee dans les réseaux de distribution, Doctorat de l’Institut National Polytechnique de Grenoble. Laboratoire d’Electrotechnique de Grenoble (2000)

Optimal Placement of Distributed Generation

259

4. Momoh, J.A., Boswell, G.: Value-based implementation of distributed generation in optimal power flow. IEEE Trans. Power Syst. 7803–9255 (2005) 5. Yingchen, L., et al.: Optimal power flow of receiving power network considering distributed generation and environment pollution IEEE Trans. Power Syst. 4244–4813 (2010) 6. Zhu, Y., Tomsovic, K.: Optimal distribution power flow for systems with distributed energy resources. Electr. Power Energ. Syst. 29(3), 260–267 (2007) 7. Djamel Mohammedi, R., Hellal, A., Arif, S., Mosbah, M.: Optimal DG placement and sizing in radial distribution systems using NSGA-II for power loss minimization and voltage stability enhancement. Int. Rev. Electr. Eng. 08(06) (2013) 8. Lin, S.Y., Chen, J.F.: Distributed optimal power flow for smart grid transmission system with renewable energy sources. Energy 56, 184–192 (2013) 9. Satish, K., Vishal, K., Barjeev, T.: Optimal placement of different type of DG sources in distribution networks. Electr. Power Energ. Syst. 53, 752–760 (2013) 10. Masoud, E.: Placement of minimum distributed generation units observing power losses and voltage stability with network constraints. IET Gen.Trans. Distrib. 7(8), 813–821 (2013) 11. Kumar, A., Gao, W.: Optimal distributed generation location using mixed integer non-linear programming in hybrid electricity markets. IEEE Trans. Power Syst. 1751–8687 (2009) 12. Zeinalzadeh, A., et al.: Optimal multi objective placement and sizing of multiple DGs and shunt capacitor banks simultaneously considering load uncertainty via MOPSO approach. Electr. Power Energ. Syst. 67, 336–349 (2015) 13. Bakelli, Y., Mosbah, M., Kaabeche, A., Acimi, M.: Voltage stability improvement by optimal location of wind sources. In: Proceedings IEEE 1st International Conference on Sustainable Renewable Energy Systems and Applications, 04–05 December 2019, Tebessa, Algeria (2019) 14. Mosbah, M., Arif, S., Zine, R., Djamel Mohammedi, R., Oudjana, S.H.: Optimal size and location of PV based DG-unit in transmission system using GA method for loss reduction. J. Electr. Eng. 17(37) (2017) 15. Moghavvemi, M., Omar, F.: Technique for contingency monitoring and voltage collapse prediction. Trans. Distrib. 145(6), 634–640 (1998) 16. Mosbah, M., et al: Optimal location and size of wind source in large power system for losses minimization, International Conference in Artificial Intelligence in Renewable Energetic System: Smart Energy Empowerment in Smart and Resilient Cities, pp 566–574 (2019)

Implementation and Optimization of PWM Technique for a Three-phase Inverter Associated with an Asynchronous Machine Chafa Mohamed1(&), Kamel Messaoudi2, and Lamri Louze1 1

2

Electric Engineering Department, LEC Laboratory, Constantine 1 University, Constantine, Algeria [email protected], [email protected] Electric Engineering Department, LEER Laboratory, Souk-Ahras University, Souk Ahras, Algeria [email protected]

Abstract. Pulse width modulation (PWM) is a technique used in power electronics to convert a continuous voltage to a continuous voltage or a continuous voltage to an alternative voltage, this technique is the most used in the control of inverters intended for the control of asynchronous machines, in this work techniques have been implemented and optimized (PWM) for a voltage inverter combined with a three-phase asynchronous machine using the Xilinx System generator tool, the random RPWM technique showed an improvement in the harmonic spectrum and voltage and current THD compared to the sine-triangular technique (SPWM). Keywords: Hardware implementation source converter
Medium-voltage


PWM
System-Generator
Voltage

1 Introduction In the field of power electronics, we are currently talking about new concepts such as real-time and hardware implementation, despite the fact that these notions were reserved for information sciences for a few years. In-fact, advancement in the numerical control methods of machines has led to complex and heavy algorithms concerning processing time. In parallel with these advancements, other advancements are also noticed in the digital systems field especially in the field of the configurable devices (FPGA: Field programmable Gate Array). These devices are generally used to realize the necessary degrees of parallelism and pipelining in order to accelerate the data processing in various algorithms and in order to realize the real time of various implementations. Several studies published in recent years concerning the optimization of PWM techniques used FPGA tools. In this paper we present a new approach based on the use of the combination Matlab/Simulink and Xilinx-System-Generator to optimize PWM techniques for a voltage inverter associated with an asynchronous machine. This new tool allows us to simulate and synthesize implementations with the advantage of avoiding the details of HDL programming and performing fast simulations. The purpose © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 260–269, 2021. https://doi.org/10.1007/978-3-030-63846-7_26

Implementation and Optimization of PWM Technique

261

of the modulation function is to determine the switching moments and the logical control commands of the switches in order to obtain a switching sequence of the switches, The choice of a modulation strategy can be made according to the performance desired by the user and all strategies have advantages and disadvantages and can be achieved by software or hardware programming. In this work, we take advantage of the Xilinx-System-Generator tool to optimize the PWM techniques in order to reduce the THD which allows reducing the noise in the machine and the losses in the inverter; finally, we present the conclusions and the work to come.

2 Simulation of PWM Techniques on Matlab/Simulink 2.1

The Pulse Width Modulation

The PWM (Pulse Width Modulation) technique is the growth and development of power electronics. It is the heart of static converter control. The choice of the PWM technique to control the voltage inverter is in order to have a fast response and high performance. The choice of technique depends on the type of machine to be controlled, the type of inverter semiconductors, the power to be applied and the algorithm to be implemented [1]. • Advantages of PWM: The advantage of PWM based switching power converter over linear power amplifier is: 1. 2. 3. 4. 5. 6.

Easy to implement and control, No temperature variation-and ageing-caused drifting or degradation in linearity, Compatible with today’s digital microprocessors, Lower power dissipation, and It allows linear amplitude control of the output voltage/current from previously not present.

• Disadvantages of PWM: 1. Attenuation of the wanted fundamental component of the PWM waveform, in this case from 1.10.866^pu. 2. Drastically increased switching frequencies (in this case from 1 pu to 21 pu)-this means greater stresses on associated switching devices and therefore derating of those devices. 3. Generation of high-frequency harmonic components [2]. 2.2

The Sinusoidal PWM

In this section, we describe the three-phase SPWM used to control a two-level threephase inverter. This technique is composed mainly by two chains: generation of a triangular signal and generation of three sinusoids displaced by 120 degrees from each other. In this technique we also use three comparators to compare the triangular signal

262

C. Mohamed et al.

to the three sinusoids generated. Figure 1 shows the SPWM schematic and a threephase inverter made under Matlab-Simulink combined with a three-phase asynchronous machine [3].

Fig. 1. Three-phase sinus-triangular PWM simulation schemes

Figure 2 shows the simulation results at different stages of the SPWM. Figures 2.a, b and c show respectively the three generated sinusoids, the triangular signal and the control signal (SPWM). Figure 2.d represents the line voltage of the inverter output and (e) represents the stator current as shown in Fig. 2 (f) and (g) shows the electromagnetic torque and the speed of rotation of the machine figures h and i represent the harmonic spectra of line and phase voltages.

a) The three sinusoids

d) Inverter output line voltage

g) machine rotation speed

b) the triangular signal

c) the command signal PWM for a inverter arm

e) phase statoric current

h) line voltage harmonic spectrum of inverter output

Fig. 2. SPWM simulation results.

f) electromagnetic torque

i) statoric current harmonic spectrum

Implementation and Optimization of PWM Technique

2.3

263

Optimization with the Random Technique (RPWM)

The simulation of a random PWM (RPWM) using Matlab-Simulink is similar to that of the SPWM with a difference in the carrier. In the RPWM command, the carrier signal is random. Figure 3 shows the block diagram of the RPWM command with a three-phase asynchronous machine [4].

Fig. 3. Random PWM simulation schemes (RPWM).

Figure 4 shows the simulation results of the random PWM technique of a threephase two-stage inverter combined with a three-phase asynchronous machine.

a) The three sinusoids

b) carrier random signal

d) inverter output line voltage

e) phase statoric current

g) Machine rotation speed

c) inverter arm random PWM control signal.

h) line voltage harmonic spectrum of inverter output

Fig. 4. RPWM simulation results.

f) electromagnetic torque

i) statoric current harmonic spectrum

264

C. Mohamed et al.

3 Hardware Implémentations of PWM In this part of our work, we describe the different steps in generating a PWM signal. We use the Xilinx blocks combined with the Simulink simulation blocks (input generation blocks and display blocks) and compare the results (Co-simulation). The Xilinx block diagram is later converted to HDL using Xilinx System-Generator. Finally, this HDL code is synthesized for an FPGA circuit and we show the synthesis results. 3.1

The Sinusoidal PWM

In this step, we will generate the sinus-triangular PWM signal using the Xilinx (System Generator) blocks. For hardware implementations, we use both Xilinx boards (ZedBoard and XUPV5) with Virtex5-LX110T and Zynq-7000 FPGA devices, respectively. FPGA boards are a good solution for the implementation of control laws including the PWM sinus-triangular control. The following figure shows the proposed schematic for the PWM Sinus-Triangular control using the Xilinx System-Generator tool. The purpose of this proposal is to perform a Co-simulation of hardware implementations by taking advantage of the simulation tools offered by Matlab-Simulink. On the other hand, we can move towards a physical implementation of these proposals on FPGA cards after a few steps of HDL conversion, input/output assignment and synthesis. We can even use these implementations as hardware accelerators using the “inthe-loop” technique [5] (Fig. 5).

Fig. 5. Proposed scheme for PWM Sinus-Triangular control using Sys-Generator

As shown in Sect. 2, the PWM Sinus-Triangular control is composed by two chains for the generation of the two signals (Sinusoidal and Triangular). These two chains ensure the inputs of a comparator with a Boolean output. In this part of our work, we only use Xilinx blocks for the generation of these two signals.

Implementation and Optimization of PWM Technique

265

• The main block “System Generator” The ‘System Generator’ block is the main block in a modelled architecture using Matlab-Simulink. By clicking on this module, we can configure mainly the link between the hardware and the simulation. It allows you to specify the FPGA card and the language used for the implementation of our hardware systems in order to generate the file to be implemented on the FPGA targets using the ISE tool. We used a Simulink system period of 10e−08 (sec), which is equivalent to a frequency of 100 MHz. This frequency is available at the input (AH15) of the Xilinx-XUPV5 board and on several other Xilinx boards. • Triangular signal generation For the triangular signal generation, we use a modulo 8 (or moduloX) counter “Counter1” with parameters (initial value = −100, explicit period Ts = 2^ −20). We also use a multiplier to ensure a real triangular signal varying between 0 and 1.the operating frequency (10 kHz) of the moduloX counter is according to the requested accuracy (we use in our case a frequency of the order of 1/10−4). The operating frequency of the modulo1 counter is calculated according to the first frequency (F2 = F1/2X). To ensure a real triangular signal varying between 0 and 1, we use a multiplication coefficient equal to 1/100 (Fig. 6).

a) Triangular signal generation block

b) the triangular signal

Fig. 6. Generation of a triangular frequency signal 50 Hz using Sys-Gen.

• Sinusoidal signal generation For the generation of the sinusoidal signal, we use the CORDIC block. Indeed, the CORDIC algorithm is an algorithm used for the generation of trigonometric functions. Xilinx offers several types of CORDIC block. We use the simple CORDIC «CORDIC SINCOS» block with one input and two outputs (sinus and cosine). Several parameters are needed for the configuration of this block: the method used, pipeline or not, latency. The input of this block is a signal varying from +p to –p provided by a 20-bit counter (−1/Ts/100 to 1/Ts/100) combined with a gain (Cmult1) with constant value 1. The CORDIC block calculates the sine/cosine of a single value, which implies that it is necessary to ensure the time axis at the input of this block a multiplier to adapt the signals, a MCODE function is used to provide the 120° offset between the three sinusoidal signals. The operating frequencies must be chosen to synchronize the operation of the different blocks and to generate a sinusoidal signal close to reality (Fig. 7).

266

C. Mohamed et al.

a) Generation block of the three sinusoidal signals

b) the three sinusoidal signals

Fig. 7. Generating a 50 Hz sine frequency signal using Sys-Gen.

We also used a Xilinx “Relational” block to compare signals. The output signal of the comparators are reversed for the generation of the 2 control signals PWM necessary for the control of an UPS with even Xilinx blocks for the management of dead time. Gateway Out blocks are used to convert output signals before they are displayed on the oscilloscope under Matlab/Simulink. 3.2

SPWM Implementation Results

In this application, we take advantage of the tools provided by Matlab-Simulink to make simulations and compare results. The following Figures show the PWM sinetriangular signal simulation results of the Xilinx block-based model for simulation and hardware implementation on the FPGA device (Fig. 8).

a)

Signal PWM Triangular Sines generated

d) Inverter output line voltage harmonic spectrum

b) Inverter output line voltage.

e) Statoric current harmonic spectrum

c) Stator current

f) electromagnetic torque

g) Machine rotation speed

Fig. 8. SPWM Implementation Results on System Generator.

3.3

The Random PWM

In order to generate a signal of the random carrier a counter which counts from −2200 to 2200 followed by a multiplier then a block CORDIC (SIN, COS) the whole as input of a multiplier (Fig. 10), the multiplier output is compared with the three sinusoidal signals generated by the same method as the SPWM to form a random control signal for the inverter switches as shown in Fig. 9.

Implementation and Optimization of PWM Technique

267

In this implementation, we used the same blocks as the SPWM technique with a difference in the carrier where we introduced the CORDIC block.

Fig. 9. Scheme proposed for the random PWM control using Sys-Gen.

a) Random signal generation block

b) random signal

Fig. 10. Carrier random signal generation block.

3.4

RPWM Implementation Results

Figure 11 shows the results of implementing the RPWM technique using XilinxSystem-Generator

268

C. Mohamed et al.

a) The three sinusoidal signals.

b) Inverter arm random PWM control signal.

d) Line voltage harmonic spectrum

c) inverter output line voltage

e) Statoric current harmonic spectrum

g) Machine rotation speed

f) Statoric current

h) electromagnetic torque

Fig. 11. RPWM Implementation Results on System Generator.

4 Synthesis and Implementation Results and Discussions System-Generator allows self-generation of HDL (VHDL or Verilog) codes from a high-level specification (Simulink model). It also allows the transition directly to the NGC level or to the generation of FPGA circuit configuration binary codes, we used the Resource Estimator block to obtain the synthesis results for the different PWM techniques implemented, the following table (Table 1) shows the results for each technique. Table 1. Synthesis results Resources SPWM RPWP Slices 945 1160 FFs 270 358 BRAMs 0 0 LUTs 1761 2152 IOBs 35 0 MULTs/DSP48s 0 0 TBUFs 0 0

The following table (Table 2) shows the THD results obtained for each technique of the two implementations.

Implementation and Optimization of PWM Technique

269

Table 2. THD results Technique Matlab/Simulink System generator THD line voltage THD phase current THD line voltage THD phase current SPWM 0.23 7.70 64.74 19.21 RPWM 4.28 11.05 27.31 10.18

By comparing the results obtained from the simulations and implementations we find a significant improvement in the waveform of the voltage composed of the inverter output for the random technique, also an improvement in the curve of the electromagnetic torque, Random technique showed us less THD compared to SPWM. On the resource side the random technique is more consumed compared to the sinusoidal, in future work we propose the practical realization of the implementations using the technique in the loop. It is noted that the simulation was performed on the same asynchronous machine for both techniques with the following parameters: p = 4kw, v = 400v, f = 50 h, speed = 1430 rpm, sample period Ts = 2^−20, resisting torque = 4NM.

5 Conclusion In this work we presented a hardware implementation of the two techniques PWM for a three-phase voltage inverter associated with an asynchronous machine, first we used the simulation software Matlab/Simulink to simulate the two techniques then we used the tool Xilinx System generator for the hardware implementation of the two techniques, a comparison between the two techniques on the THD results and synthesis results side was presented, an improvement in the harmonic spectrum of the voltage and inverter output current was achieved by the random technique.

References 1. Chaikhy, H.: Contribution Au Developpement et a l’Implantation des Strategies de Commandes Evoluees des Machines Asynchrones (2013) 2. Muthukumar, P., Mary, P.M.: A co-simulation of random pulse width modulation generation. In: 2014 International Conference on Circuits, Power and Computing Technologies [ICCPCT-2014], pp. 1296–1301. IEEE March 2014 3. Selvabharathi, P., Kannan, V.K., Kumar, S.: FPGA based speed control of three phase induction motor. Int. J. Innov. Technol. Explor. Eng. 8, 113–117 (2019) 4. Boudjerda, N.: Réduction des Perturbations Conduites dans les Convertisseurs de l’Electronique de Puissance par une Commande en MLI Aléatoire. Thèse de doctorat (2018) 5. Vinay, K.C., Shyam, H.N., Rishi, S., Moorthi, S.: FPGA based implementation of variablevoltage variable-frequency controller for a three phase induction motor. In: 2011 International Conference on Process Automation, Control and Computing, pp. 1–6. IEEE July 2011 6. Model-Based DSP Design Using System Generator 2 UG948 (v2019.1). www.xilinx.com. Accessed 22 May 2019

Real Time Implementation of Polynomial Control and Nonlinear Backstepping Strategies Integration for Motion Control of a PMSM Y. Mihoub1(&), D. Toumi1, S. Moreau2, and S. Hassaine1 1

L2GEGI Laboratory, University of Tiaret, Tiaret, Algeria [email protected] 2 LIAS Laboratory, Poitiers University, Poitiers, France [email protected]

Abstract. In this paper, a robust control scheme is proposed for the permanent magnet synchronous motor (PMSM) based on nonlinear Backstepping approach with an RST controller and observation of load torque. Based on a simple structure, a set of RST-Backstepping algorithms can be obtained such that the stability for the closed loop system and the tracking performance are guaranteed, and the disturbance rejection ability subject to a prescribed attenuation level can also be achieved. The proposed RST-Backstepping control scheme for the PMSM is implemented by a TMS320LF2407 DSP-based fully digital controller. Experimental results indicate that the proposed control is reliable and effective for the speed control over a wide range of PMSM drive operations. The proposed solution is characterized by simplicity for practical and control design in the PMSM drive. Keywords: PMSM


RST
Backstepping
DSP

1 Introduction The Permanent Magnet Synchronous Motors (PMSMs) are popular in variable-speed applications used in renewable energy production, electrical vehicles, robotic application, submarines, medical and industrial servo drives. This is due to their low inertia, high efficiency, high power density, reliability, high power factor, faster response and rugged construction. The elimination of brushes slip-rings and rotor losses in the field winding leads a higher efficiency. All these advantages allow a reduction in the machine frame size [1]. Therefore, excellent current and speed dynamics performances are important indicator of the PMSM servo drive. Historically, the (PI) conventional controllers are the largely used in the PMSM drives. The control synthesis aims to provide a good system behavior in terms of dynamic speed control response, a load torque rejection and uncertainties of parametric variation [2]. However, PI controller designed for optimum quality performs satisfactorily with the PSMSs nominal parameters but it is sensitive to the parameter variation [3]. Moreover, the PMSM servo-drive system presents a non-linear model with coupled variables. In such situation so, the linear control can't ensure high © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 270–280, 2021. https://doi.org/10.1007/978-3-030-63846-7_27

Real Time Implementation of Polynomial Control

271

dynamic performance. The PMSM servo-drive overall performance is defined by two characteristics: the system response quickness and the control technique robustness. The accuracy and real time of the system response and the robustness to model uncertainties, parameters and external disturbances are the key problems to overcome these limitations [4]. In the literature of PMSMs control, many researchers give great importance to the digital controllers providing good high closed-loop performances in rotor position or speed tracking applications via different strategies that make use of advanced control algorithms to ensure high dynamics performances. But few studies that examine the current control by using powerful algorithms, knowing that it is the current loops that generally cause problems in machine control. In this way, a lot of robust vector strategies that guaranty fast dynamic response with parameters variations and external disturbances have been developed in the design of the PMSM drive including adaptive PI controller [5], adaptive control [6], passivity-based controller [7], predictive controller [8], robust sliding mode control [9, 10], intelligent control [11, 12] and generalized predictive control [13]. In this paper, a robust digital controller solution were elaborated for the PMSMs currents and feedback speed control using conventional vector control strategy associated with load torque estimation. In this work, the digital Backstepping current control scheme is based on a non-linear composite law feedback control. The recursive and systematic synthesis methodology for the calculation of the control law of uncertain nonlinear systems represents the main advantage of the proposed approach, in particular for systems with suitable uncertainties. The objective is the robust current control and the minimization of torque ripples. To guaranty the control robustness properties, like in [14], a choice of Lyapunov function including integral action is used. The speed control is ensured by a regulator of the RST type. To guarantee the closedloop system stability, the pole placement strategy [3] is used. The tracking performance is satisfied. Another objective of this study consists in reaching a prescribed level of attenuation for the rejection of disturbance. To get load torque, an estimation algorithm is introduced. The DSP-based digital controller is used to implement the proposed RST-Backstepping control structure. Finally, the performances of the proposed controller are validated by experimental results.

2 Mathematical Modeling of PMSM Drive System Considering the assumption that the impact of eddy current and hysteresis losses are neglect, no saturation of the magnetic circuit, the three phase windings are balanced and space magnetic field have sinusoidal distribution, the PMSM model in the synchronous (d–q) reference frame reads as [15]: p2 fc dx dt ¼ J ðuf isq þ ðLsd  Lsq Þisd isq Þ  J x Lsq disd Rs 1 dt ¼  Lsd isd þ Lsd xisq þ Lsd Vsd u disq Rs Lsd f 1 dt ¼  Lsq isq  Lsq xisd  Lsq x þ Lsq Vsq

 p TJL

ð1Þ

272

Y. Mihoub et al.

In this model x is the electrical speed, where p denotes the number of pole pairs, Vsd,sq represent the stator voltages in the (dq) frame and isd,sq represent the direct and in quadrature components of stator current. Moreover, Tl denotes the load torque that is considered as an undefined constant disturbance which is added to the global control system and has to be eliminated. The electromagnetic torque Te has two components: the principal one produced by permanent magnet flux, denoted /f, and the reluctance torque produced by isd and isq, respectively due to the pole saliency. In the PMSM control drive, the reference signal for the d-axis current is fixed to zero for energy efficiency [15]. Imposing isd = 0 enable to control the electromagnet torque Te directly through isq, which is also the basic idea of field-oriented control. Furthermore, for a surface mounted PMSM, which is largely employed in many industrial applications, however the winding stator inductances respect the following relationship Lsd = Lsq. The presented work uses surface mounted PMSM and so, the electromagnetic torque Te is independent of isd, current and linearly related to the value of isq current and is given as follow: Te ¼ p/f isq

ð2Þ

3 PMSM Controllers Design 3.1

Nonlinear Backstepping Currents Control Design

Figure 1 shows the proposed the vector-structure with the motor drive system. The control strategy aims to synthesis a suitable control law for the PMSM drive system that the model is represented by (1) so that isd,sq currents state trajectory can track the reference stator currents isd;sq trajectory independently of the parameters variation and external disturbance. The Backstepping design for the uncertain PMSM drive system can be developed step-by-step specially when all PMSM dynamics are well known. 3.2

Stator Currents Loops

The design of the PSMS drive control inputs Vsq and Vsd are presented in this section. The following tracking error is introduced to design the control input Vsq: eq ¼ isq  isq

ð3Þ

Considering the variable nq Z nq ¼ eq þ Kq2 with Kq2 a twining gain. If we set n0q ¼ defined as follow:

Rt

0 eq ,

t

eq

ð4Þ

0

we can introduce the Lyaponuv function

Real Time Implementation of Polynomial Control

273

1 V1 ¼ ðn2q þ n02 qÞ 2

ð5Þ

Then the time derivative of V1 will be: 0

dnq 0 dnq d V1 ¼ nq þ nq dt dt dt   0 deq dV1 ¼ nq þ Kq2 eq þ nq Kq2 eq dt dt    disq disq 0   þ Kq2 ðisq  isq Þ þ nq Kq2 ðisq  isq Þ ¼ nq dt dt

ð6Þ

ð7Þ

By replacing disq/dt from the model (1) we obtain:    disq uf 0 dV1 Rs 1 ¼ nq þ isq þ xisd þ x Vsq þ nq Kq2 ðisq  isq Þ þ nq Kq2 ðisq  isq Þ Lsq dt dt Lsd Lsq ð8Þ 0 dV1 ¼ nq U1 þ nq Kq2 ðisq  isq Þ þ nq Kq2 ðisq  isq Þ dt

Where: U1 ¼

disq dt

þ

Rs Lsd isq

Let

þ xisd þ

uf Lsq

ð9Þ

x  L1sq Vsq .

U1 ¼ nq kq

ð10Þ

Vdc

+

i*sd = 0

ω*

RST Contro l

−

-

Backstepping Control

isq*

Vsd*

Vsq*

V sa*

Vs*α

d-q

α-β

Vs*β

α-β

a-b-c

α-β

i sd

i sq

Tˆl

ωˆ

i sβ

d-q

Load Torque

Estimation

V5

i sα

ω

d dt

a-b-c α-β

2

V s*

V0 V7

V sc*

θ

1 pφ f

βV

V3

V sb*

+

iˆqsd

-

SVPWM

V1

4

α1

α

Inverter

V6

i sa i sb

θ

SG

SM Load

Fig. 1. A nonlinear backstepping currents controller with a RST digital speed controller.

274

Y. Mihoub et al.

Where Kq is a twining gain. Then the relation given by Eq. (9) defining dV1/dt, will be written as follow: 0 dV1 ¼ Kq n2q þ nq Kq2 ðisq  isq Þ þ nq Kq2 ðisq  isq Þ dt

ð11Þ

0 dV1 ¼ Kq n2q þ ðnq þ nq ÞKq2 ðisq  isq Þ dt

ð12Þ

According to Eq. (4), we write: 0 0 dV1 ¼ Kq n2q þ Kq2 ðn2q  nq2 Þ ¼ ðKq  Kq2 Þn2q  Kq2 nq2 dt

ð13Þ

Therefore, under the constraint given by Eq. (11), and if Kq2 [ 0;

Kq [ Kq2 dV1 0 dt

)

ð14Þ ð15Þ

By solving the constraint (10), the control input Vsq can be obtained by substituting U1 from Eq. (10) in Eq. (8) which gives the following equation: disq dt

þ

/f Rs 1 isq þ xisd þ x Vsq ¼ nq Kq Lsq Lsq Lsq 

Vsq ¼ Lsq nq Kq þ Lsq

disq dt

þ

 /f Rs isq þ xisd þ x Lsq Lsq

ð16Þ ð17Þ

The same approach is used to calculate the control input Vsd. 3.3

Polynomial Speed Control: Pole Placement Synthesis

The proposed PMSM drive control strategy is based on speed and stator current loops associated to the load torque observation. The load and friction torques are considered as external disturbances. Because that the current and the torque control dynamic is more important, the speed control synthesis can be easily obtained using a RST polynomial controller represented in Fig. 2. The optimization of the dynamic response time and disturbance rejection, represent a multi-objective of the proposed solution. That is what this controller has two freedom degrees. The RST controller adjustment is based on tree polynomials R, S and T synthesis using a robust pole placement [6, 7].

Real Time Implementation of Polynomial Control Speed Controller

y*

T R

−

d +

System B (s) A(s)

R S

Te

y

isq

pφ f

+ −

θ

d/dt

Rotating arbre

ω TˆL

-

+ Mechanical

F(s)

275

ωˆ

PI

+

Model

+ p

Fig. 3. Load torque estimation

Fig. 2. RST controller structure

The transfer function from the reference current isq to the motor speed can be given by: GðsÞ ¼

p2 /f x  isq Js þ fc

ð18Þ

The RST speed controller synthesis is done so that to have a global dynamic of the second order for the system. Where; the reference model to be imposed will be given as follow: GdBF ðsÞ ¼

s2

x2n þ 2fxn s þ x2n

ð19Þ

To reach the desired specifications, the poles imposition method in closed loop is chosen depending on the natural frequency xn and the damping ratio f. According to the transfer function of the system to be control, the three polynomials of RST controller must be imposed as follow [16]. SðsÞ ¼ s0 þ s1 s;

RðsÞ ¼ r0 þ r1 s;

TðsÞ ¼ t0

ð20Þ

Mechanical PMSM drive parameters and desired specification fixe the RST speed controller coefficients. The resolution of Diophantine identity Eq. (20) gives the following gains: s0 ¼ 1;

3.4

s1 ¼

p2 uf ; J

r0 ¼ x2n ;

r1 ¼ 2fxn 

s0 J ; p2 u f

TðsÞ ¼ t0 ¼ x2n ð21Þ

Load Torque Estimation

In presence of load torque and plant parameter variation many control strategies have been suggested to achieve better control performances. In Fig. 1 a robust speed control scheme is proposed. This observer type uses the quadrature stator current and the speed obtained from the numerical position measurement to estimate the load torque T^l and ^ [17]. The observer proposed structure is shown in Fig. 3. the speed x

276

Y. Mihoub et al.

In the matrix form, the system given by the Fig. 3 takes the following formulation:    pu   pK þ f pKiob f pob ^ x ^_ x   J J c þ  J c ¼ z z_ 0 1 0 R ^  xÞdt. with: z ¼ ðx For the estimated load torque, we can write: 

T^l ¼ ½ Kpob

 Kiob 

 ^ x þ½0 z

pKpob J

1

 Kpob 

isq x



isq x

 ð22Þ

 ð23Þ

The eigenvalues of the state matrix of Eq. (22) define the poles imposed by the gains PI controller. We obtain: (

Jfc Kpob ¼ 2fob xnob p Kiob ¼ Jp x2nob

ð24Þ

4 Experimental Results The feasibility and the performance of the proposed RST-Backstepping have been verified by Matlab/Simulink software environment implementation. The experimental setup developed in LIAS laboratory of Poitiers University is based on a Dspace DS1104 control board as the digital controller and Matlab/Simulink environment. To apply the load torque to the PMSM, permanent magnet synchronous generator is mechanically coupled to the same shaft. For the modulation strategy, space vector pulse width modulation (SVPWM) technique is adopted. The used power amplifier is a 10 kHz SVPWM current-regulated inverter which is consisted of a 300 V, 5 A IGBT. The experimental setup photo and the experimental environment are presented in Fig. 10. In order to validate the performances of the proposed strategy for PMSM current and speed control, the PMSM vector control plat form based on RST-Backstepping is used. The PMSM drive system in this work contains two control loops: - velocity external loop based on RST controller and stator internal current loops realized by nonlinear Backstepping approach. The PMSM drive structure introduced the load torque for effectively ensure a good dynamics performances and reduce the ripples in the motor torque and stator current. Polynomials RST controller synthesis are easier to design and the obtained closed loop transfer function is without zero so the overshoot and the dynamic instability are eliminated in the time response. These two characteristics represent the advantage of the RST controller. Because of its simple structure, this type of controller is very appreciated and largely used for control system. Polynomials speed loop controllers are computed after the imposition of the damping coefficient f and the natural frequency xn slightly on the desired performances for the closed loop system. To control the stator current, this work uses a nonlinear control employing the Backstepping approach with feedforward compensation disturbance in the

Real Time Implementation of Polynomial Control

277

q axis of PMSM drive via the introduction of load torque estimation. The gains of the stator currents loop are chosen in staying within the overall conditions Lyapunov stability.. For obtaining a speed dynamic without overshoot response f is fixed by its optimal value (i.e. f = 0.707). Many experimental tests have been carried out in order to evaluate the dynamic performances of the proposed control strategy. In the nominal case, the experimental results of proposed control are given in Fig. 4. The machine was loaded between t = 6.5 s and t = 9 s. The decoupling between the d and q components stator current is provided by the PMSM vector control with value of isd imposed around zero. It is clearly shown that reference speed is properly followed by PMSM actual speed. A dynamics of a second order system is perfectly imposed to the speed control system as shown in Fig. 5. For a speed step response, the optimal damping ratio is equal to 0.707 where a maximum value does not exceed 5% of the reference value which is clearly observed. The isd and isq currents responses of the proposed strategy present good dynamic performances facing to Tl changing on a wide range. The obtained results shown shows the superiority of the proposed structure in this paper compared to the vector control using a PI conventional controllers (Fig. 6). The improvement affects the control performance in terms of response time. Face to the application of an external level of torque (a load torque disturbance), the speed response dynamic of the proposed control strategy is globally insensitive. The load torque observer contributes effectively to the improvement of the disturbances rejection. Parameter identification errors of the PMSM are used to verify the command law robustness. Some PMSM estimated parameters values were modified and experimentally tested. Three cases of parametric variations have been realized and tested. The variation is made by increase and decrease of 30% of Rs, Ls and J nominal values. The speed control performance against parameter's variation is showed in Fig. 7, 8, and 9. The obtained time responses of the speed, the torque, and the stator phase current are globally similar to those relative for the nominal values (Fig. 4). Generally, the proposed control guaranty an acceptable level of the dynamic performances. The proposed controller structure obtained results confirm its efficiency relative to dynamic performances and robustness properties. The analysis of the different results obtained for the closed loop system with the proposed control law shows that it can provide acceptable performances.

Fig. 4. Experimental results of RST-Backstepping with the load torque.

278

Y. Mihoub et al.

Fig. 5. Experimental results of RST-Backstepping at 100 rad/sec step reference with the load torque.

Fig. 6. Experimental results of vector control via PI conventional controller with the load torque.

A

B

Fig. 7. Experimental results of proposed control algorithm under the load torque when the stator resistance is A (0.7 * Rs), B (1.3 * Rs).

Fig. 8. Experimental results of proposed control algorithm under the load torque when the stator inductances are A (0.7 * Ls), B (1.3 * Ls)

Real Time Implementation of Polynomial Control

279

Fig. 9. Experimental results of proposed control algorithm under the load torque when the moment inertia is A (0.7 * J), B (1.3 * J)

Onduleur de Tension

Oscilloscope

Partie Commande PC + DSP

Circuit de Conversion de la position (AD2S90)

Variateur industriel

Source continue

A

MSAP

B (A) experimental set-up

(B) Experimental environnement

Fig. 10. (A) Experimental set-up. (B) Experimental environmement

5 Conclusion In this paper, a simple and practical robust RST-Backstepping control scheme with the load torque estimation is proposed for the PMSM drive. An excellent dynamics for the feedback speed and an important efficiency of PMSM drive stator current control face to load torque application and to the parameter variation are obtained. This design is governed by a Backstepping strategy that guarantees robust performances and gives satisfactory results. A polynomial speed control strategy (RST controller) parameters associated with a load torque estimation is determined using a pole placement to simultaneously achieve different control objectives which consists of robust system stability in a looped system, in terms of dynamic reference tracking performance as well as the efficiency of disturbances rejection when a prescribed attenuation level for the disturbance rejection can be achieved through the introduction of load torque estimation. Experimental results using Dspace 1104 digital control board confirm that the proposed RST-Backstepping control strategy presents good tracking on a wide range of speeds, accurately with a fast dynamic response. So the developed control

280

Y. Mihoub et al.

laws can be considered as robust. This approach can be extended to other servo systems such as renewable energy system. it can be combined with a nonlinear observer to estimate the PMSM speed and parameters.

References 1. Karthikeyan, J., Dhana, S.R.: Current control of brushless dc motor based on a common DC signal for space operated vehicles. Int. J. Electr. Power Energy Syst. 33, 1721–1727 (2011) 2. Tursini, M., Parasiliti, F., Zhang, D.: Real-time gain tuning of PI controllers for highperformance PMSM drives. IEEE Trans. Ind. Appl. 38(4), 1018–1026 (2002) 3. Pillay, P., Krishnan, R.: Control characteristics and speed controller design for a high performance permanent magnet synchronous motor drive. In: Proceedings of IEEE PESC, pp. 598–606 (1987) 4. Edwards, C., Spurgeon, S.K.: Sliding Mode Control: Theory and Applications. CRC Press, Taylor & Francis, New York (1998) 5. Li, S.H., Liu, Z.G.: Adaptive speed control for permanent magnet synchronous motor system with variations of load inertia. IEEE Trans. Industr. Electron. 56(8), 3050–3059 (2009) 6. Abdel-Rady, Y., Mohamed, I.: Adaptive self-tuning speed control for permanent-magnet synchronous motor drive with dead time. IEEE Trans. Energy Convers. 21(4), 855–862 (2006) 7. Ortega, R., vander Schaft, A., Castanos, F., Astolfi, A.: Control by interconnection and standard passivity-based control of port-Hamiltonian systems. IEEE Trans. Autom. Control, 53, 2527–2542 (2008) 8. Preindl, M., Bolognani, S.: Model predictive direct speed control with finite control set of PMSM drive systems. IEEE Trans. Power Electron. 28(2), 1007–1015 (2013) 9. Ghafari-Kashani, A.R., Faiz, J., Yazdanpanah, M.J.: Integration of non-linear H∞ and sliding mode control techniques for motion control of a permanent magnet synchronous motor. IET Electr. Power Appl. 4(4), 267–280 (2010) 10. Zhang, X., Sun, L., Zhao, K., Sun, L.: Nonlinear speed control for PMSM system using sliding-mode control and disturbance compensation technique. IEEE Trans. Power Electron. 28(3), 1358–1365 (2013) 11. Saeed, K., Mahdi, S.: Adaptive fuzzy tracking control of robot manipulators actuated by permanent magnet synchronous motors. Comput. Electr. Eng. 72(9), 110–111 (2018) 12. Güney, E., Karagöl, S., Demir, M.: A comparative real-time speed control of PMSM with fuzzy logic and ANN based vector controller. Şırnak Univ. J. Sci. 1(1), 123–143 13. Hassaine, S., Moreau, S, Ogab, C., Mazari, B.: Robust Speed Control of PMSM using Generalized Predictive and Direct Torque Control Techniques, ISIE, Vuego, Spain, paper on CD-ROM 4–6 June (2007) 14. Mohamed J.D.L.A.H., Glumineau, A.: Robust integral backstepping control for sensorless IPM synchronous motor controller. Journal of The Franklin Institute (2012) 15. Chaia, S., Wanga, L., Rogersb, E.: Model predictive control of a permanent magnet synchronous motor with experimental validation, Control Eng. Pract. 21(11), 1584–1593 (2013) 16. Doumi, M., Aissaoui, A., Tahour, A., Abid, M., Tahir, K.: Robust fuzzy gains scheduling of RST controller for a WECS based on a doubly-fed induction generator. Automatika 57(3), 617–626 (2017) 17. Hassaine, S., Moreau, S., Bensmaine, F.: Design and hardware implementation of PMSM sliding mode control in SISO and MIMO cases. In: 23rd IEEE International Symposium on Industrial Electronics (IEEE-ISIE), pp. 762–767 (2014)

Discrete Time Sensorless PMSM Control Using an Extended Kalman Filter for Electric Vehicle Traction Systems Fed by Multi Level Inverter A. Khemis R.1, T. Boutabba1(&), and S. Drid2 1

LSPIE Batna Laboratory, University of Abbès Laghrour Khenchela, Khenchela, Algeria [email protected], [email protected] 2 LSPIE Laboratory, University of Batna 2 (Mostefa Ben Boulaïd), Batna, Algeria [email protected]

Abstract. Electric vehicles are widely used today; however, the phenomenon of sticking and shaking often occurs during starting, crawling and shifting. This paper presents the three-phase switching control by an NPC three level inverter for a PMSM Drive for Electric Vehicle Traction Systems, based on a discretetime structure of control and the estimation of speed, rotor position by The Extended Kalman Filter (EKF). The cause of this phenomenon from the perspective of motor control was analyzed in this paper. The model of PMSM fed by the three level inverter and its EKF models in MATLAB/Sirnulink simulation environment is developed. The proposed EKF speed, position estimation method is also proved insensitive to the PMSM parameter variations. Simulation results demonstrate a good efficiency and robustness. Keywords: FOC
Extended Kalman Filter (EKF)
PMSM control, sensorless
Discrete time control
Three level’s inverter
Vehicle traction

1 Introduction Permanent Magnet Synchronous Machines (PMSM) they have traditionally been considered as suitable candidates for application in electric vehicles and hybrids, mainly due to their high power density and high efficiency [1, 2]. Nearly 40% of the world’s pollutant emissions are produced by cars powered by internal combustion engines and electric vehicles (VE) have been found to represent a viable alternative to the solution. of this problem (Zhang et al. 2016) and (Mademlis et al. 2011). There are previous works (Pham-Dinh et al. 2007), (Shyu et al. 2002), (Bolio 2001) that have focused on the study and design of controllers for PMSM with in order to achieve an optimal performance in applications where the control of angular speed, torque and motor consumption currents are the variables that directly impact the operation and efficient use of energy electric (Bazzi et al. 2009). In the area of electric vehicles it is very important to maintain a speed control, of developed torque as well as of the use of energy since in these it is required to maintain the constant speed and power before changes of slope or speed and acceleration profiles imposed by the operator, as well as an autonomy of operation when using battery power. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 281–293, 2021. https://doi.org/10.1007/978-3-030-63846-7_28

282

A. Khemis R. et al.

However, speed/ rotor position estimation methods based on the model and still the electromotive force are generally used for engines operating at medium and high speed (Xiaohui Ju, 2019; Yong Zhang et al. 2016). Based on the electromotive force model, several types of observers for estimating speed and rotational position are presented in the literature, such as extended Kalman filter, MRAS algorithms, sliding mode observer, among others (K. K. Shyu et al. 2002, M. Kosaka et al. 2004 T. Pham-Dinh et al. 2007, Farazdaq R al. 2018). This article proposes the sensorless discrete-time vector control using an Extended Kalman Filter applied to PMSM fed by a multilevel inverter. For the development of the speed estimation algorithm/ rotational position an Extended Kalman Filter is used in the control. To show the performance of this control system a simulation using MATLAB/Sirnulink will be presented.

2 Mathematical Model of the Electric Vehicle The analysis that has been carried out to control the speed trajectories of the PMSM, considers the operation of the PMSM under no-load conditions and the operation with mechanical load coupled to the axis, constituted by the electric drive vehicle. The following forces are considered to model the electric vehicle are: 1. 2. 3. 4.

The tensile force that exists in the mechanical transmission Fte. The frictional force between the body of the EV and the wind Fad. The frictional forces acting between the surface and the wheels of the Frr. The weight of the vehicle, distinguishing its components due to the slope of the surface on which the vehicle is driven.

We considere a simple mechanical transmission such as the one shown in Fig. 1. The analysis of the modeling starts from Eqs. 1 and 2 shown below [9]. T¼

r Fte ng G

ð1Þ

r Fte G

ð2Þ

T ¼ ng

The Eq. 1 is used when the machine operates as a motor providing an input torque. Figure 1 shows the relationship between the linear speed v of the EV and the angular speed xr of the motor in rolling conditions without sliding. xr ¼ G

v r

ð3Þ

The EV will have to operate on land where the slopes play an important role in the performance of a control system. It should be noted that these are not considered as disturbances when tracking speed paths. This topic will be treated in a future work. Figure 2 shows the forces acting on the EV on a slope.

Discrete Time Sensorless PMSM Control

283

T Motor

G

Fad

r

Fig. 1. Representation of the EV transmission.

The Frr force is the rolling resistance that exists between the tires and the surface on which the vehicle moves and is given by: Frr ¼ lr rmgcosðwÞ

ð4Þ

The aerodynamic resistance that acts throughout the vehicle is: 1 Fad ¼ qACd v2 2

ð5Þ

It is also necessary to overcome the force Fhc to move the EV up the slope with an inclination angle, which can be determined as a component of the tangential vehicle weight to the slope and mathematically express as: Fhc ¼ mgsinðwÞ

ð6Þ

Considering Fte as the total force required to move the vehicle and applying Newton’s second law to the EV (sum of forces) Fte  Frr  Fad  Fhc ¼ ma

ð7Þ

By neglecting Fte in Eq. 7 and replacing Eq. 5 and Eq. 6 in Eq. 9 you get the differential equation that relates the Fte with the speed v. Fte ¼ ma þ Frr þ Fad þ Fhc Fte ¼ ma þ lr rmgcosðwÞ þ

1 qACd v2 þ mgsinðwÞÞ 2

ð8Þ ð9Þ

284

A. Khemis R. et al.

Fte Fad

Fhc ψ

Frr

mg

ψ

Fig. 2. Acting forces in the EV.

To unify the model of the mechanical part with the electrical part we use the electromagnetic torque equation of the PMSM and expressed as: Te ¼ J

dxr þ Bxr þ TL dt

ð10Þ

The load pair T expressed in Eq. 1 is now the term TL in Eq. 10, expressing the previous equation in terms of T we have: Te ¼ J

dxr r þ Bxr þ Fte ng G dt

ð11Þ

When replacing Fte in Eq. 11 you have: Te ¼ Jð

dxr r dv Þ þ Bxr þ ðm þ lr rmgcosðwÞ ðng GÞ dt dt 1 þ qACd v2 þ mgsinðwÞÞ 2

ð12Þ

In this manner the unified mathematical model of the mechanical part and the electrical part is obtained. For the purpose of simulations from the EV, the angular acceleration of the PMSM can be cleared and expressed by Eq. 13.   Bxr þ dxr 1 Te 
 ¼ r dv 1 2 J ng G m dt þ lrr mgcosðwÞ þ 2 qACd v þ mgsinðwÞ dt

ð13Þ

Discrete Time Sensorless PMSM Control

285

3 Modeling of the PMSM The dynamic behavior of the permanent magnet synchronous motor is expressed through the differential equations of the stator currents in a synchronous reference (dq), as presented in [16–19]: 8 Lq dId Rs 1 > > ¼  þ pxr Iq þ Ud > > L dt L L > d d d > > > < dIq Rs Ld Uf 1 ¼   pxr Id  pxr þ Uq > q L dt L L L > q q q > > > > > dx 3p 1 f > : r ¼ ½ðLd  Lq ÞIq Id þ Uf Iq   TL  xr 2J J J dt

ð14Þ

And can be written as: d½X  ¼ ½ A½ X  þ ½B½U  dt

ð15Þ

Defining: " R  T  Lds ½ X  ¼ Id Iq  and ½ A ¼ wrLd ½V  ¼ Ud Uq /f L

q

" L # l wr Ldq Ld Rs ; ½B ¼  Lq 0

0

0

l Lq

wr Lq

# ð16Þ

The equation of the electromagnetic torque is given by: 

TL ¼ 3p 2 ½ðLd  Lq ÞIq Id þ Uf Iq  J dX ¼ TL  Tr  f X dt

ð17Þ

4 The Three Levels Inverter Modelling Under the same concept of the known two-level inverter and in order to have one more level, a three-level continuous voltage source is required, for example: two E/ 2, Vdc sources in series providing + E/ 2, 0 V and–E/ 2 where the source connection point is the reference, consequently an additional state possibility in relation to the two-level inverter, + E, 0 V and–E when the voltage is between the phases. Two levels, state P, state O, state N, are defined respectively. The inverter to be used in this work is the three-phase inverter with three stapled neutrals. The semiconductor power switch will be IGBT and the stapling diodes, or stapling diodes, guarantee the current flow so that state O is possible. The inverter will

286

be

A. Khemis R. et al.

activated

through 0

2 U @ 1 6 ½C ½SWith½C  ¼ 1 ð Van

Vbn

pulse

width

modulation

by

spatial

vectors.

½V  ¼

1 1 1 2 1 A and 1 2

Vcn Þt ¼ ½V ; ð S1

S3 Þt ¼ ½S; Si ¼ Ti1 Ti2  Ti3 Ti4 ; ði ¼ 1; 2; 3Þ ð18Þ

S2

5 Description of Field-Oriented Control 5.1

Field-oriented Control System

The principle of vector control is based on decoupling the electromagnetic torque of the machine and the magnetic flux. This flux is generated from the fixed rotor of the PM in relation to the rotor shaft position. This shaft sensor can find out the flux position in the coordinates. From Eq. (14) d-axis current Id = 0 and the linkage flux /d = 0 are fixed. Since, the fulx /f is constant, then the electromagnetic torque be given as fellow. 3 Ce ¼ P/f Iq 2

ð19Þ

Hence the representation follows: [6] Ce ¼ kt Iq Where: 3 kt ¼ P/f Iq 2

ð20Þ

In this way, a linear relationship is obtained between the generated torque Ce and the current Iq, this being the only variable necessary to control the torque developed by the machine.

6 Principale and Design of the Controller 6.1

Discrete Time Controller

To study the behaviour of the field oriented control by numerical simulation, and possibly for the implementation of this structure on a digital processor, it is necessary to have a model regulator in discrete time [4]. The structure of the PI controller is represented by the block diagram shown in Fig. 3. The continuous time representation of a PI controller is as follows; Z 1 eðtÞdt ð21Þ uð t Þ ¼ K p e ð t Þ þ Ki

Discrete Time Sensorless PMSM Control Controller

Ω Ω

*

DAC

Tr

ε (nT ) D(s)

T

B0 (p)

287

Process

Te +

A B

Ω

ADC T

Fig. 3. Discrete time PI controller

In this controller, KP is the proportional gain, KI is the integral gain, and e(t) is the error between the desired response and the actual system response. In the sampled time domain, with sample period T, the z-transform of this controller gives the following transfer function; DðzÞ ¼ Kp þ



Ki T z þ 1 2 z1

ð22Þ

7 Development of the EKF Algorithm In the motor-based system, it is impossible to have the exact value of the state variable disturbed by noise (a random variable). However, a set of observations can be made to estimate it statistically. Where, the inaccuracy of permanent magnet synchronous motor (PMSM) displacement sensor measurement comes from the fact that the measurement is also related to the limitation of noise and mechanical performance [12]. This article proposes the use of the Kalman filter in the system of dynamic estimation of speed based on the optimal measurement [6]. The optimality of the state estimate is obtained with the minimization of the mean estimate error. In this study, EKF, is used for the estimation of,Ids , Iqs ,xr , h, and TL [10, 11, 13] Fig. 4.

Fig. 4. Structure of Kalman filter estimator

288

A. Khemis R. et al.

Consider the system with the presence of disturbances described by the following equations, (

xðk þ 1Þ ¼ f ðxðkÞ; uðkÞÞ þ wðkÞ yðkÞ ¼ cðxðkÞÞ þ vðkÞ

ð23Þ

With: wðkÞ is the measurement noise and v (k): is the process noise.

8 Application of the Extended Kalman Filter The Kalman filter is a recursive algorithm that is used to estimate the states of a system, based on the knowledge of the signals applied to its input, the output measurement and the system model. Rewriting the equations of the motor model, so that the components of the direct and quadrature axis current are the state variables and considering, for simplification, the zero load torque, infinite inertia. As the angular velocity of the rotor varies slowly with respect to the other state variables, the assumption of zero drift or infinite inertia is consistent. The angular velocity of xr can then be treated as a variable parameter, whose value will be corrected by the Kalman algorithm itself. The noise vector v(k) expresses system errors due to, for example, inaccuracies in the model, and its covariance matrix is represented by the letter Q. the letter R represents the covariance matrix of the noise vector w(k), due to measurements using current sensors. The elements of the covariance matrices are assumed to be known, although it is the most critical part of the filter design. The solution to the Kalman problem, in this case linearized, can be put in the form of an algorithm, as shown in the following Table 1.

Table 1. Complete kalman filter algorithm

Discrete Time Sensorless PMSM Control

289

9 Sensitivity Study and Simulation Results To evaluate the proposed sensorless vector control, the algorithm was simulated in the MAT LAB software. A field-oriented control (FOC) method is implemented using discrete time PI controllers for the current control and rotor speed control loops. In the current loops, the observed disturbances are added in order to supply the coupling existing in the model. From the observed disturbances, the estimated speed is obtained and this is used in the rotor speed control circuit, allowing sensorless control. In Fig. 1, the block diagram of the implemented control system is presented. In Table 2, the parameters used for PMSM simulation are presented (Fig. 5).

Table 2. PMSM parameters Cn Iqn p Rs Rr Ld Lq /f f J

Nominal torque Nominal current Number of pole pairs Stator resistance Rotor resistance Stator inductance Rotor inductance Mutual inductance

8.5 Nm 20 A 4 0.6 X 11.6 X 1.4 mH 2.8 mH 0.12 Wb

Viscous friction coefficient 1.4e − 3 Nm/rad.s−1 Rotor inertia 1.1e − 3 kg m2


 A comparison of the obtained speeds is shown in Fig. 6. The reference wr , the measurement (xr) and the estimation ðw^r Þ of the rotational speed are plotted. Through the good convergence of the estimated speed to the measured values the algorithm for estimating the rotational speed can be validated, and thus, the control system is able to track the reference speed using the estimated information. Figure 7 shows the response of the motor when submitted to an acceleration ramp beginning at t = [0 0.4] s with x*r = 20 rad/s and the second one during the period t = [1.25 1.5] s with x*r = 30 rad/s. From this test we observe the robustness of the proposed sensorless schema at low speed value. Also, we can see the capability of good tracking and regulation with fast responses to an applied torque at 0.6 s with 4 Nm value. The evolution of the stator courant, position and the sectors succession during the PMSM DTFOC control fed by a three level inverter at low speeds region and under various load conditions are given in this simulation. On the other hand, a speed dip is noticed at the instant of step increase in load torque, but it is successfully rejected by the EKF but without it take less than 0.2 s. Figure 8 and 9 shown that the obtained current id and iq and the direct and quadrature flux are decoupled and present a good field oriented control.

290

A. Khemis R. et al.

Fig. 5. Basic DTFOC for sensorless PMSM drives.

Fig. 6. Low speed: with and without load Fig. 7. Motor speed: with and without load torque compensation torque compensation

In Fig. 10 rated mechanical load is applied to the motor between t = 0.375 s and −0.9 s after a leadless starting, to verify the performance of the EKF under loaded conditions. As shown above EKF works properly even under fully loaded case. We can see the insensibility of the control algorithm to load torque variation. Also the deference between the signal of three level’s inverter and 2 level’s inverter is clearly shown and we notice that the ripple has been minimized with the three level inverter presented in Fig. 11.

Discrete Time Sensorless PMSM Control

Fig. 8. Direct and quadrature axis stator currents

291

Fig. 9. Direct and quadrature flux magnitude stator

Figure 12 shows the measured electrical position of the rotor and the estimated position of the rotor, in steady state. The machine operates at load variation and variable speed reference of (20 to 30 rad/ s). It can be seen that the estimated position does not lag behind the measured position, with a small position estimation error.

Fig. 10. The phase current

Fig. 11. Motor torque: actual torque with 2 level and 3 level inverter.

Figure 13 shows the waveforms of the inverter line-to line output voltage and phase A current at a fundamental frequency of 50 Hz. The waveforms of the phase B and phase C currents are similar to that of the phase A current and being mutually 120° apart in phase. These test show the inverter is capable of delivering rated current at rated torque for a range of fundamental frequencies. The three level inverter have many advantage as the: a) b) c) d)

Reduce Surge Voltage- Leads to Lower Deterioration of Motor Insulation Reduce Leakage Currents (common mode current) Reduce Bearing Currents- Motor Bearing Friendly Reduced Audible Noise Level Generated by PWM

292

A. Khemis R. et al.

Fig. 12. Real and estimated Rotor position obtained in simulation

Fig. 13. Phase voltage shifted PWM “3 level inverter”

10 Conclusion The speed, rotor position and load torque estimation using the EKF was presented for an electric vehicle traction system with PMSM, in which the stability conditions of the reference monitoring of the closed-loop system were guaranteed. The simulation results obtained for the estimation of the speed and the position in the discrete time field oriented control (DTFOC) are very satisfactory from the point of view of estimation error, robustness and stability of the overall drive system in any operating condition (loaded, reversal of direction of rotation). Also, the result proves the efficiency of the three levels inverter by minimization of the torque ripples. The filtering action of EKF improves the system performance, especially at low speeds. Simulation results reveal that the tracking and regulation of the vehicle are guaranteed.

References 1. Pham-Dinh, T., Nguyen-Thanh, T.: Fuzzy speed controller for rotor flux oriented control of permanent magnet synchronous machine. In: International Symposium on Electrical & Electronics Engineering, HCM City, Vietnam, 24, 25 October (2007) 2. Mademlis, C., Agelidis, V.G.: On. IEEE Trans. Energ. Convers. 16(3), 246–252 (2001) 3. Shyu, K.K., Lai, C.K., Tsai, Y.W., Yang, D.I.: A newly robust controller design for the position control of permanent-magnet synchronous motor. IEEE Trans. Ind. Elect. 49: 558– 565 (2002). https://doi.org/10.1109/TIE.2002.1005380 4. Gan, W.-C., Li, Q.: Design and analysis of a plug-in robust compensator: an application to indirect-field-oriented-control induction machine drives. IEEE Trans. Ind. Electron. 50(2), 272–282 (2003) 5. Shu, H., Guo, C.S., Yitong, C., Xianbao, L.S.: Design of model predictive controllers for PMSM drive system based on the extended Kalman filter observer. Int. J. Electr. Hybrid Veh. 11(4), 378–394 (2019) 6. Zhang, Y., Cheng, X.F.: Sensorless control of permanent magnet synchronous motors and EKF parameter tuning research. Math. Prob. Eng. 3916231, 12–2016 (2016). https://doi.org/ 10.1155/2016/3916231

Discrete Time Sensorless PMSM Control

293

7. Ju, X., Wu, F., Shi, L.: PMSM speed control method based on Kalman filter and dynamic fuzzy control in electric vehicle CISAT 2019. J. Phys. Conf. Ser. (2019). https://doi.org/10. 1088/1742-6596/1345/2/022059 8. Qiu, A., Kojori, H.: Sensorless control of permanent magnet synchronous motor using extended Kalman filter. Electr. Comput. Eng. 3, 1557–1562 (2004) 9. Bazzi, A.M., Friedl, A.P., Choi, S., Krein, P.T.: Comparison of induction motor drives for electric vehicle applications: dynamic performance and parameter sensitivity analyses. In: 2009 IEEE International Electric Machines and Drives Conference (2009) 10. Leite, A.V., Araujo, R.E., Freitas, D.: «Full and reduced order extended kalman filter for speed estimation in induction motor drives: a comparative study» . In: Power Electronics Specialists Conference, PESC 04. IEEE, vol. 3, pp. 2293 – 2299 June 2004 11. Kosaka, M., Uda, H.: Sensorless IPMSM drive with EKF estimation of speed and rotor position. J. Low Freq. Noise Vibr. Active Control 22(4), 59–70 (2004) 12. Qiu, A., Wu, B.: Sensorless control of permanent magnet synchronous motor using extended Kalman filter. In: Proceedings Conference of CCECE, Niagara Falls, pp. 1557–1562 (2004) 13. Khalid, H., Wala, W., Yasien, F.R.: Sensorless speed estimation of permanent magnet synchronous motor using extended kalman filter. Iraqi J. Comput. Commun. Control Syst. Eng. 18(1), 64–81 (2018) 14. Shao, M., Yu, H., Yu, J. Shan, B.: Position control of permanent magnet synchronous motor speed sensorless servo system via backstepping. In: The 27th Chinese Control and Decision Conference (2015 CCDC), pp. 4089–4094 (2015) 15. Singh, S., Tiwari, A.N.: Various techniques of sensorless speed control of pmsm: a review. In: 2017 Second International Conference on Electrical, Computer and Communication Technologies (ICECCT), pp. 1–6 (2017) 16. Sira-Ramirez, H., Linares-Flores, J., Garcia Rodriguez, C., Contreras-Ordaz, M.: On the control of the permanent magnet synchronous motor: an active disturbance rejection control approach. IEEE Trans. Control Syst. Technol. 22(5), 2056–2063 (2014) 17. .Vieira, R.P., Gabbi, T.S Gründling, H.A.: Combined discrete-time sliding mode and disturbance observer for current control of induction motors. J. Control Autom. Electr. Syst. 1–9 (2017) 18. Zhang, X., Hou, B., Mei, Y.: Deadbeat predictive current control of permanentmagnet synchronous motors with stator current and disturbance observer. IEEE Trans. Power Electron. 32(5), 3818–3834 (2017) 19. Zhang, G., Wang, G., Xu, D.: Saliency based position sensorless control methods for PMSM drives - a review. Chin. J. Electr. Eng. 3(2), 14–23 (2017)

Modeling and Characteristic Analysis of Harmonics in Railway Traction Chain Meriem Aissaoui(&), M. Benidir, Hamza Bouzeria, Imen Mammeri, and Amira Chaib Ras Laboratory of Transportation Engineering and Environment - LITE, Mentouri University, Constantine 1, Constantine, Algeria [email protected], {mohamed.benidir, bouzeria.hamza}@umc.edu.dz

Abstract. Various types and forms of harmonic distortions have been reported in railway systems (ERSS) as they have attracted a lot of attention due to their severe impact on the stable performance of railways. These distortions are due to non-linear components such as converters and other elements of the ERSS. This paper presents a general study of the ERSS, taking into account the engineering requirements, where the harmonic problem consists mainly of harmonics resulting from the load increase (resistive torque) and its impact on the network and the machine. Mathematical modeling and analysis methods have been introduced. The simulation results by using the MatLab/Simulink environment were analyzed and discussed. Furthermore, the presence of harmonics was checked and the total harmonic distortion rate (THD) has therefore been observed. The focus was on providing an ideal perspective for harmonic state problems in ERSS for researchers and engineers in order to avoid these problems from occurring by giving the effective solutions. Keywords: Electrical railway systems quality
Simulation


Traction chain
Harmonics
Power

1 Introduction Rail transport is extremely important in terms of economic and environmental nature. Many countries depend on it since they have the capacity to carry large and heavy quantities of goods and people. Power system on which this type of transport depends is complicated because it contains many components of the electrical network such as (transformers, inverters, catenary and electronic devices……. etc.).Therefore, since its inception, electric rail systems have faced many challenges to harmonic problems [1], and these harmonics cause many problems and among the problems that were reported (the energy quality problem [2, 3], due to the instability of the network adapter, harmonics amplification occurs [4–6]. These negative problems of harmonic phenomena in ERSs have become a major concern. Conditions of harmonic phenomena occurrence in a system ERSs differ according to the components of the system. In order to deal with these problems, a right understanding of their nature and conditions of their occurrence is required [1]. Harmonics study in ERS should rely on mathematical © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 294–301, 2021. https://doi.org/10.1007/978-3-030-63846-7_29

Modeling and Characteristic Analysis of Harmonics

295

models or simulation models [2]. The latter makes the rate determination easy as well as its value and location. This helps its elimination and prevents its emergence as well as controlling the factors that lead to its appearance. Based on previous studies, this paper presents a study of the harmonics phenomenon resulting from overload and its effect on the railway electrical network. This work may be divided into three main sections; the first is a description of the system used along with the modeling for some of its elements. The second section deals with the study and changes been made to know the harmonics effect on the network and devices. The last section discusses the simulation results.

2 Description and Modelling of the Studied System 2.1

Description of the system

This system is a mixture of several basic components of the electric traction network which is divided into three main parts. The first part is intended for power supplies (three-phase network 25 kV/50 Hz). The second part is a substation that contains a three-phase rectifier connected to the voltage transformer and the last part is a tram composed of two inverters controlled by pulse width modulation (PWM) and each is connected to two asynchronous machines, as shown in (Fig. 1).

Fig. 1. Synoptic of proposed system

2.2

Electrical grid network

The three-phase electrical network modeled by a three-phase sine voltage and current source is illustrated in (1):

296

M. Aissaoui et al.

8 8 < Ia ¼ Im sin
ðxtÞ < Va ¼ Vm sinð
xtÞ   Vb ¼ Vm sin xt þ 2p Ib ¼ Im sin
xt  2p 3 3 : : Vc ¼ Vm sinðxt þ 4p Ic ¼ Im sin xt þ 2p 3Þ 3

2.3

ð1Þ

Three-Phase Transformer

The system constitutes a three-phase transformer modeled as shown in (2): 8 dip1 du1 > < V1 ¼ rp ip1 þ ip dt þ Np dt dip2 2 V2 ¼ rp ip2 þ ip dt þ Np du dt > : dip3 3 V3 ¼ rp ip3 þ ip dt þ Np du dt

2.4

ð2Þ

Catenary

The catenary is a mean of energy transfer between substation and the tram and is modeled as shown in Eq. (3) [8]: (

dV ðx;tÞ ¼ ðr þ jxlÞI ðx; tÞdx dt dI ðx:tÞ ¼ jxcV ðx; tÞdx dt

ð3Þ

From these equations, the input impedance of a catenary section of length connected to a load of impedance 0 can be obtained by Eq. (4) [8] ZL ¼ Z0

2.5

Z0 chðcLÞ þ Zc shðcLÞ Zc chðcLÞ þ Z0 shðcLÞ

ð4Þ

Three-Phase Inverter

In order to convert the locomotive output voltage from DC to AC, a three-lever reflectoris used and each is composed of two switches. The latter is composed of a transistor (IGBT) and a parallel diode. Each of the two switches in one arm that operates in a complementary manner. Thus, the relationship that relates these switches is given in (5) [7]: 8 0 < k1 ¼ 1  k1 k 0 ¼ 1  k2 : 02 k 3 ¼ 1  k3

ð5Þ

Modeling and Characteristic Analysis of Harmonics

297

Different voltages can also be connected, as shown in (6) [7] 2 3 2 V1 VE 4 1 4 V2 5 ¼ 3 1 V3

32 3 1 1 K1 2 1 54 K2 5 1 2 K3

2

2.6

ð6Þ

Asynchronous Machine

An asynchronous machine is known by its advantages, but it has many problems and difficulties in controlling its speed. Due to the natural separation of its field and torque, this makes machine modeling a very important step. The asynchronous machine model is given by the following Eq. (7): 8 > < Vsa ðtÞ ¼ Rs isa ðtÞ þ Vsb ðtÞ ¼ Rs isb ðtÞ þ > : Vsc ðtÞ ¼ Rs isc ðtÞ þ

8

dusa ðtÞ > dt < Vra ðt Þ dusb ðtÞ dt > Vrb ðt Þ dusc ðtÞ : Vrc ðtÞ dt

¼ Rr ira ðtÞ þ ¼ Rr irb ðtÞ þ ¼ Rr irc ðtÞ þ

dura ðtÞ dt durb ðtÞ dt durc ðtÞ dt

ð7Þ

The machine’s mechanical model can be expressed as shown in (8) [7]: Tem ¼ Tr þ f X þ J

dX dt

ð8Þ

Flux in both the stator and the rotor can also be modeled, see the following Eq. (9) 

½;s  ¼ ½Lss :½Is  þ ½Lsr :½Ir  ½;r  ¼ ½Lrs :½Is  þ ½Lrr :½Ir 

ð9Þ

Matrices of stator inductances [Ls] and rotor inductances [Lr] are given by (10) 2

Ls ½Ls  ¼ 4 Ms Ms

Ms Ls Ms

3 2 Ms Lr Ms 5½Lr  ¼ 4 Mr Ls Mr

Mr Lr Mr

3 Mr Mr 5 Lr

ð10Þ

Stator/Rotor coupling presented by Eq. (11) and (12) 2

cos h Msr ðhtÞ ¼ MrsT ðhÞ ¼ 4 cosðh þ 2p 3 cosðh  2p 3



3 cos h þ 2p cos h  2p 3 3 5 cos h cosðh  2p 3 2p cosðh þ 3 Þ cos h

Mrs ðhÞ ¼ MsrT ðhÞ ¼ Msr ðhÞ

ð11Þ ð12Þ

298

M. Aissaoui et al.

3 Simulation and Discussion The analysis is mainly based on the results of the torque resistance and speed variations which is carried out on the machine. The variation of torque resistance has been done in three stages 1, 30 and 70 Nm, respectively. The speed variation of the machine is done by the frequency change. The analyzed characteristics will help in understanding the effect of these variations such as problems and dangers that may cause in the traction network. In order to study the different changes: voltage and current in the network, speed and the electromagnetic torque and the stator current of the machine in addition to the fast Fourier transform (FFT) in currents and monitoring THD for the mains current and the machine, the MatLab software is used for simulation purposes.

Fig. 2. Grid voltage

Figures 2 and 3 show the voltage and current of the primary transformer (network). It should be noted that the network was not much affected since the load is not high enough because only one tram car is operating and the reactive capacity is respected. However, an increase in the harmonic distortion rate of THD was observed when studying FFT, when the load is weak, THD was estimated to be about 0.25%. As the load reaches 30 Nm value, an increase in TDH by 1.13% in noted. Furthermore, when 70 Nm is attained, the ratio reaches 2.18%, as shown in (Fig. 4). This indicates that an increase in load leads to the appearance of harmonics and an increase in their rate.

Modeling and Characteristic Analysis of Harmonics

299

Fig. 3. Grid current

Fig. 4. Transformer primary harmonics

Figure 5 shows three different curves (a), (b) and (c) illustrating the modifications made to the asynchronous device curve (a) that expresses velocity. Curve (c) indicates that the machine is operating normally. It can be seen that as the load increases, speed varies and the electromagnetic torque and current will also increase. It can be noted that at time 1.5 s, speed increases due to frequency that has been changed in order to maintain the speed value because any increase in load leads to a significant decrease in the Speed.

300

M. Aissaoui et al.

Fig. 5. Parameters of machine

Fig. 6. Harmonics of stator current for asynchronous machine.

However, the FFT of the device current was studied as shown in (Fig. 6), and a strong increase in the harmonics can be observed where, at a weak load, a ratio of THD is estimated at 10.15%. But when increasing the torque resistant up to 30 Nm and 70 Nm, a large increase in the harmonic distortion rate in the machine is noticed leading reaching to a high THD rate. This leads to device failure mainly due to harmonics rate increase as the load increases.

4 Conclusion The ERS system is equipped with many components represented in the transformer and the converter (AC/DC) and the inverter (DC/AC) which include devices which in turn generate harmonics and cause different effects. According to simulation results, an increase in a small load in the network is noted which in turn causes an increase in currents leading to an increase in harmonics resulting from the transformers previously mentioned. Thus, large problems can occur that may cause failures in the network and

Modeling and Characteristic Analysis of Harmonics

301

reduce the transformer lifetime as well as the machine reliability and decreases its efficiency. These problems must be taken into consideration so as to find the appropriate solutions and harmonics generated by large loads leading to high currents can be controlled.

References 1. Kaleybar, H.J., Kojabadi, H.M., Brenna, M. Foiadelli, F. Fazel, S.S., Rasi, A.: An inclusive study and classification of harmonic phenomena in electric railway systems. In: 2019 IEEE International Conference on Environment and Electrical Engineering and 2019 IEEE Industrial and Commercial Power Systems Europe (EEEIC/I&CPS Europe), pp. 1–6 (2019) 2. Hu, H., Shao, Y., Tang, L., Ma, J., He, Z., Gao, S.: Overview of harmonic and resonance in railway electrification systems. IEEE Trans. Ind. Appl. 54(5), 5227–5245 (2018) 3. Hu, H., He, Z., Gao, S.: Passive filter design for China high-speed railway with considering harmonic resonance and characteristic harmonics. IEEE Trans. Power Deliv. 30(1), 505–514 (2015) 4. Song, K., Mingli, W., Yang, S., Liu, Q., Agelidis, V.G., Konstantinou, G.: High-order harmonic resonances in traction power supplies: a review based on railway operational data, measurements, and experience. IEEE Trans. Power Electron. 35(3), 2501–2518 (2019) 5. Wang, H., Mingli, W., Sun, J.: Analysis of low-frequency oscillation in electric railways based on small-signal modeling of vehicle-grid system indqframe. IEEE Trans. Power Electron. 30(9), 5318–5330 (2015) 6. Harnefors, L., Wang, X., Yepes, A.G., Blaabjerg, F.: Passivitybased stability assessment of grid-connected vscs—an overview. IEEE J. Emerging Sel. Top. Power Electron. 4(1), 116– 125 (2016) 7. Chaouali, H., Othmani, H., Mezghani, D., Jouini, H., Mami, A.: Fuzzy logic control scheme for a 3 phased asynchronous machine fed by Kaneka GSA-60 PV panels. In: 2016 7th International Renewable Energy Congress (IREC), pp. 1–5. IEEE (2016) 8. Stackler, C., Morel, F., Ladoux, P., Dworakowski, P.: 25 kV–50 Hz railway supply modelling for medium frequencies (0–5 kHz). In: 2016 International Conference on Electrical Systems for Aircraft, Railway, Ship Propulsion and Road Vehicles & International Transportation Electrification Conference (ESARS-ITEC), pp. 1–6. IEEE (2016)

Hierarchical Control of Paralleled Voltage Source Inverters in Islanded Single Phase Microgrids Ilyas Bennia, Yacine Daili, and Abdelghani Harrag(&) Mechatronics Laboratory (LMETR), Optics and Precision Mechanics Institute, Ferhat Abbas University, Sétif 1, Cite Maabouda (ex. Travaux), 19000 Setif, Algeria {ilyas.bennia,yacine.daili,a.harrag}@univ-setif.dz

Abstract. Voltage source inverters (VSIs) are key elements of microgrids (MGs), their control presents an important challenge in order to ensure the system stability. In this paper, a modeling and hierarchical control of singlephase paralleled VSIs forming a MG is presented. The diagram of the hierarchical control consists of inner, primary, and secondary control levels. The inner control commonly referred as zero level is used to regulate the VSI output voltage. The primary control is based on the droop control method and the virtual impedance loop with the objective of sharing the power accurately regardless of the line impedance of the inverters. Secondary control is used to regulate the voltage and frequency to their rated values and eliminating the deviation caused by the primary control as well as synchronizes the MG voltage with the main grid for a smooth transition. The simulation was conducted to show the feasibility of the control system. The results demonstrate the disturbance rejection performance and the stable operation of the studied MG with the proposed control scheme. Keywords: Hierarchical control
Distributed generators (DGs)
Voltage source inverters (VSIs)
Droop control method
Islanded mode
Single phase microgrid

1 Introduction The electrical system power tends to be more intelligent and more flexible in recent years with the objective of responding to the users and copes with their needs. One of the most challenging tasks to reach this goal is to change from the centralized power generation toward the decentralization one, where the distributed generators (DGs) play the primary role of energy production due to their multiplicity resources including various generators and renewable energy sources, such as photovoltaic cells and wind power generator, which is characterized by the reduced cost of generation and the environmental friendly [1], however, the stochastic behavior of this latter referring to their prime mover cause high fluctuation in electrical parameters which makes the decentralization of power production more challenging regarding the stability of power systems. Hence, new technologies and new control concepts is required, one of these © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 302–313, 2021. https://doi.org/10.1007/978-3-030-63846-7_30

Hierarchical Control of Paralleled Voltage Source Inverters

303

concepts is the Microgrid (MG), MG has received increasing interest in the last decade to cope with the aforementioned issues [2, 3]. Microgrid is the building block of the smart grid as a small local grid, MG forming by the interconnection of different technologies such as power electronics converters, DGs which are often small units widely distributed, energy storage systems, and telecommunications infrastructure to reach the load supplying [4], MG can operate either in islanded mode or connected to the main grid [5, 6]. In recent years, many control approaches have been proposed, one of the most popular is the hierarchical control concept which is introduced to deal with the control challenge of MG [7], different hierarchical approaches are presented in literature such as the traditional PI hierarchical control, which consists of three levels [8]: primary control of this hierarchy using droop control method [9] to regulate the frequency and voltage depending on the active and reactive powers aiming to share the power without communication [8], the secondary control level is used to restore the system frequency and voltage to the nominal values, removing deviations produced by primary level [8, 10]. The tertiary control is responsible for power exchange with external grid or/and with other MGs [8], decentralized model predictive hierarchical control is presented in [11] with both primary and secondary control to address power quality and unequal power-sharing, this approach released low value of error voltage which enhance directly the reactive power-sharing, hierarchical multi-agent system was proposed and validated in [12] and stochastic hierarchical control [13]. This paper aims to present modeling of a single-phase microgrid operating in island mode with PI-hierarchical control, the control system of the studied MG consists of primary and secondary control (the tertiary control level is not studied in this work). In the primary control level, the droop control method with a virtual impedance is used to improve reactive power sharing accuracy due to the mismatched line impedances of the Microgrid. The secondary control level is tested for both decentralized control using an improved droop control based on a washout-filter studied in [14] as an equivalent secondary control, and centralized control which sends proper reference signals to each of the DGs to restore the frequency and voltage amplitude at PCC to their nominal values. The studied microgrid with the hierarchical control system has been tested in MATLAB/Simulink environment. This present paper is organized as follows. Section 2 presents the microgrid system configuration. In Sect. 3, the droop control method and the virtual impedance loop were explained. Secondary control is mentioned in Sect. 4. Simulation results were presented in Sect. 5. Section 6 concludes this paper.

2 Microgrid System Configuration A typical structure of a single-phase microgrid with n DG units is given in Fig. 1. For each DG unit, the dc power is provided by the renewable energy sources or an energy storage system. The ideal dc-link with fixed dc voltage is assumed in this paper rather than the distributed generation (DG) to simplify studies. DG are interfaced to the main grid through a single-phase voltage source inverter (VSI), which is connected at the point of common coupling (PCC) through an LCL filter and line impedance [15].

304

I. Bennia et al.

Fig. 1. Typical microgrid configuration.

The hierarchical scheme consists of two control levels. The primary level comprises DG local controllers and the secondary level is a central controller which sends proper reference signals to each of the DGs to restore the frequency and voltage amplitude at PCC to their nominal values [8]. The secondary controller can be far from DGs. Thus, the low bandwidth communication (LBC) is applied for sending the data information of the secondary controllers and the PCC voltage from the microgrid control center (MGCC) to the local controller to realize the PCC frequency and voltage amplitude compensation in DG units in a synchronized manner. Also, a static switch is used to dynamically disconnect the microgrid from the main grid in case of grid faults. Although the proposed control strategies can operate either in a grid connected mode or in an islanded mode [8, 16], only the islanded operation mode will be considered in this paper. The inner voltages control loop is used to track accurately the voltage references provided by the primary control level. In this work, the inverter output voltage control is adopted using a standard proportional regulator with a feedforward control loop to improve the controller response speed and to minimize the steady-state error before generate the reference current vector. A current controller is needed to provide active damping of the resonance created by the LCL filter and shape the voltage across the filter inductor, where current error has to be the minimum possible. A standard proportional current regulator with the current feedback filter is used for current regulation [17, 18]. The voltage control design is not detailed in this paper.

3 Droop Control and Virtual Impedance Loop The sinusoidal reference for the voltage controller is obtained from the droop method [8]. The principle of the droop control (P/f, Q/V) is described in Fig. 2, according to this method, the voltage and frequency references are generated through the

Hierarchical Control of Paralleled Voltage Source Inverters

305

measurement active and reactive powers to shape the droop curves which are described by the following equations [10]: x ¼ x  GP ðsÞðP  P Þ

ð1Þ

V ¼ V   GQ ðsÞðQ  Q Þ

ð2Þ

ω

(a)

V

(b) ω*

-Pmax

Storage

Generation

V*

+Pmax

P

-Qmax

C Load

L Load

+Qmax

Q

Fig. 2. (a) Frequency versus active power, (b) voltage versus reactive power.

being x the angular frequency and x is the angular frequency reference, P and Q are the active and reactive power references to be injected in the grid for grid-connected mode, in island mode P and Q set to zero, GP (s) and GQ(s) are the controller transfer functions, which are described follows: G P ðsÞ ¼ K w

ð3Þ

GQ ðsÞ ¼ Ka

ð4Þ

Where Kw and Ka are the static droop coefficients, the static droop coefficients can be obtained by the following equations: Kw ¼

Df Pmax

ð5Þ

Ka ¼

DV Qmax

ð6Þ

where Df is the maximum frequency deviation, Pmax is the maximum active power, DV is the maximum amplitude deviation and Qmax is the maximum active power. Figure 3 illustrates the implementation of droop control, where the instantaneous power calculated passe through a low pass filter to eliminate the ripples [17].

306

I. Bennia et al.

Fig. 3. Block diagram of the droop controller.

In addition, to overcome the problem of power coupling and current sharing caused by high R/X ratio in low voltage distribution grids, virtual impedance loop is added to improve the current sharing between the VSI by fixing and normalizing the output impedance of the VSI which will determine the P/Q power angle/amplitude relationship thus avoid using additional physical inductors/resistors [17, 19]. Figure 4 shows the additional block of the virtual impedance loop, the output impedance of the VSI must be inductive sufficiently. The additional block of the virtual impedance loop can be expressed in as shown follows: Zv ¼

s Lv ss þ 1

ð7Þ

Being Lv is the inductance of the virtual impedance and s is the time constant of the high pass filter used to approximate the derivative in the transfer function of the ideal virtual inductance (Zv = s.Lv). And the output impedance can be defined as: Zo ¼

Lf s þ GðsÞZv LCs2 þ Kc Cs þ Kv þ 1

Fig. 4. Block diagram of the droop controller and the virtual impedance loop.

ð8Þ

Hierarchical Control of Paralleled Voltage Source Inverters

307

4 Secondary Control for Frequency and Voltage Restoration In grid-connected mode the voltage and frequency are carried out by the main grid [7], however in island mode, they are set on by the MG itself, secondary control is added in this direction to remove the limitation of primary control [10] which are the voltage and frequency deviation hence maintaining the voltage and frequency in a specific range, ±0.3 Hz for frequency in UCTE (Union for the Co-ordination of Transmission of Electricity, Continental Europe) it can also enhance the power sharing in the MG [10], further ensuring the stability operating of the MG, the secondary control can be classified into three categories [20] the first one and the most popular in a centralized way is using a microgrid centralized central controller which sent the correcter signals for the local controllers to adjust their outputs as cited in [8], the second classification is the decentralized manner by local controllers without using communication infrastructure which give more resilience for MGs, the last classification is the distributed secondary control which based on a number of distributed generation cooperating together to control MG with low data rate communication infrastructure [21], in this paper only centralized and decentralized control were simulated. The main centralized restoration and regulation process start by measuring the voltage and frequency values inside the MG then compare them to the nominal values in the MGCC and this latter generate the correcter signal using PI controllers to set on all units at the nominal values and toward zero steady state-error after every change of load or generation [8]. The voltage and frequency restoration compensators are identified by the following equations [10]:
  Wrest ¼ Kpsw WMG  WMG þ Kisw
  Vrest ¼ Kpsv VMG  WMG þ Kisv

Z

Z


  WMG  WMG dt


   WMG dt VMG

ð9Þ ð10Þ

being Kpsw; Kpsv; Kisw and Kpsv are the secondary control compensators coefficients. In this case, Wrest and Vrest must be limited to not exceed the maximum allowed voltage and frequency deviation and respect the grid requirement [20]. Figure 5 depicts the overall system and control scheme including tow levels primary and secondary control. In order to connect the DGs to each other forming a MG and to connect this latter to the grid phase between the grid and the MG will be synchronized utilizing the synchronization control loop which can be seen as a conventional PLL [10].

308

I. Bennia et al.

Fig. 5. Block diagram of two VSIs forming a MG with control system (centralized secondary control).

However t is worth to note that the failure of the communication in the secondary control can affect the overall system [22] which decrease the MG reliability, in order to overcome this issue many decentralized approaches have been proposed in the literature such as decentralized model predictive hierarchical control in [11], this approach allows a successful voltage and frequency restoration and enhanced the power quality, decentralized adaptative control is presented in [23] consist of a new term of power derivative supplemented to the droop control to compensate the deviation in steadystate, furthermore an improved droop control based on bandpass filter implemented in local controllers was proposed in [14] which called also washout filter, this improved droop control is an equivalent secondary control without needing any communication as depicted in Fig. (6) it is obvious the difference between the conventional droop control in figure and the improved one. This can be expressed by: x ¼ x   G P ðsÞ

s ðP  P Þ s þ Kp

ð11Þ

V ¼ V   GQ ðsÞ

s ðQ  Q  Þ s þ Kq

ð12Þ

Fig. 6. Block diagram of the improved droop controller.

Hierarchical Control of Paralleled Voltage Source Inverters

309

Fig. 7. Block diagram of two VSIs forming a MG with control system (decentralized secondary control).

It is clear that the improved droop control utilizes dynamic feedback based on the washout filter. A washout filter is a first-order high-pass filter that rejects the DC component and passes the transient component of the signal, the improved droop control eliminates the need of communication however it decreases the observability and the monitoring of the system, Fig. (7) shows the implementation of the decentralized secondary control using the improved droop control.

5 Results and Discussion The above strategy of control has been tested in MATLAB/Simulink environment, which consists of two single-phase inverters in island mode, controlled hierarchically compromising tow level. The circuit diagram of each inverter is shown in Fig. 6. In order to evaluate the effectiveness of hierarchical control. The parameters used in simulation are listed in Table 1. Table 1. Simulation parameters. Parameter Power stage Grid voltage Grid frequency Output inductance Filter inductance Filter capacitance Line 1 Line 2 Load DC voltage

Symbol Value Vg f Lo L C R1=L1 R2=L2  Vd c

Units

311 V 50 Hz 250 lH 800 lH 60 lF 0.5/1 m X/lH 5/10 m X/lH 1000/500 W/VAR 400 V (continued)

310

I. Bennia et al. Table 1. (continued) Parameter Symbol Power stage Voltage/Current P control Voltage proportional gain Kv Current proportional gain Kc Cutoff frequency wf Primary control Proportional frequency droop Kw Proportional amplitude droop Kv Virtual inductance Lv Time constant s Centralized secondary control Frequency proportional term Kpsw Frequency integral term Kisw Amplitude proportional term Kpsv Amplitude integral term Kisv Decentralized secondary control Frequency proportional term Kp Frequency integral term Kq

Value

Units

2 2.2 200

  kHz

0.0003 0.004 650 1/1500

W/rd Var/V lH 1=s

0.2 50 0.2 50

 1=s  1=s

2.2 3

 −

From Fig. 8 and 9 show the operation of one VSI unit where Fig. 8.a and 8.b illustrate the shape of current and voltage output, as well as the active and reactive power clarifying the stable operation of the system, which achieves a satisfactory signal performance for the voltage response, reaching its nominal value.

(a)

(b)

Fig. 8. a) Current and voltage, b) Active and reactive power.

Hierarchical Control of Paralleled Voltage Source Inverters

311

Figure 9. (a) and (b) show the restoration of frequency and voltage using the secondary control, initially the microgrid is operating in the steady-state under no load the voltage and the frequency were at their set points, after connecting the load at 0.5 s the drop of the voltage was about 1.86 V and 0.05 Hz for frequency, as shown in this figure the centralized secondary control was able to eliminate the voltage and frequency deviation in 0.15 s while the decentralized is about 10 time of the centralized one.

(a)

(b)

Fig. 9. a) PCC frequency, b) PCC voltage amplitude (in this figure time in second or seconds).

(a)

(b)

(c)

Fig. 10. Active and reactive power at the output of the inverters.

312

I. Bennia et al.

Figure 10 shows the parallel operation of the tow VSI units under centralized secondary control where the second VSI is triggered at 0.5 s to analyze the dynamic of the system, as appears, the active and reactive power were shared equally with a high accuracy and a fast dynamic with a high match between signals due to the primary and the secondary control where the primary one is presented by the droop control mechanism and the virtual impedance loop which removes the mismatched inductive/resistive feeders impedance notice that the line 2 is longer than the line 1 ten times and the distribution was in low voltage, the secondary control improved the power quality by compensate the deviation of voltage and frequency.

6 Conclusion In this paper, an hierarchical control of a single-phase microgrid operating in island mode is studied, a typical structure of a single-phase microgrid is presented and explained, as well as droop control and virtual impedance loop were defined. The hierarchical control is tested until the secondary control for both centralized and decentralized secondary control, the control schemes were formulated and illustrated. The droop control loop and the virtual impedance control loop formed the primary control concept with the task of providing the voltage reference for the voltage control loop. Secondary control is responsible for removing the voltage and frequency deviation came from the primary control thus enhance the power sharing and the power quality. Hierarchical control strategy was adapted to controlling parallel single-phase VSIs and shows a high flexibility and effectiveness with a fast dynamic of the MG control system. Acknowledgement. The Algerian Ministry of Higher Education and Scientific Research via the DGRSDT supported this research (PRFU Project code: A01L07UN190120180005).

References 1. Hossain, M.A., Pota, H.R., Hossain, M.J., Blaabjerg, F.: Evolution of microgrids with converter-interfaced generations: challenges and opportunities. Int. J. Electr. Power Energy Syst. 109, 160–186 (2019) 2. Arfeen, Z.A., Khairuddin, A.B., Larik, R.M., Saeed, M.S.: Control of distributed generation systems for microgrid applications: a technological review. Int. Trans. Electr. Energy Syst. 29(9), e12072 (2019) 3. Vasquez, J.C., Guerrero, J.M., Miret, J., Castilla, M., De Vicuna, L.G.: Hierarchical control of intelligent microgrids. IEEE Ind. Electron. Mag. 4(4), 23–29 (2010) 4. Singh, A., Suhag, S.: Trends in islanded microgrid frequency regulation–a review. Smart Sci. 7(2), 91–115 (2019) 5. Lopes, J.P., Moreira, C., Madureira, A.: Defining control strategies for microgrids islanded operation. IEEE Trans. Power Syst. 21(2), 916–924 (2006) 6. Shafiee, Q.: Multi-Functional Distributed Secondary Control for Autonomous Microgrids (2014)

Hierarchical Control of Paralleled Voltage Source Inverters

313

7. Zambroni de Souza, A.C., Castilla, M. (eds.): Microgrids Design and Implementation. Springer, Cham (2019). https://doi.org/10.1007/978-3-319-98687-6 8. Guerrero, J.M., Vasquez, J.C., Matas, J., De Vicuña, L.G., Castilla, M.: Hierarchical control of droop-controlled AC and DC microgrids—A general approach toward standardization. IEEE Trans. Ind. Electron. 58(1), 158–172 (2010) 9. Guerrero, J.M., De Vicuna, L.G., Matas, J., Castilla, M., Miret, J.: A wireless controller to enhance dynamic performance of parallel inverters in distributed generation systems. IEEE Trans. Power Electron. 19(5), 1205–1213 (2004) 10. Vasquez, J.C., Guerrero, J.M., Savaghebi, M., Eloy-Garcia, J., Teodorescu, R.: Modeling, analysis, and design of stationary-reference-frame droop-controlled parallel three-phase voltage source inverters. IEEE Trans. Ind. Electron. 60(4), 1271–1280 (2012) 11. Jayachandran, M., Ravi, G.: Decentralized model predictive hierarchical control strategy for islanded AC microgrids . Electr. Power Syst. Res. 170, 92–100 (2019). https://doi.org/10. 1016/j.epsr.2019.01.010 12. Cintuglu, M.H., Youssef, T., Mohammed, O.A.: Development and application of a real-time testbed for multiagent system interoperability: a case study on hierarchical microgrid control. IEEE Trans. Smart Grid 9(3), 1759–1768 (2018). https://doi.org/10.1109/tsg.2016.2599265 13. Wang, D., Qiu, J., Reedman, L., Meng, K., Lai, L.L.: Two-stage energy management for networked microgrids with high renewable penetration. Appl. Energy 226, 39–48 (2018). https://doi.org/10.1016/j.apenergy.2018.05.112 14. Yazdanian, M., Mehrizi-Sani, A.: Washout filter-based power sharing. IEEE Trans. Smart Grid, 1−2 (2015). https://doi.org/10.1109/tsg.2015.2497964 15. Issa, W., Sharkh, S., Mallick, T., Abusara, M.: Improved reactive power sharing for paralleloperated inverters in islanded microgrids. J. Power Electron. 16(3), 1152–1162 (2016). https://doi.org/10.6113/jpe.2016.16.3.1152 16. Planas, E., Gil-de-Muro, A., Andreu, J., Kortabarria, I., de Alegría, I.M.: General aspects, hierarchical controls and droop methods in microgrids: a review. Renew. Sustain. Energy Rev. 17, 147–159 (2013) 17. Issa, W.R.: Improved control strategies for droop-controlled inverter-based microgrid (2015) 18. Mohamed, Y.A.-R.I., El-Saadany, E.F.: Adaptive decentralized droop controller to preserve power sharing stability of paralleled inverters in distributed generation microgrids. IEEE Trans. Power Electron. 23(6), 2806–2816 (2008) 19. Micallef, A., Apap, M., Spiteri-Staines, C., Guerrero, J.M.: Performance comparison for virtual impedance techniques used in droop controlled islanded microgrids, pp. 695−700 (2016) 20. Yamashita, D.Y., Vechiu, I., Gaubert, J.-P.: A review of hierarchical control for building microgrids . Renew. Sustain. Energy Rev. 118, 109523 (2020). https://doi.org/10.1016/j. rser.2019.109523 21. Khayat, Y., et al.: On the secondary control architectures of AC microgrids: an overview. IEEE Trans. Power Electron. 35(6), 6482–6500 (2020). https://doi.org/10.1109/tpel.2019. 2951694 22. Meng, L., Savaghebi, M., Andrade, F., Vasquez, J.C., Guerrero, J.M., Graells, M.: Microgrid central controller development and hierarchical control implementation in the intelligent microgrid lab of Aalborg University, pp. 2585−2592 (2015) 23. Heydari, R., Khayat, Y., Naderi, M., Anvari-Moghaddam, A., Dragicevic, T., Blaabjerg, F.: A decentralized adaptive control method for frequency regulation and power sharing in autonomous microgrids. In: 2019 IEEE 28th International Symposium on Industrial Electronics (ISIE), Vancouver, BC, Canada, June 2019, pp. 2427−2432. https://doi.org/10. 1109/ISIE.2019.8781102

Faults Detection and Diagnosis of Concentrated Photovoltaic Systems N. Kellil1(&), A. Aissat2, and A. Mellit3 1

Development Unit of Solar Equipment, Development Centre of Renewable Energies, RN 11, POB 386, 42415 Tipaza, Algeria [email protected] 2 Saad Dahleb University, POB 270, 09000 Blida, Algeria [email protected] 3 Mohamed Seddik Benyahia University, POB 98, 18000 Ouled Aïssa, Jijel, Algeria [email protected]

Abstract. Power losses and maintenance in large photovoltaic (PV) systems are the most important factors causing the price increase per kilowatt hour, also known as the Levelized Cost of Electricity (LCOE). The challenge is to detect faults and make a correct diagnosis, which permits rapid intervention (at the right time and in the right place, with the right decision) to avoid costly breakdowns or even decrease in the production. The Concentrated Photovoltaic (CPV) technology is more promising and also more complicated, compared to classic PV installations (fixed), because it requires additional tools such as mechanisms for the sun tracking and some optical components (Fresnel lenses) for the concentration of sunlight beam. This could be the origin of additional faults. This work aims to design a system for detecting and diagnosing faults occurring during the operation of CPV systems. The considered sun tracker is installed at BouIsmail, north of Algeria in Mediterranean climate. Only mechanical faults, that could affect the normal functioning of the tracking system, are considered in this paper. Keywords: CPV system


Mechanical faults
Detection and diagnosis

1 Introduction Nowadays, the search for alternative sources to fossil energies has become more obvious. This is mainly due to the adoption of some energy security strategies and also to environmental restrictions (reducing the CO2 emission rate). Photovoltaic (PV) energy is clean, sustainable and inexhaustible [1]. It has been considerably developed from the first PV cell manufactured in the 1950s, to large PV plants. Since then, the global installed PV market has been growing significantly (*650 GWp in 2019), with an annual rate of *120 GWp of installed PV [2]. The researchers aimed to reduce the costs of solar electric energy, to make it more competitive and attractive. Indeed, the LCOE decreased considerably after the rapid growth of the PV module manufacturing industry in China after 2005, reaching © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 314–324, 2021. https://doi.org/10.1007/978-3-030-63846-7_31

Faults Detection and Diagnosis of Concentrated Photovoltaic Systems

315

$0.069/kWh this last decade [3]. In the second half of 2018, the range of LCOE for fixed systems varied between $0.038/kWh and $0.147/kWh, and between $0.041/kWh and $0.083/kWh for 1-axis tracking systems [4]. The introduction of CPV systems with the new generation of high-efficiency multijunction solar cells (MJSC) on the solar electricity market has had a major impact on lowering electricity production costs. In fact, CPV technology did not make much progress until recently with the development of MJSC, although this technology has been around since the mid-1970s [5]. The 1st CPV plants were installed and connected to the grid in 2008 in Puertollano, Spain (Fig. 1) [6, 9].

Fig. 1. “Instituto Sistemas Fotovoltaica Concentrador” (ISFOC), Puertollano, Spain

In 2011, the annual installed capacity of the CPV technology did not exceed 0.26% of the total installed PV capacity [7]. In 2016, cumulative CPV installations reached 350 MW, which represents less than 0.2% of the global installed capacity over the world [8]. Figure 2 shows the evolution of the annual installed CPV capacity (left), and the distribution by country of cumulative CPV installations at the end of 2016 (right).

Fig. 2. Annual CPV installations (left) from 2001 to 2017 and cumulative CPV in grid installations at the end of 2016 by country (right) [8].

316

N. Kellil et al.

CPV prices decreases sharply between 2007 and 2013, from (7–10 $/W [9]) to (2.62 $/W [10] and 3.2 $/W [11]). This had a direct impact on the LCOE value of the CPV technology, which was in the range of 0.10 and 0.15 €/kWh in 2013, for the area where the DNI is 2000 kWh/ m2 [12]. During its operation, the CPV system may be breakdown, or have faults and anomalies which may affect its efficiency and cause power losses which reduce its performance and increase maintenance costs. Nevertheless, the detection, localization and correct diagnosis of the fault prevent expensive breakdowns or even the complete stoppage of the CPV installation production. The simplest method for detecting and diagnosing a fault in a CPV systems is the so-called prediction method. Indeed, the electrical characteristics (current and voltage) of a CPV generator are measured using sensors as well as meteorological data (solar irradiance and temperature). These data are saved in a datalogger. In this case, the measured output power of the CPV generator is compared to the simulated one. The difference tells about the presence or not of a fault (shadowing, cable cut, …). This method has had certain limitations [13]. The modelling and simulation of PV modules are essential for the analysis of normal and faulty operating situations of the PV system [14]. Most of these methods use elements of artificial intelligence, to create a functional model, so as to allow machine learning [15]. In addition to the known faults associated with conventional PV systems, CPV systems can be subject to other additional faults, such as mechanical failures (blockage of sun tracking systems), as well as faults related to optical precision of sunlight concentrators, and defects in cooling system. In this paper, we will focus on the detection, localization and diagnosis of faults related to sun trackers, as well as faults related to the deviation of the focal point of the concentrated sunlight beam due to the lack of the accuracy of sun trackers. The paper is organized as follows: Sect. 2 provides a CPV system description. Section 3 shows different faults that may occur during the CPV system operation. Section 4, describes some available methods for fault detection and diagnosis in CPV systems, and finally this work is concluded by some remarks and perspectives.

2 Concentrated Photovoltaic System The CPV systems are a new technology of PV generators, based on the use of small solar cells called Multijunction solar cells (MJSC), with high efficiency. Those MJSC are mounted under high optical concentrators (lens or mirrors) is a new generation of PV systems. Unlike the conventional PV generators which convert the global solar radiation (direct, diffuse, albedo, …) into electric current, the CPV system generate electric current by converting mainly solar Direct Norma Irradiation (DNI) by the MJSC under high concentration. The dependence on solar DNI makes the addition of a sun tracking system more than essential. As the concentration of the sunlight beam increases the temperature of the cells, some CPV systems are equipped with cooling systems.

Faults Detection and Diagnosis of Concentrated Photovoltaic Systems

317

A. Sun Tracking System The sun tracker is a mechanism that allows the active surface of a CPV system to be directly in front of the sun. In other words, it is used to minimize the angle of the solar beam incoming angle relative to the normal of the active surface of the CPV. It is used for both CPV and concentrated solar power (CSP) applications. There are two kinds of solar trackers: single-axis and two-axis trackers. In our Unit for Developing Solar Equipment (UDES – BouIsmail), a FEINA SF09 sun tracker is installed (Fig. 3.). It uses 2 types of sun tracking methods: blind tracking and real-time tracking. Each type of configuration complements the other one. The 1st method is based on the astronomical calculations using GPS data (latitude and longitude), date and time given by the user to turn the CPV system directly in front of the incoming solar beam. This configuration has considerable errors for CPV applications. These errors may be acceptable for CSP applications. While for the 2nd method, the electronic control card calculate the differences between measured values of resistances for the 4 LDR (Light Decreasing Resistance or Photoresistor) constituting le solar sensor. The 4 LDR represents the 4 direction (East, West, North and South). Another LDR measures the intensity of solar radiation, providing information of the state of the sky and commutate between the 2 tracking methods. Real-time sun tracking is more accurate.

Fig. 3. Sun tracker installed at the UDES – BouIsmail Algeria.

B. Multijunction Solar Cells The Multijunction Solar Cell (MJSC) is a stack of several sub-cells [16]. Each sub-cell is made of a specific semiconductor alloy. This allows each sub-cell having unique band gap to generate electric current by absorbing different wavelengths segments of sunlight. This allows a wider range of absorption of the sunlight by the MJSC [17].

318

N. Kellil et al.

C. Fresnel Lens It is a refractive optical concentrator, more compact and lighter than an ordinary lens. It was developed in 1819 by Augustin Fresnel. It was used for the 1st time, for maritime navigation (lighthouses). It is composed of a succession of annular prisms, which direct the incoming DNI by refraction to the focal point (active surface of the MJSC).

3 CPV System Faults In literature, PV systems faults can be classified in different categories (affected components, internal or external faults causes, permanent or temporary faults …). Some faults are caused during manufacturing process, such as encapsulation faults [18], cracking of cells [19], and damage of interconnection between cells [20]. Others faults are caused by the environment of the PV installation, such as shading [21]. In [22] the authors have summarized some types of fault that can affect the PV system in Table 1. Some defects in CPV systems are similar to those that can affect PV systems. For the CPV systems, mechanical obstacles can stuck or even stop the movement of the sun tracking system (horizontally and/or tilt). This defect increases the angle between the solar DNI and the normal of the receiving surface of the concentrator, which affects the performance of CPV modules. Table 2 regroups the two faults and their effects that can be caused by a failure in orientation motors or presence of an obstacle (Fig. 4a). The Violent storms can be also considered as an external risk or default. For this, an anemometer, measuring the wind speed, is mounted a little higher, in an open area, near the solar tracker. The tracker automatically moves to a horizontal position once the wind speed reaches a significant value, which can be defined by the user. This position will decrease the area opposite to the wind direction, and therefore minimize the risk of total collapse.

Table 1. Summary of different type of faults with connected information: affected components, causes and effects [22]. Type of fault Affected Causes components Hot spot Cells/Module External causes (HS) - Soiling, dust, snow and shadow - PVMs of different classes or technology Internal causes - Fragmentation of cells - Fragmentation of cells - Current mismatch between cells - High resistance or “cold” solder points - Aging and degradation of cells Diode faults Bypass diode - Partially shaded cells (PS) - Overheating (DF) (BpD) or blocking diode (BkD)

Effects - Damage of solar cells - Open circuits (OC) - Reduce efficiency and reliability

- Damage diodes - Short circuited diode, open circuited or Shunted diode (continued)

Faults Detection and Diagnosis of Concentrated Photovoltaic Systems

319

Table 1. (continued) Type of fault Affected Causes components Junction box Junction box - Fretting corrosion, fault (JBF) (JB) - Loosen and oxidation

PV module PVM fault (PVMF) and PV array fault (PVAF)

Ground fault PVA or PV (GF) String (PVStr)

Arc fault (AC)

PVStr

Linetoline fault (LLF)

PVA

External causes - Glass breakage of frameless PVMs caused by the clamps - Connector failure (disconnection) - Isolated from ground - Encapsulation - Wiring mistake at install fault Internal causes - Corrosion of solar cells - Manufacturing defects - Delaminated, bubbles, yellowing, scratches and burnt cells - Insulation failure of cables - Incidental short circuit between normal conductor and ground - GF within PVMs cable insulation damage during the installation - GF within the PVMs (e.g., degraded sealant and water ingress) - Insulation damage of cables. - Accidental short circuit inside the PV combiner box - A short break is created in a conductor - Two conductors of widely differing voltage are placed near one another - Degradation in solder joints, wiring or connections inside the insulation damage due to mechanical damage, aging, or wild life junction box, loosening of screws - An unintentional low impedance current path between two points. - Insulation failure of cables - Incidental short circuit between current carrying conductors - LLFs within the DC junction box

Effects - Damage and risk of fire - Reduce efficiency and reliability - Shunted module - Short circuit (SC) within a module - SC between PVMs - Leakage currents within a PVM - Damage PVM - Reduce efficiency and reliability - Reduced output power

- Risk of fire

- Damage of PVStr - Risk of fire

- Damage PVMs and conductors. - Risk of fire

320

N. Kellil et al. Table 2. Mechanical faults causes and effects affecting the sun tracker.

Type of fault Tilt axis machanical blocking (Fig. 4.b)

Affected components - Tilt motor - Electronic system - Counter switch - Motor cables

Hourly axis (Fig. 4.c)

Same affected components for the Hourly motor

(a)

Causes - The tilt motor does not work - The connections to the tilt motor have failed - The switch does not work or fails to activate - The switch connections have failed - The tracker is obstructed and therefore cannot move Same causes for the Hourly motor

(b)

Effects - The tracker gets stuck,

(c)

Fig. 4. Faults of stuck on tilt axis and/or hourly axis (a), and display error messages: (b) for tilt axis and (c) for hourly axis

Figure 5 shows a drop in solar irradiance received by the solar collector installed on the solar tracker, which is due to blockage of the sun tracking system. This fault causes a loss of power and decreases the solar system performance.

Fig. 5. Global solar irradiance received by both fixed tilted and by sun tracker system (July, 04th, 05th and 07th 2020).

Faults Detection and Diagnosis of Concentrated Photovoltaic Systems

321

The diagram below (see Fig. 6) illustrates the different causes of the failures of the 2 Sun Tracker motors. Some faults are of electrical origin and others are of mechanical origin.

Fig. 6. Schematic of motors blockage causes

4 Methods of Detection and Diganosis of Sun Tracker Systems Faults In the industrial field, monitoring systems are installed for the detection, identification and even isolation of failures that may occur, to improve production and also to protect people and their environment [23]. Similarly, solar power plants are equipped with telemonitoring systems including faults detection systems. Some of methods of faults detection are mentioned in this paper. Different methods are used for the solar systems faults detection, from PV cell to solar power plants, such as electric characterization [22], which uses methods independent of climate data. It also depends on voltage and current measurements to compare the simulated and measured values of the solar system output parameters. This technique is also based on the analysis of power loss and monitoring of residual currents (SCR) and isolation and also makes it possible to detect earth fault. Visual inspection [24], Ultrasonic inspection [25] and Infrared imaging by measuring the temperature distribution of the PV cell [26]. These methods have improved the performance and lifespan of solar systems. In this study, a field of solar trackers (See Fig. 7) is considered. So, a network of sensors should be configured for system telemonitoring. To detect mechanical faults, two methods have been suggested. The 1st one consists in measuring the differences between the light intensities received by each of the solar collectors, using the solar sensors mounted on each of the solar trackers (LDR diodes). The measured data will be then compared with those measured by the weather station (Pyrhéliomètre). The 2nd one involves in detecting the signals received by the switch counter. To do this, each switch counter switch is given an address.

322

N. Kellil et al.

Fig. 7. Diagram of field of solar trackers connected to a telemonitoring system based on IoT.

5 Concluding Remarks and Future Works In this work, blocking faults of the sun tracking system were highlighted. This type of faults are among the major faults of CPV systems. Two methods have been suggested for the detection of these faults. Solar irradiance curves, captured by both installations (fixe tilt and tracked) for three days, where the sun tracking system was stuck, showed a significant loss of power received by the sun tracking system. An IoT system for detecting and diagnosing engine blocking faults in the sun tracking system is being developed. The detection and diagnosis of other CPV system faults, which may appear during its operation, such as failures of the cooling system, lack of precision in tracking the sun, shading, accumulation of dust, etc., will be the subject of our future works. For this, IoT technique will be configured. A prototype of such telemonitoring system is in progress. Also Machine Learning (ML) for image fault classification in CPV based on collected thermal images, will be investigated.

References 1. Chela, A., Kaushik, G.: Renewable energy technologies for sustainable development of energy efficient building. Alexandria Eng. J. 57(2), 655–669 (2018) 2. Waldau, A.J.: February. Snapshot of Photovoltaics—February 2019. European Commission, Joint Research Centre (JRC), Via E. Fermi 2749, I-21027 Ispra (VA), Italy (2019) 3. Waldau, A. J.: PV Status Report 2018; Publications Office of the European Union: Luxembourg, 2018 (2018). ISBN 978-92-79-97466-3 4. Bloomberg New Energy Finance. 2H 2018 LCOE Update: PV; EU: Brussels, Belgium 5. Jo, J.H., Waszak, R., Shawgo, M.: Feasibility of Concentrated Photovoltaic Systems (CPV) in various United States geographic locations. Energy Technol. Policy 1(1), 84–90 (2014) 6. Gevorkian, P.: March. Solar Power Generation Problems, Solutions, & Monitoring. Cambridge University Press,Cambridge (2016). 1107120373, 9781107120372.38

Faults Detection and Diagnosis of Concentrated Photovoltaic Systems

323

7. Messmer, R.: CPV Market Evolution and the potential in cost reduction of CPV modules. In: Proceedings of the 9th Conference on Concentrator Photovoltaic Systems (CPV-9), pp. 1–4, April 2013 8. Fraunhofer, I.S.E.: Current status of concentrator photovoltaic (CPV) technology. Fraunhofer Institute for Solar Energy Systems ISE, & National Renewable Energy Laboratory NREL. TP-6A20-63916 (2017) 9. U.S. Department of Energy. National solar technology roadmap: concentrator PV, management report. NREL/ MP-520-41735, (2007) 10. IHS. Press Release - Concentrated PV (CPV) report – 2013. IHS Online Pressroom, 10 December 2013 11. Standard Bank - Webber Wentzel, CPV power plant No. 1 bond SPV (RF) Limited – Offering Circular, 18 April 2013 12. Fraunhofer, I.S.E.: Levelized Cost of Electricity, renewable energy technologies study. Fraunhofer Institute For Solar Energy Systems ISE (2013) 13. Ojhal, M., Chamat, N.: Solar PV module fault analysis using artificial intelligence. Int. J. Adv. Res. Electr. Electr. Instr. Eng. 8, 1507–1515 (2019) 14. Arani, M.S., Hejazi, M.A.: The comprehensive study of electrical faults in PV arrays. J. Electr. Comput. Eng. 3, 1–10 (2016) 15. Leahy, K., Hu, R.L., Konstantakopoulos, I.C., Spanos, C.J., Agogino, A.M.: Diagnosing wind turbine faults using Machine Learning techniques applied to operational data. In Proceedings of the 2016 IEEE international conference on prognostics and health management, pp. 1–8, June 2016 16. Babar, M., Rizvi, A.A., Al-Ammar, E.A., Malik, N.H.: Analytical model of multi-junction solar cell. Arabian J. Sci. Eng. 39(1), 547–555 (2013) 17. Bett, A.W., Dimroth, F., Guter, W., Hoheisel, R., Oliva, E., Philipps, S.P., Schöne, J., Siefer, G., Steiner, M., Wekkeli, A., Welser, E., Meusel, M., Köstler, W., Strobl, G.: Highest efficiency multi-junction solar cell for terrestrial and space applications. In: Proceedings of the 24th European Photovoltaic Solar Energy Conference, pp. 1–6, January 2009 18. Cai, J., Chen, X., Wu, H., Kuang, X.: Typical failure mechanisms of plastic encapsulated devices’ internal connection. In: Proceedings of the 17th International Conference on Electronic Packaging Technology, pp. 1323–1326, August 2016 19. Dhimish, M., Holmes, V., Mehrdadi, B., Dales, M.: The impact of cracks on photovoltaic power performance. J. Sci. Adv. Mater. Devices 2, 199–209 (2017) 20. Report IEA-PVPS T13-01:2014. Performance and Reliability of Photovoltaic Systems, Review of Failures of Photovoltaic Modules, March 2014. ISBN 978-3-906042-16-9 21. Bunthof, L.A.A., Kreuwel, F.P.M., Kaldenhoven, A., Kin, S., Corbeek, W.H.M., Bauhuis, G.J., Vlieg, E., Schermer, J.J.: Impact of shading on a flat CPV system for façade integration. Sol. Energy 140(162–170), 1–14 (2016) 22. Mellit, A., Tina, M., Kalogirou, S.: Fault detection and diagnosis methods for photovoltaic systems: a review. Renew. Sustain. Energy Rev. 91, 1–17 (2018) 23. Andrianajaina, T., Sambatra, E.J.R., Andrianirina, C.B., Razafimahefa, T.D., Heraud, N.: PV fault detection using the least squares method. In: Proceedings of the 2016 International Conference and Exposition on Electrical and Power Engineering, pp. 846–851, October 2016

324

N. Kellil et al.

24. Aghaei, M., Dolara, A., Leva, S., Grimaccia, F.: Image resolution and defects detection in PV inspection by unmanned technologies. In: Proceedings of the 2016 IEEE Power and Energy Society General Meeting, pp. 1–5, July 2016 25. Abdulmawjood, K., Refaat, S.S., Morsi, W.G.: Detection and prediction of faults in photovoltaic arrays: a review. In: Proceedings of the IEEE 12th International Conference on Compatibility, Power Electronics and Power Engineering, pp. 1–8, April 2018 26. Tsanakas, J.A., Ha, L., Buerhop, C.: Faults and infrared thermographic diagnosis in operating c-Si photovoltaic modules: a review of research and future challenges. Renew. Sustain. Energy Rev. 62, 695–709 (2016) 27. Mellit, A., Hamied, A., Lughi, V., Pavan, A.M.: A low-cost monitoring and fault detection system for stand-alone photovoltaic systems using IoT technique. In: ELECTRIMACS 2019. Springer, Cham, pp. 349–358, April 2020

New Droop Control Technique for Reactive Power Sharing of Parallel Inverters in Islanded Microgrid Yacine Daili, Abdelghani Harrag, and Ilyas Bennia(&) Mechatronics Laboratory (LMETR), Optics and Precision Mechanics Institute, Ferhat Abbas University, Sétif 1 Cite Maabouda (Ex. Travaux), 19000 Setif, Algeria {yacine.daili,a.harrag,ilyas.bennia}@univ-setif.dz

Abstract. This paper investigates a new droop control approach for accurate reactive power sharing of parallel inverters in islanded microgrid. The conventional active power-frequency droop and the reactive power-voltage magnitude droop control are widely used to share the total load demand among parallel distributed generators (DGs) according to their power ratings. However, even though the active power is properly shared, the reactive power sharing is inaccurate with the conventional droop control approach due to the power coupling of active - reactive power and the feeder impedance difference between DGs. In order to solve this problem, an improved droop method is proposed. The reactive power sharing error between DGs is estimated by injecting a small perturbation in the active power loop. Furthermore, an integrator is added to the classical reactive power droop controller to compensate reactive power sharing error. The effectiveness of the modified droop control strategy is verified by simulation, the results demonstrate the superior performances of the proposed control strategy. Keywords: Parallel inverters
Droop control
Microgrid
Distributed generation
Islanded mode
Load demand sharing
Voltage source inverters (VSIs)

1 Introduction In recent years, the concept of a microgrid (MG) has proposed to facilitate the integration of distributed renewable generation, such as wind turbines, fuel cell and photovoltaic [1–3] in the power distribution network. The MGs are low-voltage electrical distribution systems that connect multiple distributed generators (DGs) and storage systems to multiple customers [4]. Microgrids are able to operate in either islanded mode or connected to a larger power system. Islanded mode operation of a microgrid could be result of an emergency, such as failures of the main grid, or may be designed to the remote area power supply [5–7]. An important aspect in the islanded mode operation of MG is that the total load demands are expected to be properly shared among parallel DGs based VSIs regarding to their power ratings. To achieve this goal, the conventional active power-frequency © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 325–335, 2021. https://doi.org/10.1007/978-3-030-63846-7_32

326

Y. Daili et al.

(P-f) droop and the reactive power-voltage magnitude (Q-V) droop are widely used [8–12]. The conventional droop controller has principally been designed for centralized generation systems with high transmission line, where the line impedance is purely inductive [11]. However, DG units are commonly connected to medium and low voltage line, where the impedance line is not dominantly inductive and active and reactive powers are nonlinear and strongly coupled [13, 14]. Furthermore, there are mismatches of feeders impedances, which is typical problem of distribution grids. As a result, the accurate reactive power sharing among DG units is not easy to be realized in such situation [15, 16]. Moreover, by using the conventional droop control strategy, there are circulating currents between DGs, which increases losses, occupies the capacity of DGs, and affects the stability of the whole system. To deal with aforementioned challenge, various methods have been proposed to improve reactive power sharing among DGs in an islanded MG. The authors of [17–19] suggested to adjust the droop parameters to improve the reactive power sharing with difference of DGs feeder impedance. However, the stability of the microgrid could be affected with the high values of power coefficients. A robust droop control method has been proposed in [19] to achieve accurate reactive power sharing. The problem associated with this control method is that the controller requires accurate measurement of point of common coupling (PCC) voltage to estimate the mismatch output impedances among different feeders, which is difficult to be achieved in the practical application. The method proposed in [20] consists in compensation of voltage droop caused by line impedance of DG. However, the compensation parameters are obtained when the microgrid operates in grid-connected mode. The system could not work properly when the microgrid switch to island mode due to faults occurred in the microgrid. The authors of [20–22] propose to use the virtual impedance methods, the proper virtual impedance makes the output impedance of the DG to be mainly inductive and reduces sharing error. But to obtain the optimal virtual impedance, the MG configuration on real-time information is required which is difficult to be obtained. The main contribution of this paper is to investigate an enhanced droop control method to solve the problem of reactive power allocation of DGs in islanded MGs. The reactive power error among DGs is estimated through injecting an additional small active power signal which is activated by synchronisation signals. Then, an integrator is added to the classical reactive power droop controller to compensate reactive power sharing error. The performance of improved droop method is analyzed through simulation results. The simulation results confirm the effectiveness and the feasibility of the proposed droop control approach.

2 Configuration and Operation of Microgrid A typical configuration of a microgrid is illustrated in Fig. 1. It consists of multiple distributed generation (DG) units, storage systems, and loads. The microgrid is connected to the utility through a static transfer switch (STS) at the point of common coupling (PCC). Each DG unit is interfaced to the microgrid through voltage source

New Droop Control Technique for Reactive Power Sharing of Parallel Inverters

327

inverter (VSI) and LC filter. In this work, the DC source is used instead distributed generation (DG) and storage systems to simplify studies. This paper aims to solving an important aspect of MG operation in island mode related to the total reactive power demand sharing among DG units.

DG unit 1

Local DG 1 Controller

LC Filter

Feeder 1

PCC

RES Local Load

DG unit 2

Local DG 2 Controller

LC Filter

Feeder 2

RES

STS

DG unit n

Local DG n Controller

LC Filter

Main Grid

Feeder n

RES

Fig. 1. Typical AC microgrid configuration.

Figure 2 shows the equivalent model of the i-th DG unit of MG illustrated in Fig. 1 where two nodes have line to line voltage Vi and VPCC are interconnected by a line impedance Zi including the resistance and inductance line denoted respectively Ri and Li. The output active Pi and reactive powers Qi injected to the grid could be expressed as [23]: 8 2 di þ XL Vi VPCC sindi < Pi ¼ Ri Vi Ri Vi VPCC cos R2 þ X 2 L

L

i

i

: Qi ¼ Xi Vi Xi Vi VPCC 2cos di2Ri Vi VPCC sindi R þX 2

where di is the load angle.

ð1Þ

328

Y. Daili et al.

Inverter

Grid

Fig. 2. Equivalent circuit of an inverter connected to PCC.

Normally, for a large synchronous generator coupled to the grid through a high voltage line, the line impedance is dominantly inductive. Consequence, the expressions of the active and reactive powers (Eq. 1) becomes: (

Pi ¼ Vi VXPCC di i Qi ¼ VXii ðVi  VPCC Þ

ð2Þ

Therefore, the active power is controlled by the angle and the reactive power is regulated by changing the inverter amplitude voltage Vi. The conventional droop control is given by: 

i xi ¼ xn þ mi P  Vi ¼ Vn þ ni Qi

ð3Þ

Where xn and Vn are the nominal values of DG angular frequency and DG voltage i and amplitude, mi and ni are the active and reactive droop coefficients, respectively. P  i are the filtered measured active and reactive power values, respectively. Q The active and reactive droop coefficient mi and ni respectively are determined by the local grid standard [23, 24]. Finally, the amplitude voltage Vi and the phase angle di information are used for synthesizing the references voltages (via, vib,vic), which are the inputs of voltages controllers. The voltage references can be written as: pffiffiffi viap¼ffiffiffi 2Vi sindi  vib ¼ 2Vi sin di  2p 3 pffiffiffi  : vic ¼ 2Vi sin di þ 2p 3 8
< ci Þ Prated  ðvðvv vci  v  vr r vci Þ PWT ðvÞ ¼ ð8Þ P vr  v  vco > > : rated 0 vco  v where, vci, vci, and vco are cut-in speed, rated speed, and cut-off speed of the wind turbine, respectively. PWT is the output power of the WT generator, and v is the average wind speed over each period, which is considered as one hour in this study.

3 Problem Formulation 3.1

Multi-objective Functions

In this paper, the proposed Multi-Objective Functions (MOF) aims to solve the problem of finding the optimal allocation of DG units, through minimizing the four parameters of PLoss, VSI, SCL, and ALC on all simultaneously, which can formulate as follows: MOF ¼ Min

24 N bus N bus  X X X PtLoss ði; jÞ þ t¼1 i¼1 j¼2

1 þ SCI t ðjÞ þ ALC t ði; jÞ VSI t ðjÞ

ð9Þ

The first parameter PLoss of the distribution line connecting bus i to bus j is represented by the following equation [10–20]: PLoss ði; jÞ ¼





 Rij Rij cos di  dj Pi Pj þ Qi Qj þ sin di  dj Qi Pj þ Pi Qj Vi Vj Vi Vj

ð10Þ

where, t is the time in hours, Nbus is the number of branches, and Rij is the resistance of line. (Vi, di) and (Vj, dj) are the voltage magnitudes and angles at buses i and j, respectively. (Pi, Qi) and (Pj, Qj) are the active and reactive powers at buses i and j, respectively.

Optimal Allocation of Renewable Energy Source Integrated-Smart Distribution Systems

395

The bus for which the value of the VSI is poor is subject to voltage collapse. The equations used to formulate this index and to solve the load flow of the distribution system are as follows [15, 21]:
2
 VSIðjÞ ¼ jVi j4 4: Pj Xij  Qj Rij 4: Pj Rij þ Qj Xij jVi j2

ð11Þ

The third term is the Short Circuit Level (SCL) in VA can be defined as [22, 23]: SCLðjÞ ¼

Vj2 Zij

ð12Þ

Then, the Annual Losses Cost (ALC), which depends on the active power loss, can be calculated as follows [24]: ALCði; jÞ ¼ PLoss ði; jÞ  KP  T

ð13Þ

Where, KP is the incremental cost of power loss is equal 0.06 $/kW, and T is the number of hours per year is equal 8760 h. 3.2

Equality Constraints

Equality constraints are represented by the following power balance equations [8–15]: PG þ PDG ¼ PD þ PLoss

ð14Þ

QG þ QDG ¼ QD þ QLoss

ð15Þ

where (PG, QG) are the total active and reactive power of the generator (sub-station), respectively; PDG is total active power of DG; QDG is the total reactive power injected by DG. (PD, QD) are the total active and reactive power of demand load, respectively. (PLoss, QLoss) are the total active and reactive power losses, respectively. 3.3

Inequality Constraints of Distribution Line

Inequality constraints are represented for the distribution line ij as follows [6–13]: Vmin  jVi j  Vmax



V1  Vj  DVmax



Sij  jSmax j

ð16Þ ð17Þ ð18Þ

where (Vmin, Vmax) are the minimum and maximum voltages specified; V1 is the voltage at the sub-station, which is equal to 1.0 p.u. DVmax is the maximum voltage drop at each branch. Sij is the apparent power flow in ij branch that links bus i to bus j. Smax is the maximum apparent power.

396

3.4

M. Zellagui et al.

Inequality Constraints of DG Units

Inequality constraints represent the limits of the DG units, which can be given as [18–25]: max Pmin DG  PDG  PDG

ð19Þ

max Qmin DG  QDG  QDG

ð20Þ

NDG X

PDG ðiÞ 

i¼1 Nbus X i¼1

Nbus X

PD ðiÞ

ð21Þ

QD ðiÞ

ð22Þ

i¼1

QDG ðiÞ 

Nbus X i¼1

2  DGPosition  NBus

ð23Þ

NDG  NDG:max

ð24Þ

nDG;i =Location  1

ð25Þ

min max  PFDG  PFDG PFDG

ð26Þ

PDG PFDG ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 PDG þ Q2DG

ð27Þ

min max max where (Pmin DG , QDG , PDG , QDG ) are the minimum and maximum active and reactive power output limits of DG, respectively. DGPosition is the position of DG units. NDG is the number of DG units. NDG.max is the maximum number of DG units to be used, and nDG,i is the location of DG units at bus i; PFDG is the power factor of DG unit at each bus. In the case of PV, the DG source only delivers active power as the PF is equal to unity. However, in the case of WT, the source delivers active and reactive powers.

4 SMA Algorithm (SMA) In this section, the mathematical model and SMA algorithm proposed will be described in details [16]. 4.1

Approach Food

To model the approaching behavior of slime mould as a mathematical equation, the following rule is proposed to imitate the contraction mode:

Optimal Allocation of Renewable Energy Source Integrated-Smart Distribution Systems

8 ! ! ! ! ! ! < Xb ðtÞ þ vb
W
XA ðtÞ  XB ðtÞ ; r\p Xðt þ 1Þ ¼ : ! ! vc
XðtÞ
r [ p

397

ð28Þ

where vb is a parameter with a range of [−a, a], vc decreases linearly from one to zero. t represents the current iteration, Xb represents the individual location, X represents the location of slime mould, XA and XB represent two individuals randomly selected from the swarm, W is the weight of slime mould. The formula of p is as follows [16]: p ¼ tanhjSðiÞ  DF j

ð29Þ

where S(i) represents the fitness of X, DF represents the best fitness obtained in all iterations. The formula of vb is as follows: ! vb ¼ ½a; a     t a ¼ arctan h  þ1 t max

ð30Þ ð31Þ

The formula of W is listed as follows:

 8 bFSðiÞ ! < 1 þ r
log bFwF þ 1 ; condition

 W ðSmellindexðiÞÞ ¼ : 1  r
log bFSðiÞ þ 1 ; others bFwF

ð32Þ

Smellindex ¼ sortðSÞ

ð33Þ

where condition indicates that S(i) ranks first half of the population r denotes the random value in the interval of [0, 1], bF denotes the optimal fitness obtained in the current iterative process, wF denotes the worst fitness value obtained, SmellIndex denotes the sequence of fitness values sorted (ascends in the minimum value problem). 4.2

Wrap Food

The mathematical formula for updating the location of slime mould is as follows [16]: 8 rand
ðUB  LBÞ þ LB; rand\z > < ! ! ! ! ! !  X b ðtÞ þ vb
W
XA ðtÞ  XB ðtÞ ; r\p X ¼ > : ! ! vc
XðtÞ; r  p where LB and UB denote the lower and upper boundaries of the search range.

ð34Þ

398

M. Zellagui et al.

5 Test System, Results and Comparison The proposed SMA algorithm is applied to the IEEE standard test system to evaluate the integration of multiple renewable DGs based PV and WT sources. The SMA algorithm has been developed using MATLAB 2017.b and simulations are performed on a computer of CPU Intel Core i5-5257U, 2.7 GHz, and 8 GB RAM. The test system used in this study is the standard IEEE 33-bus SDS represented in Fig. 1, which is composed of 33 busses, 32 lines and branches with the base voltage, is 12.66 kV.

Fig. 1. Single line diagram of standard IEEE 33-bus SDS.

In the SDS test system, the following scenarios are considered to obtain the most suitable planning model and analyze the effectiveness of the proposed algorithm: Scenario 1: Base case, SDS without renewable DG units, Scenario 2: Optimized SDS planning with one PV-DG unit, Scenario 3: Optimized SDS planning with one WT-DG unit, Scenario 4: Optimized SDS planning with two PV-DGs, and Scenario 5: Optimized SDS planning with two WT-DGs. The characteristics of a PV module and WT generator used in this paper are shown in Tables 1 and 2, respectively. Table 1. Characteristics of the PV module. PV module characteristics

Value

Nominal cell operating temperature, NOT (°C) Current at maximum power point, IMPP (A) Voltage at maximum power point, VMPP (V) Short circuit current, Isc (A) Open circuit voltage, VOC (V) Current temperature coefficients, Ki (A/°C) Voltage temperature coefficients, Kv (V/°C)

43 7.76 28.36 8.38 36.96 0.00545 0.1278

Table 2. Characteristics of the WT generator. WT generator characteristics Generator type Rated power (kW) Cut-in wind speed (m/s) Cut-out wind speed (m/s) Rated wind speed (m/s) Number of blades Rotor diameter/Hub height (m)

Value DFIG 2300 4 25 14 3 93/80

Optimal Allocation of Renewable Energy Source Integrated-Smart Distribution Systems

399

Figure 2 shows 20 iterations of the convergence curve for the MOF minimization, where the best execution results are plotted in red. As shown in the figure, the algorithm converges quickly within 100 iterations, the algorithm shows quick convergence, and it can be observed that all solutions are close to the best solution.

Fig. 2. Convergence characteristic of SMA algorithm: a). PV-DG, b). WT-DG.

For solving this problem, SDS is assumed to follow a daily load power demand curve as shown in Fig. 3. The power outputs of PV and WT-based DG are assumed to follow the nominalized average output curve shown in Fig. 4.

Fig. 3. Daily load power demand curve.

Fig. 4. Renewable daily DG Profile.

Table 3 shows the parameters of multiple DGs and optimization results before and after the integration of PV-DG and WT-DG for 24 h. Table 3. Optimization results before and after the integration of DGs. DG type

DG number

WT-DG

PDG (kW)

PFDG

P

P

P

SCL (MVA)

P Vmin (p.u.)

P

P

PLoss (kW)

QLoss (kVar)

VSI (p.u.)

ALC (M.$)

–

–

–

3557.0607

2412.2452

213636.331

22.1088

646.5337

1.8696

1 DG

7

2847.8

1.000

2603.3094

1824.3591

217524.356

22.4743

672.6181

1.3683

2 DGs

13–30

709.3–961.4

1.000–1.000

2491.2307

1687.1638

217119.663

22.5465

670.6248

1.3094

1 DG

6

2119.1

0.8237

1830.7739

1334.4441

219363.090

22.6257

684.2805

0.9623

2 DGs

13–30

682.6–1031.9

0.8825–0.8000

1339.41550

904.3355

220873.813

22.9761

697.3223

0.7040

Basic case PV-DG

DG bus

400

M. Zellagui et al.

As reported in Table 3, the total PLoss, QLoss, and ALC of SDS before the integration of DGs in 24 h are 3557.0607 kW, 2412.2452 kVar, and 1.8696 M.$. These values are reduced by 26.8129%, 24.3709%, and 26.8132%, after installing one PVDG unit at bus 7 with a maximum output active power of 2847.8 kW. The actual values of those quantities are 2603.3094 kW, 1824.3591 kVar, and 1.3683 M.$, respectively. The total SCL and VSI are also maximized from 21363.6331 MVA and 646.5337 p.u. to 21752.4356 MVA and 672.6181 p.u., respectively. Two PV-DGs with maximum output active power up to 709.3 kW and 961.4 kW are located at two different locations, which are buses 13 and 30, respectively. The total PLoss, QLoss, and ALC were reduced to 2491.2307 kW, 1687.1638 kVar and 1.3094 M. $. Furthermore, SCL and VSI are maximized to 21711.9663 MVA, and 670.6248 p.u., respectively. On the other hand, one WT-DG with generated active power of up to 2119.1 kW and power factor equal to 0.8237 is located at bus 6. The WT-DG reduces PLoss, QLoss, and ALC to 1830.7739 kW, 1334.4441 kVar and 0.9623 M.$. It also maximizes SCL and VSI to 21936.3090 MVA, and 684.2805 p.u., respectively. From the results of using two WT-DGs, it is observed that the best results are obtained in this case. Best locations for the WT-DG units are the same as in the case of using two PV-DGs, i.e., at bus 13 and 30. The output of each WT-DG is 682.6 kW and 1031.9 kW with power factors of 0.8825 and 0.8000, respectively. In this case, PLoss, QLoss, and ALC are minimized to 1339.41550 kW, 904.3355 kVar, and 0.7040 M.$. When comparing these values with those of the basic case, they are reduced by 62.3448, 62.5106, and 62.3448%. Similarly, VSI and SCL are maximized to 22087.3813 MVA, and 697.3223 p.u., respectively. The total PLoss and Vmin in SDS for 24 h considering load demand and DG uncertainties in all case studies are represented in Fig. 6a. The total SCL, VSI, and ALC are represented in Figs. 6b, c, and d, respectively. From the load demand curve in Fig. 3, the consumption increases by 90% after 9:00 and decreases to less than 90% after 22:00. During this period, PLoss, SCL, VSI, and ALC profiles follow the load curve and their values are raised by 90% of their initial values during this period. From Figs. 5, the PV-DG can provide energy from approximately 5:00 until 22:00 and during this period; the power injected from PV-DG contributes to minimizing the total PLoss and ALC, and maximizing SCL and VSI. The peak of load demand is between 10:00 and 16:00. During this peak period, PV-DG injects the maximum power which contributes to the maximum reduction of PLoss and ALC, and maximizing SCL and VSI. The best result is recorded at 13:00 after the integration of two PV-DGs, which in fact is the hour of maximum PV-DG output.

Optimal Allocation of Renewable Energy Source Integrated-Smart Distribution Systems

401

Fig. 5. Optimal results with the integration of renewable DG units.

Moreover, PLoss, SCL, VSI and ALC curves follow the WT-DG output curve for 24 h. The best results are obtained while having two WT-DGs, where the best PLoss, SCL, VSI, and ALC are obtained within the period of 13:00 to 19:00. The maximum output occurs at 16:00, which contributes to the best minimization of PLoss and ALC, and best maximization of SCL and VSI. Unlike PV-DG, WT-DG has two features, the first is its generating output powers during 24 h and secondly, its ability to provide both active and reactive powers. Because of this, the best curves are obtained after the integration of WT-DGs.

402

M. Zellagui et al.

The bus voltage profile for 24 h time-varying load in the following cases: the base case (without DG), the installation of two PV-DGs, and with the installation of two WT-DGs in SDS are represented in Figs. 6a, b, and c, respectively.

Fig. 6. Voltage profile for 24 h time-varying load: a). Without DGs, b). With two PV-DGs, c). With two WT-DGs.

From Fig. 6a, it is noted that the voltage profile of the basic case has the minimum values. The worst profiles are recorded between 10:00 and 19:00, and the minimum voltage value is at bus 18, which is close to 0.9 p.u. From Fig. 6b, the installation of two PV-DGs provides an improvement for all profiles. For example, the minimum voltage at bus 18 is raised to 0.91 p.u. The worst profiles mentioned in the base case are no longer happening between 10:00 and 19:00 in this case. During this interval, the PV-DGs have their best performance, while the worst profiles happen between 20:00 and 23:00. On the other hand, using two WT-DGs provides more improvement, as shown in Fig. 6c. The minimum voltage at bus number 18 is improved as it becomes 0.93 p.u. In general, all profiles for the 24 h have improved and the worst profiles are obtained between 22:00 and 23:00.

6 Conclusions In this paper, a relatively new optimization algorithm called SMA has been applied to determine the optimal allocation of single and multiple DGs based on PV and WT sources. This has been performed to IEEE 33-bus SDS as a test case, while taking into account the load demand and the power output of DG uncertainties during every hour of the day. This implementation is carried out for the purpose of improving system performance by optimizing various technical and economic parameters. The results obtained show the advantages of using DGs, which lead to improving voltage profiles, and increasing SCL. Furthermore, we observed the direct impacts of DG, load demand uncertainties and the number of DG units installed on the parameters of the distribution system. The integration of WT-DGs showed superiority over using PV-DGs in improving SDS performance. Finally, the results foster further investigations on these topics and their future application to practical networks which can lead to an even more

Optimal Allocation of Renewable Energy Source Integrated-Smart Distribution Systems

403

improved performance of SDS. Therefore, future work will involve the formulation of the combined planning of DG and electric vehicles scheduling problems for abnormal operating scenarios for a longer period and in different sessions of the year.

References 1. Ahmadi, M., Lotfy, M.E., Shigenobu, R., Yona, A., Senjyu, T.: Optimal sizing and placement of rooftop solar photovoltaic at Kabul city real distribution network. IET Gener. Transm. Distrib. 12(2), 303–309 (2018) 2. Poornazaryan, B., Karimyan, P., Gharehpetian, G.B., Abedi, M.: Optimal allocation and sizing of DG units considering voltage stability, losses and load variations. Int. J. Electr. Power Energy Syst. 79, 42–52 (2016) 3. Pecas Lopes, A., Hatziargyriou, N., Mutale, J., Djapic, P., Jenkins, N.: Integrating distributed generation into electric power systems: a review of drivers, challenges and opportunities. Electr. Power Syst. Res. 77(9), 1189–1203 (2007) 4. Ganguly, S., Samajpati, D.: Distributed generation allocation on radial distribution networks under uncertainties of load and generation using genetic algorithm. IEEE Trans. Sustain. Energy 6(3), 688–697 (2015) 5. Ehsanab, A., Yanga, Q.: State-of-the-art techniques for modelling of uncertainties in active distribution network planning: a review. Appl. Energy 239, 1509–1523 (2019) 6. Notton, G., Nivet, M.L., Voyant, C., Paoli, C., Darras, C., Motte, F., Fouilloy, A.: Intermittent and stochastic character of renewable energy sources: consequences, cost of intermittence and benefit of forecasting. Renew. Sustain. Energy Rev. 87, 96–105 (2018) 7. Ghanegaonkar, S.P., Pande, V.N.: Optimal hourly scheduling of distributed generation and capacitors for minimisation of energy loss and reduction in capacitors switching operations. IET Gener. Transm. Distrib. 11(9), 2244–2250 (2017) 8. Gholami, K., Dehnavi, E.: A modified particle swarm optimization algorithm for scheduling renewable generation in a micro-grid under load uncertainty. Appl. Soft Comput. 78, 496– 514 (2019) 9. Barik, S., Das, D.: Impact of FFC distributed generations in a DNR in the presence of renewable and load uncertainties by mixed-discrete particle swarm-based point estimation method. IET Renew. Power Gener. 13(9), 1431–1445 (2019) 10. Abou El-Ela, A.A., El-Sehiemy, R.A., Ali, E.S., Kinawy, A.M.: Minimisation of voltage fluctuation resulted from renewable energy sources uncertainty in distribution systems. IET Gener. Transm. Distrib. 13(12), 2339–2351 (2019) 11. Souza, S.S.F., Romero, R., Pereira, J., Saraiva, J.T.: Artificial immune algorithm applied to distribution system reconfiguration with variable demand. Int. J. Electr. Power Energy Syst. 82, 561–568 (2016) 12. Yong, C., Kong, X., Chen, Y., Xu, Q., Yu, L.: An optimization method of active distribution network considering uncertainties of renewable DGs. Energy Procedia 158, 934–939 (2019) 13. Murty, V.S.N., Kumar, A.: Optimal DG integration and network reconfiguration in microgrid system with realistic time varying load model using hybrid optimization. IET Smart Grid 2(2), 192–202 (2019) 14. Elsakaan, A.A., El-Sehiemy, R.A., Kaddah, S.S., Elsaid, M.I.: Optimal economic-emission power scheduling of RERs in MGs with uncertainty. IET Gener. Transm. Distrib. 14(1), 37– 52 (2020) 15. Selim, A., Kamel, S., Jurado, F.: Efficient optimization technique for multiple DG allocation in distribution networks. Appl. Soft Comput. 86, 1–20 (2020)

404

M. Zellagui et al.

16. Li, S., Chen, H., Wang, M., Heidari, A.A., Mirjalili, S.: Slime mould algorithm: a new method for stochastic optimization. Future Gener. Comput. Syst. 111, 300–323 (2020) 17. Atwa, Y.M., El-Saadany, E.F., Salama, M.M.A., Seethapathy, R.: Optimal renewable resources mix for distribution system energy loss minimization. IEEE Trans. Power Syst. 25 (1), 360–370 (2010) 18. Soroudi, A., Aien, M., Ehsan, M.: A probabilistic modelling of photo voltaic modules and wind power generation impact on distribution networks. IEEE Syst. J. 6(2), 254–259 (2012) 19. Haque, M.H.: Evaluation of power flow solutions with fixed speed wind turbine generating systems. Energy Convers. Manag. 79, 511–518 (2014) 20. Lokman, M.H., Musirin, I., Suliman, S.I., Suyono, H., Hasanah, R.N., Mustafa, S.A.S., Zellagui, M.: Multi-verse optimization based evolutionary programming technique for power scheduling in loss minimization scheme. IAES Int. J. Artif. Intell. 8(3), 292–298 (2019) 21. Settoul, S., Chenni, R., Hassan, H.A., Zellagui, M., Kraimia, M.N.: MFO algorithm for optimal location and sizing of multiple photovoltaic distributed generations units for loss reduction in distribution systems. In: Proceedings of the 7th International Renewable and Sustainable Energy Conference (IRSEC), Agadir, Morocco, 27–30 November 2019 (2019) 22. Parizad, A., Baghaee, H.R., Yazdani, A., Gharehpetian, G.B.: Optimal distribution systems reconfiguration for short circuit level reduction using PSO algorithm. In: Proceedings of the IEEE Power and Energy Conference at Illinois (PECI), Illinois, USA, 22–23 February 2018 (2018) 23. Zellagui, M., Hassan, H.A.: Modeling the effects of PWMSC and fault resistance on ground fault system in MV distribution line. WSEAS Trans. Syst. Control 12, 114–122 (2017) 24. Hassan, H.A., Zellagui, M.: MVO algorithm for optimal simultaneous integration of DG and DSTATCOM in standard radial distribution systems based on technical-economic indices. In: Proceedings of the 21st IEEE International Middle East Power Systems Conference (MEPCON), Tanta, Egypt, 17–19 December 2019 (2019) 25. Lasmari, A., Zellagui, M., Chenni, R., Semaoui, S., El-Bayeh, C.Z., Hassan, H.A.: Optimal energy management system for distribution systems using simultaneous integration of PVbased DG and DSTATCOM units. Energetika 66(1), 1–14 (2020)

People Counter with Area Occupancy Control for Covid-19 E. Khoumeri1(&), H. Fraoucene2, El Hadi Khoumeri1,2, C. Hamouda1,2, and R. Cheggou1 1

Laboratoire des Technologies Innovantes, GEII Department, Ecole Nationale Supérieure de Technologie, Algiers, Algeria {elhadi.khoumeri,rabea.cheggou}@enst.dz 2 Laboratory of Advanced Technologies of Genie Electrics, Faculty of Electrical and Computer Engineering, Mouloud Mammeri University, Tizi-Ouzou, Algeria [email protected]

Abstract. Governments have imposed social distancing regulations to counter the coronavirus pandemic. An automated occupancy control system to provide a more cost-effective and efficient way to abide with these safety regulations. With the generalization of the use of digital images, the analysis of movement in video sequences has proved to be an essential tool for various applications such as video surveillance, robotics etc. The advance in video processing algorithms and the fast computational capability, give a possibility to use a video tracking and counting people in real time. Estimating the number of people in real time is useful information for several applications such as security and health management. With the COVID-19 pandemic, the counting of people present in a region of interest is important to control the area occupancy in order to minimize human virus transmission. In this paper we present a finished solution for counting people present in the same area. The system is based on Raspberry Pi and a common camera. The method called BLOB (Binary Large Object) analysis is used. The performance of the system, achieving an average count rate between 95% and 98%. Keywords: People counting Raspberry Pi
BLOB


COVID-19
OpenCv
Occupancy control


1 Introduction Today, a lot of research has been published to try to solve the problem of counting people using a video camera. This is not a simple task, there are some difficult issues to solve. One of the problems with using the video camera for people counting is the occlusion between people, when a group enters or leaves the field of view of the camera. It is very difficult to distinguish between people in the same group. Existing solutions offering video cameras based on people counting are efficient and reliable but very expensive. This work proposes an efficient and inexpensive solution for counting people and discusses the problem of detecting overlapping people shapes in the video. In this project, we will propose the design and an implementation of a counting © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 405–415, 2021. https://doi.org/10.1007/978-3-030-63846-7_38

406

E. Khoumeri et al.

algorithm based on image processing using the Raspberry-Pi in real time. The extraction of the characteristics of an image containing the objects to be detected is given using a method called BLOB (Binary Large Object) analysis [1]. The number of people counted is displayed on the screen directly. The Raspberry Pi, running with the Linux operating system installed on an SD memory card, a video camera used as an image acquisition device to capture it continuously. The coronavirus transmission mechanisms identified in [2] this prompted us to find a way to limit close face-to-face contact in closed area. Our system accurately counts the number of people entering and leaving the closed area, giving the number of customers inside in real time. A tablet installed at the entrance shows people if they can enter or if the maximum occupancy has been reached and they must wait. When other people leave, the screen updates accordingly to the occupancy situation. This allows an effective occupancy control for essential businesses such as supermarkets, pharmacies, banks and hospitals as well as for government institutions. 1.1

Internet of Things

The term Internet of Things (IoT) was first introduced in 1999 [3], IoT is an essential part of modern embedded systems that describes a network of physical objects connected to the internet in order to communicate and share data between them. The goal of IoT is to make use of these large amounts of data to make systems more efficient and automated. There are several examples of IoT solutions for various problems, such as using light beams and infrared sensors to analyze the occupancy level of a house and to adjust the airflow of the ventilation system [4], using video and location data to enhance counting the number of people in an area [5], and for fire safety in buildings [6]. The Internet of Things is also used for remote patient health management with the sending of vital information directly to the attending physician [7]. 1.2

Why People Counting?

Due to the coronavirus pandemic, travel is limited and people are advised to stay at home. At the time of deconfinement, the medical authorities described distancing protocols. The idea of making a person counter at the entrance of a building welcoming people was born from the limitation of the number of people present at the same time in a closed area. People have to do their shopping, go to a doctor or to the bank. To monitor occupancy with the number of people present we need an automated system, our application will display the number of people entering and leaving so, the person managing security can be sure that they meet the requested health standards. 1.3

Literature Survey

The great difficulty of counting people by camera is knowing how to count a person hidden behind another person. Different research has been proposed, as in [8] and [9] Uses method for facial recognition to detect people. with a single camera, the system counts the number of people. Counting is done by analyzing the image to detect faces. Ching-Tang and All [10] offered a bi-directional counting system for the flow of people

People Counter with Area Occupancy Control for Covid-19

407

with Kinect. They put the Kinect camera above the door to capture the situation of the pedestrian flow. Then, this system detects people in the coverage area using the depth image information from the Kinect system. Borislav [11] presents an efficient and reliable method of automatic segmentation, tracking and counting of people, designed for a system with an aerial (overhead) camera. [12] This work presents an implementation of a people counting system based on a modified gradient histogram (HoG), a support vector machine (SVM) and a novel counting technique using an FPGA implementation. Another automatic bidirectional method [13] for counting passers-by based on image processing. To reduce the overall overlap, a camera set with a straight down view is adopted. And an experimental formula based on the area is adopted to derive the rough estimate of pedestrians with simple and fast tracking An algorithm is used to track pedestrians. Using this, track information, merge and split, cases can be easily judged. Some companies [14, 15] and [16] sell products similar to our application, we differentiate ourselves from these products, by their prohibitive prices which sometimes exceed 1000 dollars for one system [17].

2 Methodology 2.1

The Characteristics of the Human Body

We can recognize a human by using different information. We see the silhouette of a person with our eyes, we will collect other types of information that will confirm that it is a person. The person will undoubtedly breathe, blink, so to identify a person, we can use technologies that provide information that will identify the person [18]. Although each individual is different and even unique, we all have in common our morphology. A person normally made up of two arms, two legs, a head, etc. which implies that a human silhouette can be differentiated from the silhouette of an animal. 2.2

The People Counting Algorithm

Design is the process of representing the various functions of the system, so one or more programs quickly perform those functions. This is the most important part of our job. In the following we will describe an algorithm for counting people in a video sequence, first we describe the design of our system then we implement it, then we present the results. Our goal is to design a system for counting people entering and leaving a building or an area. The general scheme of the design of our system is represented by the following flowchart: Referring to (Fig. 1), the system detects pedestrians passing in a visual field, the video stream is captured with a camera placed on the ceiling with a top view, this video is split into several frames (images), one image per 30 ms, each image is processed separately so that people can be detected. For the people tracking part, the system follows people in the video sequences by memorizing their coordinates and their movement (entering or leaving). In the part counting people we will count the number of people entering and leaving.

408

E. Khoumeri et al.

Fig. 1. General system diagram.

2.3

Detection Algorithm

The people detection is summarized in the algorithm (Fig. 4), in this algorithm we find the following functions: Video (Picture): stream playback by a webcam located in the ceiling with top view, as shown in Fig. 3. Transformation: transform images to grayscale, Filtering: in this step we will apply a Gaussian filter to eliminate the noise in the image in order to have better detection, Detection of the Background: memorization of the first image that we will consider it as the background, Background Subtraction: the subtraction of each frame from the recorded background, Threshold: binarize the image to have a better result, transforming the gray level image into a black and white image such that the parts in white are moving as shown in (Figs. 2 and 3).

Fig. 2. Camera location.

Fig. 3. Thresholding.

People Counter with Area Occupancy Control for Covid-19

409

Fig. 4. People detection algorithm.

Find Contours: detected all contours of the borders of moving objects, For Each Contour C in the Set of Contours: the area of each contour is calculated if it is outside the detection interval (if less than min or greater than the max as indicated in the algorithm, it will be eliminated). Then, we find the contour characteristics (person) (x coordinate, y coordinate, width and height), Calculate the Center of Gravity Cg: for each contour, we memorize its coordinates x (Cx) and y(Cy), see (Fig. 5).

410

E. Khoumeri et al.

Fig. 5. People detection algorithm.

2.4

Tracking Algorithm

It is proposed as showing in Fig. 6:

Fig. 6. Tracking algorithm.

After having determined the coordinates Cx and Cy, we store them in two tables cx-current and Cy-current. In the following image, if the distance between the center of gravity found Cg (cx and cy) and the points stored in Cx-current and Cy-current is less than a threshold (threshold which determines the step of a normal person who is about 45 cm, which was found by tests), then We consider that the person detected in this

People Counter with Area Occupancy Control for Covid-19

411

image is the same as that detected in the previous image; otherwise, if the distance is great, we consider that this person is a new person who appears in the video and we memorize these coordinates. 2.5

Counting Algorithm

The algorithm is presented in Fig. 7

Fig. 7. Counting algorithm.

This algorithm uses two virtual lines (the entry and exit direction). We take H as the height of the image and L as the width (see Fig. 7). For each contour (a person), which will be detected, has a center of gravity Cg (Cx, Cy). To know if the detected person will enter or will leave, we proceed as follows: if Cy is greater than H/2, the person is consigned as outgoing, however, if Cy is less than H/2, the person will enter (Fig. 8).

412

E. Khoumeri et al.

Fig. 8. Image parameter.

To be able to count exactly the number of people, after each disappearance of a contour which had already existed in the previous image, the coordinates of Cg are compared with two thresholds Min and Max, these two thresholds determine the precision of the entry or of the exit of persons, if Cy is less than the Min threshold, then the person is considered to be outgoing, otherwise if Cy is greater than the Max threshold, then the person is considered as entering.

3 Experimental Results Our system is developed under the Raspberry Pi 2 environment which meets the needs of our application, its characteristics: CPU Quad cortex A7 900 MHz, 1G RAM, GPU 250 MHz Video Core IV, for a price of 40 dollars. For the implementation of algorithms we have chosen the Python language and for image processing we have used functions from the OpenCV library. We used a webcam for a price of $ 10. Note that we can use any camera. A tablet is connected to the Raspberry-Pi via Wi-Fi to display the authorization to enter the store, the number of people present in the store in this case has been set at 10. Once this threshold is reached, the tablet displays STOP, if a person goes out, so there will be 9 people, the tablet displays GO, it gives permission for a person to enter inside. The system calculates the number of people entering minus the number of people leaving to define the number of people present in the store. Once the number of people inside is 10, it displays STOP again. To test our system, we placed the webcam at the entrance of a building as in Fig. 9. It shows the principle of detecting a moving person by keeping its center of gravity with the calculation of the difference between the images (current and old image).

People Counter with Area Occupancy Control for Covid-19

413

Fig. 9. Detection of a person going out.

Figure 10 shown, when a person goes out, the OUT counter is incremented by +1.

Fig. 10. Counter increase.

414

E. Khoumeri et al.

Another case presented in Fig. 11a and b, where two people enter at the same time.

Fig. 11. a. Two persons detected. b. Counter increased +2.

We put the tablet at the store entrance, to regulate the entrance. If the number of people is less than ten in our case, the system displays the GO message on the tablet in green, as in Fig. 12a. If the number of people reaches exactly ten, the system displays the STOP message in red on the tablet as in Fig. 12b.

Fig. 12. a. Occupancy less than ten. b. Occupancy reached ten people.

4 Conclusion This article presents a simple and automatic method based on bi-directional processing to count people. It offers an inexpensive solution to manage the flow of visitors to shops and buildings. To reduce the overall overlap, a single camera set to a downward view is adopted. An experimental formula based on area was used to obtain a pedestrian body estimate by determining the center of gravity Cg with the fast tracking algorithm to track pedestrians. Using this information, merging and splitting cases can be handled quickly. This system has already been applied in a shopping center. Based on these experimental results, he showed that the proposed method works very reliably and achieves high accuracy rates. Different approaches can be used to improve the system; like designing a database to display input/output statistics, adding other cameras to improve counting.

People Counter with Area Occupancy Control for Covid-19

415

References 1. Cazasnoves, A., et al.: Statistical content-adapted sampling (SCAS) for 3D computed tomography. Comput. Biol. Med. 92, 9–21 (2018) 2. Chamola, V., et al.: A comprehensive review of the COVID-19 pandemic and the role of IoT, drones, AI, blockchain, and 5G in managing its impact. IEEE Access 8, 90225–90265 (2020) 3. Suresh, P., et al.: A state of the art review on the Internet of Things (IoT) history, technology and fields of deployment. In: International Conference on Science Engineering and Management Research (ICSEMR), Chennai, pp. 1–8 (2014) 4. Cheggou, R., et al.: Energy-saving through smart home concept. In: International Conference in Artificial Intelligence in Renewable Energetic Systems, pp. 50–58. Springer, Cham, March 2017 5. Stenbrunn, A.: Hosting a building management system on a smart network camera: on the development of an IoT system. Bachelor thesis, Malmo University, November 2015 6. Khoumeri, E., et al.: An IoT system for smart building. In: Future of Information and Communication Conference, pp. 522–533. Springer, Cham, December 2018 7. Khoumeri, E., et al.: An IoT based e-health platform using Raspberry Pi. In: Future of Information and Communication Conference, pp. 257–271. Springer, Cham (2020) 8. Chen, T.Y., et al.: A people counting system based on face-detection. In: Proceedings of the Fourth International Conference on Genetic and Evolutionary Computing, pp. 699–702, December 2010 9. Surbhi, S., Dalpat, S.: Design of people counting system using MATLAB. In: Proceedings of Tenth International Conference on Contemporary Computing (IC3), August 2017 10. Heish, C.-T., et al.: A kinect-based people-flow counting system. In: IEEE International Symposium on Intelligent Signal Processing and Communication Systems (ISPACS 2012), pp. 146–150, November 2012 11. Antić, B., Letić, D., Ćulibrk, D., Crnojević, V.: K-means based segmentation for real-time zenithal people counting. In: 2009 16th IEEE International Conference on Image Processing (ICIP), Cairo, pp. 2565–2568 (2009) 12. Ahmad, I., et al.: An FPGA based approach for people counting using image processing techniques. In: 2019 11th International Conference on Knowledge and Smart Technology (KST), Phuket, Thailand, pp. 148–152 (2019) 13. Cao, J., Sun, L., Odoom, M.G., Luan, F., Song, X.: Counting people by using a single camera without calibration. In: 2016 Chinese Control and Decision Conference (CCDC), Yinchuan, pp. 2048–2051 (2016) 14. Covid-19 Automated Occupancy Control System. FootfallCam. https://www.footfallcam. com/Industries/Covid-19-Automated-Occupancy-Control-System. Accessed 28 July 2020 15. Digital versus COVID 19. Digital versus COVID 19 - People counting: Alphabird GmbH. https://www.alphabird.at/en/. Accessed 28 July 2020 16. People counter project for Covid-19. https://www.rs-online.com/designspark/peoplecounter-project-for-covid-19. Accessed 28 July 2020 17. People Counting Camera: 3D SCOPE II. StoreTraffic, 06 July 2020. https://storetraffic.com/ product/3d-scope-ii-people-counting-camera/. Accessed 28 July 2020 18. Belconde, A.: Modélisation de la détection de presence humaine. PhD thesis, University of Orlean, France, September 2010

Finite Element Modelling and Analysis for Modal Investigation of a Blade H-Type Darrieus Rotor F. Ferroudji1,2(&), L. Saihi1, and K. Roummani1 1

Unité de Recherche en Energies Renouvelables en Milieu Saharien, URERMS, Centre de Développement des Energies Renouvelables, CDER, 01000 Adrar, Algeria [email protected] 2 Laboratoire de Mécanique des Structures et Matériaux, Université Batna 2, Fesdis, Algeria

Abstract. The blade is considered as the most significant element in wind turbine system which is expensive and the reliability of wind turbine is related directly to its quality. Therefore, its structure has been receiving a great consideration to comprehend the behaviour of blade by researchers and engineers. This research article presents the 3D numerical modal analysis of a straight blade for 10-kW H-type Darrieus rotor. Numerical analysis was carried out in order to determine modal parameters (natural frequencies and corresponding mode shapes) for the blade structure by Finite Element Analysis (FEA) technique with the help of SolidWorks Simulation software program. The results indicate that no resonant problem during the different phases of the wind turbine operation. Keywords: H-Darrieus wind turbine participation factors


Numerical modal analysis
Modal

1 Introduction In the last few years, research interest in the small vertical axis wind turbines, in particular, H-type Darrieus rotor has been increasing rapidly considering due to their advantages, such as its relatively low cost, low cut-in wind speed, ease of maintenance, and low noise output, [1–3]. Wind Blade is considered as the most significant part in wind system which is expensive and the reliability of wind turbine is related directly to its quality. Therefore, its structure has been receiving a great consideration to comprehend the behavior of blade by researchers and engineers [4]. During operation, the wind turbine is subjected to dynamic environments, static studies are not enough to evaluate the safety, stability or reliability of the structure. Frequency or modal studies can help engineers to detect structural damage and to avoiding resonance during regular wind turbine operation [5, 6]. Due to its versatility and numerical efficiency, Finite element analysis (FEA) technique has become a key © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 416–422, 2021. https://doi.org/10.1007/978-3-030-63846-7_39

Finite Element Modelling and Analysis for Modal Investigation

417

indispensable technology in the simulation and it dominates the advanced engineering analysis software like SolidWorks, Pro/ENGINEER, and ANSYS [7–9]. In this present works, the 3D modelling and structural modal analysis on a straight blade for a 10-kW H-type Darrieus wind turbine is performed by the FEA technique by using SolidWorks Simulation software program [10]. In this analysis, Modal parameters (natural frequencies and corresponding mode shapes) are determined and analyzed. Finally, the structural safety is confirmed through the analysis results.

2 Blade Structure Model The H-type Darrieus rotor having rated power capacity of 10-KW. Modelling of blade was done using SolidWorks software. Its diameter and height are 8.9 and 7.6 m. The baseline airfoil of the blade profile is NACA0018 symmetric (National Advisory Committee for Aeronautics) [11] which commonly used in Darrieus rotor [12], and its chord length of 470 mm. The blade consists four sections. Figure 1 shows a 3D blade model used for modal analysis.

Fig. 1. Wind blade model.

3 Numerical Investigation FEA technique is an efficient, power and accept tool for tool for the design of wind turbine blade. FEA simulation is generally composed of three steps; pre-processing (the modelling the geometry, generate the mesh and define the boundary condition and loads), processing (computing the solution), and post-processing (obtain the results) [13]. The rounded corners and chamfers are removed from the blade structure, and the

418

F. Ferroudji et al.

simplified model is finally obtained. Material chosen for the blade model employed in this analysis is aluminum. Its material properties are presented in Table 1. Table 1. Aluminium properties. N°. 1 2 3 4

Property Young’s modulus Yield stress (f y ) Density (q) Poisson ratio (m)

Value 71.9 (Gpa) 10.792 (Mpa) 2810 (kg/m3) 0.33

In the next, the loading and boundary conditions are applied in the blade model, as shown in Fig. 2: (i) the roots (strut attachment) of the blade model is fixed constraint, which is at the distance of 1.9 m (H/4) from each end; (ii) The welding areas and joining were replaced by the appropriate connections; and (iii) the except of the gravity no any loads were applied.

Fig. 2. Mesh of the blade model.

After material properties and boundary conditions are defined in FEA program, meshing of the blade model which directly influence the results are completed. The blade model in SolidWorks simulation comprises 3D solid tetrahedron elements with 10-node. The mesh model has 1,542,072 elements and 1,742,877 nodes for a total 4,622,520 degree of freedom (DoF). The size of the element is chosen as small as possible for accuracy of numerical results, which was from 28.357 mm to 141.789 mm). Due to the blade model has more than 2 million of degrees of freedom (large problem), the Intel Direct Sparse solver was used for the numerical solution [11].

Finite Element Modelling and Analysis for Modal Investigation

419

This solver needs large memory (1 GB of RAM for every 200,000 degrees of freedom). A workstation (using 12 processor (INTEL (R) Core (TM) i7-4770 k CPU, RAM 16.0 Go) at a 3.5 GHz clock frequency computer) was used in order to running the simulation.

4 Resultants and Discussions The structural dynamic characteristics of blade structure include the natural frequencies and their corresponding mode shapes. For the multi-DOF blade structure, the frequency of resonance is mainly in the low frequency band. That is because the nodes of the main vibration corresponding to the lower order are fewer than the ones corresponding to high order, so the vibration of lower order is more dangerous [14]. The number of significant modes to be calculated for the dynamic analysis shall be based on mass participation ratio. FEA software programs require the number of modes that excite at least 90% of the total structure mass should be considered in the analysis for each principal direction (X, Y and Z) [14]. In this study, it would require the calculation of 50 modes to obtain the 90% mass participation. Figure 3 shows the mode shapes of the blade structure, and the description of those modes is illustrated in Table 2. As indicate that the mode shapes for this blade from first to tenth orders are located in the frequency range of 20 Hz to 33 Hz, which are distributed evenly on each order. The frequency of 2nd and 3rd, 4th and 5th, 6th and 7th and 8th and 9th are close to each other (almost the same value), respectively. That is because the blade is with the axial-symmetrical structure, so the mode shape is also symmetrical. The results show that the six first modes are bending along the x-direction with exception of the fifth mode is along the z-direction. The natural frequency (excited frequency) of a wind turbine at wind speed can be calculated by the following equation: f¼

kV pD

ð1Þ

Where k ¼ 2:8 is the tip-speed ration (TSR) [15], V is the wind speed (m/s) and D ¼ 8:6 m is rotor diameter. The values of rotor speed of the wind turbine at cut-in and cut-out are 3 and 20 m/s, respectively [16]. This means that the corresponding operational frequency lies between 0.3 and 2 Hz. Obviously, the 1st natural frequency of the blade structure is 20.366 Hz (Table 2), which indicates that no resonant phenomenon during the different phases of the wind turbine operation.

420

F. Ferroudji et al.

The 1st mode

The 4th mode

The 2nd mode

The 5th mode

The 3rd mode

The 6th mode

Fig. 3. Mode shapes of the wind turbine blade.

Table 2. Natural frequency and its mode shapes of the blade structure. Order 1st 2nd 3rd 4th

Natural frequency (Hz) Mode shapes 20.366 1st Bending of the blade (x-direction) 23.529 2nd Bending of the blade (x-direction) 23.624 3rd Bending of the blade (x-direction) 32.170 4th Bending of the blade (x-direction) (continued)

Finite Element Modelling and Analysis for Modal Investigation

421

Table 2. (continued) Order 5th 6th 7th 8th 9th 10th

Natural frequency (Hz) Mode shapes 32.261 1st Bending of the blade (z-direction) 32.358 5th Bending of the blade (x-direction) 32.451 Higher bending of the blade 32.536 Higher bending of the blade 32.618 Higher bending of the blade 32.723 Higher bending of the blade

5 Conclusion In this article finite element modelling of a straight blade for a 10-kW H-Darrieus wind turbine is done in order to investigate the modal behavior. Natural frequencies and their mode shapes are acquired using SolidWorks software. The natural frequencies are close to each other (almost the same value). That is because the blade is with the axialsymmetrical structure, so the mode shape is also symmetrical. The six first modes are bending along the x-direction with exception of the fifth mode is along the z-direction. The 1st natural frequency of the blade is greater than excited frequency, which indicates that no resonant problem during the different phases of the wind turbine operation.

References 1. Kim, C.K., Ali, S., Lee, S.M., Jang, C.M.: Blade optimization of a small vertical-axis wind turbine using the response surface method. In: Sayigh, A. (ed.) Renewable Energy and Sustainable Buildings Innovative Renewable Energy. Springer Chapter 66, pp. 801–812 (2020) 2. Ferroudji, F., Khelifi, C., Meguellatic, F., Koussa, K.: Design and static structural analysis of a 2.5 kW combined Darrieus-Savonius wind turbine. Int. J. Eng. Res. Africa 30, 94–99 (2017) 3. Bianchini, A., Giovanni, F., Ferrari, L.: Design guidelines for H-Darrieus wind turbines: optimization of the annual energy yield. Energy Convers. Manag. 89, 690–707 (2015) 4. Ferroudji, F., Khelifi, C.: Structural strength analysis and fabrication of a straight blade for a H-Darrieus wind turbine. J. Appl. Comput. Mech. 7(2) (2021). https://doi.org/10.22055/ jacm.2020.31452.1876 5. Verma, G., Weber, M.: SolidWorks Simulation 2017 Black Book 4th Cadcamcae Works, USA (2016) 6. Sellami, T., Berriri, H., Darcherif, A.M., Jelassi, S., Mimouni, M.F.: Modal and harmonic analysis of three-dimensional wind turbine models. Wind Eng., 690–707. https://doi.org/10. 1177/0309524x16671093 7. Ferroudji, F., Outtas, T., Khelifi, C.: Design, modeling and finite element static analysis of a new two axis solar tracker using SolidWorks/COSMOSWorks. Appl. Mech. Mater., 446– 447, 738–743 (2020)

422

F. Ferroudji et al.

8. Cizikova, A., Monkova, K., Markovic, J., Ciencala, J.: Numerical and experimental modal analysis of gear wheel. MM Sci. J. (2016). https://doi.org/10.17973/mmsj.2016_11_201654 9. Khelifi, C.H., Ferroudji, F.: Stress and fatigue analyses under wind loading of the dual axis sun tracking system via finite element analysis. J. Mech. Eng. Sci. 10(2), 2008–2015 (2016) 10. SolidWorks, SolidWorks Corporation, 300 Baker Avenue, Concord, MA 01742 (2016). http://www.solidworks.com/ 11. Jacobes, E.N., Ward, K.E., Pinkerton, R.M.: The characteristics of 78 related airfoil sections from tests in the variable-density wind tunnel national advisory committee for aeronautics, Report No. 460 (1933) 12. Mohamed, O.S., Ibrahim, A., Etman, A., Abdelkader, A., Elbaz, A.: Numerical investigation of Darrieus wind turbine with slotted airfoil blades. Energy Convers. Manag. X (2016). https://doi.org/10.1016/j.ecmx.2019.100026 13. Ferroudji, F., Khelifi, C., Meguellati, F.: Modal analysis of a small H-Darrieus wind turbine based on 3D CAD, FEA. Int. J. Renew. Energy Res. 6(2), 637–643 (2016) 14. Ferroudji, F., Khelifi, C., Outtas, T.: Structural dynamics analysis of three-dimensional biaxial sun-tracking system structure determined by numerical modal analysis. J. Sol. Energy Eng. 140, 031004-1-11 (2018) 15. Roummani, K., Koussa, K., Ferroudji, F., Meguellati, F., Bakou, Y., Saihi, L.: A new study of direct-driven wind energy conversion system under variable wind speed. In: International Renewable and Sustainable Energy Conference (IRSEC), Rabat (2018). https://doi.org/10. 1109/IRSEC.2018.8703023 16. Roummani, K., Hamouda, M., Mazari, B., Bendjebbar, M., Koussa, K., Ferroudji, F., Neçaibia, A.: A new concept in direct-driven vertical axis wind energy conversion system under real wind speed with robust stator power control. Renew. Energy 143, 478–487 (2019)

Engineering Applications of Artificial Intelligence

Heating Load Energy Performance of Residential Building: Machine LearningCluster K-Nearest Neighbor CKNN (Part I) Aissa Boudjella1,2(&) and Manal Y. Boudjella3 1

Bircham International University, Avda Sierra-2 (Urb. Guadamonte), Villanueva de la Cañada, Madrid 28691, Spain [email protected] 2 Bircham in Ernational University, 1221 Brickel Av., Suite 900, Miami, FL 33131, USA 3 University of Sciences and Technology of Oran Mohamed Boudiaf USTO-MB, BP 1505, El M’naouar, 31000 Oran, Algeria [email protected], [email protected]

Abstract. In this paper, we perform energy analysis in assessing the heating load of building shapes system based on Cluster K-Nearest Neighbor (CKNN) classifier. The system is implemented and simulated in Anaconda, and its performance is tested on real dataset that contains 8 features with 768 instances to classify the heating load magnitude into four (04) classes (4 target name labels). Classes were created separately based on the magnitude of captured heating load. Various training and test sizes are used to compare the performance measurement of the cooling load energy to predict the class label description, when k, the nearest neighbor changes from 1 to 19. The simulation results show that the accuracy depends on both independent parameters, the test size and kneighbors, which gives better training accuracy, slightly higher than the test accuracy in the range of [85–100%] and [79%–89%], respectively. When k is in the interval [2–6], the training accuracy is approximately equal to 0.917
0.0575, and the test accuracy is about 0.83
0.05. Each class shows three (03) regions: 1) Region (I), in the range of k [1–5], the accuracy increases or decreases as k increases; Region (II), in the interval of k [5–10], the accuracy decreases as k increases, and 3) finally, in the region (III), where the accuracy remains approximately constant when k increases from 10 to 19. In this investigation, the prediction of the heating load magnitude under different classes maybe optimized in the interval of k [2–6] by combining the test size in the range of [10%–50%]. The present proposed methodology can serve as a platform how to utilize the machine learning techniques for measurement and verification of the energy heating load performance. This approach defines a boundary of analysis based on the independent parameters such as k-neighbors and test size and makes use of all data recorded across the facility for the purposes of maximizing the accuracy. Keywords: Heating load  Accuracy  Test size  Machine learning  Cluster  k-Nearest  Neighbor  Classifier

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 425–435, 2021. https://doi.org/10.1007/978-3-030-63846-7_40

426

A. Boudjella and M. Y. Boudjella

1 Introduction The k-Nearest Neighbor (KNN) is a supervised learning algorithm where the result of a new instance query is classified based on majority of K-nearest neighbor category. The purpose of the algorithm is to classify a new object based on attributes and training samples. The classification is based on memory and on the majority vote among the classification of the k objects without using any model for fitting. The k number of objects or training points closest to the query point can be found when the query point is known. The KNN algorithm uses neighborhood classification as the prediction value of the new query instance (Teknomo, K., 2006) [1]. This approach has many applications, in particular, pattern recognition, machine learning (Duda R. O. And Hart P. E., 1973) and database querying (P. Indyk and R. Motwani, 1998). Supervised machine learning is one of the most commonly used and successful types of machine learning. It is used whenever we want to predict a certain outcome (ouput) from a given input, these input/output pairs, that contain the training and test set. A model is build on the training data and then be able to make accurate predictions on a new, unseen data that has the same characteristics as the training set. If a model is able to make accurate predictions on unseen data, it is able to generalize from the training set to the test set. The goal is to make accurate predictions for a new, never-before-seen data, with a model that is able to generalize as accurately as possible. The only measure of whether an algorithm will perform well on a new data is the evaluation on the test set. Therefore, is to find the simplest model to avoid overfitting or underfiting. Building a model that is too complex for the amount of information, is called overfitting. Overfitting occurs when you fit a model too closely to the particularities of the training set and obtain a model that works well on the training set but is not able to generalize to new data. On the other hand, if a model is too simple, then it might not be able to capture all the aspects of and variability in the data, and a model will do badly even on the training set. Choosing too simple model is called underfitting. In recent years, artificial intelligence (AI) and machine learning (ML) techniques have been proposed for forecasting of building energy consumption and performance [4–11]. Srihari et al. (2018) presented energy efficiency techniques using four (04) evaluation metrics for predicting Heating and Cooling Load energy which consist of eight (08) input features and 768 rows of residential buildings data. They used four (04) linear regression ((linear regression, Lasso, Ridge, and Elastic-Net) and three (03) gradient boosting models (GBM, XGBoost, and LightGBM) to compare the performance of these models. The results showed that the gradient boosting models perform significantly better than the standard regression models for both Heating Load (HL) and Cooling Load (CL). XGBoost achieves the highest R-squared score of 0.99 for Heating Load and Cooling Load. Grasiele et al. (2017) evaluate the performance of four (04) machine learning methods (decision trees, random forests, multi-layer perceptron neural networks and support vector machines) to predict cooling and heating loads of residential buildings. The dataset consists of 768 samples with eight (08) input variables and two (02) output variables derived from building designs. Four (04) statistical measures and one

Heating Load Energy Performance of Residential Building

427

(01) synthesis index were used for assessment and comparison of the performance. From the obtained results, random forests proved to be the best option for predicting heating loads, while multi-layer perceptron neural networks produced the most accurate results for cooling loads. Support vector machines showed accurate predictions for cooling loads with a slightly lower performance. In order to analyze the effects of the variables to predict heating and cooling loads of a building, Sholahudin et al. (2015) used an artificial neural network method for design of experiment and conducted ANOVA analysis. The model gives a very good prediction in comparison with the original data sets (A. Tsanas, A. Xifara, 2012), but it does not give any information on the effect of individual variables on the results of Heating and Cooling Load. Comparison also shows quite good agreements between predicted and original data values especially for Heating Load. Muhammad Fayaz et al. (2018) have proposed a methodology for energy consumption prediction in residential buildings. They used three (03) prediction algorithms, the deep extreme learning machine (DELM) the adaptive neuro-fuzzy inference system (ANFIS) and artificial neural network (ANN). For a comparative analysis of performance measurements, different statistical measures have been used: The mean absolute error (MAE), the root mean square error (RMSE) and the mean absolute percentage error (MAPE). The results indicate that the performance of DELM is far better than ANN and ANFIS. A simplified calculation model for building envelope cooling loads has been presented (Ping Wang, Guangcai Gong, Yan Zhou and Bin Qin, 2018). It is based on dimensional analysis and dynamic hourly calculation results obtained from EnergyPlus. These authors reported that the simplified proposed model is accurate enough to predict building envelope cooling loads. In addition, the new concept of EWWR (The equivalent window to wall ratio) and building orientation factor have been defined. They conclude: 1) EWWR is more performant than traditional WWR (The window to wall ratio) in cooling load calculation and energy efficiency evaluation because it takes window orientation into account; and 2) Building orientation factor depends on the building orientation and its length to width ratio. General regression neural networks (GRNN) were designed and trained to investigate the feasibility of using the technology to optimize HVAC (heating, ventilating, and air conditioning) systems thermal energy storage in public buildings as well as office buildings (Abdullatif E. Ben-Nakhi, Mohamed A. Mahmoud, 2004). The performance of the NN analysis was evaluated using a statistical indicator (the coefficient of multiple determination) and a statistical analysis of the error patterns, including confidence intervals of regression lines, as well as an examination of the error patterns. The results showed that a properly designed NN is a powerful instrument for optimizing thermal energy storage in buildings based only on external temperature records. An artificial neural network (ANN) model was established to predict the heat and cooling requirements of buildings (Asif Rabbani, 2018). The aim of this investigation was to develop ANN to see the best training parameters that can predict the energy efficiency. The main conclusion is that the prediction of heating and cooling loads with multiple output ANN are much better than the single output model. The ANN also shows that the number of hidden layers should be greater than 5. The best results were obtained in the hidden layers interval of [10,11,12,13,14,15] with the best R values

428

A. Boudjella and M. Y. Boudjella

observed with number of layers 15, R = 0.99128, R = 0.98796, R = 0.98456, and R = 0.98989, for training, validation, test, all data, respectively. Hosein et al. (2019) investigated the design of energy-efficient buildings using six (06) machine learning techniques: Multi-layer Perceptron Regressor (MLPr), Lazy Locally Weighted Learning (LLWL), Alternating Model Tree (AMT), Random Forest (RF), ElasticNet (ENet), and Radial Basis Function Regression (RBFr). They compute R2 (square of the correlation coefficient), MAE (mean absolute error), RMSE (root mean squared error), RAE (relative absolute error), and RRSE (root relative squared error) for the training dataset and for testing dataset. They found that RF was the most predictive with the best values (R2 = 0.9997, MAE = 0.19, RMSE = 0.2399, RAE = 2.078, and RRSE = 2.3795) and (R2 = 0.9989, MAE = 0.3385, RMSE = 0.4649, RAE = 3.6813, and RRSE = 4.5995) for the training dataset and the test dataset, respectively. the goal in this investigation, is to build a machine learning model, applied to data that does not contain any label information (Unsupervised learning algorithms and can learn from the measurements of eight (08) input variables whose features are known, so that we can predict the range of the heating load (HL) for a new eight (08) input dataset of residential building shapes. In this problem, we want to predict one of several options of the heating load as classes with 4 target label names that define the heating loadd magnitude. The possible outputs with different Heating Load (HL) ranges are called classes (class 1, class 2, class 3, and clss 4). The aim of this investigation is to implement platform for measurement and verification of the energy heating load by defining a boundary analysis based on the independent parameters such as k-neighbors and test size and makes use of all data recorded across the facility for the purposes of maximizing the accuracy.

2 Methodology The data we will use for this investigation is the Energy efficiency Data Set (A. Tsanas, A. Xifara, 2012) (https://archive.ics.uci.edu/ml/datasets/Energy+efficiency), a classical dataset in machine learning and statistics. The dataset consists of eight (08) features, denoted by (X1, X2,…X8) and one response (or outcome, denoted by y1, Heating Load, HL) to build our model, making this model unsupervised learning task. The short description of the dataset is reported in reference (https://archive.ics.uci.edu/ml/ datasets/Energy+efficiency) (A. Tsanas, A. Xifara, 2012. Data fields contains eight (08) numeric measurements: X1: Relative Compactness; X2: Surface Area; X3: Wall Area; X4: Roof Area; X5: Overall Height; X6: Orientation; X7: Glazing Area; X8: Glazing Area Distribution; and one outcome y1: Heating Load in a NumPy array. The value of feature names is a list of strings, giving the description of each feature. The value of the key target names is an array of strings. The rows in the data array correspond to HL measurement. The aim is to use the eight (08) features to predict the Heating Load magnitude defined by each class target name label. We want to build a machine learning model from this given data without any label that we can predict HL magnitude in certain range of a new set of measurements of residential building shapes. Unsupervised learning algorithms are usually applied to data that does not contain any

Heating Load Energy Performance of Residential Building

429

label information (without class target name), so we don’t know what the right output should be. The authors proposed to complete this data by defining four (04) classes comprise four (04) target names. The possible outputs with different HL ranges are called classes: class 1, class 2, class 3, and class 4. The outcome y1 is a list of 4 strings, based on the heating load energy magnitude. Each class is defined by the minimum, maximum HL magnitude and range as indicated in Table 1 class 1 is defined in the range of [6.01–19.95], class 2 [20.71–30], class 3 [30.05–40], and class 4 [40.03–43.1]. The system is implemented and simulated in Anaconda (IPython 7.8.0. An enhanced Interactive Python. Python version: 3.7.4 [MSC v.1915 64 bit (AMD64)] and its performance is tested on real dataset that contains(0 8) features with 768 instances to classify the heating load magnitude into eight (04) classes (4 target name labels). Table 1. Heating load range for each class Classes labels Class 1 Class 2 Class 3 Class 4

Minimum heating load 6.01 20.71 30.05 40.03

Maximum heating load 19.95 30 40 43.1

Range of each class [6.01–19.95] [20.71–30] [30.05–40] [40.03–43.1]

Number of instances 397 170 167 34

In this problem, we want to predict one of several options of the heating load as four (04) class ranges of heating lead magnitude. Every eight (08) attributes in the dataset belongs to one of 4 classes. This is an example of four (04) classification problems. The desired output for a single data point is the HL class of this dataset. For a particular data point, the class with defined range it belongs to, is called class 1, class 2, class 3, and class 4. This would certainly be a possible way to divide a collection of data of HL energy, maybe the one we are looking for. As a consequence, unsupervised algorithms are used often in an exploratory setting to understand the data better, rather than as part of a larger automatic system. Another common application for unsupervised algorithms is the preprocessing step for supervised algorithms. Learning a new representation of the data can sometimes improve the accuracy of supervised algorithms, or can lead to reduce memory and time consumption. One way of deciding which performance measure is suitable for the task is to consider the confusion matrix. A confusion matrix is a table of contingencies. In the context of statistical modeling, they typically describe the label prediction versus actual label. It is common to output a confusion matrix (particularly for multiclass problems with more classes) for a trained model as it can yield valuable information about classification failures by failure type and class.

430

A. Boudjella and M. Y. Boudjella Table 2. Confusion matrix 4 by 4 array Prediction Result Class Class Class Class

A B C D

Actual result Class A Class B Class C Class D AA AB AC AD BA BB BC BD CA CB CC CD DA DB DC DD

The output of confusion matrix is a four-by-four array, where the rows correspond to the true classes and the columns correspond to the predicted classes. Table 2 illustrates this meaning by computing accuracy, which can be expressed as indicated in Eq. (1) Accuracy ¼

AA þ BB þ CC þ DD ð1 Þ ðAA þ AB þ AC þ ADÞ þ ðBA þ BB þ BC þ BDÞ þ ðCA þ CB þ CC þ CDÞ þ ðDA þ DB þ DC þ DDÞ

ð1Þ In other words, accuracy is the number of correct predictions (AA þ BB þ CC þ DD) diagonal entries of the confusion matrix summed up divided by the number of all samples (all entries of the confusion matrix summed up).

3 Simulation Results and Discussion Figures 1, 2, 3, 4, 5, 6, 7, 8 and 9 show the accuracy versus k, the nearest neighbors variable under various test and training sizes. The highest accuracy of the training data is observed when k = 1, while the test data shows its lowest value for each test size. In the interval of k [1–5], for both training and test data, the accuracy decreases or increases. When k goes from 5 to 10, the accuracy decreases. When k is larger than 1, the maximum values of the training and the test accuracy are shown in the range of k [3–5]. When k increases in the interval [2–5], the training and the test accuracy increases to reach its maximum value. The position of the maximum depends on the test size or the training size. When k goes from 5 to 10, the accuracy in general decreases. As k becomes larger than 10, it remains approximately constant. Increasing the test or the training size by keeping k constant does not affect significantly the accuracy. Based on the magnitude of accuracy rate, we can define three (03) regions: Region (I), in the interval of k [1–5], the accuracy increases or decreases as k increases. Region (II), in the interval of k [5–10], the accuracy decreases as k increases. Finally, Region (III), the accuracy remains approximately constant when k increases from 10 to 19.

Heating Load Energy Performance of Residential Building

Fig. 1. Training and test accuracy versus k, nearest neighbor Test size: 15%, Training size: 85%

431

Fig. 2. Training and test accuracy versus k, nearest Neighbor. Test size: 20%, Training size: 80%.

Figure 9 shows the test and the training accuracy versus k, under various training and test sizes. The results are summarized in Table 3. For k = 1, the training accuracy is not affected by the test or the training size. It is always equal to unity. While for the test size, the accuracy varies from 0.80 to 0.82. In the range of k [2–10], the accuracy for the training and the test size are observed in the range of [0.86–0.97] and [0.78–0:98], respectively. When k becomes larger than 10, the simulation results show that the accuracy for the traning and the test size remains approximately contant in the range of [0.85–0.87] and [0.78–0.84], respectively.

Fig. 3. Training and test accuracy versus k, nearest neighbor Test size: 25%, Training size: 75%.

Fig. 4. Training and test accuracy versus k, nearest Neighbor. Test size: 30%, Training size: 70%.

432

A. Boudjella and M. Y. Boudjella

Fig. 5. Training and test accuracy versus k, nearest neighbor Test size: 35%, Training size: 65%.

Fig. 6. Training and test accuracy versus k, nearest Neighbor. Test size: 40%, Training size: 60%.

Fig. 7. Training and test accuracy versus k, nearest neighbor Test size: 45%, Training size: 55%

Fig. 8. Training and test accuracy versus k, nearest Neighbor. Test size: 50%, Training size: 50%.

The ratios of the training size accuracy to the test size accuracy versus k, are indicated in Fig. 10. The maximum value is observed when k = 1. When k becomes larger than 1, the ratio decreases and remain approximately constant in the interval [1.14–0.95]. For Q = 90/10, Q = 85/15 and Q = 80/20, the ratio is larger than 1. While for Q = 70/30, it is approximately equal to unity, which means that under this condition the training accuracy is approximately equal to the test accuracy.

Heating Load Energy Performance of Residential Building

433

Fig. 9. Training and test accuracy versus k, neighbors. For various Test and Training sizes (50  90).

Table 3. Accuracy (Fig. 9) and ratio (Fig. 10) under various test training and test sizes. k

Accuracy training set

Min Max Range k = 1 NA 1 NA [2–10] 0.86 0.97 [0.86– 0.97] >10 0.85 0.87 [0.85–0:87]

Accuracy test set Min 0.80 0.78 0.78

Fig. 10. Ratio of training size accuracy to test size accuracy versus k, neighbors under various test and training sizes (50  90).

Max 0.83 0.88 0.84

accury ratio ¼ training training test

Range Min Max Range [0.80–0.82] 1.17 1.25 [1.17–1:25] [0.78–0:88] 1 1.12 [1–1:12] [0.78–0:84]

Fig. 11. Time processing vs k, neighbors under various test and training sizes

434

A. Boudjella and M. Y. Boudjella

The processing time (t) versus k under various test sizes is illustrated in Fig. 11. The results show that the range of the processing time t is in the interval [0.111–0.335] s, which decreases or increases depending on the test size and k parameters. The highest and the lowest values are observed when (k = 2, test-size = 25%, tmax= 0.35 s) and (k = [1–19], test-size = 10%, tmin= 0.1.6 s). Based on the simulation results, this model is able to make accurate predictions from the training set to the test set. We build a machine learning model, applied to data that does not contain any label information (Unsupervised learning algorithms) that works well on the training set and does not do badly even on the test set. A model that can learn from the measurements of eight (08) input variables whose features are known, so that we can predict the range of the heating load (HL) for a new eight (08) input dataset of residential buildings. The present research is only an approach how to utilise the machine learning techniques for measurement and verification of energy heating load. The methodology maybe considered as a primary guidance on the application of machine learning for the purposes of maximizing the accuracy. This approach defines a boundary of analysis based on the independent parameters such as k-neighbors and test size and makes use of all data recorded across the facility. The proposed methodology seeks to realise the achievable uncertainty by utilising the available resources. With quantify energy heating load, the application of this methodology may demonstrates an acceptable level of uncertainty to overcome certain impredictible issues such as poor or complex quality data. The methodology can be designed to be robust enough to be compatible across the spectrum of measurement and verification of dataset. This is a confirmation that measurement and verification of the present Heating Load accuracy (Table 3) and Cooling Load accuracy reported by the author in reference (14-Table 3) (Aissa Boudjella, 2020) indicate that the level of the uncertainty achievable is dependent on the individual dataset based on the k-neighbors and test size independent parameters.

4 Conclusion Quantitative estimation classification using statistical machine learning tool based on KNN classifier is proposed to predict the Heating Load performance energy of residential building shapes by creating four (04) classes. The working system was tested successfully, which is able to detect and recognize the heating load magnitude on real dataset features consisting of 8 features and containing 768 instances. The experimental results show a high training accuracy when k = 1, while the test accuracy reaches its minimum value. In the range of k [1–10], the accuracy increases or decreases. When k becomes larger than 10, it remains approximately constant. The test size combined with the CKNN method can be used to control the accuracy rate. For the whole test size, when k = 4 or 5, the class label extraction of this data succeeds in high accuracy larger than 0.91% for the traning test and 0.85% for the test size. To conclude, the prediction of the heating load magnitude class can be optimized in the range of k [2–6] (region I) by combining the test size from 10% to 50%.

Heating Load Energy Performance of Residential Building

435

This model works well on the training set as well as the test set and maybe is able to generalize to a new data. It can learn from the measurements of eight (08) input variables whose features are known, so that we can predict the range of the heating load for a new eight (08) input dataset of residential buildings.

References 1. Teknomo, K.: K-Nearest Neighbors Tutorial (2006). http://www.people.revoledu.com\kardi \tutorial\KNN\ 2. Duda, R.O., Hart, P.E.: Pattern Classification and Scene Analysis. John Wiley & Sons (1973) 3. Indyk, P., Motwani, R.: Approximate nearest neighbors: towards removing the curse of dimensionality. In: STOC, pp. 604–613 (1998) 4. Srihari, J., Santhi, B.: Prediction of heating and cooling load to improve energy efficiency of buildings using machine learning. Tech. J. Mech. Cont. Math. Sci. 3(5), 97–113 (2018) 5. Regina, G., Leonardo, D., da Fonseca, G., Vanessa, P., Capriles, Z., Afonso, G.: Celso de Castro Lemonge Ambiente Construído, Porto Alegre 17(3), 103–115 (2017).. https://doi.org/ 10.1590/s1678-86212017000300165 6. Sholahudin, A.G.A, Baek, C.-I., Han, H.: Prediction and Analysis of Building Energy Efficiency using Artificial Neural Network and Design of Experiments Applied Mechanics and Materials Submitted: 2014–09-01. ISSN: 1662-7482, vol. 819, pp. 541–545 Revised: 2015-01-3. doi:10.4028/ www.scientific.net/AMM.819.541 7. Tsanas, A., Xifara, A.: Accurate quantitative estimation of energy performance of residential buildings using statistical machine learning tools. Energy Build. (2012). https://doi.org/10. 1016/j.enbuild.2012.03.003 8. Fayaz, M., Kim, D.: A prediction methodology of energy consumption based on deep extreme learning machine and comparative analysis. Residential Build. Electron. 7(10), 222 (2018). https://doi.org/10.3390/electronics7100222 9. Wang, P., Gong, G., Zhou, Y., Qin, B.: A Simplified Calculation Method for Building Envelope Cooling Loads in Central South China. Energies 11, 1708 (2018). https://doi.org/ 10.3390/en11071708.www.mdpi.com/journal/energies 10. Ben-Nakhi, A.E., Mahmoud, M.A.: Cooling load prediction for buildings using general regression neural networks. Energy Conversion Manage. 45, 2127–2141 (2004). www. elsevier.com/locate/enconman 11. Rabbani, A.: Energy efficiency: prediction of the heat and cooling requirements of buildings. Int. J. Eng. Res. 7(12), 207–210 (2018). ISSN:2319-6890 (online), 2347-5013(print), IJER@2018 Page 207

Cooling Load Energy Performance of Residential Building: Machine LearningCluster K-Nearest Neighbor CKNN (Part I) Aissa Boudjella1,2(&) and Manal Y. Boudjella3 1

Bircham International University, Avda Sierra-2 (Urb.Guadamonte), 28691 Villanueva de la Cañada, Madrid, Spain [email protected] 2 Bircham in Ernational University, 1221 Brickel Av., Suite 900, Miami, FL 33131, USA 3 University of Sciences and Technology of Oran Mohamed Boudiaf USTO-MB, BP 1505 El M’naouar, 31000 Oran, Algeria [email protected]

Abstract. In this paper, we perform energy analysis in assessing the cooling load of building shapes system based on the Cluster K-Nearest Neighbor (CKNN) method for classification. The system is implemented and simulated in Anaconda, and its performance is tested on real dataset that contains 8 features with 768 instances to classify the cooling load magnitude into 8 classes (8 target name labels). Classes were created separately based on the magnitude of captured cooling load. Various training and test sizes are used to compare the performance measurement of the cooling load energy to predict the class label description, when k, the nearest neighbor changes from 1 to 19. The simulation results show that the accuracy depends on both independent parameters, the test size and k-neighbors, which gives better training and no bad test accuracy rate with slightly lower in the range of [55%–100%] and [35%–75%], respectively. When k is in the interval [3–6], the training accuracy is approximately equal to [0.80
0.08], and the test accuracy, except for the test size = 10%, 40% and 50% is about [0.65
0.10]. Each class shows three (03) regions: 1) Region (I), in the range of k [1–6], where the accuracy increases or decreases when k increases; 2) Region (II), in the interval of k [6–10], the accuracy decreases as k increases, and 3) finally, in the region (III), where the accuracy remains approximately constant when k increases from 10 to 19. In this investigation, the prediction of the cooling load magnitude under different classes maybe optimized in the interval of k [4, 6] by combining the test size in the range of [10%– 50%]. The present proposed methodology can serve as a platform how to utilize the machine learning techniques for measurement and verification of the energy cooling load, by defining a boundary of analysis based on the independent parameters such as k-neighbors and test size and makes use of all data recorded across the facility for the purposes of maximizing the accuracy. Keywords: Cooling load  Accuracy K-Nearest neighbor classifier

 Test size  Machine learning  Cluster

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 436–446, 2021. https://doi.org/10.1007/978-3-030-63846-7_41

Cooling Load Energy Performance of Residential Building

437

1 Introduction The k-Nearest Neighbor (KNN) is a supervised learning algorithm where the result of a new instance query is classified based on majority of K-nearest neighbor category. The purpose of this algorithm is to classify a new object based on attributes and training samples. The classification is based on memory and on the majority vote among the classification of the k objects without using any model for fitting. The k number of objects or training points closest to the query point can be found when the query point is known. The KNN algorithm uses neighborhood classification as the prediction value of the new query instance (Teknomo 2006). This approach has many applications, in particular, pattern recognition, machine learning (Duda and Hart 1973) and database querying (Indyk and Motwani 1998). Supervised machine learning is one of the most commonly used and successful types of machine learning. It is used whenever we want to predict a certain outcome (output) from a given input, these input/output pairs, that contain the training and test set. A model is build on the training data and then be able to make accurate predictions on a new, unseen data that has the same characteristics as the training set. If a model is able to make accurate predictions on unseen data, it is able to generalize from the training set to the test set. The goal is to make accurate predictions for a new, never-before-seen data, with a model that is able to generalize as accurately as possible. The only measure of whether an algorithm will perform well on a new data is the evaluation on the test set. Therefore, is to find the simplest model to avoid overfitting or underfiting. Building a model that is too complex for the amount of information, is called overfitting. Overfitting occurs when you fit a model too closely to the particularities of the training set and obtain a model that works well on the training set but is not able to generalize to new data. On the other hand, if a model is too simple, then it might not be able to capture all the aspects of and variability in the data, and a model will do badly even on the training set. Choosing too simple model is called underfitting. In recent years, artificial intelligence (AI) and machine learning (ML) techniques have been proposed for forecasting of building energy consumption and performance [4–11]. Srihari et al. (2018) presented energy efficiency techniques using four evaluation metrics for predicting the Heating and Cooling Load energy which consist of eight (08) input features and 768 rows of residential buildings data. They used four (04) linear regression ((linear regression, Lasso, Ridge, and Elastic-Net) and three (03) gradient boosting models (GBM, XGBoost, and LightGBM) to compare the performance of these models. Their results showed that the gradient boosting models perform significantly better than the standard regression models for both Heating Load (HL) and Cooling Load (CL). XGBoost achieves the highest R-squared score of 0.99 for Heating Load and for Cooling Load. Grasiele et al. (2017) evaluate the performance of four (04) machine learning methods (decision trees, random forests, multi-layer perceptron neural networks and support vector machines) to predict cooling and heating loads of residential buildings. The dataset consists of 768 samples with eight (08) input variables and two (02) output variables derived from building designs. Four (04) statistical measures and one (01) synthesis index were used for assessment and comparison of the performance.

438

A. Boudjella and M. Y. Boudjella

From the obtained results, random forests proved to be the best option for predicting heating loads while multi-layer perceptron neural networks produced the most accurate results for cooling loads. Support vector machines showed accurate predictions for cooling loads with a slightly lower performance. In order to analyze the effects of the variables to predict heating and cooling loads of a building, Sholahudin et al. (2015) used an artificial neural network method for design of experiment and conducted ANOVA analysis. The model gives a very good prediction in comparison with the original data sets (Tsanas and Xifara 2012), but it does not give any information on the effect of individual variables on the results of HL and CL. Comparison also shows quite good agreements between predicted and original data values especially for HL. Muhammad Fayaz et al. (2018) have proposed a methodology for energy consumption prediction in residential buildings. In the prediction layer, they used three (03) prediction algorithms, the deep extreme learning machine (DELM) the adaptive neuro-fuzzy inference system (ANFIS) and artificial neural network (ANN). For a comparative analysis of performance measurements, different statistical measures have been used: the mean absolute error (MAE), the root mean square error (RMSE) and the mean absolute percentage error (MAPE). The results indicate that the performance of DELM is far better than ANN and ANFIS. A simplified calculation model for building envelope cooling loads has been presented (Wang et al. 2018). It is based on dimensional analysis and dynamic hourly calculation results obtained from EnergyPlus. These authors reported that the simplified proposed model is accurate enough to predict building envelope cooling loads. In addition, the new concept of EWWR (The equivalent window to wall ratio) and building orientation factor have been defined. They conclude: 1) EWWR is more performance than traditional WWR (The window to wall ratio) in cooling load calculation and energy efficiency evaluation because it takes window orientation into account; 2) Building orientation factor depends on the building orientation and its length to width ratio. General regression neural networks (GRNN) were designed and trained to investigate the feasibility of using the technology to optimize HVAC (heating, ventilating, and air conditioning) systems thermal energy storage in public buildings as well as office buildings (Abdullatif et al. (2004)). The performance of the NN analysis was evaluated using a statistical indicator (the coefficient of multiple determination) and a statistical analysis of the error patterns, including confidence intervals of regression lines, as well as an examination of the error patterns. The results showed that a properly designed NN is a powerful instrument for optimizing thermal energy storage in buildings based only on external temperature records. An artificial neural network (ANN) model was established to predict the heat and cooling requirements of buildings (Rabbani 2018). The aim of this investigation was to develop ANN to see the best training parameters that can predict the energy efficiency. The main conclusion is that the prediction of heating and cooling loads with multiple output ANN are much better than the single output model. The ANN also shows that the number of hidden layers should be greater than 5. The best results were obtained in the hidden layers interval of [10–15] with the best R values observed with number of

Cooling Load Energy Performance of Residential Building

439

layers 15, R = 0.99128, R = 0.98796, R = 0.98456, and R = 0.98989, for training, validation, test, all data, respectively. In this investigation, the goal is to build a machine learning model, applied to data that does not contain any label information (Unsupervised learning algorithms) that can learn from the measurements of eight (08) input variables whose features are known, so that we can predict the range of the cooling load (CL) for a new eight (08) input variables dataset of residential building shapes. In this problem, we want to predict one of several options of the cooling load as classes with 8 target label names that define the cooling lead magnitude. The possible outputs with different Cooling Load ranges are called classes (class 1, class 2, class 3, class 4, class 5, class 6, class 7, and class 8). The aim of this investigation is to implement platform for measurement and verification of the energy cooling load by defining a boundary analysis based on the independent parameters such as k-neighbors and test size and makes use of all data recorded across the facility for the purposes of maximizing the accuracy.

2 Methodology The data we will use for this investigation is the Energy efficiency Data Set (Tsanas and Xifara 2012) (https://archive.ics.uci.edu/ml/datasets/Energy+efficiency), a classical dataset in machine learning and statistics. The dataset consists of eight (08) features, denoted by(X1, X2, … X8) and one response (or outcome, denoted by y1, Cooling Load, CL) to build our model, making this model unsupervised learning task. The short description of the dataset is reported in reference (https://archive.ics.uci.edu/ml/ datasets/Energy+efficiency) (Tsanas and Xifara 2012). Data fields contains eight (08) numeric measurements: X1: Relative Compactness; X2: Surface Area; X3: Wall Area; X4: Roof Area; X5: Overall Height; X6: Orientation; X7: Glazing Area; X8: Glazing Area Distribution; and one outcome y1: Heating Load in a NumPy array. The value of feature names is a list of strings, giving the description of each feature. The value of the key target names is an array of strings. The value of feature names is a list of strings, giving the description of each feature. The rows in the data array correspond to CL measurement. The aim is to use the eight (08) features to predict the Cooling Load magnitude defined by each class target name label. We want to build a machine learning model from this given data without any label that we can predict CL magnitude in certain range of a new set of measurements. Unsupervised learning algorithms are usually applied to data that does not contain any label information (without class target name), so we don’t know what the right output should be. The authors proposed to complete this data by defining eight (08) classes comprise eight (08) target names. The possible outputs with different CL ranges are called classes: class 1, class 2, class 3,.. class 8. The outcome y1 is a list of 8 strings, based on the cooling load energy magnitude. Each class is defined by the minimum, maximum CL magnitude and range as indicated in Table 1. Class 1 is defined in the range of [10.90–14.58], class 2 [15. 03–19.90], class 3 [20.01–24.93], class 4 [25.02–30], class 5 [30.02–34.99], class 6 [35.4–39.92], class 7 [40.10–44.87], and class 8 [45.52–47.01]. The system is implemented and simulated in Anaconda (IPython 7.8.0. An enhanced Interactive Python. Python version: 3.7.4 [MSC v.1915 64 bit (AMD64)] and its performance is tested on

440

A. Boudjella and M. Y. Boudjella

real dataset that contains 8 features with 768 instances to classify the cooling load magnitude into 8 classes (8 target name labels). Table 1. Cooling load range for each class Classeslabels Class 1 Class 2 Class 3 Class 4 Class 5 Class 6 Class 7 Class 8

Minimum Cooling load 10.90 15.03 20.01 25.02 30.02 35.4 40.10 45.52

Maximum Cooling load 14.58 19.90 24.93 30 34.99 39.92 48.87 47.01

Range of each class [10.90 14.58 ] [15.03 19.90 ] [20.01 24.93] [25.02 30] [30.02 34.99] [35.4 39.92] [40.10 44.87] [45.52 47.01]

Number of instances 150 198 53 114 126 89 26 13

In this problem, we want to predict one of several options of the cooling load as eight (08) classes of cooling lead magnitude. Every eight (08) attributes in the dataset belongs to one of eight (08) classes. This is an example of eight (08) classification problems. The desired output for a single data point is the CL class of this dataset. For a particular data point, the class with defined range it belongs to, is called class 1, class 2, class 3, … class 8. This would certainly be a possible way to divide a collection of data of CL energy, maybe the one we are looking for. As a consequence, unsupervised algorithms are used often in an exploratory setting to understand the data better, rather than as part of a larger automatic system. Another common application for unsupervised algorithms is the preprocessing step for supervised algorithms. Learning a new representation of the data can sometimes improve the accuracy of supervised algorithms, or can lead to reduce memory and time consumption. One way of deciding which performance measure is suitable for the task is to consider the confusion matrix. A confusion matrix is a table of contingencies. In the context of statistical modeling, they typically describe the label prediction versus actual label. It is common to output a confusion matrix (particularly for multiclass problems with more classes) for a trained model as it can yield valuable information about classification failures by failure type and class. The output of confusion matrix is a 8by-8 array, where the rows correspond to the true classes and the columns correspond to the predicted classes. Table 2 illustrates this meaning by computing accuracy which can be expressed as indicated in Eq. (1) Pi¼8 Pj¼8   i¼1 : j¼1 aij i¼j Accuracy ¼ Pi¼8 Pj¼8 i¼1 : j¼1 aij In other words, accuracy is the number of correct predictions

ð1Þ iP ¼8 j¼8 P

:

i¼1 j¼1

aij

 i¼j

(all

diagonal entries of the confusion matrix summed up) divided by the number of all iP ¼8 j¼8 P : aij (all entries of the confusion matrix summed up). samples i¼1 j¼1

Cooling Load Energy Performance of Residential Building

441

Table 2. Confusion matrix 8 by 8 array Actual result

Prediction Result

Class 1 Class 2 Class 3 Class 4 Class 5 Class 6 Class 7 Class 8

Class 1 Class 2 Class 3

Class 4

Class 5

Class 6

Class 7

Class 8

a11 a21 a31 a41 a51 a61 a71 a81

a14 a24 a34 a44 a54 a64 a74 a84

a15 a25 a35 a45 a55 a65 a75 a85

a16 a26 a36 a46 a56 a66 a76 a86

a17 a27 a37 a47 a57 a67 a77 a87

a18 a28 a38 a48 a58 a68 a78 a88

a12 a22 a32 a42 a52 a62 a72 a82

a13 a23 a33 a43 a53 a63 a73 a83

3 Simulation Results and Discussion Figures 1, 2, 3, 4, 5, 6, 7, 8, 9 and 10 show the accuracy versus k, the nearest neighbor variable under various test and training sizes. The highest accuracy of the training data is observed when k = 1, while the test data shows its second lowest accuracy for each test size. In the interval of k [2–6], for both training data and test data, the accuracy decreases or increases. When k goes from 6 to 10, the accuracy decreases. When k is larger than 1, the maximum values of the training and the test accuracy are shown in the range of k [3–5]. When k increases in the interval [2–5], the training and the test accuracy increases to reach its maximum value. The position of the maximum value depends on the test size or the training size. As k becomes larger than 10, it remains approximately constant. In the interval of k [2–6], when keeping k constant and increasing the test or the training size, the accuracy changes significantly from 70% to 88% for the training set and from 35% to 75% for the test set. For k larger than 10, the accuracy for the training set and the test set is observed in the range of [055%–70%] and [50%–60%], respectively.

Fig. 1. Training and test accuracy versus k, nearest neighbor. Test size: 10%, Training size: 90%.

Fig. 2. Training and test accuracy versus k, nearest neighbor. Test size: 15%, Training size: 85%.

442

A. Boudjella and M. Y. Boudjella

Fig. 3. Training and test accuracy versus k, nearest neighbor. Test size: 20%, Training size: 80%.

Fig. 4. Training and test accuracy versus k, nearest neighbor. Test size: 25%, Training size: 75%.

Based on the magnitude of the accuracy rate, we can define three regions: Region (I), in the interval of k [1–6], the accuracy increases or decreases as k increases. Region (II), in the interval of k [6–10], the accuracy decreases as k increases. Finally, Region (III), the accuracy remains approximately constant when k increases from 10 to 19. Figure 10 show the test and the training accuracy versus k, under various test and training sizes. The results are summarized in Table 3. For k = 1, the training accuracy is not affected by the test size or the training size, it is always equal to 1. While the test size the accuracy varies from 44% to 48%. In the range of k [2–10] the accuracy for the training and the test size is observed in the range of [0.65%–0.89%] and [0.36%– 0.75%], respectively. When k becomes larger than 10, the simulation results shows that the accuracy for the training and the test size remains approximately constant in the interval [0.65%–0.68%] and [0.48%–0.60%], respectively.

Fig. 5. Training and test accuracy versus k, nearest neighbor. Test size: 30%, Training size: 70%.

Fig. 6. Training and test accuracy versus k, nearest Neighbor. Test size: 35%, Training size: 75%.

Cooling Load Energy Performance of Residential Building

Fig. 7. Training and test accuracy versus k, nearest neighbor. Test size: 40%, Training size: 60%.

443

Fig. 8. Training and test accuracy versus k, nearest neighbor. Test size: 45%, Training size: 65%.

The ratios of the training size accuracy to the test size accuracy versus k, are indicated in Fig. 11. The highest value is observed when k = 1 and 2 in the range of [1.9–2.28] and [1.5–2.28]. When the test size is larger than 35% in the range of k [2– 10], the training accuracy varies from 65% to 85%. While the test size shows lower accuracy in comparison with the training size. The accuracy changes from 36% to 75% with its highest accuracy (70%) when the test size becomes larger than 30%. Considering a single nearest neighbor, the prediction on the training set is better. But when more neighbors are considered, the model becomes simpler and the training accuracy drops. The test set accuracy for using a single neighbor is lower than when using more neighbors, indicating that using the single nearest neighbor leads to a model that is too complex. On the other hand, when considering more than 10 neighbors, the model is too simple and performance is even worse. The best performance is somewhere in the range of k [4–6], using around 5 ± 1 neighbors. The processing time t versus k under various test sizes are illustrated in Fig. 12. The results show that the range of the processing time is in the interval [0.175–0.325] s which decreases or increases depending on the test size and k parameters. The highest and the lowest values are observed when (k = 5, test-size = 45%, t = 3.175 ns) and (k = 16, 17, test-size = 25%, t = 3 ms, and t = 2.375 ns), respectively. Based on the simulation results, this model is not able to make accurate predictions from the training set to the test for the whole range of k and the test size. When k = 1, the model performs well for the training set (accuracy = 100%), while the test set shows lower accuracy in the range of [44%–49%]. In the range of k [2–10], the model performs well with the training set with the accuracy in the range of [65%–89%]. But, the test size shows the accuracy in the range of [36%–75%]. The best performance for this model is when k is in the range of k [4–6], the training size is less than 35%, with the training set and the test set accuracy larger than 80% and 70%, respectively. We see that our model is about 75% accurate, which might still be acceptable.

444

A. Boudjella and M. Y. Boudjella

Fig. 9. Training and test accuracy versus k, nearest neighbors. Test size: 50%, Training size: 50%.

Fig. 10. Training and test accuracy versus k, neighbors. For various Test and Training sizes (50–90).

The present research is only an approach how to utilise the machine learning techniques for measurement and verification of energy cooling load. The methodology maybe considered as a primary guidance on the application of machine learning for the purposes of maximizing the accuracy. The approach in the present investigation defines a boundary of analysis based on the independent parameters such as k-neighbors and test size and makes use of all data recorded across the facility. The proposed methodology seeks to realise the achievable uncertainty by utilising the available resources. With quantify energy cooling load, the application of this methodology may demonstrates an acceptable level of uncertainty to overcome certain issues. The methodology can be designed to be robust enough to be compatible across the spectrum of measurement and verification of dataset. This is a confirmation that measurement and verification based on the k-neighbors and test size independent parameters of the present Coling Load accuracy (Table 3) and Heating Load accuracy reported by the author in reference [13]-Table 3 (Boudjella 2020) indicate that the level of the uncertainty achievable is dependent on the individual dataset.

Cooling Load Energy Performance of Residential Building

445

Table 3. Accuracy (Fig. 11) and ratio (Fig. 12) under various test and training sizes.

k k=1 [2–10] >10

Accuracy training set

Accuracy test set

Min NA 0.65 0.55

Min 0.44 0.36 0.48

Max 1 0.89 0.68

Range NA [0.65–0.89] [0.55–0.68]

Fig. 11. Ratio training size accuracy to test size accuracy versus k, neighbors

Max 0.49 0.75 0.60

accuracy ratio ¼ training training test

Range Min Max Range [0.44, 0.49] 1.9 2.23 [1.9–2.2] [0.36–0.75] 1.05 1.5 [1.1–1.5] [0.48–0.60]

Fig. 12. Time processing vs k, neighbors under various test sizes

4 Conclusion In this investigation, we build a machine learning model, applied to data that does not contain any label information (Unsupervised learning algorithms). Quantitative estimation classification using statistical machine learning tool based on KNN classifier is proposed to predict the Cooling Load performance of energy of residential building shapes by creating eight (08) classes. The working system is tested successfully, which is able to detect and recognize the cooling load magnitude on real dataset features consisting of 8 features and containing 768 instances. The experimental results show a high training accuracy when k = 1, while the test accuracy reaches its minimum value. In the range of k [1–10], the accuracy increases or decreases. When k becomes larger than 10, it remains approximately constant. The test size combined with the CKNN method can be used to control the accuracy rate. To conclude, the prediction of the cooling load magnitude class can be optimized in the range of k [4–6] (region I) by combining the test size less than 30%. This model works well on the training set, but does not perform badly on the test set. But, it is still good, which might still be acceptable that can learn from the measurements of eight (08) input variables whose features are known, so that we can predict the range of the cooling load (CL) for a new eight (08) input dataset of residential buildings.

446

A. Boudjella and M. Y. Boudjella

References Teknomo, K.: K-nearest neighbors tutorial (2006). http://people.revoledu.com/kardi/tutorial/ KNN/ Duda, R.O., Hart P.E.: Pattern Classification and Scene Analysis. Wiley (1973) Indyk, P., Motwani, R.: Approximate nearest neighbors: towards removing the curse of dimensionality. In: STOC, pp. 604–613 (1998) Srihari, J., Santhi, B.: Prediction of heating and cooling load to improve energy efficiency of buildings using machine learning techniques. J. Mech. Cont. Math. Sci. 13(5), 97–113 (2018) Duarte, G.R., Goliatt da Fonseca, L., Zabala Capriles Goliatt, P.V., de Castro Lemonge, A.C.: Ambiente Construído 17(3), 103–115 (2017). https://doi.org/10.1590/s1678-86212017000300165 Sholahudin, S., Alam, A.G., Baek, C.I., Han, H.: Prediction and analysis of building energy efficiency using artificial neural network and design of experiments. Appl. Mech. Mater. 819, 541–545. https://doi.org/10.4028/www.scientific.net/AMM.819.541. Revised: 2015-01-3. Submitted: 2014-09-01. ISSN: 1662-7482 Tsanas, A., Xifara, A.: Accurate quantitative estimation of energy performance of residential buildings using statistical machine learning tools. Energy Build. (2012). https://doi.org/10. 1016/j.enbuild.2012.03.003 Fayaz, M., Kim, D.: A prediction methodology of energy consumption based on deep extreme learning machine and comparative analysis in residential 222; buildings. Electronics 7(10) (2018), https://doi.org/10.3390/electronics7100222 Wang, P., Gong, G., Zhou, Y., Qin, B.: A simplified calculation method for building envelope cooling loads in Central South China. Energies 11, 1708 (2018). https://doi.org/10.3390/ en11071708www.mdpi.com/journal/energies Ben-Nakhi, A.E., Mahmoud, M.A.: Cooling load prediction for buildings using general regression neural networks. Energy Convers. Manage. 45, 2127–2141 (2004). www.elsevier. com/locate/enconman Rabbani, A.: Energy efficiency: prediction of the heat and cooling requirements of buildings. Int. J. Eng. Res. 7(12), 207–210 (2018). IJER@2018 Page 207. ISSN: 2319-6890 (online), 23475013 (print) https://archive.ics.uci.edu/ml/datasets/Energy+efficiency Boudjella, A.: Heating load energy performance of residential building: machine learning-cluster k-nearest neighbor (part I). In: 4th International Conference on Artificial Intelligence in Renewable Energetic Systems-IC-AIRES2020. Will be published

Fuzzy Traffic Lights Controller Based on PLC Mounir Bouhedda1(&), Hamza Benyezza1, Yassine Toumi1, and Samia Rebouh2 1 Laboratory of Advanced Electronic Systems (LSEA), Department of Electrical Engineering, Faculty of Technology, University of Medea, Nouveau Pôle Urbain, 26000 Medea, Algeria [email protected], [email protected], [email protected] 2 Laboratory of Experimental Biology and Pharmacology, Department of Material Sciences, Faculty of Sciences, University of Medea, Nouveau Pôle Urbain, 26000 Medea, Algeria [email protected]

Abstract. This paper aims to setup of an intelligent system based on fuzzy logic for controlling the traffic light of an intersection using a programmable logic controller (PLC). A set of data is collected by the controller to determine the best switching time of the lights by minimizing the average waiting time of the drivers. The controller, which contains two internal fuzzy controllers, is used to obtain the adequate phase of the traffic light cycle and its duration. After validation of the proposed architecture using MATLAB, the synthesized intelligent controller is mathematically described in detail and simplified in order to perform the corresponding PLC program where modular programming is adopted. An intersection model with a Human-Man Interface is realized and connected to the programmed PLC (S7-300 of Siemens). The results of tests and comparisons against the time interval algorithm systems gave good performance, where the average waiting time is reduced considerably, for over than 25 s and 11 s less for a flow of 800 and 1200 vehicles/hour respectively. Keywords: Fuzzy logic
Programmable logic controller Intelligent system
Intersection
Waiting time


Traffic lights


1 Introduction The first and most critical urban development strategy must satisfy the different requirements of traffic [1]. Moving has become an integral part of daily life nowadays: by using public transit or personal cars, handling the extensive network of these modes of transport is incredibly difficult. Serious traffic and pollution issues have arisen in the world, resulting in substantial human-level costs for the community; challenges in accessing the labor market and public services, and lack of economic time and money [2]. This work is a contribution in the traffic light control systems for improving the fluidity of urban road traffic. Many traffic control systems have been used in different countries in the world, such as SCATS [2, 3] and TRANSYT (Traffic Network Study Tool) [4] where the switching time of traffic lights depends on the traffic conditions obtained by measuring © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 447–456, 2021. https://doi.org/10.1007/978-3-030-63846-7_42

448

M. Bouhedda et al.

systems designed for this purpose [3]. Recently, advanced mathematical models are proposed to deal with the problem of road traffic like Markov decision control and game theory models [5]. Because of their success in many applications in the field of control engineering, artificial intelligent techniques are widely introduced in traffic control systems [6] such as artificial neural networks [7], genetic algorithms [8], ant colony system [9] and fuzzy logic [10]. For efficient management of the traffic at an intersection according to the changing traffic conditions, this work proposed a fuzzy approach for traffic light control in an intersection. The main objective is to minimize the drivers’ waiting time at the intersection, the system acquired continuously traffic data for evaluating traffic condition, the fuzzy system used these data for reasoning using the fuzzy controller to obtain the switching time of traffic light in each cycle, so that the traffic can be controlled efficiently. In this paper, the implementation on PLC (S7-300 of SIEMENS) of an intelligent traffic light control system based on fuzzy logic technique, which can mimic the human reasoning, for controlling the lights is done. First, a fuzzy mathematical model is developed and validated using Matlab. In the second step, the obtained fuzzy model is implemented on a PLC; S7-300 model (CPU 312). The intelligent system is evaluated against the time interval algorithm [11] by adopting multiple tests.

2 Traffic Fuzzy Controller Description 2.1

Overview

Figure 1 illustrates the intersection model considered in this study. Each path has two inputs lanes and two outgoing lanes. Only straight and turning vehicles are considered in this model. However, at this time, pedestrian crossings are not considered. In result, four phases can be obtained which are clearly indicated by Fig. 2.

D

N

Sensor

Fig. 1. Considered four paths intersection

Fuzzy Traffic Lights Controller Based on PLC Phase 1

Phase 2

Phase 3

449

Phase 4

Fig. 2. Four paths intersection phases

As a solution to this problem, an intelligent controller based on fuzzy logic (Fig. 3) is proposed for the control of green light of the phases intelligently such that the queue duration of each phase is optimized and the priority is given to the most overloaded phase and the time of the green light is variable according to the number of vehicles present at the intersection. As a result, the average waiting time of the drivers is minimized. For this purpose, the controller needs two inputs (Fig. 3):

Number of vehicles (NV) Vehicles flow(F)

Intelligent Traffic Light Controller

Phase (Ph) Phase duration (Ph_d)

Fig. 3. External structure of the intelligent traffic light controller

• The number of vehicles at the intersection (NV): Two sensors are used (Fig. 1). The number of vehicles passing through the lights is obtained using the first sensor which is placed behind each traffic light and the number of vehicles arriving at the intersection is counted using the second sensor placed behind the first sensor at a distance D. The number of vehicles (NV) in one lane is determined using data given by the two sensors (Fig. 1). • The flow of vehicles (F): It is the rate of vehicles which move in a unit of time corresponding to a phase. 2.2

Fuzzy Controller Structure

The fuzzy controller is composed of two fuzzy systems denoted FS1 and FS2 (Fig. 4). FS1 can determine the urgency degree (Ur) of a phase according to the number (NV) and the flow (F) of vehicles (vehicles/minute) of the corresponding lanes (Fig. 4). FS2 gives the green duration (Gd) of a phase according to the number (NV) and the flow (F) of vehicles of the corresponding phase. For the two fuzzy systems, each input and each output has five membership functions (two trapezoidal and three triangular) corresponding to the linguistic values very small, small, middle, big and very big denoted vs, s, m, b and vb respectively. Details about universe of discourse, linguistic variables, linguistic values and membership functions parameters are given in Fig. 5 and Fig. 6. Each of the two fuzzy systems is described by using 25 possible combinations of AND rules. Table 1 shows some of these rules.

450

M. Bouhedda et al. FUZZY TRAFFIC CONTROLLER Urgency (Ur)

Number of vehicles (NV)

FS1

Phase (Ph)

Max

Vehicles flow(F)

Phase duration (Ph_d)

FS2

Fig. 4. Internal structure of the fuzzy traffic controller

m

s

b

s

m

1

Membership degree

Membership degree

vs

vb

0,8

0,6

0,4

0,2

b

vs

vb

0,8

0,6

0,4

2

4

6

8

10

12

14

16

Number of vehicles (NV)

18

20

vb

0,8

0,6

0,4

0

0 0

b

0,2

0,2

0

m

s

1

Membership degree

vs 1

0

10

20

30

Flow of vehicles (F)

40

0

50

0,1

0,2

0,3

0,4

0,5

0,6

Urgency (Ur)

0,7

0,8

0,9

1

Fig. 5. Membership functions of FS1 m

s

b

s

m

1

Membership degree

Membership degree

vs

vb

0,8

0,6

0,4

b

vb

0,8

0,6

0,4

0 2

4

6

8

10

12

14

16

Number of vehicles (NV)

18

20

5

10

m

b

vb

0,8

0,6

0,4

0

0 0

s

0,2

0,2

0,2

vs 1

Membership degree

vs 1

0

10

20

30

Flow of vehicles (F)

40

50

0

15

20

25

30

35

40

Time (s)

Fig. 6. Membership functions of FS2

2.3

Software Implementation of the Fuzzy Controllers

Before proceeding to the hardware implementation on the PLC, it is necessary to check the proposed two fuzzy systems using Matlab but without using Fuzzy Logic Toolbox. The reason is to consider all the necessary operations and thus simplify the transition into the implementation of the fuzzy systems on the PLC. A Matlab code is written with being careful to use only functions that can be converted easily to the PLC language programming. Below are some ideas used to get the output for the fuzzy system.

Fuzzy Traffic Lights Controller Based on PLC

451

Table 1. Some of the control rules FS1 Rule R1 R2 R3 R4 R5 R6 R7 …

VehNbr s vs m b vb s b …

Flow m m m vs s vb vb …

Urgency s vs m s m b vb …

FS2 Rule R1 R2 R3 R4 R5 R6 R7 …

VehNbr vs m m b vb vb s …

Flow vs vs b b vs m b …

Time vs vs b b m vb m …

• Each of the trapezoidal and triangular membership function are described by a written function, the mathematical expressions are given by (1) where a, b, c and d are the parameters corresponding to the triangular or trapezoidal function [12]. • The membership degree corresponding to each linguistic value is calculated and stored in a vector. • The rules are organized in a matrix • The active rules are determined using the vectors corresponding to the membership degrees. • Since the AND rules are used, the minimum between the two degrees corresponding to an active rule is obtained • The active rules are compared to the matrix rule in order to get the linguistic values of the output and the maximum value in case of two rules or more result on the same linguistic value. • To obtain the numerical value of the fuzzy system, a defuzzification is done using the center of gravity method (COG). Three possible geometric shapes are obtained (Fig. 7). The abscissa of the COG of the final obtained form is the numerical value of the output. Only the abscissas of the COG of each form is needed. [12] propose a very simple method using basic calculus to obtain the numerical value of the fuzzy system output from the basic geometric shapes.

b

a

c

e

Form 1

b

d

a

f

c

e

Form 2

b

d

a

c

d

e

Form 3

Fig. 7. Possible geometric shapes obtained after the inference

452

M. Bouhedda et al.

TriðxÞ ¼

8 0 > > < xa ba

cx > > : cb 0

if if if if

8 0 > > xa > xa > < ba a\x  b TraðxÞ ¼ 1 b\x  c > dx > > > x[c : dc 0

if if if if if

xa a\x\b bxc c\x\d xd

ð1Þ

3 Hardware Implementation of the Traffic Fuzzy Controller The traffic fuzzy controller is implemented around a PLC (S7-300 of SIEMENS with CPU 312). Figure 8 shows the realization of this work on the laboratory where an intersection model is realized and connected to the PLC.

Traffic light model

S7-300 PLC

Fig. 8. Hardware realization of the fuzzy traffic controller

The PLC programming using ladder language and Structured Control Language (SCL) is done under “Totally Integrated Automation Portal” Software (TIA Portal Version 12). Modular programming is adopted by using subroutines and database blocks. • A subroutine (Fig. 9) is created for each of the membership function corresponding to (1). • A database block is used to store the membership degrees of each input (Fig. 10). • A subroutine is created to get the membership degree of each linguistic value of the outputs (Fig. 11). • A subroutine using SCL language is created in order to get the index and maximum value of the membership degrees of the output. • The deffuzification is done using SCL language to get the numerical value of the fuzzy systems. • Figure 12 illustrates the two blocks corresponding to FS1 and FS2. These blocks contain all subroutines and database blocks cited above. • The output type of the FS2 (Green duration) is real and needs to be converted to a type that can be recognized by the timer of S7-300, this is done using special ladder instructions. • A Human Man Interface (HMI) is realized for the visualization of the intersection lights on a screen (Fig. 13).

Fuzzy Traffic Lights Controller Based on PLC

Fig. 9. Function blocks of the triangular and trapezoidal membership functions

Fig. 10. Database block for the storage membership degrees

Fig. 11. Outputs’ membership degree function block

Fig. 12. FS1 and FS2 blocks

Fig. 13. HMI of the intersection traffic controller

453

454

M. Bouhedda et al.

4 Tests and Evaluation In a real intersection, the parameters (flow and number of vehicles) cannot be set to some desired values. For this reason, a database block is used on the PLC to give different values to the system. Two sets of tests are done where the given values of the intersection parameters are presented in Table 2. The obtained results of the tests are presented in Table 3. The results show that the duration of the green light and the phase given by the fuzzy controller depends on the intersection parameters favorizing the minimum waiting time of the drivers.

Table 2. Parameters of the two tests Test Parameters Lane1 Lane2 Test 1 NV 7 5 F (veh/mn) 20 18 Test2 NV 7 5 F (veh/mn) 20 18

Lane3 6 15 13 40

Lane4 5 14 14 35

Lane5 Lane6 Lane7 11 9 4 25 20 17 3 4 10 12 13 36

Lane8 3 15 12 35

Table 3. Results of the two tests Test

Parameters

State of the lights Test 1 Phase Orange Phase 3 Orange Duration (s) 3 22 3 Test2 Phase Orange Phase 2 Orange Duration (s) 3 33 3

Phase 1 Orange 13 3 Phase 4 Orange 31 3

Phase 2 Orange 10 3 Phase 1 Orange 13 3

Phase 4 6 Phase 3 6

The realized fuzzy controller is evaluated against classical time interval algorithm [3] with taking different values of the flow. The comparison results are given in Table 4 and represented by Fig. 14. The results show the advantage of using the proposed fuzzy controller system where the average waiting time is reduced considerably, over than 11 s for a flow of 1200 vehicle/hour.

Table 4. Average waiting time for multiple tests based fuzzy controller and TIA algorithm. F(Veh/h) 400 600 800 1000 1200 1600 TIA algorithm (s) 25.60 31.04 37.07 39.80 42.35 44.82 Fuzzy controller (s) 2.67 4.52 12.72 21.99 31.16 41.13

Fuzzy Traffic Lights Controller Based on PLC

455

50

Fuzzy controller TIA Algorithm

Average waiting time (s)

45 40 35 30 25 20 15 10 5 0 400

600

800

1000

1200

1400

1600

Traffic flow (veh/h)

Fig. 14. Average waiting time based on vehicle flow for fuzzy controller and TIA algorithms

5 Conclusion The proposed fuzzy traffic light controller is designed and implemented on PLC of SIEMENS (CPU 312) with success. The results obtained by comparing the proposed system to TIA algorithm show the superiority of the fuzzy controller in term of the average waiting time, where the proposed system gives always a very low time comparing to TIA waiting time. As it can be seen in Table 4, Fuzzy controller gives widely good results comparing to TIA algorithm. Acknowledgment. This work is supported by Directorate General for Scientific Research and Technological Development (DGRSDT) and partially funded by the Thematic Research Agency on Science and Technology (ATRST) of the Algerian Ministry of Higher Education and Scientific Research (MESRS).

References 1. Durning, A.T.: The Car and the City, 1st Thus edn. Northwest Environment Watch, Seattle (1996) 2. Buchanan, C.: Traffic in Towns: A Study of the Long Term Problems of Traffic in Urban Areas, 1st edn. Routledge, London, New York (2015) 3. Gala, P.R., Verma, S., Kumar, U., Ojha, H.: A survey of intelligent traffic light control systems. Int. J. Comput. Appl. 180(21), 31–36 (2018) 4. Wallace, C.E., Courage, K.G., Reaves, D.P., Schoene, G.W., Euler, G.W.: TRANSYT-7F User’s Manual, June 1984. https://trid.trb.org/view.aspx?id=213009. Accessed 20 July 2020 5. Bazzan, A.L.C.: A distributed approach for coordination of traffic signal agents. Auton. Agents Multi-Agent Syst. 10(1), 131–164 (2005). https://doi.org/10.1007/s10458-004-6975-9 6. Zhao, D., Dai, Y., Zhang, Z.: Computational intelligence in urban traffic signal control: a survey. IEEE Trans. Syst. Man Cybern. Part C Appl. Rev. 42(4), 485–494 (2012). https:// doi.org/10.1109/tsmcc.2011.2161577 7. Kareem, E.I.A., Jantan, A.: An intelligent traffic light monitor system using an adaptive associative memory. IJIPM Int. J. Inf. Process. Manag. 2(2), 23–39 (2011)

456

M. Bouhedda et al.

8. Villagra, A., Alba, E., Luque, G.: A better understanding on traffic light scheduling: new cellular GAs and new in-depth analysis of solutions. J. Comput. Sci. 41, 101085 (2020). https://doi.org/10.1016/j.jocs.2020.101085 9. Bouhedda, M., Bellatreche, S., Ahmed-Serier, R.: Smart traffic signal controller design and hardware implementation based ant colony system. In: ICMIC 2016 IEEE Proceedings, pp. 1110–1116, November 2016. https://doi.org/10.1109/icmic.2016.7804278 10. Acharya, S., Dash, K.K., Chaini, R.: Fuzzy logic: an advanced approach to traffic control. In: Applications of Robotics in Industry Using Advanced Mechanisms, Cham, pp. 176–186 (2020). https://doi.org/10.1007/978-3-030-30271-9_17 11. Papageorgiou, M., Diakaki, C., Dinopoulou, V., Kotsialos, A., Wang, Y.: Review of road traffic control strategies. Proc. IEEE 91(12), 2043–2067 (2003). https://doi.org/10.1109/ jproc.2003.819610 12. Wang, W.-J., Luoh, L.: Simple computation for the defuzzifications of center of sum and center of gravity. J. Intell. Fuzzy Syst. 9(1,2), 53–59 (2000)

Implementation of a Smart Traffic Light Controller Based on Multi-agent System Mounir Bouhedda1(&), Abderrezak Aggoun1, Samia Rebouh2, and Abderrahmane Oudjouadj1 1

Laboratory of Advanced Electronic Systems (LSEA), Department of Electrical Engineering, Faculty of Technology, University of Medea, Nouveau pôle urbain, 26000 Medea, Algeria [email protected], [email protected], [email protected] 2 Laboratory of Experimental Biology and Pharmacology, Department of Material Sciences, Faculty of Sciences, University of Medea, Nouveau pôle urbain, 26000 Medea, Algeria [email protected] Abstract. A novel, intelligent, cooperative, multi-agent approach to control traffic lights in an intersection is presented in this paper. The system is composed of agents functioning in their environment. In this context, the world involves the traffic light, road information, cars, etc. Each agent observes repeatedly its environment and uses this information to collaborate with other agents so they can minimize the waiting time of the vehicles which enter the intersection. The performance of the proposed system is tested with famous system of traffic light control. The results demonstrate the efficiency of the multi-agent system in dealing with the other traffic light control system where the waiting time is reduced to 75% and 25% for flows of 3200 and 4800 vehicle per hour respectively. Keywords: Multi-agent system Intersection
Waiting time


Traffic lights control
Intelligent system


1 Introduction In recent times, the transport industry is facing challenges in regard to time spent waiting on the roads as opposed to movement. This could be attributable to situations like unfavorable rush hour vehicles loading on the highway, or a vehicle that fails to start at a highway etc., leading to pilings of vehicles on the roads. Over the years, population growth and economic prosperity in urban centers has resulted into increased acquisition of automobiles in urban centers. Hence increase on the number of vehicles on the roads and continual expansion of urban centers. According to a report published by the United Nations Population Foundation [1], For the first time, more than half of the world’s population lives in urban areas and the balance of people continues to move to cities. Therefore, drivers and passengers spend a large time of their day in traffic. In different countries in the world, famous control systems are used which take into consideration the traffic condition to calculate the phases duration [2, 3] like SCATS [3], SCOOT [2] and TRANSYT [4]. Many researches works introduced the use of the © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 457–466, 2021. https://doi.org/10.1007/978-3-030-63846-7_43

458

M. Bouhedda et al.

artificial intelligence techniques, Markov decision control and the game theory approach in the control of the traffic lights [3, 5–9]. A multi-agent universe is of a set of computing elementary processes taking place at the same time, and therefore of several agents living at the same time and environment, communicating with each other and sharing common resources. The formalization of the necessary coordination between agents is the key point of multi-agent universe theory. The theory of agents is thus can be considered a theory of control what are the hierarchical relations between agents? How are they synchronized? - and communication - what kinds of messages are they sending? What syntax do these messages obey? - for which it proposes elaborate formalisms. The answers to these questions can set a multi-agent system. Multi-agent universes theory has many applications in the field of artificial intelligence where it reduces the problem complexity solution by dividing it into elementary tasks by associating an independent intelligent agent with each of these tasks and coordinating the activity of these agents. This is called distributed artificial intelligence. This theory applies, for example, to the monitoring of an industrial process where it implements the common-sense solution of coordinating several specialized supervisors rather than considering a single omniscient supervisor. This principle of the distribution of knowledge can be implemented in an even more systematic way in order to solve famous combinatorial problems [10]. For efficient management of the traffic at intersection according to the changing traffic conditions, this work proposes the traffic light control of an intersection using a multi-agent approach. It aims to introduce agents at the intersection to manage traffic in order to minimize the waiting time of the drivers. The agent continually observes the state of traffic by collecting data, this information is then used to reason with the traffic light control rules by the agent inference engine to determine the best switching time, so that traffic can be controlled effectively. The developed intelligent controller is implemented on an Arduino microcontroller. The phases sequence of the traffic lights depends on the flow and the arrival time of vehicles at the intersection in each direction. The system based multi-agent is evaluated against the time interval algorithm with adopting multiple tests.

2 Problem Description The main objective of traffic lights at the intersections is to manage traffic efficiently. This is to ensure minimum queue length at the roads’ intersection. This can be achieved by ensuring efficient coordination of the traffic lights at the intersections with respect to density and arrival times of vehicles. In this day traffic light control is based on interval time algorithm, these time intervals are mostly scheduled to have equal intervals during morning, midday and evening hours, with the assumption that these are peak hours. Between the peak hours, they show different intervals with the assumption that these are off-peak hours. This does not solve the problem effectively since such changes are still pretimed. There is also poor interaction between the existing traffic control systems of various intersections to the adjacent ones, thereby resulting into traffic snarl-up. The police officers are used in such circumstances, to coordinate the movement of traffic, hence usually leading to biasness from the part of the officers resulting to massive time

Implementation of a Smart Traffic Light Controller

459

wastage to motorists waiting on the roads not released. This pre-timing therefore, presents traffic snarl-up as a static problem, not a dynamic in nature of which is a false impression. 2.1

Proposed Solution

The aforementioned problem therefore presents a scenario, which could easily be solved by an agent-coordinated traffic light system. This work seeks to present a system of agents that coordinate, cooperate and share information based on the average waiting time and queue lengths on the roads to the intersections’ controllers. The agents are used to control the traffic lights and determine for how long to set the lights to achieve minimum waiting time and reduced queue lengths based on the prevailing traffic conditions. The conventional intersection model shown in Fig. 1 is composed of a set of four directions or roads, this model is used in most of the literature, but nothing prevents it from being extended. Each direction can be decomposed into one direction ingoing and outgoing direction, or both. An incoming direction allows the vehicles, distributed on one or more paths, to cross the intersection to an outgoing direction, which makes the link with a neighboring intersection. On each incoming direction, the leftmost lane is used by the vehicles turning to the left, While the vehicles going straight or turning right use the rightmost lane. It is assumed that the drivers drive on the right side of the road. The outgoing tracks can make neighboring with other tracks, in another meaning two cars can get out from the intersection safely and without crossing each other. Table 1 shows the cases in which tracks have the green light to move at the same time (showed by gray color).

Fig. 1. Adopted intersection model

460

M. Bouhedda et al. Table 1. Green light table. 1B 2B 3B 4B 1A 2A 3A 4A

2.2

1B

2B

3B

4B

1A

2A

3A

4A

Terminology

Several terms must be defined in order to describe adequately the situation and the problem. This section explains, among other things, the terms common to each traffic signal management model. • The intersection is an area where several streams share a conflict space represented by the gray rectangle in the Fig. 1. • The number of vehicles waiting in a lane is called “tail”. • A signal cycle is a complete set of green plus red light for each path its length is called cycle time. • The “all-red” time is the length of time when all the tracks have a red light. It is considered for security reasons. • The distance between vehicle is two seconds, it is called advancement and denoted dm . • The waiting time of a vehicle is the time of departure minus the time of arrival. • Once a vehicle has arrived at an intersection, the time it takes for the next arriving vehicle is called arrival interval. 2.3

Arrival of Vehicles

The vehicles arrive from the different tracks at the intersection in a random and uncorrelated manner, making the Poisson process a good arrival model. In a Poisson process, the arrival intervals follow an exponential distribution relation (1) [11]. f ðxÞ ¼ kekx uðxÞ

ð1Þ

Where k is the arrival rate of vehicles per hour per track, uðxÞ is the unit function given by relation (2). This function is used to avoid using negative time intervals.  uðxÞ ¼

1 if 0 if

x0 x\0

ð2Þ

The Poisson distribution given by relation (1) allows for calculating instantaneous time intervals. Because of the geometric and physical considerations of the traffic network

Implementation of a Smart Traffic Light Controller

461

road, a minimum distance must be considered between vehicles. To avoid this problem, a shifted exponential distribution is used for the arrival intervals. Where relation (3) can give the probability density function. fx ðxÞ ¼

k k e1kdm ðxdm Þ uðx  dm Þ 1  kdm

ð3Þ

Note that for the function given by the relation (3), to have a meaning it is necessary that 1  kdm [ 0. Vehicle arrival times can be generated by relationship (4). 

ta2

 1  dm lnðrÞ ¼ ta1 þ dm  k

ð4Þ

Where: ta2 : Next arrival time. ta1 : Previous arrival time. r: Random value uniformly distributed over [0 1]. The number of scheduled vehicles arriving during an interval of time is required to calculate the arrival time of a vehicle during a signal cycle. This is difficult to do with an exponential distribution shifted from the relation (3). The reason is that there is a loss of the property. The duration of a signal is longer, so the expected number of vehicles can be approximated by Poisson distribution function. The probability of arrival of n vehicles during Dt seconds is given by the Eq. (5). In a Poisson distribution, the predictable number of new incoming vehicles during Dt seconds is kDt. pðnÞ ¼

2.4

kDtekDt n!

ð5Þ

Multi-agent System

A multi-agent system (MAS) is composed by multiple intelligent agents that can interact within an environment around a computerized system. Working and interacting together, multi-agent systems have the ability to solve complex problems that for an individual is impossible or difficult to solve. The objective of an agent-based model (ABM) is to explore into the collective work of the agents following elementary rules [12]. An extensive discussion and investigation have been focused on using agent technology within the scientific community [13]. Actually, multi-agent systems are classified as a subfield of artificial intelligence and have been the subject of hardware and software implementation for control application in different fields [12]. The main features of agents can be cited namely: • Autonomy: the agent executes its own actions independently. • Reactivity: the agent interacts quickly with other agents in its environment to the requests from other agents.

462

M. Bouhedda et al.

• Intentionality: the agent is skilled for immediate and long-term intentions. • Social capability: the agent is able to form a team by interacting, collaborating and performing different levels of perception about the other agents. • Reactiveness: Agents can also manifest by taking the initiative of a goal-directed behavior and do not act in response to their environment only. Improving traffic flow is very important, as the urban traffic has shown that how congestion is managed and can either slow or accelerate the economic output of a population [14]. However, we can conclude the difficulty of improving traffic control when it is modelled with some degree of realism. It is a complex, nonlinear system, in which substandard control actions can easily lead to congestions that extend quickly and that are difficult to liberate. This therefore called for the intelligent approaches towards management of traffic in major areas, since traffic control is basically a sequential decision-making problem, and at the same time, is a complex task that is challenging to the uncomplicated approaches such as the pre-timed interval traffic lights.

3 Hardware Implementation Figure 2 gives a global description of the hardware implementation, two main information are given to the system: the arrival time of each vehicle at each path and the vehicles flow, this system can calculate the minimum time period to switch to green light and release vehicles from the path with minimizing the average waiting time of the vehicles.

Vehicles flow Arrival time of vehicles

Intelligent multi-agent system for traffic light control

Optimal switching time

Fig. 2. Description of the hardware implementation

3.1

Multi-agent System for Traffic Control

The diagram of Fig. 3 illustrates the communication lines between the agents of the system, agent main_path and agent second_path generate vehicles arriving time and send it to agent_crossroad, therefore that agent makes a priority of the paths to get a main path and second path agents and they can release vehicles in parallel and without crossing each other. The obtained results go to agent_manipulation to schedule the green lights period of times. C++ modular programming language is used for the hardware implementation where the MAS uses four classes as follows,

Implementation of a Smart Traffic Light Controller

463

Fig. 3. Agents diagram

• Vhl_arrival: this class function (Funct_Matrix) generates a matrix of vehicles arriving time according to the path traffic flow, the number vehicles can also be calculated using the matrix. • Path: each path has identifier (id) and vehicles number, this class gives second paths id (3 of the paths can release vehicles in parallel with main path without crossing each other), and receives vehicles from Vhl_arrival class and releases them according to the results of Crossroad class. • Crossroad: this class receives vehicles arriving time to select the priority of the main path and the second path which can be chosen from 3 paths giving by Path class and send this results to Manipulation class. • Manipulation: when it receives the results from Crossroad class, this class do the scheduling of the vehicles and the results is a period of green light for each path so they can release vehicles accordingly. 3.2

System Scheduling

The number of vehicles arriving in each period of time [t1 t2] is called DT (data time). This period can be changed according to the path flow. From those periods, the load of vehicles in the path can be calculated, so the system can choose a main path, it is the path that have a high load of vehicles, after that the system chooses the second path from 3 paths that can release vehicles in parallel with the main path, the chosen path has also the high load of vehicles. With this process, the system schedules a table of paths priorities classifying from the high to the low load of vehicles in that period DT and generates the green light time, which is the number of vehicles in the path multiply by 2 s (the distance between two successive vehicles) and the system gives to that time a maximum and changeable value of vehicles called MV (maximum vehicle) to be released until the end of the period DT. The same process happens to the next period DT until the system releases all vehicles in question.

464

M. Bouhedda et al.

4 Test and Evaluation To test the efficiency of the multi-agent system in its performance, a comparison has to be made between the average waiting time of the multi-agent system and time interval algorithm [15]. This latter can be described briefly as follows: In the Fig. 1, Assuming that the green light is initially turned on in the two opposite directions (1B) and (3B), There is a motion sensor that sends information to the system to trigger the green light in the other paths, after a five seconds without passing of vehicles. When the sensor detects a movement of a vehicle, it sends information to the system to increase the time for 5 s to each vehicle arrives until reaching the maximum time value of 30 s, at this moment the green light automatically triggers to another two lanes. Many tests have been done to observe the waiting time of the vehicles and to choose perfect conditions to get a minimal waiting time, and those condition are the data time periods DT and the maximum value of vehicles MV. In this test, the data time period DT is changeable for a maximum value of vehicles MV and compare the system (MAS) with the time interval algorithm (TIA) to see the difference of the waiting time. The tests are done with taking four values of DT = 30, 60, 90 and 120 s for each MV = 10, 15, 20, 25 and 30 vehicles. The observation from these tests shows that the system gives a remarkable optimized result for MV = 15 vehicles and for DT = 30 s. For the traffic flow from 1600 vehicles/h to 3200 vehicles/h. For MV = 10, 20 or 30 vehicles and DT = 120 s, the results are optimized for the flow from 3600 vehicles/h to 4800 vehicles/h. The realised MAS is evaluated against TIA algorithm [15] with taking different values of the flow for DT = 30 and 120 s. The comparison results are given in Table 2 and represented by Fig. 4.

Average waiting time (s)

80

MAS Algorithm TIA Algorithm 60

40

20

0 1500

2000

2500

3000

3500

4000

4500

5000

Traffic flow (veh/h)

Fig. 4. Average waiting time based on vehicle flow for MAS and TIA algorithms.

Implementation of a Smart Traffic Light Controller

465

Table 2. Average waiting time for multiple tests based MAS and TIA algorithms. DT = 30 s Veh/h 1600 2000 2400 2800 3200 MAS 1.110 2.130 2.867 4.473 12.769 TIA 31.401 33.211 39.950 40.828 47.007

DT = 120 s 3600 4000 4400 4800 24.226 48.976 54.254 60.526 53.933 67.044 76.713 79.164

The results in Fig. 4 show the advantage of using multi-agent systems in the control of traffic signal where a considerable reduction of the average waiting time can be observed, over than 19 s for a flow of 4800 vehicle/h.

5 Conclusion A new multi-agent system approach is proposed for the control traffic light signal. The system is successfully implemented on an Arduino microcontroller. The designed system is evaluated using Matlab to perform the generating of the vehicles moving and arrivals where Matlab/Arduino communication is necessary. The intelligent system is compared to another traffic light controller based on TIA algorithm and demonstrate the performance of MAS approach comparing to TIA algorithm. The proposed controller gives always a very low average waiting time comparing to TIA waiting time where it is reduced to 75% and 25% for flows of 3200 and 4800 vehicle per hour respectively. Acknowledgment. This work is supported by Directorate General for Scientific Research and Technological Development (DGRSDT) and partially funded by the Thematic Research Agency on Science and Technology (ATRST) of the Algerian Ministry of Higher Education and Scientific Research (MESRS).

References 1. Lahariya, C.: The state of the world population 2007: unleashing the potential of urban growth. Indian Pediatr. 45(6), 481–482 (2008) 2. Bretherton, D.: SCOOT urban traffic control system—philosophy and evaluation. In: Perrin, J.-P. (ed.) Control, Computers, Communications in Transportation, pp. 237–239. Pergamon, Oxford (1990) 3. Zhao, D., Dai, Y., Zhang, Z.: Computational intelligence in urban traffic signal control: a survey. IEEE Trans. Syst. Man Cybern. Part C Appl. Rev. 42(4), 485–494 (2012). https:// doi.org/10.1109/tsmcc.2011.2161577 4. Li, M.-T., Gan, A.C.: Signal timing optimization for oversaturated networks using TRANSYT7F. Transp. Res. Rec. 1683(1), 118–126 (1999). https://doi.org/10.3141/1683-15 5. Bazzan, A.L.C.: A distributed approach for coordination of traffic signal agents. Auton. Agents Multi Agent Syst. 10(1), 131–164 (2005). https://doi.org/10.1007/s10458-004-6975-9

466

M. Bouhedda et al.

6. Bouhedda, M., Bellatreche, S., Ahmed-Serier, R.: Smart traffic signal controller design and hardware implementation based ant colony system. In: ICMIC 2016 IEEE Proceedings, pp. 1110–1116, November 2016. https://doi.org/10.1109/icmic.2016.7804278 7. Youcef-Toumi, K., Bouhedda, M., Tchokecth-Kebir, S.: Smart cooperative control of an intersection group based on fuzzy logic. In: 2018 International Conference on Applied Smart Systems (ICASS), pp. 1–4, November 2018. https://doi.org/10.1109/icass.2018.8652045 8. Kareem, E.I.A., Jantan, A.: An intelligent traffic light monitor system using an adaptive associative memory. IJIPM Int. J. Inf. Process. Manag. 2(2), 23–39 (2011) 9. Rahman, S.M., Ratrout, N.T.: Review of the Fuzzy logic based approach in traffic signal control: prospects in Saudi Arabia. J. Transp. Syst. Eng. Inf. Technol. 9(5), 58–70 (2009). https://doi.org/10.1016/S1570-6672(08)60080-X 10. Joo, H., Ahmed, S.H., Lim, Y.: Traffic signal control for smart cities using reinforcement learning. Comput. Commun. 154, 324–330 (2020). https://doi.org/10.1016/j.comcom.2020. 03.005 11. Ross, S.M.: Introduction to Probability Models, 10th edn. Academic Press, Amsterdam, Boston (2009) 12. Wang, Y., Garcia, E., Casbeer, D., Zhang, F. (eds.): Cooperative Control of Multi-agent Systems: Theory and Applications, 1st edn. Wiley, Hoboken (2017) 13. Pechoucek, M., Thompson, S.G., Voos, H.: Defense Industry Applications of Autonomous Agents and Multi-agent Systems. Springer (2008) 14. Bakker, B., Whiteson, S., Kester, L., Groen, F.C.A.: Traffic light control by multiagent reinforcement learning systems. In: Interactive Collaborative Information Systems, pp. 475– 510. Springer, Heidelberg (2010) 15. Papageorgiou, M., Diakaki, C., Dinopoulou, V., Kotsialos, A., Wang, Y.: Review of road traffic control strategies. Proc. IEEE 91(12), 2043–2067 (2003). https://doi.org/10.1109/ JPROC.2003.819610

Intelligent Solar Shunt Active Power Filter Based on Direct Power Control Strategy Ghania Boudechiche1, Mustapha Sarra1, Oualid Aissa2, and Abderezak Lashab3(&) 1

2

Electronics Department, University of Mohamed El-Bachir El-Ibrahimi, Bordj Bou Arreridj, Algeria [email protected] Electromechanical Department, University of Mohamed El-Bachir El-Ibrahimi, Bordj Bou Arreridj, Algeria [email protected] 3 Center for Research on Microgrids (CROM), Department of Energy Technology, Aalborg University, Aalborg, Denmark [email protected]

Abstract. This paper deals with a double-stage grid-connected photovoltaic (PV) system, operating as a shunt active power filter in order to improve the power quality and to supply the extracted PV power to the utility, simultaneously. On the grid side, a direct power control (DPC) is developed to supply the harvested PV energy into the electrical network based on the provided reference, which is estimated for unwanted harmonics and reactive power eliminations. On the PV side, an intelligent method of tracking the maximum power point based on fuzzy logic has been adopted to eventually solve the problem of the rapidly changing weather conditions. The overall control scheme is examined by simulation using MATLAB/Simulink software. The obtained simulation results demonstrate the feasibility, performance, and robustness of these control strategies under different test conditions. Keywords: Direct power control
Parallel active power filter
IP controller Solar photovoltaic system
Fuzzy logic MPPT controller




1 Introduction So far, most of the world’s energy is produced from fossil fuels (coal, oil, and gas). The consumption of these energy sources contributes to harmful gas emissions which are heavily involved in the global warming, as well as inducing pollution of the earth and organisms [1]. The excessive consumption of non-renewable natural resources systematically leads to the reduction of reserves of this kind of energy potential whose repercussions will be harmful to future generations. Moreover, energy production is still a challenge of great importance for the coming years since it is employed almost everywhere, i.e., in residential, commercial, and industrial areas. On the other hand, the regeneration of renewable energies such as hydroelectric, geothermal, biomass, wind, and PV operates without pollution effects on the atmosphere after using one of these © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 467–477, 2021. https://doi.org/10.1007/978-3-030-63846-7_44

468

G. Boudechiche et al.

energies [2]. The availability of solar energy as an environment-friendly, unlimited, and free energy on the entire globe surface [1] has prompted researchers to select it among other existing sources of renewable energy for study and investigation. Meanwhile, the rapid growth of nonlinear loads integration causes serious problems in the electrical power systems, such as the harmonic currents and reactive power [3–6]. Active power filters (APF) are the most popular solution for mitigating these unwanted issues in the electrical grids [7]. Within the APF family, the shunt active power filter (SAPF), which is paralleled to the grid to inject a current that is opposing both the current harmonics and reactive power emitted by the load, is commonly employed. After the APF is integrated, it eventually makes the current supplied by the electrical power system sinusoidal and in phase with its voltage [8]. In the literature, many control strategies have been presented to control the APF. One among the popular ones is Direct Power Control (DPC), which is does not require current control loops or pulse-widthmodulator (PWM) blocks. The switching table based on the correction of the reactive and active powers, as well as based on the sector indicating the angular position of the source voltage vector, is intended to select the switching states of the converter [7–9]. In most cases, the DPC is fed by a reference of zero reactive power and active one produced via the Integral-Proportional (IP) regulator of the converter dc-link voltage [7]. As the solar insolation varies, several algorithms of maximum power point tracking (MPPT) [1, 2, 10–12] such as perturb and observe (P&O), incremental conductance (IC) and hill climbing (HC) have been proposed [1, 2, 10]. Due to the deficiencies of the aforementioned algorithms especially during dynamically changing weather conditions, intelligent controllers like fuzzy logic has been used in tracking the maximum power point in PV systems [11, 12]. In this paper, a new control scheme composed of two layers, where the first layer consists of controlling the SAPF through DPC approach, while the second layer consists of tracking the MPP of the PV array using fuzzy logic, is proposed. The introduced modification offers greatly improved performance of the overall system as will be shown later. This paper includes the following sections: Description of the operating principle of the SAPF is presented in Sect. 2; The principle of the DPC controller applied to the SAPF, integral-proportional to regulates the DC- link voltage, and fuzzy logic controller (FLC) for tracking the maximum power point, are presented in Sect. 3. Simulation results are given and discussed in Sect. 4. Finally, the presented work is concluded in Sect. 5.

2 General Description of the SAPF APFs are, in simple words, systems that are used to eliminate the harmonics pollution along the power line caused by the non linear loads, as well as the reactive power induced by the loads regardless of their nature [3]. The voltage source connected in parallel with the non-linear load, becomes almost sinusoidal since the SAPF injects harmonic currents with the same amplitude and opposite in phase with the load’s one. Regarding the reactive power, it is compensated by injecting filtering current with a

Intelligent Solar Shunt Active Power Filter

469

phase that is opposed to the line’s one [4–8]. In this way, the source’s current would have the following form [7–13]: Is ¼ II þ If

ð1Þ

where Is is the current of source; Il is the current of the load; and If is the compensation current.

3 Description of the Proposed Control System The basic principle of DPC was inspired from the direct torque control (DTC) of electrical machines [7, 9]. In the DPC approach, reactive and active powers imitate, respectively, the electromagnetic torque and the amplitude of the stator flux of the DTC. This non-linear method is known as a direct power control technique because it chooses the appropriate voltage vector without need for any modulation technique or coordinates transformation. The basic concept of the DPC is to select the appropriate switching state from the switching table based on localisation of the source voltage vector and errors [3]. These errors are limited by a hysteresis band corresponding to the reactive and active powers as depicted in Fig. 1. The instantaneous active and reactive powers are calculated starting from the following equations: Ps ¼ Vsa Isa þ Vsb Isb þ Vsc Isc

ð2Þ

1 Qs ¼ pffiffiffi ½ðVsb  Vsc ÞIsa þ ðVsc  Vsa ÞIsb þ ðVsa  Vsb ÞIsc  3

ð3Þ

Ss ¼ Ps þ jQs

ð4Þ

The reference of the reactive power is maintained at zero to ensure a unity power factor, whereas the reference of the active power is developed by multiplying the peak value of the current source generated by the IP regulator and the optimal value of the PV generator voltage. This method is based on two hysteresis regulators using as input the error signal between the references values and the calculated reactive and active powers. Then, the powers are compared with their respective references and the obtained errors are applied to the hysteresis regulators [7, 8, 13]. The used hysteresis regulators permit maintaining the errors of the instantaneous reactive and active powers in predefined bands. The output of the controller switches between 0 and 1. If the error is positive, the controller output is equal to 1, otherwise it is 0. The switching table has as inputs the signals from the two outputs of the hysteresis comparators and an information on the localization of the source voltage vector.

470

G. Boudechiche et al.

The angle between the a axis and the inverter output voltage reference is determined by an inverse trigonometric function based on the vector components of the voltage in the fixed reference frame (a, b) and the output one as follows: hp ¼ tan1



Vsb Vsa

 : p ¼ 1; 2;. . .; 12:

ð5Þ

Fig. 1. General structure of the SAPF controlled by the proposed DPC approach, considering the inclusion of a PV system. Table 1. Switching table for the DPC strategy. Cp Cq h1 h2 h3 h4 h5 h6 h7 h8 h9 h10 h11 h12 1 1 0 0

1 0 1 0

v6 v7 v6 v1

v7 v7 v1 v2

v1 v0 v1 v2

v0 v0 v2 v3

v2 v7 v2 v3

v7 v7 v3 v4

v3 v0 v3 v4

v0 v0 v4 v5

v4 v7 v4 v5

v7 v7 v5 v6

v5 v0 v5 v6

v0 v0 v6 v1

The switching table indicated in Table 1 is a paramount part in the direct power control [7–13]. It selects the appropriate voltage vector of the inverter in order to set the instantaneous reactive and active powers in their desired values.

Intelligent Solar Shunt Active Power Filter

3.1

471

IP Controller for the DC Bus Voltage Regulation

In order to control the dc-link voltage to the desired reference, an integral-proportional (IP) controller is adopted, as shown in Fig. 2. It can be noted from the same figure that, the adopted IP control embraces anti-windup capability. The dc-bus voltage is sensed and then compared to its reference vdcref. The resulted error of comparing From Fig. 2, the DC voltage closed loop transfer function can be expressed as follows: GVdcðipÞ ¼

Vdcref

+

-

Ki

k
k =k Vdc p i ¼ 2 Vdcref s þ kp k
s þ kp
ki =k

+-

1 S

+-

Vdc

ð6Þ

Ismax

Kp

+ K

Fig. 2. Anti-windup IP controller system.

where: pffiffiffi 2
Cdc
Vdcref k¼ 3
Vs

ð7Þ

From (2) it can be seen that the relation between Vdc and Vdcref is a second order transfer function (TF) in the form of: k :k =k Vdc w2n p i ¼ 2 ¼ 2 Vdcref s þ kp k:s þ kp :ki =k s þ 2:n:wn :s þ w2n

ð8Þ

Where wn is the natural frequency and n is a damping coefficient. The transfer function contains two poles and does not possess a zero. This proves that the IP controller ensures a fast response and good stability during transient states compared to the PI controller [7]. By comparing (6) to (8) considering equal poles, the IP parameters can be chosen as: ki = wn/2n and kp = 2nkwn. 3.2

Fuzzy Logic MPPT Controller

The fuzzy logic method is employed for tracking the MPP of the PV array in order to achieve good efficiency under any weather conditions. The fuzzy logic approach is very efficient for both linear and nonlinear systems and without the need of a mathematical model. The FLC has three functional blocks: fuzzification, rules inference, and defuzzification [11, 12, 14]. Fuzzification step is the process of changing the digital input variables into linguistic equivalent, which are achieved by using membership

472

G. Boudechiche et al.

functions. Rules inference step determines the output of the fuzzy logic controller by Mamdani method with a max-min technique according to the set belonging to the rule base. Defuzzification step converts the linguistic variables into a crisp value which computes the incremental duty cycle. The inputs of FLC-based MPPT are usually an error E and a change in that error DE: EðKÞ ¼

PðkÞ  Pðk  1Þ VðkÞ  Vðk  1Þ

ð9Þ

here p(k), p(k − 1), v (k), and v (k − 1) are respectively the PV power and voltage at the sampling times k and (k − 1). The proposed algorithm has two input variables, DP(k) and DV(k), while the output variable is the duty cycle DD(k). The input variables are assessed in the following way [14, 15]: DP ¼ PðkÞ  Pðk  1Þ

ð11Þ

DV ¼ VðkÞ  Vðk  1Þ

ð12Þ

where at the MPP of the PV array, DP(k)/DV(k) is null. The input variables DP(k) and DV(k) are divided into five fuzzy sets which are denoted as: Negative Big (NB), Negative Small (NS), Zero (Z), Positive Small (PS) and Positive Big (PB). The rule base connects the fuzzy inputs to the fuzzy output by the master rule of syntax: “if X and Y, then Z.” [11, 15]. For example: if DP is PB and DV is NB then: DD is NS, as given in Table 2. For ease of calculation, equilateral triangle membership functions are chosen. The center of gravity method for defuzzification step is used to calculate the incremental duty cycle DD as: Pn DD ¼

wj DDj j¼0 wj

j¼0 P n

ð13Þ

Finally, the duty cycle is achieved by adding this change to the previous value of the control duty cycle as: DðK þ 1Þ ¼ DðK Þ þ DDðK Þ

Table 2. Decision table DP\DV NB NS Z PS PB

NB PS Z Z Z NS

NS PB PS Z NS NB

Z PB PS Z NS NB

PS PB PS Z Z Z

PB NS Z Z Z PS

ð14Þ

Intelligent Solar Shunt Active Power Filter

473

4 Simulation Result and Discussion The whole system depicted in Fig. 1 has been simulated using the MATLAB/Simulink software in order to test the proposed control techniques performance. The implementation parameters used for these tests are shown in Table 3. Table 3. Simulation parameters Parameters Vs, Fs Ls, Rs

Ll, Rl Lf, Rf,, Cdc 70 V, 0.1 0.566 2.5 mH, 50 Hz mH, mH, 0.1 0.01 Ω 0.01Ω, 2200 Ω µF

L, R

Cpv, Lpv

Values

10 20 mH, µF, 40 Ω 3 mH

Vdcref

Fswitching (DC/DC converter) 226 V 5 kHz

Fswitching (DC/AC converter) 20 kHz

Figure 3 shows the PV array current and power under varying solar irradiation profile obtained by the proposed control system. Firstly, the system started with null solar irradiance until 0.4 s. Then, from 0.4 s to 2 s they follow their trajectories imposed by the applied irradiation profile. Consequently, the irradiance increases from 0 to 600 W/m2 until 0.8 s passes, providing 3 kW with 25 A by applying the FLC based MPPT algorithm. At 0.8 s, the solar irradiance decreases from 600 to 400 W/m2 tailed by a power decrease to 1.99 kW with decreasing current to 15A. At 0.9 s, the solar irradiance increases gradually until it reaches 1000 W/m2 at 1 s, and continues at this level until the end of the profile by generating 5 kW with 40 A. Figure 4 shows the powers and DC bus voltage variations, obtained by the DPC equipped with IP regulator and fuzzy MPPT controller. When the irradiation (G) is null, the electrical network supplies all the power (Ps) to the load (Pl). After the PV array starts in the time interval [0.4, 2] s, it simultaneously supplies the demanded power (Pl) by the non-linear load and the rest of the energy is transferred to the network “Pl + Ps”. During the time interval [0.1, 2] s, while the (SAPF) is inserted, the reactive power of the network (Qs) becomes null since the reactive power demanded by the load is ensured by the SAPF. While, before filtering it was the grid who provides the reactive power to the non-linear load. On the other hand, the DC bus voltage stabilizes to its reference value (Vdcref) during the insertion of the SAPF, and it returns to Vdcref at each irradiation variation due to the change in exchange of power between the grid, the non-linear load, and the APF, as shown in Fig. 4.

474

G. Boudechiche et al.

Fig. 3. Irradiation profile, current, and power of the PV array.

Fig. 4. Powers and DC bus voltage of the SAPF based on DPC strategy associated with fuzzy logic MPPT controller.

Intelligent Solar Shunt Active Power Filter

475

Fig. 5. Simulation results of the SAPF based on DPC strategy associated with fuzzy logic MPPT controller: source voltage and currents, filter and load currents.

Fig. 6. Zoomed-in view on the simulation results of the SAPF based on DPC strategy associated with fuzzy logic MPPT controller: source voltages and currents, filter and load currents.

476

G. Boudechiche et al.

(a)

(b)

(c)

Fig. 7. Source voltage and current with FFT of the latter of the SAPF based on DPC strategy associated with fuzzy logic MPPT controller: (a) without SAPF, (b) with SAPF and (c) with solar SAPF.

Figures 5, 6 and 7 show the waveforms of the voltage (Vs) and current of the source (Is) along with their FFT Analysis, the current of the filter (If), and the current of the load (Il), before and after filtering, and without and with PV array. Before filtering and when G = 0 W/m2 between 0 s and 0.1 s, the form of the source current is distorted and rich in harmonics, which are generated by the nonlinear load. The value of current harmonics distortion was measured as 28.89%. However, the source current becomes sinusoidal and in phase with the network voltage after the insertion of the SAPF at the instant 0.1 s, where the total harmonic distortion (THD) decreased to 2.42%. Then from 0.4 to 2 s, the solar irradiance level increases, where the SPAF is supposed to injected power to the load and grid. The source current remains sinusoidal despite the change in the irradiation and in opposed phase with the corresponding voltages. Consequently, THD is 0.72%, meaning compliance with IEEE-519 standard [8].

5 Conclusion This paper presented a multifunctional grid connected PV system, which includes a SAPF that is controlled by DPC for power quality improvement. On the grid side, a DPC strategy is applied to ensure both supplying a part of the load demand through the extracted PV power, and compensating both of the harmonics and reactive power caused by the nonlinear load on the grid current. Whereas on the PV side, an intelligent method of MPPT using FLC strategy has been presented to mitigate from the power loss during fast changing weather conditions. The results of simulation obtained by the MATLAB/ Simulink software confirmed the performance of the proposed system: harmonic and reactive power compensation while feeding PV power into the utility grid.

Intelligent Solar Shunt Active Power Filter

477

References 1. Piegari, L., Rizzo, R.: Adaptive perturb and observe algorithm for photovoltaic maximum power point tracking. IET Renew. Power Gener. 4(4), 317–328 (2010) 2. Loukriz, A., Haddadi, M., Messalti, S.: Simulation and experimental design of a new advanced variable step size Incremental Conductance MPPT algorithm for PV systems. ISA Trans. 62, 30–38 (2016) 3. Aissa, O., Moulahoum, S., Colak, I., Babes, B., Kabache, N.: Analysis and experimental evaluation of shunt active power filter for power quality improvement based on predictive direct power control. Environ. Sci. Pollut. Res. 25(25), 24548–24560 (2017) 4. Terriche, Y., et al.: Adaptive CDSC-based open-loop synchronization technique for dynamic response enhancement of active power filters. IEEE Access 7, 96743–96752 (2019) 5. Arunsankar, G., Srinath, S.: Optimal controller for mitigation of harmonics in hybrid shunt active power filter connected distribution system: an EGOANN technique. J. Renew. Sustain. Energy 11(2), 025507-1–025507-16 (2019) 6. Sarra, M., Gaubert, J.P., Chaoui, A., Krim, F.: Experimental validation of two control techniques applied to a three phase shunt active power filter for power quality improvement. Int. Rev. Electr. Eng. 6, 2825–2836 (2011) 7. Chaoui, A., Gaubert, J.P., Krim, F.: Power quality improvement using DPC controlled threephase shunt active filter. Electr. Power Syst. Res. 80, 657–666 (2010) 8. Krama, A., Zellouma, L., Rabhi, B.: Anti-windup proportional integral strategy for shunt active power filter interfaced by photovoltaic system using technique of direct power control. Revue Roumaine des Sciences Techniques Series Electrotechnique et Energetique 62(3), 252–257 (2017) 9. Noguchi, T., Tomiki, H., Kondo, S., Takahashi, I.: Direct power control of PWM converter without power-source voltage sensors. IEEE Trans. Ind. Appl. 34(3), 473–479 (1998) 10. Kjær, S.B.: Evaluation of the “Hill Climbing” and the “Incremental Conductance” maximum power point trackers for photovoltaic power systems. IEEE Trans. Energy Convers. 27(4), 922–929 (2012) 11. Benlahbib, B., Bouarroudj, N., Mekhilef, S., Abdelkrim, T., Bouchafaa, F.: A fuzzy logic controller based on maximum power point tracking algorithm for partially shaded pv arrayexperimental validation. Elektronikair Elektrotechnika 24 (2018) 12. Zadeh, L.A.: Fuzzy sets. In: Conference on Information and Control (1965) 13. Mesbahi, N., Ouari, A., Abdeslam, D.O., Djamah, T., Omeiri, A.: Direct power control of shunt active filter using high selectivity filter (HSF) under distorted or unbalanced conditions. Electr. Power Syst. Res. 108, 113–123 (2014) 14. Lashab, A.B., Snani, H.: Comparative study of three MPPT algorithms for a photovoltaic system control. In: World Congress on Information Technology and Computer Applications (WCITCA), Hammamet, pp. 1–5 (2015) 15. Boukezata, B., Chaoui, A., Gaubert, J.P., Hachemi, M.: An improved fuzzy logic control MPPT based P&O method to solve fast irradiation change problem. J. Renew. Sustain. Energy 8(4), 043505-1–043505-14 (2016)

MPPT - Based Improved Salp Swarm Algorithm for Improving Performance and Efficiency of Photovoltaic System Under Partial Shading Condition H. Azli1(&), S. Titri2, and C. Larbes1 1

2

Laboratoire Des Dispositifs de Communication et de Conversion Photovoltaïque LDCCP, Département D’Electronique, Ecole Nationale Polytechnique, El-Harrach, Algérie {hadjer.azli,cherif.larbes}@g.enp.edu.dz Division Micro-électroniques et Nanotechnologie, Centre de Développement Des Technologies Avancées CDTA, Baba Hassen, Algiers, Algérie [email protected]

Abstract. The aim of this paper is to evaluate the performance of the photovoltaic system under partial shading condition, by implementing an improved salp swarm algorithm (ISSA) on MPPT controller. The problem that resides with solar panels is the non-linear characteristics of their P-V curve. In the uniform condition the characteristics has one global maximum, and finding this latter can be done with just a conventional methods like P&O. Whereas for partial shading condition, the characteristics becomes a multi-modal curve, this exhibit an intelligent algorithms to track the global peak and escape the locals. In this paper we evaluated the performance of ISSA by comparing it with the original SSA and the simple PSO algorithm under uniform and partial shading condition and under a rapid variation of irradiance. The system is modeled in MATLAB/SIMULINK by using four solar panels (MSX60) connected in series. The results of 50 tests confirms that the improved algorithm performs better in response time, quality and efficiency over PSO and SSA. Keywords: Maximum Power Point Tracking (MPPT)
Bio-inspired algorithms
Partial shading
Photovoltaic system
Salp Swarm Algorithm Particle swarm optimization




1 Introduction The world is facing energy crisis problem, caused by the increasing consumption of fossil fuels over the last two decades. This problem drove the human been to make a transition into renewable energies to fulfill their needs of power. Photovoltaic (PV) cells convert solar energy to electricity, and it can be used as a connected grid system or a standalone application. PV arrays which is a combination of a connected panels in series, parallel, or both has a non-linear P-V characteristics that may change with environment condition as: temperature, irradiance. The distribution of sunlight on © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 478–486, 2021. https://doi.org/10.1007/978-3-030-63846-7_45

MPPT - Based Improved Salp Swarm Algorithm

479

PV arrays may not be uniform for all the times, a moving clouds, birds or an accumulated dust on the surface of the panels can produce non-uniform distribution or Partial shading condition (PSC). This shading prevents the solar energy to across the PN junction making a multimodal characteristics with multiple peaks of power. The PV system is generally connected with a DC-DC converter, this latter works as a regulator or in other terms it fixes the mismatch of impedance between the PV and the load. The converter is accompanied with MPPT charge controller for tracking and extracting the maximum power available from the PV. In the literature, there is a high research areas on the methods used for MPPT, enormous techniques that differs from each other’s in complexity, robustness and cost are studied and implemented to improve the performance of PV systems. Conventional techniques like P&O, IC, are simple, easy to implement, and they have a fast convergence, but their performance is limited to uniform radiation and they suffer from the steady state oscillations. Another pack of algorithms in the research area that has a better performance over Conventional techniques are the intelligent algorithms. Artificial intelligence (AI) techniques are many and they are classified and categorized differently from author to another. In (Mirjalili et al. 2019) the author summed all the AI algorithms into one name: Computational Intelligence (CI). In his book CI techniques are divided into three fields: Fuzzy Logics, Machine Learning (ML) and Evolutionary Computation, where the last one concerns the nature inspired algorithms. As part of this study we are interested by the bio-inspired techniques, more specifically the swarm intelligent algorithm (SI) which is a sub-field of Evolutionary algorithms. This sub-field is a population based technique inspired from the foraging behavior of animals and insects like Ant Colony Optimizer (Titri et al. 2017), Particle swarm algorithms and his modified versions (Azli et al. 2020; Gavhane et al. 2017; Kaced et al. 2020), Bat Swarm algorithm (Kaced et al. 2017; Titri et al. 2020), Grey Wolf Optimizer (Mirjalili et al. 2014), Cuckoo Search (Mosaad et al. 2019) and many others. For MPPT problem, SI algorithms has proved their highly performance in improving the energy efficiency of PV under uniform und partial shading condition. Regarding the Learning based algorithm (ANN, ML) (Hsu et al. 2015; Yaichi et al. 2014) those methods are simple and don’t require much computational time, or any external equipment, thus making them suitable for such a problem. In the present work, an implementation of MPPT based ISSA algorithm is performed and compared with the popular PSO algorithm in term of convergence speed and efficiency where ISSA is improved version of Salp Swarm Algorithm proposed by (Mirjalili et al. 2017). To summarize, the paper is organized as follows: an overview on PV architecture and modification introduced in Sect. 2, followed by the concept of ISSA algorithm and its implementation on MPPT controller in Sect. 3, after that Sect. 4 with simulation, results and discussion, and finally the conclusion.

2 Photovoltaic System Overview The architecture of the PV system used in this paper see Fig. 1 contains a PV array connected with buck boost converter, MPPT controller and a load. The PV array consists of four modules from MSX-60 solar panels connected in series with a bypass

480

H. Azli et al.

diode. MSX-60 has the following specifications: Pmax ¼ 60 W; Vmpp ¼ 17:1 V; Impp ¼ 3:5 A ; Isc ¼ 3:8 A ; Voc ¼ 21:1 V. The solar cell model parameters are: KI ¼ 3 mA=C ; Kv ¼ 80 mV=C ; Ns ¼ 36; Rp ¼ 176:4 X; Rs ¼ 0:35 X:

Fig. 1. General schematic of PV system with buck boost converter

A two diode model proposed by (Ishaque et al. 2011) is used to simulate the functionality of the solar cell. The output current I is given as follows: 

 
 V þ IRs V þ IRs V þ IRs I ¼ Iph  I0 exp þ exp þ2  ; VT ðp  1ÞVT Rp

p  2:2 ð1Þ

Iph ¼ ðIsc þ KI DT Þ Io ¼ Io1 ¼ Io2 ¼

G GSTC

ðI þ KI DT Þ  sc  exp

ðVoc þ KV DT Þ VT

ð2Þ

1

ð3Þ

Where: V: The terminal voltage. k: Boltzmann’s constant, 1:3865  1023 J=K. q: Electron charge, 1:6021  1019 C. T: Temperature in Kelvin. p ¼ a1 þ a2 : a1 ; a2 are Diffusion and recombination current component of the diodes, a1 ¼ 1. VT ¼ NsqKT : The thermal voltage of the PV having Ns cells connected in series; Rp ; Rs : are shunt resistor and series resistor. KV : Open circuit voltage coefficient. KI : short-circuit current coefficient.

MPPT - Based Improved Salp Swarm Algorithm

481

3 Improved Salp Swarm Algorithm: Overview, and Application on MPPT A. SSA concept The SSA algorithm is a swarm intelligent algorithm proposed by (Mirjalili et al. 2017) inspired from the swarming behavior of the salp fishes in the ocean. The slaps searches for food in a collaboration with each other by making a chain. According to the biological researchers, the reason of this foraging behavior isn’t clear yet because of the hard living environment that restrict having them in a laboratory. The salps individuals create a chain shown in Fig. 2, where the salp in the head is called the leader and the other salps are followers.

Follower

Leader Individual salp

Fig. 2. Swarm of salps

The mathematical model of this algorithms is not complex and easy to implement for engineering problems. To stimulate the chain chasing behavior, the author introduced two types of movement for two groups: leader and followers. The leader update its position x1j according to the food source Fj with the following equation:

x1j ðl þ 1Þ

¼

 Fj þ c1  ubj  lbj  c2 þ lbj ; Fj  c1  ubj  lbj  c2 þ lbj ;

c3  0:5 c3 \0:5

 ð4Þ

Where x1j is the position of the first salp in the chain (leader), Fj is the source food and represent the best solution found so far, c2 and c3 are random number in ½0; 1. c1 is the parameter responsible for balancing between the exploration and exploitation of the algorithm, it is defined as follows:
2 ! c
l c1 ¼ a exp  L

ð5Þ

482

H. Azli et al.

Where l is the current iteration and L is the maximum iteration. a and c are constants equals to 0.2 and 6, and can be changed according to the applied problem. The follower’s positions is updated with following equation: xij ðl þ 1Þ ¼

 1 i xj ðlÞ þ xi1 ð l Þ j 2

ð6Þ

is the best salp Where, xij is the position of the ith follower in the jth dimension, and xi1 j right after the current one in the chain. The original salp algorithm has a simple mathematical model. In the first iterations the algorithm tend to explore the space by moving with large steps. When the food region is found the salps will move with small steps by creating a chain where the leader will be the closest to the food. In this phase the followers will not contribute in the exploitation making the algorithm slowly converges, see Fig. 3. To fix this issue a small modification is made in Eq. 6. We assume that the algorithm will always find the food region after the m iterations. In this case we switch from Eq. 6 to the following one:   i1 i xij ðl þ 1Þ ¼ xi1 ð l Þ þ r  x ð l Þ  x ð l Þ ; j j j

ð7Þ

Iterations

Fig. 3. Movements of the salps over iterations

Where, r is a random number among ½0:2; 0:3: B. MPPT based ISSA algorithm We used an open circuit system, where the input is the duty cycle and the output is the power of the PV. The PV output power serves as the objective function and the duty cycle as the positions of the salps: Ppv ¼ f ðd Þ; d 2 ½dmin ; dmax  The flow chart in Fig. 4 shows the process of ISSA algorithm. In the beginning, the algorithm generates a population of N particles with a predefined uniform distribution in the search space ½0:1; 0:9. The population is sorted in descend way according to their fitness values, then the new position is updated with Eq. (4), Eq. 6 and Eq. (7). When the convergence criteria is met the algorithm output the best value found so far and stops the searching process. If a climatic change happened, the algorithm restarts from the beginning.

MPPT - Based Improved Salp Swarm Algorithm

483

Fig. 4. Flow chart of ISSA

4 Simulation, Results and Discussion The simulation is carried out on MATLAB/SIMULINK, where the proposed algorithm is implemented on MPPT controller with buck boost converter. The parameters of the converter chosen, by satisfying the continuous conduction current mode and an output ripple voltage of 5% are: f ¼ 100 kHz; Cout ¼ 220 lF; Cin ¼ 0:5 mF; L ¼ 0:4 mH. The values of ISSA parameters are selected properly to balance between the exploration and exploitation process. In this paper we have chosen: the population size N = 3 with 30 maximum iterations a ¼ 0:2 and c ¼ 6. The proposed method was tested and compared with PSO and original SSA under different patterns presented in Fig. 5. At first the PV modules were subjected to the  standard test condition STC G ¼ 1000 W/m2 ; T ¼ 25  C , then under partial shading condition with a change of patterns from PSC1 to PSC3 every 3 s. The temperature is fixed to 25  C. In the first simulation, depicted in Fig. 6, the proposed method reached 99% of the maximum power (239.54 W) at t = 1 s, at this point the oscillation was gradually diminishing to zero until t = 1.6 s. Regarding PSO method, the results shows a large fluctuation in the searching process and converged with 99% of maximum power at t = 1.7 s.

484

H. Azli et al.

Fig. 5. P-V curve of different patterns

Fig. 6. Performance of ISSA and PSO under standard condition

In the second test, the two algorithms were subjected to a rapid changing in the radiation with a constant temperature T = 25°. At t = 0, the four panels received an equal radiation of 800 W=m2 (PSC1), after 3 s the pattern changed from PSC1 to PSC2, and after the next 3 s, it changed from PSC2 to PSC3. Figure 7 shows that ISSA algorithm has successfully found the global maximum for the three patterns with high efficiency and in short time compared with PSO. It’s clearly viewed that the convergence time was respectively 1.25 s, 1.1 s, 1.3 s for ISSA and 1.8 s, 1.5 s, 1.9 s for PSO. Because SI algorithms are stochastic processes, their performance can’t be observed clearly from one test, therefore a trail of 50 runs is performed for the three algorithms: ISSA, SSA and PSO, under STC, PSCA, PSC2 and PSC3. The convergence time, settling time, average best power, efficiency (%) and standard deviation are calculated and stored in Table 1. From the above results, it is noticed that ISSA has fast convergence to the global maximum with low dispersion of values and a remarkable efficiency, thus proved its highly performance over PSO and SSA in the cases STC, PSC21 and PSC2. Whereas it was ranked the second after PSO in the efficiency and standard deviation for the case PSC3.

MPPT - Based Improved Salp Swarm Algorithm

485

Fig. 7. Performance of ISSA and PSO under PSC1, PSC2 and PSC3 successively

Table 1. Results of 50 time test under different patterns

STC: ISSA ‘1000 1000 1000 1000’ SSA PSO PSC1: ISSA ‘800 800 800 800’ SSA PSO PSC2: ISSA ‘700 500 200 100’ SSA PSO PSC3: ISSA ‘900 550 550 280’ SSA PSO

Max. Average best Power (W) power (W)

Convergence time (s)

Settling time (s)

Static eff. (%)

Standard deviation

239.55

1,368 1,426 1,916 1,314 1,451 1,928 1,354 1,427 1,811 1,379 1,452 1,975

1,866 1,911 2,087 1,679 1,922 2,233 1,838 1,905 2,154 1,859 1,918 2,291

99,99 99,99 99,99 99.98 99,97 99,95 99,99 99,99 99,99 99,98 99,90 99,99

8,04E−03 1,92E−02 2,26E−02 1,42E−01 3,36E−01 5,65E−01 1,13E−02 2,01E−02 2,24E−02 4,75E−02 2,74E−01 2,68E−03

190.82

100.10

59.39

239,55 239,54 239,54 190.79 190.77 190.73 100,10 100,10 100,10 59,38 59,33 59,39

5 Conclusion In this study an evaluation of MPPT controller based on ISSA algorithm is performed under different shading condition. The system was composed of a string of four modules connected in series, where each one has a bypass diode in parallel to be protected from hotspot problem, along with that, the system is attached to a buck boost converter controlled by MPPT charge controller. The PV was subjected to uniform shading and two partial shading condition PSC. The results shows the ability of ISSA to track the global maximum with a notable efficiency and a fast convergence speed regarding PSO and SSA. A test of 50 runs is also established in order to examine the statistical effectiveness. It was clearly observed that the proposed algorithm has a rapid convergence, also an accuracy in around of 99.98% for all cases, and on top of that, the low values of standard deviation validate the high precision of the algorithm. To summarize, ISSA algorithm has proved its outstanding performance over other algorithm under different cases of shading condition successfully.

486

H. Azli et al.

References Azli, H., Titri, S., Larbes, C.: Modified particle swarm optimization based MPPT with adaptive inertia weight. In: Hatti, M. (ed.) Smart Energy Empowerment in Smart and Resilient Cities, pp. 115–123. Springer (2020) Gavhane, P.S., Krishnamurthy, S., Dixit, R., Ram, J.P., Rajasekar, N.: EL-PSO based MPPT for solar PV under partial shaded condition. Energy Procedia 117, 1047–1053 (2017) Hsu, R.C., Liu, C.-T., Chen, W.-Y., Hsieh, H.-I., Wang, H.-L.: A reinforcement learning-based maximum power point tracking method for photovoltaic array. Int. J. Photoenergy 2015, 1–12 (2015) Ishaque, K., Salam, Z., Taheri, H.: Simple, fast and accurate two-diode model for photovoltaic modules. Solar Energy Mater. Solar Cells 95(2), 586–594 (2011) Kaced, K., Larbes, C., Ramzan, N., Bounabi, M., Elabadine Dahmane, Z.: Bat algorithm based maximum power point tracking for photovoltaic system under partial shading conditions. Sol. Energy 158, 490–503 (2017) Kaced, K., Titri, S., Larbes, C.: Enhancement of extracted power from photovoltaic systems through accelerated particle swarm optimisation based MPPT. In: Hatti, M. (ed.) Smart Energy Empowerment in Smart and Resilient Cities, pp. 94–102. Springer (2020) Mirjalili, S., Gandomi, A.H., Mirjalili, S.Z., Saremi, S., Faris, H., Mirjalili, S.M.: Salp swarm algorithm: a bio-inspired optimizer for engineering design problems. Adv. Eng. Softw. 114, 163–191 (2017) Mirjalili, S., Mirjalili, S.M., Lewis, A.: Grey wolf optimizer. Adv. Eng. Softw. 69, 46–61 (2014) Mirjalili, S., Dong, J.S., Lewis, A.: Nature-Inspired Optimizers: Theories, Literature Reviews and Applications, vol. 811. Springer, Cham (2019) Mosaad, M.I., Abed el-Raouf, M.O., Al-Ahmar, M.A., Banakher, F.A.: Maximum power point tracking of PV system based cuckoo search algorithm; review and comparison. Energy Procedia 162, 117–126 (2019) Titri, S., Kaced, K., Larbes, C.: Maximum power point tracking based on the bio inspired BAT algorithm. In: Hatti, M. (ed.) Smart Energy Empowerment in Smart and Resilient Cities, pp. 22–29. Springer (2020) Titri, S., Larbes, C., Toumi, K.Y., Benatchba, K.: A new MPPT controller based on the Ant colony optimization algorithm for Photovoltaic systems under partial shading conditions. Appl. Soft Comput. 58, 465–479 (2017) Yaichi, M., Fellah, M.-K., Mammeri, A.: A neural network based MPPT technique controller for photovoltaic pumping system. Int. J. Power Electron. Drive Syst. (IJPEDS) 4, 241–255 (2014)

Multi-agent System for Voltage Regulation in Smart Grid Hadjira Belaidi(&), Hamid Bentarzi, Zakaria Rabiai, and Abdelkader Abdelmoumene Signals and Systems Laboratory, Institute of Electrical and Electronic Engineering, University M’Hamed BOUGARA of Boumerdes, Boumerdès, Algeria [email protected], [email protected]

Abstract. In this research work, a new approach of decentralized energy management for smart grid is proposed to solve the problem of distributed voltage regulation. Where, micro-grids and aggregators are used as smart agents that can communicate with each other to share information, distribute energy and control their own energy consummation. Aggregators make the link between flexible resources. Smart-agents are an emerging technology for decentralized computation and data storage, secured by a combination of cryptographic signatures and a distributed consensus mechanism. So, two types of agents: energy Generation AGent (GAG) and Bus Agent (BAG) are used to regulate the voltage levels by injecting more power at some buses using the renewable energy sources. The interaction between the two types of agents is based on communication and exchange of information about the parameters and the state of the power grid. For testing this approach, a developed tester by our laboratory has been used that gives a good result. Keywords: Distributed energy resources
Internet of Things
Decentralized energy management system
Renewable resources
Smart grid
Multi-agent system

1 Introduction In the literature, the energy management of smart grids (SG) is based on centralized approaches [1]. Thus, centralized SG energy management issue has been largely studied and several works have sought to improve these approaches [2, 3]. In [4] the master-slave strategy has been used for a rule-based management system applied to a microgrid composed of multiple energy resources including a PV system, a fuel cell, and a battery bank. This approach is suitable solution for energy resources near to each other. In the same context, gravitational search algorithm (GSA) method was applied in [5–8] to optimize power flow in given systems and the concept of centralized microgrid management system (MMS) was used to make decisions when information was collected in a central point, while cooperation and prioritization are achieved more easily when using a central controller [9]. Other works had addressed the problem of multiagent power grid operation using centralized methods [10, 11]; however, the best © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 487–499, 2021. https://doi.org/10.1007/978-3-030-63846-7_46

488

H. Belaidi et al.

results they got even near optimal, time consuming, higher control complexity or higher communication cycles to discover the information. On the other hand, the researchers start to reshape it by using distributed energy resources (DERs). DERs are compensated for providing energy services by an aggregator or a utility: a central authority that is trusted to act fairly in scheduling generators, satisfying loads, and rendering payments [12, 13]. DERs are often remotely controlled by the Internet of Things. When they are used intelligently, these DERs can reduce cost, improve reliability, and integrate renewable resources in the electric grid—features which have led regulators to introduce policies promoting their adoption. In this research work, a new approach of decentralized energy management for SG has been proposed. Where, micro-grids and aggregators are used as smart agents that can communicate with each other to share information, distribute energy and control their own energy consummation. Aggregators make the link between flexible resources. Multi-agents are an emerging technology for decentralized computation and data storage, secured by a combination of cryptographic signatures and a distributed consensus mechanism.

2 Distributed Energy Resources (DERS): An Overview With SG, the planning, investment, and operation of the distribution system (DS) change dramatically. Historically, utility investment in distribution systems ensured circuit capacity was adequate to deliver power from the bulk grid to the customer. Now, customer-owned solar PV delivers power to the distribution system, and distributed resource (DR) from customers provides energy and capacity reduction at the bulk grid level. A host of other distributed resources, including fuel cells and energy storage, provide power that is injected at the low-voltage level and may create reverse power flows on the grid, moving power away from the customer. Platforms are being designed to host DERs at lower voltage levels to explicitly supply customers at the distribution level and to wholesale markets. An immediate objective is to monetize the option value of DERs, which translates to more flexible DER uses in multiple markets. Multiple opportunities have emerged, and more will result as DER needs increase across the grid. We examine both the voltage context and future opportunities to provide greater understanding of these new resources. Distributed resources can reduce peak demand, which can eliminate or defer new transmission and distribution capacity, and decrease total energy costs. Enhanced onsite peaking generation resources also improve the security and hence the reliability. Moreover, with the intensifying incorporation of DER in the SG, decentralized and multi-agent approach uses become inevitable for arrangement and allocation of resources in a SG.

Multi-agent System for Voltage Regulation in Smart Grid

489

3 Smart Grid (SG) Broadly defined, a SG combines electrical infrastructure with digital technologies that analyze and transmit the received information. These technologies are used at different levels of the network: production, transport, distribution and consumption. Hence, SG can be seen as the successor of the conventional electrical infrastructure which uses information and communications technology to automate the production and distribution of electricity. SG can enhance the conventional grid in several factors such that: • Network operators can reorient energy flows according to demand and send price signals to individuals to adapt their consumption (voluntarily or automatically); thus, maintaining real-time flow control. This can be ensured basing on the information recorded from the sensors installed through the grid and which indicate, instantly, electrical flows and consumption levels. • By the instantaneous exchange of information, smart grids promote interoperability between the operators of the transmission network (which links the electricity production sites to the consumption areas) and those of the distribution network (which delivers electricity to end consumers). • SG are based on an information system that allows the level of production and consumption to be predicted in the short and long term. Renewable energies which often operate intermittently and in an unpredictable way (e.g. wind power and solar energy) can thus be better managed. • More responsible management of individual consumption: smart meters (or advanced meters, “Linky” for electricity) are the first application versions of SG. Installed at consumers, they provide information on prices, peak consumption times, quality and level of electricity consumption in the home. Consumers can then regulate their consumption themselves during the day. For their part, network operators can detect faults faster. Therefore, the most amazing innovation of the SG is that consumers of the network can become an energy supplier as well, that is consumers can become prosumers.

4 IoT in SG In its universal meaning, IoT (Internet of Things) depicts the notion of inter-relating the virtual world of computers with the real world of physical objects [14, 15]. Everything in IoT Smart grid is based on networking because the grids must capable of sensing (through their sensors) and reacting (through their actuators); thus, creating smart environments surround them, this can be via the integration of communication networking, the internet, sensors (PMUs), smart meters and remotely controlled switches, hardware (embedded systems) and software technologies. IoT helps SG systems to support various network functions throughout the generation, transmission, distribution and consumption of energy by providing the connectivity, automation and tracking for such devices. Figure 1 summarizes the integration of IoT in SG.

490

H. Belaidi et al.

Intermediate interaction Interaction Protocols Agent organization

Sensors/actuators Power Grid

Fig. 1. Multi-agents system structure

The IoT focuses on the realization of three main concepts, namely things-oriented, Internet-oriented and semantic-oriented. The things oriented concept involves smart devices, such as RFID tags, sensors, actuators, smart meters, the Global Positioning System (GPS) and NFC. The Internet oriented concept enables communication among smart devices through various communication technologies, such as ZigBee, Wi-Fi, Bluetooth and cellular communications and connects them to the Internet. The semantic oriented concept realizes a variety of applications with the help of smart devices [14]. Connecting things to the Internet involves the devices to use an IP (Internet Protocol) address as distinctive identifier. IPv4 is definitively bushed owing to its insufficient address range against to the huge address range requirements that leads to unavoidable passage from IP v4 to IP v6 [16, 22].

5 Multi-agent System in SG A multi-agent system (MAS or “self-organized system”) is a computerized system composed of multiple interacting intelligent agents [17]. It can be divided to many different sub-systems as shown in Fig. 1. One agent can play one or more tasks. All agents coordinate with subsystems. Interactions among agents with the power grid can be ensured by the Remote Terminals Unit (RTU)s and smart actuators such as RES (Renewable Energy Source) and SVC (Static VAR Compensator) via communication network or internet as explained in previous section. The role of the smart agent determines the part of the power grid in which can receive and send the data. Previously, conventional grids and electrical distribution systems were monitored and automated using SCADA system. SCADA stands for Supervisory Control and Data Acquisition; it is an industrial computer-based control system employed to collect and evaluate data remotely and in real-time to maintain, supervise and check technical installations in power systems for better functioning and control.

Multi-agent System for Voltage Regulation in Smart Grid

491

In industrial organizations, SCADA systems are required to preserve efficiency, process data for smarter decisions, coordinate and communicate system issues to help mitigate downtime. Since 2010, Ignition HMI/SCADA Software has been installed in many companies in over 100 countries. SCADA system is powerful and robust permits integrators to reach the demands of their customers, Fig. 2 summarizes the Ignition HMI/SCADA architecture which allows SCADA to gathers all system knowledge in central processor then makes the decision (remotely). However, in the micro-grid, uncertainty in SCADA systems arises when sensor data or inferred knowledge cannot be deemed accurate due to intermittent nature of renewable energy resources [18].

PLC 1

Igni

n Server in the cloud

Web-based designers

PLC n Web-launched clients Database in the cloud

Touch panels Mobile phone

Fig. 2. Ignition SCADA standard architecture sample.

Generally, sensors information can be noisy, erroneous or uncertain; moreover, applications using multiple sources can suffer from incoherent, uncertain or divergent data; which cannot be solved even by using SCADA unless by the supervision of the human. Whereas, MAS approach for energy management does not suffer from such situations and offers better results, more energy efficiency and time reduction. MAS automates the micro-grid by associating protocols skills and communication facilities between the different agents of the micro-grid to exchange the data. Micro-grid based on MAS can replicate easily and seamlessly basing on plug and play adaptability to connect to external grids. MAS are useful for designing distributed systems requiring autonomy of their entities. MAS use new programming paradigm to implement agents, which is bringing about new programming paradigm for software engineering called Agent-Oriented-Programming AOP. MAS has the ability to add new agents (resources or loads), or dissolve them and associate them to a new environment with new resources and new abilities. Each agent in MAS can have the following utilities: communication tools, control unit, security, localization unit … Figure 3 illustrates a standard diagram of MAS for energy management.

492

H. Belaidi et al.

- Communication tools - Control unit - Security - Localization unit - Ö

- Communication tools - Control unit - Security - Localization unit - Ö

Agent 2

Agent 1

- Communication tools - Control unit - Security - Localization unit - Ö

- Communication tools - Control unit - Security - Localization unit - Ö

Agent 3

Agent n

Fig. 3. Standard diagram of MAS for energy management

6 Energy Management System (EMS) Centralized or decentralized energy management may be used in SG. However, the centralized scheme has a clear drawback; a failure in one of the control centers might result in the total collapse of the system. Moreover, the ambiguity related with renewable energy sources has made resource distribution matters even more difficult for grid users. The next smart-grid generation may combine different renewable energy sources and may have bi-directional power flow. Therefore, it is highly desirable to have enough intelligence and redundancy throughout the system to survive failures, to resource allocation problem and to permit inter-node communication and decision making. Multi-agent systems (MAS) is a promising platform to decentralize the traditional centralized resource allocation aspects of smart grid [19]. Our approach proposes a multi-agent system consists of several Bus Agent (BAGs) and power Generation Agent (GAGs) capable of regulating voltage and keeping it within the permissible limits, based on local information. This being a way of decentralizing the information, because the loss of information can lead to a cascade of overloads that can lead to voltage collapse. The decomposition of problem (P) into m sub-problems is considered, with most of the equality or inequality constraints from electric grid are expressed in terms of only few variables and local variables come from a small geographic area. Each sub-problem (assigned to one agent) contains a part of the objective of (P) and some of its constraints. The agents will work on their sub-problems asynchronously. Agents can be used to improve the control devices, relays, Flexible AC Transmission Systems (FACTS) devices or voltage regulators. The later is treated in our case study. The connection between software entities and automation subsystems are fixed (generally defined at design-time), but the systems should deal with unanticipated requests.

Multi-agent System for Voltage Regulation in Smart Grid

493

A. Voltage Regulation Based on Smart Agents The application of smart agents to the power grid is new research field. In the power grid, the voltage level can be affected by different factors such as the load variation or the grid reconfiguration, thus, a rapid control may be needed for solving the problem that may be caused by the disturbances. At some buses, the voltage may decrease below the allowable limits. We discuss here their capability to create a coherent structure that can guarantee an efficient way to manage and control multiple distributed energy resources (DERs) [20, 21]. A:1. Agents Types In our approach, two types of agents are used such as bus agent (BAG) and renewable energy generation agent (GAG) as shown in Fig. 4.

BAG 3

BAG 4

Bus agent (BAG) Generation agent (GAG) Information message BAG 2

BAG 1

Contingency message

GAG

Fig. 4. Different types of agents and messages.

a) Bus Agent (BAG) Bus agent exists in each bus and respects to the following rules: Rule 1: When BAG detects a decrease in the voltage level under the limit, a message “power demand” is sent to all neighbour agents. Rule 2: When BAG receives a message “reject to power demand” from the first, it sends the same message to the second agent. Rule 3: When it receives a message “generation limit” and the voltage does not attain the permitted level, BAG sends another message to the renewable energy generation agent (GAG) of the second priority. Rule 4: When the voltage enters within the allowable limits, BAG sends a message “stop” to the agents.

494

H. Belaidi et al.

b) Renewable energy generation agent (GAG) The agent GAG plays an important role in the system; it determines the stored power energy and the transmission power line thermal capacity. It acts according to the simple rules for solving the local optimization problem of generation. Rule 1: When GAG receives a message “power demand”, it will verify two constraints such as the stored power energy (generation limit) and thermal capacity of the power line. If the two constraints are verified, it will send a positive message to BAG. Rule 2: if one of the two constraints is not satisfied, it sends a rejection message to BAG. Rule 3: if GAG receives more than one message “power demand”, all messages are classified in a vector with priority order starting by the low power amount and low price. Rule 4: When GAG receives a message “stop”, it confirms the stop.

A:2. Message Types The main objective of the messages is to maintain the agents informed by the neighbour agents’ conditions. Messages can be classified according to interaction and communication of agents. They are classified into two types to facilitate the distribution, control and coordination by agents. a) Information Messages These types of messages are designed for giving information exchange among agents during the normal condition. They are message state request and message state reply. b) Contingency Messages When the voltage is outside the limits, this type of information may be exchanged among the agents in order to recover the situation. These messages are: “power demand”, “reject to power demand”, “power generation limit”, “stop”, and “confirmation”. B. Voltage Regulation by Multi-agents System The objective of this study is to decentralize energy management using MAS. Each BAG collects information about the grid from neighbour agents as shown in Table 1; hence, it determines during contingency the nearest energy source having the smallest electrical amount. For example, if an agent is located at bus (i, j) that is represented by: AGij

for

i ¼ 1; . . .m

and

j ¼ 1; . . .n

ð1Þ

Multi-agent System for Voltage Regulation in Smart Grid

495

Table 1. Agent arrangement j′ j i′ AG(i′j′) AG(i′j) i AG(ij′) AG(ij) i″ AG(i″j′) AG(i″j)

j″ AG(i′j″) AG(ij″) AG(i″j″)

Each agent has a maximum 8 neighbours, illustrated in Fig. 5, and which are defined as follows:  AGij ¼ AGi0 j0 ; AGij0 ; AGi00 j0 ; AGi0 j ; AGi00 j00 ; AGi00 j ; AGij00 ; AGi0 j00 g 

ð2Þ

 0 i0 ¼ i  1 j ¼j1 where, and; i00 ¼ i þ 1 j00 ¼ j þ 1 Each bus can have a maximum of information about its Environment exchange among agents. Data base of agent consists of the system information that can be classified into two categories of structure arrays. One concerns the branches and other structure arrays of buses as shown in Figs. 6 and 7 respectively.

branch branch(1,1) Name

Name

Z

Z

B

B

P

P

branch(2,1)

branch(2,4)

Name

Name

Z

Z

B

B

P

Fig. 5. Multi-agents compensation proposition

branch(1,4)

P

Fig. 6. Branch data structure arrays

496

H. Belaidi et al. Bus

Bus 1

Bus 2

Name

Name

V

V

P

P

Q

Q

Bus 3

Bus 4

Name

Q

Name

Q

Fig. 7. Bus data structure.

Fig. 8. IEEE 9 bus power grid.

7 Simulation Results and Discussion A standard IEEE 9 bus shown in Fig. 8 has been implemented for testing our approach concerning multi-agents by applying the previous mentioned rules using Matlab. Bus 1 is taken as reference and the others are PQ buses. Renewable energy sources may be used to inject power at buses 4 and 7. Thus, bus 7 is considered as customer-owned solar PV. Simulink model of the test bench that has been developed in our laboratory for testing our approach is shown in Fig. 9. The load at bus 9 has been increased from 125 MW to 425 MW, all voltage levels of buses remain within allowable limits, except bus 9 voltage level is reduced to 0.79 pu (see Fig. 10). Agent of bus 9 detects that the voltage of its bus has been reduced below the allowable limit. Then, it communicates with near neighbours which have power source. In this case, bus 4 may reply. GAG 4 verifies the two constraints which are thermal capacity of power transmission line and the stored electrical energy of the renewable source. If the two conditions have been satisfied, a positive reply may be sent to BAG 9. The obtained result is illustrated in Fig. 11.

Multi-agent System for Voltage Regulation in Smart Grid

497

Fig. 9. IEEE 9 bus power grid Simulink model with two PV sources.

Fig. 10. Buses voltage levels during the contingency (bus 9 changes from 125 MW to 425 MW).

Fig. 11. Bus 9 voltage as function of power injected at bus 4.

Fig. 12. Bus 9 voltage levels as function of power injected at bus 4 and bus 7.

498

H. Belaidi et al.

When bus 4 attains its maximum of power energy that can be injected to the power grid, its agent will send a message “Stop” to BAG 9. If the voltage level does not return back within the limits, BAG 9 determines the next generation agent (GAG) that is in the second order which can contribute to increase its voltage level. GAG 7 may receive a message and which in turn verifies the two constraints. It can be noted that the last satisfies the conditions and inject the required energy for returning back the voltage level within the limits as shown in Fig. 12.

8 Conclusion This paper proposed a distributed multiagent approach to mitigate the problem of voltage regulation that may be caused by the disturbances in the microgrid. In this research work, two types of agents are created the GAG and the BAG. The GAG determines the stored power energy and the transmission power line thermal capacity and BAG collects information about the grid from neighbour agents. The two kinds of agents collaborate together to regulate the voltage levels by injecting more power at some buses using the renewable energy sources. The interaction between the two types of agents is based on communication and exchange of information about the parameters and the state of the power grid. The voltage regulation can be controlled by bus agents and assured by generation agent starting by the low power generation order. Data base of each agent may contribute more in the coordination and the successful solution without conflict that may lead to time delay during the critical contingency. A coherent structure that guarantees an efficient way to manage and control multiple distributed energy resources (DERs) was created. A standard IEEE 9 bus was implemented for testing our approach; hence, simulation was used to verify the proposed approach and it was showed a satisfactory results. However, to ensure the efficiency of this approach, validation on realistic data is of must. Thus, future work will be on testing the developed approach using realistic systems with field data.

References 1. Farhangi, H.: The path of the smart grid. IEEE Power Energy Mag. 8(1), 18–28 (2010) 2. Olivares, D.E., Cañizares, C.A., Kazerani, M.: A centralized energy management system for isolated microgrids. IEEE Trans. Smart Grid 5(4), 1864–1875 (2014) 3. Storey, H.L.: Implementing an integrated centralized model-based distribution management system. In: 2011 IEEE Power and Energy Society General Meeting, Detroit, MI, USA, pp. 1–2 (2011) 4. Almada, J.B., Leão, R.P.S., Sampaio, R.F., Barroso, G.C.: A centralized and heuristic approach for energy management of an AC microgrid. Renew. Sustain. Energy Rev. 60, 1396–1404 (2016) 5. Singh, S.P., Singh, S.P.: A multi-objective PMU placement method in power system via binary gravitational search algorithm. Electr. Power Compon. Syst. 45(16), 1832–1845 (2017)

Multi-agent System for Voltage Regulation in Smart Grid

499

6. Pani, A.K., Nayak, N.: Forecasting solar irradiance with weather classification and chaotic gravitational search algorithm based wavelet kernel extreme learning machine. Int. J. Renew. Energy Res. 9(4), 1650–1659 (2019) 7. Ji, B., Yuan, X., Li, X., Huang, Y., Li, W.: Application of quantum-inspired binary gravitational search algorithm for thermal unit commitment with wind power integration. Energy Convers. Manag. 87, 589–598 (2014) 8. Radosavljević, J., Jevtić, M., Arsić, N., Klimenta, D.: Optimal power flow for distribution networks using gravitational search algorithm. Electr. Eng. 96(4), 335–345 (2014) 9. Planas, E., Gil-de-Muro, A., Andreu, J., Kortabarria, I., Martínez de Alegría, I.: General aspects, hierarchical controls and droop methods in microgrids: a review. Renew. Sustain. Energy Rev. 17, 147–159 (2013) 10. Sujil, A., Agarwal, S.K., Kumar, R.: Centralized multi-agent implementation for securing critical loads in PV based microgrid. J. Mod. Power Syst. Clean Energy 2(1), 77–86 (2014) 11. Sharma, A., Srinivasan, D., Kumar, D.S.: A comparative analysis of centralized and decentralized multi-agent architecture for service restoration. In: 2016 IEEE Congress on Evolutionary Computation (CEC), Vancouver, BC, pp. 311–318 (2016) 12. Borlase, S.: Smart Grids: Advanced Technologies and solutions. CRC Press, New York (2018) 13. Katiraei, F., Iravani, M.R.: Power management strategies for a microgrid with multiple distributed generation units. IEEE Trans. Power Syst. 21(4), 1821–1831 (2006) 14. Saleem, Y., Crespi, N., Rehmani, M.H., Copeland, R.: Internet of things-aided smart grid: technologies, architectures, applications, prototypes, and future research directions. IEEE Access 7, 62962–63003 (2019) 15. Kovatsh, F.M.: Scalable web technology for the internet of things. Ph.D. thesis, ETH Zurich (2015) 16. Belaidi, H., Belkacem, J., Abed, M.A., Bentarzi, H.: IoT path planning approach for mobile robots. In: ICASS 2018, Média, Algeria, 24–25 November 2018 (IEEExplorer) (2018) 17. Priyadarshana, H.V.V., Kalhan Sandaru, M.A., Hemapala, K.T.M.U., Wijayapala, W.D.A.S.: A review on multi-agent system based energy management systems for micro grids. AIMS Energy 7(6), 924–943 (2019) 18. Raju, L., Milton, R.S., Mahadevan, S.: Multiagent systems based modeling and implementation of dynamic energy management of smart microgrid using MACSimJX. Sci. World J. 2016, Article ID 9858101, 14 p. (2016) 19. Nair, A.S., Hossen, T., Campion, M., Selvaraj, D.F., Goveas, N., Kaabouch, N., Ranganathan, P.: Multi-agent systems for resource allocation and scheduling in a smart grid. Technol. Econ. Smart Grids Sustain. Energy 3(1), 1–15 (2018) 20. Galanis, I., Olsen, D., Anagnostopoulos, I.: A multi-agent based system for run-time distributed resource management. In: 2017 IEEE International Symposium on Circuits and Systems (ISCAS), pp. 1–4 (2017) 21. Sun, H., Guo, Q., Qi, J., Ajjarapu, V., Bravo, R., Chow, J., Li, Z., Moghe, R., Nasr-Azadani, E., Tamrakar, U., Taranto, G.N., Tonkoski, R., Valverde, G., Wu, Q., Yang, G.: Review of challenges and research opportunities for voltage control in smart grids. IEEE Trans. Power Syst. 34(4), 2790–2801 (2019) 22. Belaidi, H., Hentout, A., Bentarzi, H.: Human–robot shared control for path generation and execution. Int. J. Social Robot. 11(4), 609–620 (2019)

Neural Network-Based Attitude Estimation Djamel Dhahbane1(&), Abdelkrim Nemra1, and Samir Sakhi2 1

Laboratoire Véhicules Autonomes Intelligents, Ecole Militaire Polytechnique, BP 17, Bordj El Bahri, Alger, Algérie [email protected], [email protected] 2 Laboratoire Systèmes Embarqués, Ecole Militaire Polytechnique, BP 17, Bordj El Bahri, Alger, Algérie [email protected]

Abstract. Have an accurate information about the system state is a big challenge in several applications, such as guidance, navigation and control of drones, aircrafts and autonomous robots. INS (Inertial Navigation System) can provide information about attitude of vehicle on which is attached. However, it is recognized by error accumulation (drift) over time. Many filters are proposed in literature to overcome this problem and achieve more accurate estimation. Complementary Filter (CF) is one of the best proposed solutions for attitude estimation because of its reliability and suitability for real time applications. Nevertheless, the performance of this filter depends on the choice of gains values. In this paper, Neural Network (NN) framework is developed to enhance the quality of this filter. The BRB (Bayesian Regularization Backpropagation) training algorithm is adopted to improve the generalization qualities and solve the overfitting problem. Simulation results shows that the proposed technique exhibits high capabilities in terms of estimation robustness and accuracy. Keywords: Inertial navigation system
Complementary filter network
Bayesian regularization training algorithm


Neural

1 Introduction Sensing and estimation are prominent aspects for various applications in robotics, aircrafts, or in spacecraft. For such kind of application, accurate information about system states is required to achieve high level of control performance and guarantee a reliable navigation. Attitude estimation consists on the determination of rigid body orientation in 3D space. This research area involves the combination of information from several sources to enhance system reliability [1]. Inertial Navigation System (INS) is one of the most devices used in the field of estimation. It is comprized of IMU (Inertial Measurement Unit) and a navigation processor. IMU exploits measurements of gyroscopes, accelerometers and magnetometers to estimate the full translation and orientation of vehicle in motion [2]. MicroElectro-Mechanical System (MEMS) based IMU is the last technology in the field of INS development [3]. In spite of the INS autonomy in attitude determination, the error accumulation (drift) remains one of the major shortcomings with inertial navigation. This problem is mainly caused by the integration effect [4]. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 500–511, 2021. https://doi.org/10.1007/978-3-030-63846-7_47

Neural Network-Based Attitude Estimation

501

One of the common filter used to improve the estimation accuracy, is called complementary filter. This filter exploits information from multiple noisy sources [5]. The filter output is the sum of many pondered measurements. Many works of attitude estimation using complementary filter are presented in literature. The work in [6] have focused on robust sensor fusion with nonlinear complementary filter for robot attitude estimation. Attitude Estimation and Control Using Linear-Like Complementary Filters is proposed in [5]. The Authors in [7] have designed a fast complementary filter for attitude estimation using Low-Cost MARG Sensors. Comparison of complementary and Kalman filter is presented in [8]. In [9], the authors have proposed a Modified Madgwick Filter for Orientation Estimation Using AHRS. Authors in [3] have used a Fuzzy Tuned Complementary Filter to estimate the attitude by using MEMS IMU. In this approach filter gains are selected using a fuzzy logic controller. However, whatever the used technique, the performance of the complementary filter, absolutely depends on the choice of gain parameters. Therefore, the optimal ponderation remains the big challenge for this filter. In the last few years, Computational Intelligence (CI) based on neural networks is widely studied and incorporated in the field of estimation and control. These algorithms may overcome the issues of the classical estimators [10]. Deep Learning (DL) algorithm employs the multilayer neural network which consists of input layer, output layer and hidden layers. It has exhibited a considerable performance in several applications [11, 12]. In this paper, a new Neural Network-based Complementary Filter (NNCF) algorithm is designed to enhance the attitude estimation provided by the classical complementary filter. The Bayesian Regularization based on Backpropagation algorithm for training is adopted to improve the generalization qualities and solve the overfitting problem. A comparison between the obtained simulation results, demonstrates the performance of the proposed estimator. The rest of this paper is organized as follows: in Sect. 1, some related works to attitude estimation are introduced. Section 2 presents the INS model and architecture. Section 3 describes the complementary filter. After that, neural network for attitude estimation is explained in Sect. 4. Section 5 is reserved to the validation and simulation results. Finally, the paper is concluded in Sect. 6.

2 Inertial Measurement Unit (IMU) 2.1

Gyroscope

The gyroscope measures the angle rates around roll axis, pitch axis and yaw axis p; q; r. Euler angles ðu; h; wÞ can be obtained by the relation [13]. 2

3 2 u_ 1 4 h_ 5 ¼ 4 0 _ W 0

tan h sinu cosu sin u=cosh

32 3 tan hcos u p  sin u 54 q 5 cosu=cosh r

ð1Þ

502

2.2

D. Dhahbane et al.

Accelerometer

Accelerometers are used to measure the acceleration of the vehicles, then acceleration is transformed to the navigation coordinate system. The position and velocity parameters of vehicles can be obtained after integral operations [14]. The roll and pitch angles of the vehicle can be also determined by the following relations [15].   ua ¼ tan1 ay =az

ð2Þ

ha ¼ tan1 ðax =ðay sinðua Þ þ az cosðua ÞÞÞ

ð3Þ

Where ax ; ay ; az are the components of acceleration vector; ua ; ha : are the roll and pitch given by the accelerometer. 2.3

Magnetometer

The magnetometer measures the terrestrial magnetic field; it is a relevant complementary sensor used to improve the precision of attitude calculation [16]. The yaw angle wm can be calculated by the following equation [17]. wm ¼ tan1 ðmy =mx Þ

ð4Þ

Where mx ; my are the components of the Earth magnetic field toward North.

Fig. 1. Example of IMU sensor

As all sensors, IMU measurements are affected by bias, nonlinearity effect, scale factor effect and noise. These additional effects cause errors between the real value and the measured value (details on these effects are clearly described in [18]).

3 Complementary Filter Angle estimates based on gyro alone drift overtime, making them unreliable in the long term. Angle estimates based on accelerometers do not cause the drift, but they are sensitive to external forces like vibration, making short-term estimates unreliable.

Neural Network-Based Attitude Estimation

503

Therefore, gyro has a low-frequency error characteristic as compared with the accelerometer and magnetometer, which have high frequency error characteristics [15]. Combing the outputs of these sensors by a complementary filter, can provide angles estimates that have good behavior in low and high frequencies. Complementary filter (CF) approaches are popular alternatives solutions. Due to its simplicity, low cost and low consumption, this filter is commonly used in attitude estimation [19]. This type of filter blends heterogeneous and independent sources of data in a complementary manner. In our case, the filter fuses the static low frequency information provided by accelerometers and magnetometers, and the dynamic high frequency information provided by the gyroscope [15]. The basic structure of complementary filter is illustrated in the Fig. 2. The output of the complementary filter is the sum of the gyro measurement filtered by a High Pass Filter (HPF) and the accelerometer/magnetometer measurement filtered by a Low Pass Filter (LPF) [15].

x1

HPF +

x2

xˆ = x 1 HPF + x 2 LPF

LPF Fig. 2. Basic structure of complementary filter

The expressions of Roll, Pitch and Yaw are given by the following equations: uCF ¼ a1 ug þ ð1  a1 Þua

ð5Þ

hCF ¼ a2 hg þ ð1  a2 Þ ha

ð6Þ

wCF ¼ a3 wg þ ð1  a3 Þ wm

ð7Þ

When: ug ; hg ; wg are the gyro measurements, ua ; ha are the accelerometer measurements and wm is the magnetometer measurements; a1 ; a2 ; a3 are positive coefficients.

4 Neural Networks for Attitude Estimation Neural networks represent a potential technology widely exploited in several domains, such as: computer science, mathematics, and engineering. Neural networks are distributed structures that have the capability to learn, update and generalize [20]. The main idea of Artificial Neural networks (ANN) is based on the simulation of the human neurons by the design of learning model, applied on machines to execute the required tasks in a manner similar to process of human nervous system [21]. The basic principal of neural networks is to mapping input data into desired output. This concept is performed by the determination of the network architecture, which

504

D. Dhahbane et al.

includes the type of the neural network, the number of hidden layers and weights, and the set data for the training [22]. 4.1

Artificial Neural Networks Architecture

An Artificial Neuron (AN) operates with nonlinear application from RI usually to [0, 1] or [−1, 1], depending on the function used to activate the network, where I is the number of inputs [23]. In Fig. 3, an architecture of AN is illustrated [23]. The input signal net of an AN is usually calculated as the weighted sum of all input signals or as the weighted product of all input signals [23], net ¼

I X

z i vi

ð9Þ

zvi i

ð10Þ

i¼1

net ¼

I Y i¼1

With two input signals: the net input signal and bias b, the activation function generates the output of the neuron. Many forms of activation functions can be used (Step function, Ramp function, Sigmoid function, Hyperbolic function, Gaussian function…). In general, activation functions satisfy the following conditions given in Eq. (9), (10) [23]. f ð1Þ ¼ 0 or f ð1Þ ¼ 1

ð11Þ

f ð þ 1Þ ¼ þ 1

ð12Þ

z1

v1

z2

v2

v3

Input z3

f (net ) − b

... .. vI zI

Fig. 3. Architecture of an artificial neuron

Output

Neural Network-Based Attitude Estimation

505

To put the network in specific application, it needs to be trained. There are two approaches to training or learning; supervised and unsupervised. Supervised training consists on providing the network with the inputs and the desired output. In Unsupervised training, only inputs are given [24]. There are many training algorithms available in neural network. In this paper, Bayesian Regularization algorithm is used. 4.2

Bayesian Regularization Backpropagation Algorithm

Bayesian Regularization Backpropagation (trainbr in Matlab software) is one of the training algorithm used in neural network. It updates the weight and bias values according to Levenberg-Marquardt and generates a network that generalizes well [25]. The Bayesian approach for neural networks is based on probabilistic distribution of network weights, this formalism changes our belief about the weights from the prior probability p ðvi Þ, to the posterior probability, pðvi =DÞ, as a consequence of seeing the data D [26].

5 Simulation Results and Discussion The simulation process is shown in the Fig. 4. takes the weighted sum of the attitude measurements coming from the gyroscope and accelerometer (roll and pitch). The yaw angle is obtained by the fusion between gyroscope and magnetometer measurements.

Three Axis Gyroscope ϕg

ψg

θg

Three Axis Accelerometer ϕa

Three Axis Magnetometer ψm

θa

Complementary Filter (CF) ψ CF

θCF

ϕCF

Neural Network (NN)

ϕNN

θ NN Fig. 4. Simulation process

ψ NN

506

D. Dhahbane et al.

The outputs of the complementary filter are used to train the neural network with a supervised approach. The main contribution of this paper is to use neural networks to improve the attitude estimation given from the complementary filter. The Neural Network is trained using Bayesian Regularization Backpropagation algorithm. The process of our approach is detailed in Fig. 4. Noisy measurements from gyroscope, accelerometer and magnetometer of MEMS IMU, are combined with conventional complementary filter to get the full attitude (Euler angles). Then, the output of this filter is trained in neural network using the Bayesian Regularization Backpropagation) training algorithm. Table 1. IMU specification Gyroscope specification Parameters Units Standard full range [°/s] In-run bias stability [°/h] Bandwidth (−3 dB) [Hz] Noise density [°/s/√Hz] Non-linearity [%FS] Scale factor variation [%] Accelerometer specification Parameters Units Standard full range [g] In-run bias stability [mg] Bandwidth (−3 dB) [Hz] Noise density [lg/√Hz] Non-linearity [%FS] Magnetometer specification Parameters Units Standard full range [G] Non-linearity [%] Total RMS noise [mG] Resolution [mG]

Values ±2000 10 255 0.007 0.1 0.5 Values ±16 0.03 324 120 0.5 Values 8 0.2 0.5 0.25

The neural network used is composed of input layer, output layer and ten (10) hidden layers. The training algorithm adopted (trainbr) use 70% of data for training, 15% of data for validation, and 15% of data for test.

Neural Network-Based Attitude Estimation

507

Roll Estimation 21

GYR

50

ACC CF

25.45

25.5

30

20 10 0 0

30

20

10 0 0

20

10

40

30

50

0.2 0

60

0

Time(s)

20

0

Error__GYR(deg)

CF

14.5 25.4

NN

25.45

25.5

30

10

20

0 0

20

40

60

60

40

60

1 0 0

20

2

0 20

10

40

30

50

Error__NN(deg)

Error__CF(deg)

10

2 1 0

60

0

Time(s)

20

40

1

0 0

60

20

Time(s)

Fig. 7. Pitch angle estimate

Time(s)

Fig. 8. Pitch angle error

Yaw Estimation

30

0.14

GYR

13

MAG CF

12 25.4

NN

25.45

25.5

20 10 0 0

30

20

40

60

Error__MAG(deg)

Error__GYR(deg)

14

True

40

40

2

Time(s)

3

50

60

3

Time(s)

60

20

Time(s)

20

15

ACC

0

40

0

60

15.5

GYR

40

40

Fig. 6. Roll angle error

Pitch Estimation True

60

0.2

Time(s)

Fig. 5. Roll angle estimate

50

40

0.4

0.4

Error__ACC(deg)

0

20

Time(s)

Error__NN(deg)

0

Pitch(deg)

60

0.6

20

10

0.13 0.12 0.11 0

Time(s)

20

40

60

40

60

Time(s)

0.4

0.3

10

0 0

10

20

30

40

Time(s)

Fig. 9. Yaw angle estimate

50

60

Error__NN(deg)

20

Error__CF(deg)

Yaw(deg)

40

20

Time(s)

Error__CF(deg)

Roll(degree)

20 25.4

NN

40

30

20.5

Error__ACC(deg)

True

Error__GYR(deg)

60

0.3 0.2 0.1 0

20

Time(s)

40

60

0.2 0.1 0 0

20

Time(s)

Fig. 10. Yaw angle error

508

D. Dhahbane et al.

The parameters of IMU given in Table 1 are taken from Xsense MTi 1-series Datasheet [27].

Best Training Performance is 0.10944 at epoch 834 10

4

Train Test

Mean Squared Error (mse)

Best

10

2

10

0

10

-2

0

200

100

300

400

500

700

600

900

800

1000

1000 Epochs

Fig. 11. Training performance

Error Histogram with 20 Bins Training Test

50

Zero Error

30

20

10

Errors = Targets - Outputs

Fig. 12. Training error histogram

0.8235

0.9609

0.6861

0.4112

0.5487

0.2738

0.1364

-0.1384

-0.00099

-0.2758

-0.4132

-0.5506

-0.8255

-0.6881

-1.1

-0.9629

-1.238

-1.375

-1.65

0

-1.513

Instances

40

Neural Network-Based Attitude Estimation Test: R=0.99969

Training: R=0.99924 Data

25

Data

25

Fit

Fit

Y=T

Y=T

20

Output ~= 1*Target + -0.062

Output ~= 1*Target + 0.026

509

15

10

5

5

10

20

15

25

20

15

10

5

5

Target

10

15

20

25

Target

All: R=0.9993 Data

25

Fit

Output ~= 1*Target + 0.012

Y=T

20

15

10

5

5

10

20

15

25

Target

Fig. 13. Training regression

The angle estimates by INS, complementary filter and neural network for Roll, Pitch and Yaw are shown in Fig. 5, Fig. 7, Fig. 9 respectively. In Fig. 6, Fig. 8, Fig. 10, the errors of angle estimates (Roll, Pitch and yaw) are illustrated. In Fig. 11, Fig. 12 and Fig. 13, we present the simulation results of the neural network training (training performance, the training error histogram and the training regression). Discussion of the Simulation Results From the simulation results, it can be seen that the proposed algorithm based on neural network has demonstrated its estimation capability in presence of associated noise to the IMU. Comparing with IMU and complementary filter, the neural network method provides high accuracy in angles estimation (cleary shown in Fig. 5, Fig. 7, Fig. 9). By comparing the estimation error, presented by Fig. 6, Fig. 8, Fig. 10, it can be concluded that the neural network approach exhibits the best behavior in convergence to the real value (about 0.1° in Roll, 0.5° in Pitch, 0.1° in Yaw) in average value. According to Fig. 11, Fig. 12 and Fig. 13, the Bayesian Regularization Backpropagation training algorithm has given satisfactory results. It presents a value of 0.10944 on RMS, with good matching with the training data (R = 0.99924), the test data (R = 0.99969) and the whole data (R = 0.9993).

510

D. Dhahbane et al.

6 Conclusion The deep learning method based on neural networks is a complex nonlinear architecture, used in several applications. In this work, we have proposed this algorithm to enhance the estimation of Euler angles (Roll, Pitch and Yaw). The Bayesian Regularization Algorithm is used to train the network. This last can solve the overfitting problem effectively and generates an optimal set of weights by the approach of probability distribution of neural weights. Comparing to IMU and the complementary filter, simulation results given by the proposed estimator have demonstrated the efficiency of the neural network approach and its high performance in terms of estimation accuracy and noise reduction.

References 1. Fourati, H.: Multisensor Attitude Estimation: Fundamental Concepts and Applications, 1st edn. CRC Press (2016). https://doi.org/10.1201/9781315368795 2. Siciliano, B., Khatib, O. (eds.): Springer Handbook of Robotics. Springer, Cham (2016). https://doi.org/10.1007/978-3-319-32552-1 3. Duong, D.Q., Nguyen, T.P., Sun, J., Luo, L.: Attitude Estimation by Using MEMS IMU with Fuzzy Tuned Complementary Filter (2016) 4. Malmstrom, J.: Robust Navigation with GPS/INS and Adaptive Beamforming. Swedish Defence Research Agency System Technology Division SE-172 90 STOCKHOLM Sweden (2003) 5. Benziane, L., Hadri, A.E., Seba, A., Benallegue, A., Chitour, Y.: Attitude Estimation and Control Using Linear-Like Complementary Filters: Theory and Experiment. arXiv150302718 Math 2015 6. Allgeuer, P., Behnke, S.: Robust Sensor Fusion for Robot Attitude Estimation. arXiv180910669 Cs (2018) 7. Wu, J., Zhou, Z., Chen, J., Fourati, H., Li, R.: Fast complementary filter for attitude estimation using low-cost MARG sensors. IEEE Sens. J. 16, 6997–7007 (2016). https://doi. org/10.1109/JSEN.2016.2589660 8. Islam, T., Islam, Md.S., Shajid-Ul-Mahmud, Md., Hossam-E-Haider, M.: Comparison of complementary and Kalman filter based data fusion for attitude heading reference system. Presented at the Proceedings of the 1st International Conference on Mechanical Engineering and Applied Science (ICMEAS 2017), Dhaka, Bangladesh, p. 020002 (2017). https://doi. org/10.1063/1.5018520 9. Al Hussein Technical University, Amman, 11831, Jordan, Al-Fahoum, A.S., Abadir, M.S.: Design of a modified madgwick filter for quaternion-based orientation estimation using AHRS. Int. J. Comput. Electr. Eng. 10, 174–186 (2018). https://doi.org/10.17706/IJCEE. 2018.10.3.174-186 10. Farrell, J.: Aided Navigation: GPS with High Rate Sensors, Electronic Engineering. McGraw-Hill, New York (2008) 11. Al-Sharman, M.K., Zweiri, Y., Jaradat, M.A.K., Al-Husari, R., Gan, D., Seneviratne, L.D.: Deep-learning-based neural network training for state estimation enhancement: application to attitude estimation. IEEE Trans. Instrum. Meas. 69, 24–34 (2020). https://doi.org/10.1109/ TIM.2019.2895495

Neural Network-Based Attitude Estimation

511

12. Brossard, M., Bonnabel, S., Barrau, A.: Denoising IMU Gyroscopes with Deep Learning for Open-Loop Attitude Estimation. arXiv200210718 Cs Stat (2020) 13. Siouris, G.M.: Missile Guidance and Control Systems. Springer, New York (2004) 14. Quan, W., Li, J., Gong, X., Fang, J.: INS/CNS/GNSS Integrated Navigation Technology. Springer, Heidelberg (2015). https://doi.org/10.1007/978-3-662-45159-5 15. Narkhede, P., Joseph Raj, A.N., Kumar, V., Karar, V., Poddar, S.: Least square estimationbased adaptive complimentary filter for attitude estimation. Trans. Inst. Meas. Control 41, 235–245 (2019). https://doi.org/10.1177/0142331218755234 16. Sakhi, S.: Centrale d’Acquisition Temps Réel pour le Trace d’Engins Mobiles (2013) 17. Yoo, T.S., Hong, S.K., Yoon, H.M., Park, S.: Gain-scheduled complementary filter design for a MEMS based attitude and heading reference system. Sensors 11, 3816–3830 (2011). https://doi.org/10.3390/s110403816 18. Kumar, N.V.: Integration of Inertial Navigation System and Global Positioning System Using Kalman Filtering. Department of Aerospace Engineering Indian Institute of Technology, Bombay Mumbai (2004) 19. Wu, J., Zhou, Z., Fourati, H., Li, R., Liu, M.: Generalized linear quaternion complementary filter for attitude estimation from multisensor observations: an optimization approach. IEEE Trans. Autom. Sci. Eng. 16, 1330–1343 (2019). https://doi.org/10.1109/TASE.2018. 2888908 20. Keller, J., Liu, D., Fogel, D.: Fundamentals of Computational Intelligence Neural Networks, Fuzzy Systems and Evolutionary Computation. Wiley, Hoboken (2016) 21. Aggarwal, C.C.: Neural Networks and Deep Learning: A Textbook. Springer, Cham (2018). https://doi.org/10.1007/978-3-319-94463-0 22. Fourati, H.: Multisensor Data Fusion From Algorithms and Architectural Design to Applications. Department of Control Systems University Grenoble Alpes Grenoble, France (2016) 23. Engelbrecht, A.P.: Computational Intelligence An Introduction Second Edition, John Wiley & Sons Ltd, The Atrium, Southern Gate, Chichester, West Sussex PO19 8SQ. ed. England (2007) 24. Sumathi, S., Paneerselvam, S.: Computational Intelligence Paradigms: Theory & Applications using MATLAB 844 (2010) 25. Kaur, H., Salaria, D.S.: Bayesian Regularization Based Neural Network Tool for Software Effort Estimation (2013) 26. Kayri, M.: Predictive abilities of bayesian regularization and levenberg–marquardt algorithms in artificial neural networks: a comparative empirical study on social data. Math. Comput. Appl. 21, 20 (2016). https://doi.org/10.3390/mca21020020 27. https://www.xsens.com/hubfs/Downloads/usermanual/MTi_usermanual.pdf

Developing an Improved ANN Algorithm Assisted by a Colony of Foraging Ants for MPP Tracking of Grid Interactive Solar Powered Arc Welding Machine S. Kahla1, B. Babes1(&), N. Hamouda1, A. Boutaghane1, and A. Bouafassa2 1

2

Research Center, Industrial Technologies (CRTI), Algiers, Algeria [email protected] Department of Electronics, Electrical Engineering and Automation, National Polytechnic School of Constantine, Constantine, Algeria [email protected]

Abstract. In this article, a metaheuristic optimized multilayer Feed-forward Artificial Neural Network (ANN) controller is proposed to extract the maximum power from available solar energy for a three-phase shunt active power filter (APF) grid connected photovoltaic (PV) system supplying an arc welding machine. Firstly, in order to improve the maximum power point (MPP) delivered by PV arrays and to overcome the drawbacks in the Incremental Conductance (INC) method, a hybrid MPPT controller is designed. The proposed approach abbreviated as ANN-ACO MPPT controller is based on an ant colony optimization (ACO) algorithm which is useful to train the developed ANN and to evolve the connection weights and biases to get the optimal values of duty cycle converter corresponding to the MPP of PV array. Secondly, aiming to meet the various grid requirements such as power quality (PQ) improvement, distortion free signals etc., a shunt APF is utilized, and a direct power control (DPC) is designed for distributing the solar energy between the DC-link capacitor, arc welding machine and the AC grid. Finally, the performance of proposed control system is confirmed by simulation tests on a 12.2 kW PV system. The simulation results demonstrate that the deigned ANN-ACO MPPT controller can provide a better MPP tracking with a faster speed and a high robustness with a minimal steady-state oscillation than those obtained with the conventional INC method. Keywords: Solar photovoltaic (PV) system
DC/DC boost converter
Threephase shunt APF
Hybrid MPPT control
Multilayer Feed-forward artificial neural network (ANN)
Ant colony optimization (ACO) algorithm
Arc welding machine
Total harmonic distortion

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 512–520, 2021. https://doi.org/10.1007/978-3-030-63846-7_48

Developing an Improved ANN Algorithm

513

1 Introduction Arc welding process has been widely utilized in manufacturing technologies and it gained more importance in industries [1]. Arc welding power supply (AWPS) is a device that provides an electric current to perform welding. The main task of a modern arc welding is to develop an economical and efficient AWPS with unity power factor (PF) [2], which process the limit for the input harmonic current emissions. Different topologies based-AWPS have been introduced to meet the PQ requirements and the stringent international restrictions, such as EN 61000-3-2 and IEEE 519-1992. Generally, these topologies cans be classified into three broad categories; passive filters, active power filters (APFs) and power factor corrections (PFCs). The purpose of these topologies is to make the input current a pure sinusoidal waveform, so as to reduce the THD. The use of passive filters at the input side reduces the THD of the input line current, but the PF at the input AC-side remains low. In addition, the losses as well the size of the AWPS also get augmented, which is accentuated for higher power ratings. Boost circuit based AWPS has been investigated by Casanueva et al. [3] to get PF correction at the input AC-side. Diode bridge rectifier being a rich source of harmonics, is unfit to be utilized in any system such as arc welding, where stringent international power quality standard norms are to be followed. It has also been evolved from the existing literature that so far a 3-phase shunt (APFs) based AWPS has not been considered by the researchers. Thus, an effort has been made in this article to decrease the source current harmonics and to achieve a high PF at the PCC via a 3-phase shunt APF circuit at the front end of the arc welding machine. Shunt APF [4] is a shunt coupled device effectively for mitigating power quality issues related to AWPSs along with reduction in the rating. Usage of such devices for AWPS makes it possible to ensure the power supply electromagnetic compatibility with the power grid and to improve its energy efficiency. In this way, the purpose of this paper can be formulated as to develop a novel topology of AWPS associated with a 3-phase shunt APF, which is utilizes Direct Power Control (DPC) to achieve both harmonic and reactive power compensations. The DPC method has received a wide interest due to its simplicity and very fast power dynamic response [5]. With this control strategy, the considered shunt APF is able of providing fast corrective action, even with dynamically changing. In recent years, it has been marked that grid integrated PV system has proved to be prominent in providing continuity of power supply under serious power quality problems of the power grid [6]. With respect to increasing rate of distributed generation penetration, this work also aims to investigate the application of shunt APF in arc welding machine associated with a distribution network that employs solar photovoltaic (SPV) systems as the source of electrical power. To date, various methods have been developed and implemented in literature to regulate the duty cycle of the DC/DC boost converter for maximum power point tracking (MPPT), such as Hill Climbing (HC) technique [7], Perturb-and-Observe (P&O) technique [6], Incremental Conductance (INC) technique [8], constant or fractional current/voltage [9], Fuzzy Logic (FL) technique [10], Artificial Neural Network (ANN) technique [11], etc. Unfortunately, the majority of these MPPT algorithms lack accurate convergence analysis, and thus, only approximate MPPT is achieved. Although ANN algorithms can offer better

514

S. Kahla et al.

MPPT convergence than conventional and FLC algorithms, ANN algorithms need large training data to track strict optimal power under adverse operating conditions. Numerous researchers are interested to reduce the data dimension. The transformation and reduction of the redundant or irrelevant training data may shorten the running time and yield more generalized results [12]. Most training algorithm of ANNs, e.g., the back-propagation (BP) technique, is based on gradient descent algorithm and has been successfully applied to train ANNs. However, the BP method has restriction since it often gets trapped in local optimum of error function, because it is a local optimization algorithm [12]. One way to overcome this restriction is to adopt evolutionary ANNs (EANNs), i.e., to adjust the value of the connection weights and biases of ANNs, in which global optimization algorithms are employed to find near to the global optimum combinations for the connection weights unlike the BP algorithm in which local optimum. Motivated by the aforementioned weaknesses of existing MPPT methods in the literature and training algorithms of ANNs, this study aims to formulate a new hybrid MPPT control methodology for the solar PV generator in order to overcome the above limitations. This is based on combining intelligent ANN system and ant colony optimization (ACO) algorithm, herein named ACO-ANN MPPT controller. To the best of the author’s knowledge, the feasibility of the proposed ACO-ANN MPPT controller is investigated firstly to the solar PV system MPPT control issue of thee-phase shunt APF supplying arc welding machine.

2 Overall System Configuration Figure 1 depicts the circuit topology of the proposed system that contains of a solar PV generator connected to a three-phase shunt APF for interfacing to the AC grid via an energy storage DC-link capacitor (Cdc) in order to satisfy the high requirement of the arc welding machine.

Fig. 1. System configuration employed with the hybrid MPPT algorithm.

Developing an Improved ANN Algorithm

515

3 Control Strategies 3.1

Control Scheme Design for DC/DC Boost Converter

In the present study, a multilayer feed-forward ANN with ACO learning algorithm was developed and utilized to evaluate and achieve MPP of a PV array. According to the analysis in introduction section, two inputs of the ANN were the PV array voltage and current. The output of the ANN was the duty cycle of the DC/DC boost converter (Fig. 2). The artificial neural network toolbox developed by MATLAB software was utilized to build the considered ANN model of (2-20-1) neurons. The transfer function between the input layer and the hidden layer was set to be log-sigmoid function ‘logsig’, while the transfer functions between the hidden layer and the output layer were set be linear function ‘purelin’.

Fig. 2. Optimal ANN structure for MPPT control.

3.2

Control Scheme Design for Three-Phase Shunt-APF

The purpose of the proposed system is to supply adequate power to the arc welding machine, and to transfer the excess power to the utility with a high quality current. To guarantee these conditions, the 3-phase shunt APF is appropriately controlled in order to ensure the injection of the power in the grid and the active power filtering to eliminate harmonic and reactive currents introduced by the arc welding machine. In this aim, the DPC algorithm of the 3-phase shunt APF is used as depicted in Fig. 3 below. The instantaneous active and reactive powers of the shunt APF indicated as Pg and Qg, respectively, are computed in the synchronous frame as follows: 8  2
> Vgd Igd þ Vgq Igq < Pg ¼ 3 > : Q ¼ 2
V I þ V I  g gq gd gd gq 3

ð1Þ

where Vgd and Vgq are the d-coordinate and q-coordinate of the grid-side voltage, respectively. Igd and Igq are d-coordinate and q-coordinate of the grid-side current,

516

S. Kahla et al.

Fig. 3. Control algorithm for three-phase SAPF system using DPC algorithm [5].

respectively. The relation between the inverter vector (Vk), inverter vector position (hk), grid vector and inverter power change can be defined as follows: 8 h i < dPg ¼ 3 V 2  V Vk cosðhg  hk Þ gd 2Lg dt gd ð2Þ : dQg ¼ 3 V Vk sinðhg  hk Þ dt

2Lg

gd

The other parameter to choose the next vector from the switching table is the sector of the present voltage vector. The hexagonal space vector is divided to 12 sectors as expressed in the following equation: ðn  2Þ  p=6\hn \ðn  1Þ  p=6; n ¼ 1; 2; . . .; 12

ð3Þ

Depending on the grid vector position, the corresponding sector can be estimated from  p Sector ¼ floor hg = þ 1 6

ð4Þ

The DC-link voltage (Vdc) must be maintained constant by a conventional PI controller at the reference value. The relation of the reference active power (P*g) absorbed by the shunt APF and Vdc can be described in the following equation: Pg ðkÞ ¼ Pg ðk  1Þ þ Kp1 fVer ðkÞ  Ver ðk  1Þg þ Ki1 Ver ðkÞ

ð5Þ

Developing an Improved ANN Algorithm

3.3

517

Control Scheme Design for Arc Welding Machine

Figure 4 illustrates system configuration of a 3-phase arc welding machine using fullbridge isolated DC/DC converter. On one side, the proposed arc welding machine is desired to maintain constant arc voltage during the rated and light load conditions. On the other side, it must adjust the welding current during severe overloading situation to guarantee excellent weld bead. Owing to this, dual loop control scheme is utilized for the isolated full-bridge buck converter as illustrated in Fig. 4 below.

Fig. 4. Block diagram of control scheme for arc welding machine.

4 Simulations and Discussions The performance of the proposed system is evaluated in both transient and steady-state conditions by simulating this system in MATLAB/Simulink software under the welding load changing and the variations of the solar irradiation (G). 4.1

PV System Performance Under MPPT Mode

Figure 5 presents the waveforms of solar irradiation (G), PV output power (Ppv), PV current (Ipv), and PV voltage (Vpv) at MPP, which is being tracked by the proposed ACO-ANN MPPT controller at constant temperature (T) and different irradiation conditions (G). As illustrated in Fig. 5, the proposed ACO-ANN tracker showed a good performance tracking than the traditional INC tracker. 4.2

Global System Performance Under Steady State Operation

The steady-state operation when the welding load is constant, and is supplied power by utility only, is illustrated in Fig. 6. The waveforms of the instantaneous parameters i.e. DC-link voltage (Vdc), active power (Pg), reactive power (Qg), load power (Pload), welding current (Iw), arc voltage (Vw), load current (IL) and grid current (Ig) can be seen in Fig. 6. It can be renowned that PV system is not available (G = 0 W/m2) in this test. The measured THD of the Ig is negligible as clearly presented in Fig. 6.

518

S. Kahla et al.

Fig. 5. PV array output by proposed ACO-ANN tracker and conventional INC method.

Fig. 6. Performance of the proposed system when the PV system is not available (G = 0 W/m2).

4.3

Global System Performance Under Welding Load Step-Change

The simulated response of the system under step-change in the welding load resistance at 0.3 s is presented in Fig. 7. The irradiation (G) is kept constant at 500 W/m2. It is observed that the shunt APF compensates for the harmonic currents (IF) and reactive power drawn by the arc welding machine (Qload) and thus it maintains the grid current (Ig) sinusoidal at unity power factor (PF). It can be noticed, that the Vdc is regulated at its preferred value V*dc = 600 V. It can be seen that the regulation of the welding current (Iw) of arc welding machine is accomplished successfully in a transient state. 4.4

Global System Performance Under Varying Solar Irradiation

The transient response of the overall operating waveforms of the proposed system under gradually decreasing in solar irradiation with constant welding load is given in Fig. 8. In Fig. 8, at time t = 0.3 s, the system is subjected to a step-change in the solar irradiation from 1000 to 500 W/m2. The DC-link voltage can be kept constant at 600 V, although the solar irradiation is decreased from 1000 to 500 W/m2. As seen

Developing an Improved ANN Algorithm

519

Fig. 7. Performance of the proposed system under step-change in welding load.

Fig. 8. Performance of the proposed system under step-change in solar irradiation.

from Fig. 8, during irradiation change, the PV power delivered by PV array decreases, thus the shunt APF current (IF) as well as the grid current (Ig) decreases. In addition to this, the harmonics contained in the grid current (Ig) can be cancelled successfully, as is obviously illustrated in the Ig’s waveforms.

5 Conclusions A comprehensive investigation on a new configuration of grid interfaced PV power generating system via a 3-phase shunt APF for high-strength arc welding machine has been performed throughout this article. The performance of a proposed system has been investigated for arc welding reactive power compensation and harmonic elimination under unity power factor operation along with MPPT control. The MPP of solar PV array has been achieved using the proposed ACO-ANN MPPT method. Moreover, the proposed DPC algorithm of shunt APF inject the solar power delivered by PV array into the AC grid with high grid current quality and in conformity with international normal (IEEE-519) in all irradiance variations levels.

520

S. Kahla et al.

References 1. Hamouda, N., Babes, B., Boutaghane, A.: Design and analysis of robust nonlinear synergetic controller for a PMDC motor driven wire-feeder system (WFS). In: Proceedings of the 4th International Conference on Electrical Engineering and Control Applications (ICEECA 2019), 19–21 November 2019, Constantine, Algeria (2019) 2. Narula, S., Singh, B., Bhuvaneswari, G., Pandey, R.: Improved power quality bridgeless converter-based SMPS for arc welding. IEEE Trans. Ind. Electron. 64(1), 275–284 (2017) 3. Kim, Y.S., Sung, W.Y., Lee, B.K.: Comparative performance analysis of high density and efficiency PFC topologies. IEEE Trans. Power Electron. 29(6), 2666–2679 (2014) 4. Ferhat, M., Rahmani, L., Babes, B.: DSP-based implementation of improved deadbeat control for three-phase shunt active power flters. J. Power Electron. 20, 188–197 (2020). https://doi.org/10.1007/s43236-019-00029-y 5. Aissa, O., Moulahoum, S., Colak, I., Babes, B.: Analysis, design and real-time implementation of shunt active power filter for power quality improvement based on predictive direct power control. In: Proceedings of the 5th International Conference on Renewable Energy Research and Applications, 20–23 Nov 2016, Birmingham, UK (2016) 6. Afghoul, H., Krim, F., Beddar, A., Babes, B. Real-time implementation of robust controller for PV emulator supplied shunt active power filter. In: Proceedings of the 6th International Renewable and Sustainable Energy Conference, Rabat, Morocco (2018) 7. Ali Akbar, G., Seyed Mohammad, S., Asma, S.: A high performance maximum power point tracker for PV systems. Elect. Power Energy Syst. 53, 237–243 (2013) 8. Zoua, Y., Yua, Y., Zhangb, Y., Lu, J.: MPPT control for PV generation system based on an improved IncCond algorithm. Procedia Eng. 29, 105–109 (2012) 9. Sher, H.A., et al.: A new sensorless hybrid MPPT algorithm based on fractional short-circuit current measurement and P&O MPPT. IEEE Trans. Sustain. Energy 6(4), 1426–1434 (2015) 10. Tang, S., Sun, Y., Chen, Y., Zhao, Y., Yang, Y., Szeto, W.: An enhanced MPPT method combining fractional-order and fuzzy logic control. IEEE J. Photovol. 7(2), 640–650 (2017) 11. Al-Majidi, S.D., Abbod, F.M., Raweshidy, H.S.: A particle swarm optimization-trained feedforward neural network for predicting the maximum power point of a photovoltaic array. Eng. Appl. Artif. Intell. 92, 103688 (2020) 12. Whitley, D., Starkweather, T., Bogart, C.: Genetic algorithms and neural networks: optimizing connections and connectivity. Parallel Comput. 14(3), 347–361 (1990)

Optimal Siting and Sizing of DG Units Using a Decomposition Based Multiobjective Evolutionary Algorithm Yaaqoub Latreche1(&), Houssem R. E. H. Bouchekara2, Muhammad S. Javaid2, Mohammad S. Shahriar2, Yusuf A. Sha’aban2, and Fouad Kerrour1 1

Modeling of Energy Renewable Devices and Nano-Metric (MoDERNa) Laboratory, University of Freres Mentouri Constantine, 25000 Constantine, Algeria [email protected] 2 Department of Electrical Engineering, University of Hafr al Batin, Hafr al Batin 31991, Saudi Arabia

Abstract. The Optimal Incorporation Distributed Generation (OIDG) problem can be treated as a single objective or multi-objective optimization problem. The advantage of the second option is to give a set of solutions in one run while optimizing more than one objective at the time. This paper aims to investigate the OIDG for single and multiple units in radial distribution systems using a Decomposition Based Multiobjective Evolutionary Algorithm. Seven case studies have been investigated considering three objectives which are active power losses minimization, voltage stability enhancement, and voltage profile improvement. All these case studies are tested on the 33-bus radial system. Keywords: Distributed generation
Distribution systems
Multi-objective optimization
Power losses
Voltage profile
Voltage stability

1 Introduction In recent years, vertically integrated power systems are being deregulated to compensate for the increasing demand for electrical power. The concept of distributed generation (DG) focuses on placing the distributed resources around the load center, which is not feasible for the case of conventional power plants [1]. The key characteristics of a power system that include speedy load growth, the flexibility of power transmission, improvement of power quality, reduction of loss, environmental impact, transmission congestion, and distribution reliability are directly linked with successful incorporation of DGs [2, 3]. However, the placement and sizing of critical assets play an essential role in the penetration of DGs. Non-optimal installation and sizing of DGs could also lead to several issues such as improper relay operation, fall of voltage profile, operational instability, loss of power quality, and more. Therefore, the optimization of the siting and sizing of DG units is receiving considerable research attention in the field of modern-day power systems [4]. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 521–532, 2021. https://doi.org/10.1007/978-3-030-63846-7_49

522

Y. Latreche et al.

A reasonable amount of works has been reported in the literature on the optimization of DG units, both with single and multiple objective functions [4–9]. Moreover, a wide range of methods and tools are tried to solve such problems like analytical techniques (e.g., Eigen-value based analysis, index method, sensitivity-based analysis, classical optimization techniques (e.g. linear programming, dynamic programming, sequential quadratic programming, mixed integer nonlinear programming and metaheuristic artificial intelligent methods. Within the meta-heuristic methods, genetic algorithm (GA), particle swarm optimization (PSO), fuzzy logic, ant colony search (ACS) algorithm, artificial bee colony (ABC), and several others are used to improve the voltage profile, stability, reliability of the power network. All the mentioned approaches, along with the corresponding target areas, are well summarized in [4, 5]. A multi-objective approach is dealt with weight algorithm in [6] to increase the power quality of the power system and considered the DG uncertainties. The double trade-off method is tried in a multi-objective problem of minimizing the network cost and maximizing the power quality in [7]. Similar work is reported in [8], where Monte Carlo simulation-based approach is used to reduce the cost, technical risk, and economic risk of a DG oriented power network. In [9], an easy and quick approach of optimization, by forming the priority list of more demanding bus locations for unit installation, was proposed. This paper proposes a decomposition-based multi-objective evolutionary algorithm (MOEA/D) for the optimal positioning and sizing of the DG units of a power system. The considered algorithm has been significantly successful in several application fields and therefore, chosen for DG optimization in this paper. To the best of our knowledge, MOEA/D has not been applied to solve the considered problem, before now. The maximum power providing capacity is considered as the size of the DG unit, whereas the bus location represents the siting. The performance of the proposed approach is tested through multiple indices to ensure the reduction of active power loss, voltage profile improvement, voltage stability enhancement, etc. The remaining paper is structured in the following way: Section 2 discusses the formulation of the optimization problem and associated constraints. The performance indicators were also discussed in this section. Section 3 describes the MOEA in detail, and the results are presented in Sect. 4. Finally, Sect. 5 concludes the paper.

2 Formulation of the OIDG Problem The optimization formulation presented in this paper aims at finding the optimal location and size of DGs in a power system such that the system shows superior performance in terms of specific evaluative indices, discussed in this section. Explicitly, the size of a DG refers to its maximum power supply capability; whereby, only the real power is considered as the design variable, and the reactive power is then calculated using the power factor of the bus of interest. The location - second design variable - is represented by the bus number. In the case of multiple DGs, the formulation takes into consideration that multiple DGs are not connected on the same bus. In this work, the OIDG problem is addressed by formulating a multi-objective optimization function. The formulation will attempt to minimize two of the three

Optimal Siting and Sizing of DG Units

523

objectives at a time, namely: power losses, voltage stability index, and voltage profile. These three objectives are discussed below as performance indices. Generally, any multi-objective optimization problem (MOP) can be mathematically represented as: minimize F ð xÞ ¼ ½f1 ð xÞ; . . .; fm ð xÞT

subject to x 2 X

ð1Þ

where: X is the decision variable space, fi represent the ith objective function. For the OIDG problem considered in this work, the specific mathematical formulation is given as: FðxÞ ¼ ½Plosses ; VSI; VPT

Minimize

x ¼ ½PDG ; DGlocation 

where Subject to

Pgi ¼ Pi þ Vi

nb X

  Vj Yij cos di  dj  hij

ð2Þ ð3Þ ð4Þ

j¼1

Qgi ¼ Qi þ Vi

nb X

  Vj Yij sin di  dj  hij

ð5Þ

j¼1

Vmin  Vi  Vmax   ; i ¼ 1; . . .; N jSi j  Smax i SDGi  0

NDG X

ðDGPL  Sload Þ NDG PDGk 

X

PLoad

ð6Þ ð7Þ ð8Þ

ð9Þ

k¼1

2  DGlocation  N

ð10Þ

DGlocation1 6¼ DGlocation2 6¼ . . . 6¼ DGlocationNDG

ð11Þ

where FðxÞ is the vector of the objective function, x ¼ ½x1 ; . . .; xn  is the vector of n design variables, PDG is a vector containing the sizing capacity of all DGs, DGlocation is a vector enlisting the number of buses at which DGs are to be optimally placed, Pgi is generator active power output at bus ni, Qgi is generator reactive power output at bus ni, Pi is active power demand at bus ni , Qi is reactive power demand at bus ni; Vi is the voltage of bus ni; di is phase voltage angle at bus ni; hij is the power factor angle, Vmin and Vmax represent the minimum and maximum allowable voltages at bus ni , respectively, Si is the apparent power at branch i and Smax is the maximum apparent power at i branch i, N ð¼ ni  1Þ is the total number of branches in the given radial distribution system, N represents the total number of buses in a given radial distribution system,

524

Y. Latreche et al.

NDG is the total number of DG units incorporated in the power system, DGPL represents maximum penetration level as a percentage of the peak load of the system, Pload and Sload are denoting total active and apparent power loads of the system, respectively and penetration level ðDGPL Þ. This optimization problem is characterized by both equality and inequality constraints. Equality constraints are imposed by the conservation of real power (3) and reactive power (4) in the system. In contrast, the voltage limits (5), thermal limit (6), maximum capacity (7), demand balance (8), and the number of buses (9) enforce the inequality constraints. Equation (10) accounts for the condition so that multiple DGs cannot be placed on the same bus. By formulating the optimization problem with the three-dimensional multiobjective function, the system is optimized to perform well on these three indices: active losses reduction (ALR) index, voltage stability enhancement index (VSIM), and voltage profile improvement index (VPI). The active losses reduction is set as follows: ALR ¼

PLosses0  PLossesDG % PLosses0

ð12Þ

In (11), subscript 0 denotes the respective quantity before the installation of DGs. If Rk and Ik represent resistance and the current passing through branch k in total N branches, PLosses will be given by: PLosses ¼

N X

  Rk Ik2 

ð13Þ

k¼1

Similarly, VSIM is given by: VSIM ¼

VSIDG  VSI0 % VSI0

ð14Þ

Here VSI, voltage stability index, is be expressed as [10]: 

1 VSI ¼ min SI ðniÞ

 ni ¼ 2; 3; . . .; nb

SI ðniÞ ¼ jVmi j4 4½Pni ðniÞRni þ Qni ðniÞXni jVmi j4 4½Pni ðniÞXni þ Qni ðniÞRni 2

ð15Þ ð16Þ

where Pni ðni Þ, Qni ðni Þ denote real and reactive power load, respectively fed through bus ni, Vmi is the voltage of bus mi, Rni and Xni is the resistance and reactance of branch i. Finally, VPI is given by: VPI ¼

VP0  VPDG % VP0

ð17Þ

where VP denotes the voltage profile and is given by the squared difference of rated voltage, Vrated (1 p.u.), with the voltage Vi at bus ni.

Optimal Siting and Sizing of DG Units

VP ¼

nb X

ðVi  Vrated Þ2

525

ð18Þ

i¼1

In addition to these indices, the formulation also takes into account the number of buses violating the upper voltage limit (NBVV), as a performance index which is simply the count of those buses where the upper limit of allowed voltage is not respected.

3 Problem Solution Using MOEA/D In MOEA/D [11], a multi-objective optimization problem (MOP) is decomposed into N objective functions by defining a pareto front (PF). MOEA/D solves these objective functions simultaneously, in a single step, by exploring trade-offs among the set of optimal vectors that make up the PF. The decomposition method affects the nature of the solution of the MOP. As such, several methods have been developed for the decomposition of MOP. Some commonly used decomposition methods include the penalty-based intersection (PBI) approach and the Tchebycheff approach. The optimal solution for MOP using the PBI approach is known to have proper distribution and convergence. On the other hand, solutions using the Tchebycheff approach has a less uniform distribution as compared to the PBI methods but offers the additional advantage of robustness to the shapes of the PF. Therefore, the Tchebycheff method is well known for its ability to handle MOPs with non-convex PFs. Additionally, the aggregation function for a continuous MOP solved using the Tchebycheff method is hardly smooth. For disparately scaled objectives, solutions are typically improved by normalization. Consider the MOP (1). Then the Tchebycheff decomposition approach is mathematically expressed as follows: minimize gte ðxjk; z Þ ¼ max

1im

   ki fi ð xÞ  zi  subject to x 2 X

ð19Þ

z is where  the reference point consisting of the smallest value for each objective,  z1 ; . . .; zm i.e. zi ¼ minffi ð xÞjx 2 Xg for each i ¼ 1; . . .m: Normalization can be achieved by replacing each of the objectives f i ði ¼ 1; 2; . . .; mÞ with f i :  f i ¼ fi  zi znad  zi i

ð20Þ

where znad ¼ maxffi ð xÞjx 2 PSg such that the nadir point in the objective space, znad ¼  nad i nad T highlights the upper bound of the PF. z1 ; . . .; zm The non-uniform distribution of optimal solutions, in the Tchebycheff approach, over the PF is due to the nonlinear relationship between the weight vector and the direction of its optimal solution. Therefore, the modified Tchebycheff method was proposed [12, 13] to eliminate the inconsistency between the weight vector of the

526

Y. Latreche et al.

subproblem and the direction of its optimal solution. The modified Tchebycheff approach is given as [13]: 

j

g xjk ; z te





¼ min

( ) fi ð xÞ  z  i

1im

kij

ð21Þ

Where, kj is a direction vector as well as a weight vector. Note that the division by a 0 when kij ¼ 0; i ¼ 1; . . .m can be addressed by using a small number (e.g. 106 Þ instead of zero. Using the modified Tchebycheff approach to decompose the MOP of interest, only the latest set of solutions to neighboring subproblems of the reference point z are utilized to optimize any subproblem. Moreover, the population comprises of the best solution for each subproblem. To minimize each subproblem, MOEA/D uses information from neighboring subproblems. The neighborhood subproblems for each subproblem are defined as a set of the closest subproblems having the nearest weight vectors to it using the Euclidean distance. The detailed algorithm for MOEA/D is available from previous works [11, 13]. At each iteration in MOEA/D, the following information is maintained [13]: 1. The population of the best solution to the N subproblems obtained so far, x1 ; . . .; xN . 2. The population of the objective space, FV i ; i ¼ 1; . . .N corresponding to the solutions xi : FV i ¼ F ðxi Þ 3. The reference point z ¼ ðz1 ; . . .zm Þ, corresponding to the optimal values of objectives obtained so far 4. The set of weight vectors for each of the N scalar subproblems ki for i ¼ 1; . . .N .

4 Application and Results In this paper, seven case studies have been investigated using the 33-bus radial system. These cases are summarized in Table 1. CASE 1, is the base case needed for performance comparison. Cases 1 to 4 investigate the incorporation of one DG unit while cases from 5 to 7 investigate the incorporation of two DG units. In cases 2 and 5, PLosses and VSI are optimized. In cases 3 and 6 PLosses VP are optimized and in cases 4 and 7 VSI and VP are optimized. In all cases, the DG units are assumed to be supplying active power only. The DGPL is assumed to be 1 for the first four cases, and 0.5 for the last three cases. The main characteristics of the 33-bus radial system are given in Table 2, while its single line diagrams are shown in Fig. 1.

Optimal Siting and Sizing of DG Units Table 1. Summary of the studied cases. Case Case Case Case Case Case Case Case

1 2 3 4 5 6 7

Objective functions Base case PLosses +VSI PLosses +VP VSI+VP PLosses +VSI PLosses +VP VSI+VP

No. DG DGPL – – 1 1 1 1 1 1 2 0.5 2 0.5 2 0.5

Test radial system 33-bus 33-bus 33-bus 33-bus 33-bus 33-bus 33-bus

Table 2. The main characteristics of the investigated test systems. System characteristics Buses Branches Total active load (MW) Total reactive load (Mvar) PLosses (MW) QLosses (Mvar) VSI VP Vmax Vmin Smax (MVA)

Data is given in [14] 33 32 3.7150 2.3000 0.2110 0.1430 1 0.1338 1.1 0.95 5

Fig. 1. Single line diagram of the 33-bus RTS

527

528

Y. Latreche et al.

5 Results and Discussion The developed program has been implemented using commercial MATLAB software (version R2015a). The simulation runs were performed using the proposed MOE/D approach with n ¼ 100, and a maximum of 200 iterations. The PFs obtained for different cases are plotted in Fig. 2. From Fig. 2, it can be noticed that: – For CASE 2 losses vary between around 0.11 and 0.25 while the VSI varies between around 0.82 and 1. – For CASE 3 losses vary between around 0.11 and 0.21 while the VP varies between around 0.005 and 0.03. – For CASE 4 VSI varies between around 0.82 and 0.92 and VP varies between around 0.005 and 0.026. – For CASE 5 losses vary between around 0.09 and 0.22 while the VSI varies between around 0.81 and 1. – For CASE 6 losses vary between around 0.09 and 0.13 while the VP varies between around 0.005 and 0.02. – For CASE 7 VSI vary between around 0.82 and 0.95 and VP varies between around 0.005 and 0.015. Furthermore, some selected solutions for each case are tabulated in Table 3, for cases with a single DG unit, and in Table 4 for cases with two DG units. Based on these, the following comments can be made: 1) For a single-unit DG incorporation: – For CASE 2 solution # 1, for example, the obtained losses are 0.241 while the obtained VSI is 0.828. Compared to the base case (Plosses =0.211 and VSI=1) we can say that the losses have been increased (ALR=-14.270%) while the VSI has been reduced (VSIM=17.190%). The optimal DG location and size for this solution are 15 and 2.498 MW, respectively. – For CASE 2 solution #3, for example, the obtained losses are 0.191 while the obtained VSI is 0.897. Compared to the base case, the losses have been reduced (ALR=9.630%) while the VSI has been reduced (VSIM=10.285%). The optimal DG location and size for this solution are 12 and 2.62 MW, respectively. – For CASE 2 solution #5, for example, the obtained losses are 0.128 while the obtained VSI is 0.997. When compared to the base case, the losses have been reduced (ALR=39.350%) while the VSI has been slightly reduced (VSIM=0.344%). The optimal DG location and size for this solution are 6 and 3.715 MW, respectively. – For CASE 3 solution #1, for example, the obtained losses are 0.203 while the obtained VP is 0.007. When compared to the base case, the losses have been decreased by 3.686%, while the VP has been reduced by 94.830%. The optimal DG location and size for this solution are 8 and 3.674 MW, respectively.

Optimal Siting and Sizing of DG Units

529

Fig. 2. PF obtained for different cases.

– For CASE 3 solution #3, for example. Compared to the base case, the losses have been reduced by 32.008% while the VP has been reduced by 90.395%. The optimal DG location and size for this solution are 8 and 2.799 MW, respectively. – For CASE 3 solution #5, for example. Compared to the base case, the losses have been reduced by 44.385% while the VP has been reduced by 82.413%. The optimal DG location and size for this solution are 6 and 3.273 MW, respectively.

530

Y. Latreche et al.

– For CASE 4 solution #1, for example, the obtained VSI is 0.908 while the obtained VP is 0.007. Compared to the base case, the VSI has been reduced by 9.157% while the VP has been highly reduced by 94.821%. The optimal DG location and size for this solution are 8 and 3.715 MW, respectively. – For CASE 4 solution #3, for example, compared to the base case, the VSI has been reduced by 11.841% while the VP has been highly reduced by 90.966%. The optimal DG location and size for this solution are 9 and 3.390 MW, respectively. – For CASE 4 solution #5, for example, compared to the base case, the VSI has been reduced by 17.233% while the VP has been highly reduced by 82.077%. The optimal DG location and size for this solution are 10 and 3.470 MW, respectively.

2) The same analysis can be done for two units DG incorporation.

Table 3. Obtained results for the 33-bus RTS for one DG unit. Location CASE # CASE 1

CASE 2

CASE 3

CASE 4

Solution # 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5

DG1 15 13 12 9 6 8 8 8 7 6 8 9 9 9 10

Size (P [MW]) DG1 2.498 2.437 2.620 2.622 3.715 3.674 3.108 2.799 3.663 3.273 3.715 3.181 3.390 3.532 3.470

Objective functions PLosses 0.211 0.241 0.205 0.191 0.156 0.128 0.203 0.161 0.143 0.133 0.117 0.207 0.200 0.220 0.235 0.271

VSI 1 0.828 0.874 0.897 0.959 0.997 0.912 0.957 0.984 0.996 1.000 0.908 0.901 0.882 0.869 0.828

VP 0.134 0.027 0.021 0.017 0.013 0.017 0.007 0.009 0.013 0.015 0.024 0.007 0.011 0.012 0.013 0.024

Performance evaluation indices ALR -14.270 2.714 9.630 26.102 39.350 3.686 23.804 32.008 36.804 44.385 2.001 5.435 -4.250 -11.395 -28.639

VPI 79.676 83.959 87.550 90.407 87.650 94.830 93.007 90.395 89.101 82.413 94.821 91.705 90.966 90.108 82.077

VSIM 17.190 12.563 10.285 4.131 0.344 8.850 4.321 1.644 0.395 0.000 9.157 9.862 11.841 13.137 17.233

NBVV 21 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

Table 4. Obtained results for the 33-bus RTS for two DG units. Location CASE # CASE 1

CASE 5

CASE 6

CASE 7

Solution # 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5

DG1 8 6 6 6 6 6 6 6 6 6 6 6 6 6 6

Size (P [MW]) DG1 16 18 18 16 14 12 12 12 14 14 12 18 18 18 9

DG2

Size (P [MW]) DG2

1.507 1.812 1.835 1.845 1.853 1.857 1.854 1.857 1.857 1.857 1.857 1.857 1.857 1.857 1.857

1.699 1.390 1.284 1.191 0.662 1.613 1.311 1.242 0.902 0.658 1.818 1.567 1.440 1.336 1.857

Location

Objective functions PLosses 0.211 0.209 0.148 0.137 0.114 0.091 0.125 0.107 0.104 0.095 0.091 0.141 0.171 0.155 0.143 0.133

VSI 1 0.823 0.876 0.899 0.960 1.000 0.947 0.993 1.000 1.000 1.000 0.918 0.837 0.864 0.886 0.951

VP 0.134 0.022 0.010 0.010 0.009 0.020 0.006 0.008 0.009 0.013 0.020 0.007 0.012 0.011 0.010 0.005

Performance evaluation indices ALR 1.004 29.786 35.224 46.037 56.712 40.602 49.117 50.585 55.112 56.715 33.023 18.730 26.623 32.412 36.917

VPI 83.349 92.226 92.668 93.280 85.168 95.260 94.074 93.482 90.405 85.124 94.757 90.859 92.040 92.612 96.231

VSIM 17.697 12.378 10.105 4.026 0.000 5.316 0.721 0.000 0.000 0.000 8.216 16.266 13.639 11.379 4.924

NBVV 21 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

Optimal Siting and Sizing of DG Units

531

6 Conclusion This paper has successfully demonstrated the optimal integration of distributed generation in a power system. It also presents the multi-objective optimization formulation, where at a time, two of the three objectives are being optimized simultaneously. The performance of the optimization formulation has been evaluated on four crucial performance indices; active losses reduction index, voltage stability enhancement index, voltage profile improvement index, and the number of buses violating the voltage limits. The paper also includes the explanation and mathematical representation of the Tchebycheff method, used to decompose a multi-objective optimization problem using Pareto Fronts. In total, seven case-studies have been evaluated on a 33-bus radial system, including the base case where no DG is installed. It has been found that despite the incorporation of multiple DGs, the optimization formulation successfully finds the optimal trade-off between the conflicting performance indices and hence ensuring the efficacy of the proposed algorithm. In the future, more objective functions can be added to the multi-objective formulation, and the overall performance can be evaluated on the enhanced set of indices. Acknowledgment. The authors extend their appreciation to the Deanship of Scientific Research, University of Hafr Al Batin for funding this work through the research group project No G-1092020.

References 1. Acharya, N., Mahat, P., Mithulananthan, N.: An analytical approach for DG allocation in primary distribution network. Int. J. Electr. Power Energy Syst. 28(10), 669–678 (2006). https://doi.org/10.1016/j.ijepes.2006.02.013 2. Olabode E.O., Ajewole, T.O., Okakwu, I.K., Ade-Ikuesan, O.O.: Optimal sitting and sizing of shunt capacitor for real power loss reduction on radial distribution system using firefly algorithm: a case study of Nigerian system. Energy Sources Part A Recover. Util. Environ. Eff. (2019). https://doi.org/10.1080/15567036.2019.1673507 3. Hadidian-Moghaddam, M.J., Arabi-Nowdeh, S., Bigdeli, M., Azizian, D.: A multi-objective optimal sizing and siting of distributed generation using ant lion optimization technique. Ain Shams Eng. J. 9(4), 2101–2109 (2018). https://doi.org/10.1016/j.asej.2017.03.001 4. Latreche, Y., Bouchekara, H.R.E.H., Kerrour, F., Naidu, K., Mokhlis, H., Javaid, M.S.: Comprehensive review on the optimal integration of distributed generation in distribution systems. J. Renew. Sustain. Energy 10(5), 055303 (2018). https://doi.org/10.1063/1. 5020190 5. Prakash, P., Khatod, D.K.: Optimal sizing and siting techniques for distributed generation in distribution systems: a review. Renew. Sustain. Energy Rev. 57, 111–130 (2016). https://doi. org/10.1016/j.rser.2015.12.099 6. Ugranli, F., Karatepe, E.: Multiple-distributed generation planning under load uncertainty and different penetration levels. Int. J. Electr. Power Energy Syst. 46(1), 132–144 (2013). https://doi.org/10.1016/j.ijepes.2012.10.043 7. Falaghi, H., Haghifam, M.R.: ACO based algorithm for distributed generation sources allocation and sizing in distribution systems. In: 2007 IEEE Lausanne POWERTECH, Proceedings, pp. 555–560 (2007). https://doi.org/10.1109/pct.2007.4538377

532

Y. Latreche et al.

8. Hadian, A., Haghifam, M.R., Zohrevand, J., Akhavan-Rezai, E.: Probabilistic approach for renewable DG placement in distribution systems with uncertain and time varying loads (2009). https://doi.org/10.1109/pes.2009.5275458 9. Kaya, P., Chanda, C.K.: ª simple and fast approach for allocation and size evaluation of distributed generation. Int. J. Energy Environ. Eng. 4(1), 1–9 (2013). https://doi.org/10.1186/ 2251-6832-4-7 10. Sharma, S., Bhattacharjee, S., Bhattacharya, A.: Quasi-Oppositional Swine Influenza Model Based Optimization with Quarantine for optimal allocation of DG in radial distribution network. Int. J. Electr. Power Energy Syst. 74, 348–373 (2016). https://doi.org/10.1016/j. ijepes.2015.07.034 11. Zhang, Q., Li, H.: MOEA/D: A multi-objective evolutionary algorithm based on decomposition. IEEE Trans. Evol. Comput. 11(6), 712–731 (2007). https://doi.org/10. 1109/TEVC.2007.892759 12. Deb, K., Jain, H.: An evolutionary many-objective optimization algorithm using referencepoint-based nondominated sorting approach, Part I: Solving problems with box constraints. IEEE Trans. Evol. Comput. 18(4), 577–601 (2014). https://doi.org/10.1109/TEVC.2013. 2281535 13. Ma, X., Qi, Y., Li, L., Liu, F., Jiao, L., Wu, J.: MOEA/D with uniform decomposition measurement for many-objective problems. Soft. Comput. 18(12), 2541–2564 (2014). https://doi.org/10.1007/s00500-014-1234-8 14. Baran, M.E., Wu, F.F.: Network reconfiguration in distribution systems for loss reduction and load balancing. Power Deliv. IEEE Trans. 4(2), 1401–1407 (1989). https://doi.org/10. 1109/61.25627

Task Scheduling-Energy Efficient in Cloud Computing Karima Saidi(&) and Ouassila Hioual Abbes Laghror University, 40004 Khenchela, Algeria [email protected], [email protected]

Abstract. Cloud Computing is a computing model that provides computing resources as distributed on-demand services over a network. The data centers are on the rise and are projected to increase and lead to more energy consumption with the grown use of this technology. In fact, in recent years, the scientific community has tackled the issue of reducing data center power consumption. Our solution, which is essentially focused on task scheduling strategy, is critical to the overall success of cloud workflow energy systems by proposing a new model based on a deep neural network for regression. It aims to reduce energy consumption, to select the best placement with the minimum energy consumption allocated to the VM, and that by supporting the TOPSIS method. Keywords: Cloud computing
Energy consumption
Task scheduling
Deep neural network
TOPSIS

1 Introduction Cloud computing has become popular over the last few years. Cloud service providers such as Azure, Amazon, IBM, and GOOGLE rented massive quantities of resources due to high user demand. This process needs a lot of energy consumption in their data centers. In this work, we focus on the task scheduling of a variety of requests submitted by users taking into account energy usage. Therefore, it is useful to find a strategy to distribute these resources efficiently by reducing the energy and allowing for the best use of resources. The issue of energy efficiency is an important challenge that has been fully addressed. Because energy consumption is still very high in data centers and the forecasts for the next few years are very worrying. Although, it has different solutions to resolve but is not yet fully resolved. Therefore, many techniques have been proposed for task scheduling such as heuristic, and meta-heuristic (Thaman and Singh 2016). To solve the problem of resource allocation, a lot of solutions are reminded in our work (Saidi et al. 2019), but in this work, we focus on another broad class of approaches, namely regression based on deep neural network (DNN). We divide our proposed work on multiple regression requires only one output variable for multiple input variables to predict, which is also the subject of this study. In this paper, we focus on deep neural networks for regression, which is a bit difficult to understand, but it gives incomparable predictive power much to the most complex regression models. The DNN-based regression is generally more efficient to © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 533–540, 2021. https://doi.org/10.1007/978-3-030-63846-7_50

534

K. Saidi and O. Hioual

manage large-scale training data with high-dimensional input and output and has a strong generalization capability (Du and Xu 2017). The key contribution is summarized as minimizing energy consumption by scheduling tasks using the DNN for regression with promoting the multi-use of the technique for order preference by similarity to ideal solution (TOPSIS) is a multicriteria decision analysis method used to rank a set of alternatives with different criteria (Greco et al. 2016). The rest of this paper is structured as follows. Section 2 presents some related work. The proposed work is presented in Sect. 3. The experimental results are given in Sect. 4. Section 5 addresses the conclusion and future work.

2 Related Work In this section, we will review current studies that use the various techniques for scheduling tasks taking energy-efficient into account. In (Marahatta et al. 2019), the authors proposed a complete scheduling model for cloud data centers (CDCs) to define the system, resource, task, energy consumption, task, and virtual machine mark. Moreover, using task merging, optimization problems and constraints, and migration model to achieve energy-efficiency, and optimize task guarantee ratio, mean response time, and resource utilization for cloud data centers. The authors in (Chu et al. 2020), used the new combination of artificial intelligence technologies and graph database theories. They, first, propose a novel long short-term memory (LSTM) deep learning method to predict the execution time for query tasks in the graph database. In (Khorsand and Ramezanpour 2020), the researchers focused on reducing energy consumption, taking into account the multi-criteria of the user’s QoS preferences, to ensure the user’s service level agreement (SLA) which is the target of energy efficiency. The authors have developed a task-scheduling strategy based on the combination of the best-worst and TOPSIS multi-criteria decision-making algorithm (BWMTOPSIS) allowed to determine which more important cloud scheduling solution to be more comprehensive and principled. The main solution in (Ben alla et al. 2019) is allowed to reduce energy consumption. The authors considered the priorities of different users and optimize the makespan under deadlines constraint, used the ELECTREIII method to rank and prioritize the tasks, then these ranked tasks are dispatched in four classes (level priorities). To evaluate this solution, the authors used the CloudSim and they achieved the best performance to provider and users of cloud computing. The authors in (Prabhpreet Kaur and Monika Sachdeva 2017) are based on task scheduling by proposing the Clonal Selection Algorithm (CSA) which is a special class of Artificial Immune System. The proposed task scheduling algorithm is responsible for efficiently allocating user tasks to different available Processing Elements to optimize energy consumption and time. In (Li and Hu 2019), the authors proposed a job scheduling algorithm based on deep reinforcement learning (DeepJS) within the bin packing problem. DeepJS can automatically obtain a fitness calculation method (describe the match between the

Task Scheduling-Energy Efficient in Cloud Computing

535

machines and tasks to maximize the throughput) that reduces the makespan. To evaluate DeepJS, the authors used a trace-driven simulation. The authors, in (Zhao et al. 2016), proposed a new approach to task scheduling depending on the degree of the task requirement by calculating it to improve energy efficiency. Both the data sets and the cloud system were integrated by data correlationbased clustering into a tree-structure model. Moreover, they built a model to energyefficient by taking into consideration applied the principles of scheduling by decreasing network traffic, improving the utilization of servers in the cloud, and reduce the generation of inefficient energy. The contribution in (Panda and Jana 2019) is based on proposed the energyefficient task scheduling algorithm (ETSA) for minimizing energy consumption and makespan. The authors developed an on-line energy-efficient task scheduling for heterogeneous cloud computing systems. They used the cloud, application, energy, scheduling models in the proposed algorithm. This work is based on a combination of the use of DNN for regression and the TOPSIS method for a more efficient mixture between them. Another significant main idea of the proposed work in this paper is to first select the server and then select the appropriate virtual machine to minimize the data center’s energy consumption.

3 Proposed Work The regression model is fit to our problem. So, we need complex data to more prediction power with deep learning, because the deep neural network can extract automatically the features to construct a prediction function. In the first step, a certain number of tasks arrive in online mode; we rank all these tasks in the queue in descending order according to different parameters. We consider d resource types such as CPU, memory, storage, bandwidth, and so on. The tasks arrive in an online manner and are ranked in the tasks queue waiting for scheduling (using TOPSIS). The second step is allowed to predict the placement of these tasks by building a regression model of energy consumption in the smallest value. We have defined a set of tasks as T = {T 1 ,T 2 ; T 3 ; :::; T K }, which includes a set of n s parameters T uK = {T pK , T m K , TK , TK } Cloud providers assign task parameters when users send tasks to them, we specify other parameters according to that for each task submitted in a real-time like the time required to finish this task (deadline time), the size of tasks, the time of arrival, and the execution time. We assume that there are N virtual machine group VMG =fVMG1 ,VMG2 ; VMG3 :::; VMGN g, VMG is the number of VMs running on one server. In the supervised learning, the pair of task-machines is known, we consider the tasks are inputs and machines are the output of these tasks, but in our proposal, we assume the input is the list of machines-task pair, so it allows predicting from this model a new task arrives, as shown in Fig. 1. We refer to machine group j appropriate to task i, and (VMGj , T i ) as appropriate machines-task pair. We will choose the value of Energy Consumption (EC) which has the smallest EC from the appropriate machines-task pair list as a scheduling decision.

536

K. Saidi and O. Hioual

In the proposed model: we need to calculate: • ECij is the energy consumption produced by the task T i running on the virtual machine group VMGj • The available resources of each VMG to compare the task undertaken. So, the scheduling of tasks should be carried out in such a way as to ensure the proper use of resources, because the high use of resources leads to high energy consumption. The availability of resources of each VMG is the sum of available resources of all VMs in the VMG • Time of execution of each task running on the VMG And, we take some of the parameters of each task: • Deadline time • Size of tasks by defining the requirements of resource (we distinct the amount of CPU and memory) • Time of arrival Also, we take some of the parameters of each VMG: • The Cost Scheduling techniques are used to dynamically schedule real-time tasks using a model to predict the new placement of each task in the minimum energy consumption of VMG. After that, we can assign this task on one VM that belongs to the chosen VMG as the smallest energy consumption. This process requires another method to rank the VMs at the chosen VMG, which we presume that it uses the same method for ranking used in the first step and choose the best placement of this task by selecting the first one. The criteria of each VM in a chosen VMG are: • • • •

The completion time of this task in each VM The availability of resources of each VM to assess the task taken The cost Time of execution of this task in each VM

Fig. 1. The architecture of the proposed work

Task Scheduling-Energy Efficient in Cloud Computing

The algorithm of the proposed work can be described as follows: Algorithm of the proposed work 1

The tasks arrive

2

Rank the tasks arrive using TOPSIS

3

Put the tasks in the global queue

4

Take a task from the global queue of ranking tasks

5

task pair with each one from the list of VMG

6

Using prediction model proposed (The input of this model is

7

Schedule task to machine group VMG (The output selects the smallest value of energy consumption)

)

8

Choose one of the VMs in VMG by ranking the VMs under different criteria.

9

Take another task to schedule (Return to 4)

The functionality of this algorithm is presented in Fig. 2:

Fig. 2. Flowchart of the proposed work

537

538

K. Saidi and O. Hioual

4 Implementation and Results The planned work is carried out using PYTHON. To obtain higher accuracy and better predict, the new entry of the task is the dataset which is an exigency. Collecting the dataset is important to our work in which we want to construct a model. So, we need to use establishing dataset to include certain parameters that are similar to our work. We suggest using the dataset of the individual household electric power consumption consumption (Georges Hebrail and Alice Berard 2012) to evaluate our work. This dataset includes 2075259 measurements collected between December 2006 and November 2010 (47 months) in a house located in Sceaux (7 km from Paris, France). We assume that the house is the datacenter and the measurements taken of electric power consumption in this house are the energy consumption of the datacenter. Firstly, we need to evaluate the DNN model for regression, we need to calculate the accuracy for the different number of instances (500, 800, 1000, 5000, 10000, 70000, 100000), we note that the accuracy is improved in the deep learning model when augmenting the data, as shown in Fig. 3.

Fig. 3. The accuracy of our model

Secondly, in our work, task scheduling is executed to improve performances in the VM system and minimizing energy consumption in the data center. In Fig. 4, we note that the general energy consumption in the datacenter is reduced.

Task Scheduling-Energy Efficient in Cloud Computing

539

Fig. 4. The energy consumption in the data center before and after our proposed work

5 Conclusion We have introduced the new model, a task scheduling algorithm based on deep neural learning with the TOPSIS method. It can, automatically, obtain the best virtual machine to each task which will select, directly, the minimum energy consumption from regression. We have evaluated this model through a Python language. The results show that it finds good accuracy. Finally, while this analysis provides some important findings for the application of deep learning techniques in this field of study, it is suggested that future studies be based on testing our model in real-time by using the online model according to the nature of this field.

References Ben Alla, S., Ben Alla, H., Touhafi, A., Ezzati, A.: An efficient energy-aware tasks scheduling with deadline-constrained in cloud computing. Computers 8(2), 46 (2019) Chu, Z., Yu, J., Hamdulla, A.: A novel deep learning method for query task execution time prediction in graph database. Future Gen. Comput. Syst. 112, 534–548 (2020) Du, J., Xu, Y.: Hierarchical deep neural network for multivariate regression. Pattern Recogn. 63, 149–157 (2017) Hebrail, G., Berard, A.: UCI Machine Learning Repository: Individual household electric power consumption Data Set [WWW Document] 2012. http://archive.ics.uci.edu/ml Greco, S., Figueira, J., Ehrgott, M.: Multiple Criteria Decision Analysis. Springer, Berlin (2016) Khorsand, R., Ramezanpour, M.: An energy-efficient task-scheduling algorithm based on a multi-criteria decision-making method in cloud computing. Int. J. Commun Syst 33, e4379 (2020)

540

K. Saidi and O. Hioual

Li, F., Hu, B.: DeepJS: job scheduling based on deep reinforcement learning in cloud data center. In: Proceedings of the 4th International Conference on Big Data and Computing, ICBDC, Guangzhou, China, pp. 48–53. Association for Computing Machinery (2019) Marahatta, A., Pirbhulal, S., Zhang, F., Parizi, R.M., Choo, K.-K.R., Liu, Z.: Classification-based and energy-efficient dynamic task scheduling scheme for virtualized cloud data center. IEEE Trans. Cloud Comput. 1 (2019) Panda, S.K., Jana, P.K.: An energy-efficient task scheduling algorithm for heterogeneous cloud computing systems. Cluster Comput. 22, 509–527 (2019) Kaur, P., Sachdeva, M.: Energy efficient task scheduling in cloud computing. In: Proceedings of the 4th International Conference on Power and Energy Systems Engineering, CPESE, pp. 25– 29 (2017) Saidi, K., Hioual, O., Siam, A.: Resources allocation in cloud computing: a survey. In: Proceedings of the 3rd International Conference in Artificial Intelligence in Renewable Energetic Systems, Springer, Cham, pp. 356–364 (2019) Thaman, J., Singh, M.: Current perspective in task scheduling techniques in cloud computing: a review. Int. J. Found. Comput. Sci. Technol. 6, 65–85 (2016) Zhao, Q., Xiong, C., Yu, C., Zhang, C., Zhao, X.: A new energy-aware task scheduling method for data-intensive applications in the cloud. J. Netw. Comput. Appl. 59, 14–27 (2016)

An Efficient Hybrid Meta-heuristic Approach for Solving the K-Shortest Paths Problem Over Weighted Large Graphs Mohamed Yassine Hayi(&) and Chouiref Zahira Computer Science Department, LIMPAF Laboratory, Bouira University, Bouira, Algeria [email protected], [email protected], [email protected]

Abstract. The problem of finding the shortest path is a combinatorial optimization problem which has been well studied in the last years, optimization allow to save time for solve problems in various fields, like the road traffic network to find the shortest route between two places, or to optimize the consumption of a sensitive product in order to minimize the degree of the pollution, or other several areas. This paper presents an Improved Optimization Genetic Algorithm (IOGA) to calculate K shortest optimal paths. This algorithm uses a strategy that has proven to be effective in solving the shortest path problem. The tests were carried out using an exact algorithm (Dijkstra) and a Meta heuristic algorithm (genetic algorithm) on a large random network that we have generated. An experimental analysis and empirical results with an algorithm known in the domain (Dijkstra) demonstrate the efficiency of our proposed algorithm in terms of execution and quality of the result. Keywords: Optimization
K-shortest paths algorithm
Routing problem
Big data


Genetic algorithm
Dijkstra’s

1 Introduction With the massive growth of the Information, optimization on big data has become one of the main researches. Complex networks such as the Internet, transportation, communication and distribution networks, etc. have attracted increasing attention from various fields of science and engineering. The graph is an important complex network to describe the relationship among a set of interconnected nodes (entities). Shortest path computation is a frequent operation. In these situations, it is often valuable to be able to find the k optimal shortest paths. The problem of finding the shortest paths is one of the most widely-studied combinatorial optimization problems in the last decade and is a challenging task over large graphs. Already existing graph algorithms are not suitable for graphs of larger sizes. In our work, we study traffic network in large scale routing problems as a field of application. The determination of an appropriate route, reaching the destination in time and energy saving has significance on real world applications. The vertices (nodes) of the graph correspond to spatial units (cities) of the road network. Edges connect vertices and are weighted with non-negative weights according to © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 541–550, 2021. https://doi.org/10.1007/978-3-030-63846-7_51

542

M. Y. Hayi and C. Zahira

the distance between vertices. Operations on a graph can take a long time due to the complexity of structural connectivity and the size of the graph. In addition, a large graph makes efficiency even more difficult [10]. Therefore, in this paper, we attempt to solve the problem about finding the shortest route starting from the source to the destination where the distances between each city are known. Because the K- shortest path is NP-hard, we hybridize a metaheuristic algorithm (namely Genetic Algorithm - GA) with an exact algorithm (namely Dijkstra’s algorithm) for the given problem. In this study, the new hybrid algorithm namely IOGA (Improved Optimization Genetic Algorithm) is developed to finds one or k optimal paths containing the minimal cost between two vertices in an oriented graph. So, hybriding a genetic algorithm with Dijkstra’s algorithm is an interesting idea. The experimental study makes a comparison of IOGA with the Dijkstra’s algorithm in order to ensure the following specific contributions: i) find one or several shortest paths; and ii) minimize the execution time by comparing with the Dijkstra’s algorithm. The effectiveness of our approach that employ the heuristic algorithm is confirmed. The final results illustrate that this novel approach with the optimization strategies achieves high scalability and performance. The rest of this paper is organized as follows: Some approaches related to the shortest path problem are briefly introduced in Sect. 2. In Sect. 3 we briefly discuss the K-shortest path problem. Section 4 presents an overview of our algorithm, along with an example for illustration. Section 5 presents the experiments designed to assess the performance of the IOGA. Section 6 presents the results obtained by the experiment. A brief conclusion and future researchs are finally given in Sect. 7.

2 Related Work The problem of determining the k shortest paths has been studied in several works. Due to the problems numerous and diverse applications, there has been a surge of research in shortest-path algorithms. The computation of shortest-path can generate either exact or approximate solutions by using exact algorithms or metaheuristic algorithms. The aim of the heuristic algorithms is to minimize the computation time. Many works focus on algorithms that target traffic applications, where we summarize some of them here. Goldberg et al. [6] reviewed the classical Dijkstra’s algorithm and an over road network and they illustrated heuristic techniques for computing the shortest-path that was given by a subset of the graph. JunWoo et al. [2] described a new algorithm to enumerate the k shortest simple paths in a directed graph and they reported on its implementation. Pettie et al. [5] proposed a new algorithm out-performs Dijkstra’s algorithm on sparse graphs for the all-pairs shortest path problem and the presented results which show the new algorithm to run faster than the Dijkstra’s algorithm on a variety of sparse graphs when the number of vertices ranges from a few thousand to a few million. Lee et al. [1] made a comparison between the Dijkstra’s algorithm and the genetic algorithm and its own DGA algorithm. The results showed that the Dijkstra’s algorithm does not give the best solutions in the graph which contains more than 10 000 knots, on the other hand the DGA algorithm gave good results in acceptable execution time. In [7] and [8], the authors presented a review of various heuristic

An Efficient Hybrid Meta-heuristic Approach

543

shortest-path algorithms and they proposed the main distinguishing features of different heuristic algorithms as well as their computational costs. Analysing above mentioned works, it can be said that: – Dijkstra’s algorithm is one of the best optimization algorithms and it has been successfully applied to a variety of optimization problems and also it is a fast algorithm in the execution time but several articles and researches have said that the performance of the algorithm of Dijkstra is not top in the big graph [1, 3, 4, 9]. – The complexity of the exact algorithm for solving the shortest path problem increases with the number of vertices in the graph. – The genetic algorithm is a very well known and it has a character of biology. It works with the random function, and it is one of the best known heuristic algorithms and it is useful in several areas of optimization [11, 12]. – For these reasons genetic algorithm is offered as the optimal solution of k-shortest path problem.

3 The K-Shortest Paths Problem The K-shortest paths problem is the base for a lot of combinatorial optimization problems. The problem of finding the shortest path between starting vertex and terminal vertex (that exist in a given network diagram) is widely used in various fields, such as: computer network routing algorithm, the robot Pathfinder, route navigation, game design, and so on. To formulate a mathematical model for finding the K-shortest paths from the origin node to a destination node, assuming that G = (N, E, C), a weighted directed graph (G) with a set of vertices (nodes) (N), and a set of directed edges (E) and the edge cost values (C), such as: (i) c(s, t): cost of directed edge (s ! t) from a source node “s” to a target node “t” (s and t both in N) so that costs are non-negative; (ii) the returned paths are as short as possible and always returned Links that do not satisfy constraints on the shortest path are removed from the graph. The k shortest paths problem is a natural and long studied generalization of the shortest path problem, in which not one but several paths in increasing order of length are sought. The problem of determining the k shortest paths has proved to be more challenging. To find the shortest path one can use shortest path algorithms. The Dijikstra’s algorithm requires the calculation of shortest paths from the origin nodes to all nearest nodes in the Origin-Destination matrix. Dijkstra’s algorithm can solve this problem and it can be extended to find more than one path. The genetic algorithm has excellent performance in searching spaces with large solutions, and can compromise between efficiency and accuracy. For this reason, we tried to solve this problem by applying the Dijkstra’s algorithm and the genetic algorithm method presented on Sect. 4. In the next section, we discussed our proposed approach and then we presented the results with some comparisons to illustrate the performance of our algorithm.

544

M. Y. Hayi and C. Zahira

4 Computing the K Shortest Paths The K-shortest path problem is solved by implementing a new technique through a heuristic approach in which such hybrid heuristic can combine the power of both classic and heuristic techniques in order to find the optimal solution in a few iterations. We named this algorithm by Improved Optimization Genetic Algorithm (IOGA) or shortly IOGA. In what follows, we first introduced the hybridization principle allowing to enumerate the k shortest paths in a directed graph. Then the structure of IOGA was addressed in which the basic steps are customized. A. Hybridization Principle The IOGA is a hybrid algorithm between the Dijkstra’s algorithm and the genetic algorithm. This proposition is a mixture between the two algorithms, from where we worked to build a population, which is a part of GA, from the same principle of the Dijkstra’s algorithm. This principle represents the unfolding method of the Dijkstra’s algorithm. In this last,, we always calculate the nearest node going from one node to another. On the other hand in IOGA, we calculate the TOP nodes nearest (2,3,4,…), so that we will have several proposed routes, we put them in the population. After each iteration, we calculate the distance of each road exists in the population, we leave only the lines which can be an optimal solution, and we suppress the other lines, so that we will not have a single route, but we will have several routes of the optimal solution and quickly. Dijkstra’s algorithm: Dijkstra’s algorithm is the most popular algorithm in the shortest path problem. It is often used in orienting graphs weighted by positive real. It was created in 1959 by the Dutch computer scientist Edsger Dijkstra. This algorithm always searches for the shortest path from one node to all the other nodes. It takes an initial knot, and it searches for the shortest path to all the other nodes, as a second iteration, according to the nodes visited by the first node. It chooses the shortest of the first node, and it repeats the first step, and so on until it turns on all the nodes of the graph. Genetic algorithm: The GA is an evolutionary algorithm that is used to obtain an optimal solution in a reasonable time. And it uses a natural selection (random) to have a population. This last will be evolving after each iteration and when the algorithm reaches a stop condition, we will take the best solution exist in the population as a final solution for the problem [14]. In what follows, we introduced the IOGA to enumerate the k shortest paths in a directed graph. B. Improved Optimization Genetic Algorithm (IOGA) Given a directed graph G with non negative edge weights, and two vertices s and t, the problem asks for the k shortest paths from s to t in increasing order of length. A positive integer k is returned at the end of the iteration. The structure of IOGA in this section is addressed in which the basic steps are customized as below in algorithm 1:

An Efficient Hybrid Meta-heuristic Approach

545

Algorithm. 1. IOGA algorithm.

IOGA find many solutions exists in a logical way, it takes the source of the route, and it adds the TOP (it is a variable) nodes closest to the destination, so that it builds a population of size TOP. Figure 1 represents an explanatory example on the functioning of IOGA: The algorithm that we created is very complicated, so to better understand the flow of the algorithm we made a simple example to understand the different stages of our proposal. The parameter TOP = 3 i.e. from each node, we will take the closest TOP (3) nodes. The parameter SIZE = 4, i.e. in each row of the population we have a source and two knots and a destination.

Fig. 1. An explanatory example on the functioning of IOGA.

546

M. Y. Hayi and C. Zahira

IOGA contains some part and some function; we explain them in the next lines: A. Population It is a matrix which contains several routes, in the genetic algorithm, this matrix is generated at random, on the other hand in our algorithm, we generate this matrix in a logical way, that is to say according to rules, for example if we take the top parameter equal to 6, so from the source we have 6 first path proposed for the population, the 6 paths starting with the 6 node closest to the destination, and so on, like that we will have a population which has a dynamic size, because after each iteration (size) we delete the paths which do not allow giving optimal solutions by using the Cleaning function [13]. B. Cleaning Function The aim of the cleaning function is avoiding the non optimal solutions or that gives complicated and larger solutions of the solution already found. After the end of each iteration (SIZE) we apply the cleaning function on the population to remove the solutions which can never be less than the MIN solution that we have found so far. We apply the Cleaning function at the end of each part of size, for example if we have the parameter size which equal to 8, we apply the Cleaning function 7 times, (we do not apply this function in the first phase that contains only the source). The Cleaning function has two objectives: • Fitness: it is an objective function which calculates the solution of each line of the population in a given moment, and compares this solution with the best solution find for now. • Cleaning: it is to delete each line who exist in the population and which gives a bigger fitness result than the best solution find for now. In general, the Cleaning function works to calculate the solutions of the different routes that exist in the population, and to delete each route which cannot be an optimal solution, to reduce the size of the population and therefore to save time. Algorithm 2 represents the Cleaning function:

Algorithm. 2. Cleaning function.

An Efficient Hybrid Meta-heuristic Approach

547

C. Final Function The final function has the purpose of detecting the lines that give workable solutions, and of finding the best solution to date. This function traverses all the lines of the population in a given time, and check if there is a direct path between the last element of the line with the destination, if so we say that the line represents a feasible solution, and compare this solution achievable with our best solution, and of course if the feasible solution is lower than the best solution, we keep the feasible solution as a better solution, and also we save this population line because it represents the path of the best solution to find now. Algorithm 4 represents the final function:

5 Experiments In order to verify the performance of the proposed approach, both Dijkstra’s algorithm and IOGA were implemented in python and all the experiments were conducted on Intel Core ™ i3 CPU processors (2.53 GHz clock rate). We generated random oriented networks of size 10 000 nodes, and the distance dist(i, j) between node i and j is assigned with a random integer between 1 and 99999. There are two parameters in IOGA: TOP and SIZE. The database is in two formats: excel and Txt and to generate the source/destination we created by the Random function 500 couple sources/destination to calculate the distance between the different couples.

6 Result We choose to explain the results of the 9 couples that we have chosen at random, to see in detail the result obtained, and we cannot explain the results of the 500 couples. We notice that Dijkstra’s algorithm always give a single solution, it is say a single route, on the other hand IOGA give between one up to 5 solutions, and an average of 2.66 solutions for each couple (AVG of 500 source/destination pairs). For the run time, Dijkstra’s algorithm have 1.98 s as an average for the 500 couples, and IOGA scored 3.07 s, but it recovers several routes for each couple. Table 1 illustrates the number of solution and the run time in seconds of each algorithm for 9 source/destination pairs: Table 1. Execution time result and number of solutions for 9 couples

1 2

Run time for many solutions avg for 500 couples

548

M. Y. Hayi and C. Zahira

We compared the execution time by contribution or each solution of the 9 couples. For example, the couple 375 which obtained by the algorithm IOGA 5 solutions in 4.3 s, that is to say an average of 0.86 s for each solution, knowing that by Dijkstra’s algorithm it had only one solution in 2.2 s. Figure 2 represents a graph of comparison of the execution time between the two algorithms IOGA and Dijkstra.

Fig. 2. Graph representing the execution time for the first 9 couples

Figure 3 represents the graph of the path number obtained by the IOGA for the first 100 pairs (source/destination) by the parameters (TOP = 8 and SIZE = 10). We note that the IOGA did not find 4 solutions out of 100 by contribution to the algorithm of Dijkstra’s algorithm (73,82, 93,96), on the other hand it found the same solutions as Dijkstra’s algorithm in 62 cases (one solution), but most importantly, IOGA to find in 34 cases several solutions (from 2 to 8 paths) knowing that Dijkstra’s algorithm always returns a single route. The best solution is marked for couple number 87, hence the IOGA returns 8 different paths of the same solution as the Dijkstra’s algorithm, and we can have other solutions if we execute the program with higher values of both parameter TOP and SIZE.

Fig. 3. Graph representing the number of solutions for the first 100 couples

We have also presented our results in Table 4 which contains the number of solutions per contribution to the parameter of the IOGA (TOP and SIZE). The results of Table 2 represent the number of path solutions for the first 100 pairs (source/destination). For the parameter (TOP = 10 and SIZE = 10), we had a case where our algorithm did not give the same result as Dijkstra’s algorithm (a result less than that of Dijkstra’s algorithm). Also we had 64 cases out of the 100 cases where

An Efficient Hybrid Meta-heuristic Approach

549

IOGA had to give only one path of the same solution as Dijkstra’s algorithm and most importantly. IOGA returnes two solutions in 15 out of 100 cases, knowing that Dijkstra’s algorithm only gave one path. Finally for 20 cases, IOGA found several paths (between 3 and 8 paths). We had focused on the example of (SIZE = 10 and TOP = 10) because the execution time of the IOGA does this in almost the same execution time of Dijkstra’s algorithm, on the other hand if we used TOP = 35 we will have a great run time.

Table 2. Number of solutions result table.

7 Conclusion The K-shortest paths problem is the base for a lot of combinatorial optimization problems. In this paper, we focused on calculation K-shortest path problem on a road network. The K-shortest paths problem in which we seek a path corresponding to the minimum cost from an origin to a destination node in a network under some constraints. An enhanced optimization approach based on the integration of the exact algorithm (Dijkstra) and the metaheurstic algorithm (GA) is proposed to improve the performance of shortest path calculations for the road network. GA is one of the most powerful meta-heuristic methods. Through the use of a hybrid algorithm, the execution time of IOGA is decreased when solving the shortest path problem. The IOGA finds one or k paths containing the minimal cost between two vertices in a directed weighted large graph. This algorithm utilizes strategies which have proved to be efficient in solving shortest path problems. Testing was performed using generated and random networks. We foresee that the proposed algorithm will be very useful for the current intensive studies of real applications of complex networks.

550

M. Y. Hayi and C. Zahira

In the next researches we pursue the following directions: – Comparing the results of our proposal with others hybrid heuristics. – Taking different real constraints such as the problem of the shortest paths, further minimize the execution time. – Applying IOGA on the travelling salesman problem or/and Knapsack problem. – Comparing our approach with IOGA applied on non directed graph in order to enhance better performance.

References 1. Lee, J., Yang, J.: A fast and scalable re-routing algorithm based on shortest path and genetic algorithms. Seoul National University Research Park, vol. 7, September 2012 2. Kim, J., Kyun, S.: Genetic algorithms for solving shortest path problem in maze-type network with precedence constraints. Springer Science + Business Media, LLC, part of Springer Nature, 2018 3. Sun, X., Wang, J., et al.: Genetic algorithm for optimizing rout- ing design and fleet allocation of freeway service overlapping patrol. School of Transportation Science and Engineering, China, https://doi.org/10.3390/su10114120, November 2018 4. Singh, Y., Sharma, S., Sutton, R., Hatton, D.: Towards use of Dijkstra Algorithm for optimal navigation of an unmanned surface vehicle in a real- time marine environment with results from artificial potential field. Autonomous Marine Systems Group, Plymouth University, United Kingdom, vol. 12, March 2018 5. Pettie, S., Ramachandran, V., Sridhar, S.: Experimental evaluation of a new shortest path algorithm*, Department of Computer Sciences, The University of Texas at Austin, Austin, TX 78712 6. Bast, H., Delling, D., Goldberg, A., Müller-Hannemann, M., Pajor, T., Sanders, P., Wagner, D., Werneck, R.: Route planning in transportation networks (2014) 7. Khan, Z.: Comparison of Dijkstra’s Algorithm with other proposed algorithms. Virtual University of Pakistan, August 2016. https://doi.org/10.13140/RG.2.2.22743.88480 8. Johnson, S., Han, J., et al.: Hybrid Approach with Improved Genetic Algorithm and Simulated Annealing for Thesis Sampling, July 2018 9. Dib, O., Manier, M.-A., Caminada, A.: Memetic algorithm for computing shortest pathsin multimodal transportation networks. Transp. Res. Procedia 10, 745–755 (2015). https://doi. org/10.1016/j.trpro.2015.09.028.hal-01520123 10. O’Luing,M., Prestwich, S., Armagan Tarim, S.: Algorithme génétique de regroupement pour la stratification et la répartition simultanée de l’échantillon dans les plans de sondage”, No 12-001-X au catalogue ISSN 1712-5685, décembre 2019 11. Ma, C., He, R., Zhang, W.: Path optimization of taxi carpooling. School of Traffic and Transportation, Lanzhou Jiaotong University, Lanzhou, China, August 2018 12. Hamed, A.Y.: A genetic algorithm for finding the k shortest paths in a network, Department of Computer Science, Faculty of Science, Sohag University, Egypt, Egyptian Informatics Journal 11, 75–79 (2010) 13. Pettie,S., Ramachandran, V.O.,Sridhar, S.: Experimental evaluation of a new shortest path algorithm*, Department of Computer Sciences The University of Texas at Austin Austin, TX 78712 14. Hosseinabadi, A., Vahidi, J., et al.: Extended Genetic Algorithm for solving open-shop scheduling problem. Springer-Verlag GmbH Germany, part of Springer Nature, April 2018

A New Mutated-Firefly Algorithm for Parameters Extraction of Solar Photovoltaic Cell Model B. Aoufi1(&) and O. Hachana2 1

Department of Electronics and Communications, Faculty of New Information Technologies and Communication, Kasdi Merbah University, Ouargla, Algeria [email protected] 2 Department of Drilling and Rig Mechanics, Faculty of Hydrocarbons, Renewable Energies and Earth and Universe Sciences, Kasdi Merbah University, Ouargla, Algeria [email protected]

Abstract. Based on experimental data and metaheuristic algorithms, the extraction of the solar photovoltaic (PV) model remains an active research domain during the past few years. However, it is still a challenge to estimate those parameters accurately using metaheuristic algorithms. In this paper, a new hybridized approach is presented to extract the photovoltaic model parameters called Mutated Firefly (Mu-FA). This new algorithm is used to identify both the single-diode and double-diode PV model, while the estimation is based on the experimental data of the nonlinear I-V curve of the (PV) cell. The fitness function used to perform the optimization is the Root Mean Square Error (RMSE), which is a function of estimated and experimental data. In order to assess this new approach, a comparison study with the latest published metaheuristic methods has been presented. Results have shown the efficiency, robustness, and reliability of the (Mu-FA), thus it is concluded that it is a promising optimization technique for PV model parameters estimation. Keywords: Metaheuristic Differential Evolution


Photovoltaic
Optimization
Firefly algorithm


1 Introduction Solar energy is the principal type among the renewable sources in the world, because of environmental and economic considerations such as cleanliness in production facilities (no CO2 emission) and low cost in installation and maintenance operations. Algeria, being the largest country in Africa, in the Mediterranean and in the Arab world, has one of the highest solar irradiance in the world, estimated to exceed 5 billion GWh/year. The annual sunshine duration is estimated to be around 2.500 h average, and could exceed 3.600 h in some parts of the country [1]. In both industrial and academic research projects related to solar energy, an important effort has been conducted to identify the photovoltaic (PV) model parameters. In that context, theoretical studies are required, and it is obvious that this kind of study is a very attractive area for © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 551–561, 2021. https://doi.org/10.1007/978-3-030-63846-7_52

552

B. Aoufi and O. Hachana

researchers. Although, it is still a challenge to propose new accurate and reliable optimization methods to calculate the (PV) model parameters. So far, various electrical models have been developed for extracting the I–V curve of solar cells. Among others, the so-called single- and double-diode models are more often used in practice [2]. Wide variety of metaheuristic algorithms have been developed to estimate the (PV) cell/module models, such as particle swarm optimization (PSO) [3], ant colony optimization (ACO) [4], artificial bee colony (ABC) [5], genetic algorithm (GA) [6], differential evolution [10, 13], Firefly Algorithm [8]. However, combining two or more methods (hybridization) is a well-known strategy to overcome the deficiency of a single approach, and to enhance the performance of exploration and exploitation features to accurately find the global optima. The remainder of this paper describes briefly reviews the well-known models used for PV cells parameters optimization, followed by presentation of the obtained results of the proposed technique.

2 Problem Formulation This work aims to estimate the (5) five parameters of the equivalent circuit model illustrated in Fig. 1, and the (7) seven parameters of the equivalent circuit model demonstrated in Fig. 2. These two electrical models are widely used to estimate the unknown parameters by minimizing the cost function called root mean square error RMSE.

Fig. 1. The equivalent circuit of the single-diode model.

2.1

PV Cell Single-Diode Model

As can be seen, the output current of the solar cell can be written as follows [9]: IL ¼ Iph  Id  I sh

ð1Þ


  q:ðVL þ Rs :IL Þ VL þ Rs :IL IL ¼ Iph  Isd
exp 1  n:k:T Rsh

ð2Þ

where Iph denotes the photo-generated current, Isd is the reverse saturation current, VL denotes the output voltage, IL denotes the output current, RS is the series resistance, q = 1.60217646  10−19 C denotes the electron charge, n denotes the ideality constant, Rsh represents the shunt resistance, T is the cell temperature in Kelvin, and k = 1.380653  10−23 J/K, is the Boltzmann constant. For the single diode (SD) model, as can be noted from Eq. 2, the five unknown parameters to be extracted are (Iph, Isd, RS, Rsh and n).

A New Mutated-Firefly Algorithm for Parameters Extraction

2.2

553

PV Cell Double-Diode Model

The SD model does not consider the effects of recombination current loss in the depletion region. To improve the accuracy of SD model, a double diode (DD) model is obtained when considering the recombination current loss In DD model Fig. 2, the output current is calculated as follows [9]: IL ¼ Iph  Id1  Id2  I sh

ð3Þ




  
  q:ðVL þ Rs :IL Þ q:ðVL þ Rs :IL Þ IL ¼ Iph  Isd1
exp  1  Isd2
exp 1 n1 :k:T n2 :k:T VL þ Rs :IL  ð4Þ Rsh

Fig. 2. The equivalent circuit of the double-diode model.

where Id1 and Id2 are the first and second diode currents, n1 and n2 represent the recombination and diffusion diode ideality constants, Isd1 and Isd2 are the saturation and diffusion currents, respectively. As can be seen from Eq. 4, in DD model, seven parameters (Iph, Isd1, Isd2, RS, Rsh, n1 and n2) need to be extracted [9]. 2.3

Cost Function

The solar PV model parameter extraction problem is generally transformed into a numerical optimization problem by minimizing the difference between the measured data and simulated ones. The error function is generally defined as the root mean square error (RMSE) as follows [9]: vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi u N u1 X RMSEðxÞ ¼ t f ðVL ; IL ; xÞ2 N K¼1 where N denotes the number of measured I-V data.

ð5Þ

554

B. Aoufi and O. Hachana

For the (SD) model, the RMSE is obtained from Eq. 5 as follow [9]: 



f ðVL ; IL ; xÞ ¼ Iph Isd
exp

q:ðV

L þ Rs :IL Þ n:k: T



1 

x¼ðIph ; Isd ; Rs ;Rsh ; nÞ

VL þ Rs :IL IL Rsh

ð6Þ

For the (DD) model, the RMSE is obtained as shown in Eq. 7 [9]: h

i h

i 8

> r ¼x x ¼ ðxi;k xj;k Þ2 < ij i j k¼1

xi ¼xi þ bðxj xi Þ þ a 2i > > :

ð9Þ

Where a is a parameter that controls the step and can gradually vary with time until reach zero. 2i is a vector of random values from a uniform random variable in the range of [− 1, 1].

A New Mutated-Firefly Algorithm for Parameters Extraction

3.2

555

Mutated-Firefly

The mutation operation introduced in this work is inspired from the differential evolution (DE) algorithm. This algorithm is an evolutionary optimization method [10, 13]; it is based on the two operations of mutation and crossover. – Bx is the vector to be mutated (it is chosen from the current population as the best cost vector). – By is the number of difference vectors used. The donor vector Hi created by mutation process is defined as shown in Eq. 10: Hi ¼ Bx þ Q:ðBy  Bz Þ

ð10Þ

where Q denotes the mutation factor; and x, y, z, and i are different. The donor vector is adjusted according to the Eq. 11 to avoid exceeding the search space range limits: Hi;d ¼

n

Xmax;d if Hi;d iHmax;d Xmin;d if Hi;d hXmin;d

ð11Þ

To introduce this mutation operation to the (FA) firefly algorithm, the ith firefly (xi) with the best cost function is chosen to be mutated as follows: Hi ¼ Bxi þ Q:ðBxy  Bxz Þ

ð12Þ

The next step is to evaluate the new mutated vector Hi, if it has a better cost than the one chosen from the fireflies population, it will be kept. If it is not the case, then the ith firefly (xi) will be kept for the next generation. The different steps of the algorithm are illustrated in the flowchart Fig. 3.

4 Mutated-Firefly Algorithm for Parameter Extraction of the PV Cell Model For examination of the proposed algorithm (Mu-FA), the optimization problem of extracting the PV solar models is solved using this algorithm. The search range of the parameters is listed in Table 1 [9]. Table 1. Bounds of the solar cell models Parameter Iph (A) Isd (lA), Isd1 (lA) and Isd2 Rs (X) Rsh (X) n, n1, and n2

Single/double diode Photowatt-PWP-201 Lower bound Upper bound Lower bound Upper bound 0 1 0 2 (lA) 0 1 0 50 0 0.5 0 2 0 100 0 2000 1 2 1 100

556

B. Aoufi and O. Hachana

The proposed (Mu-FA) algorithm is compared with many metaheuristic algorithms well known in literature [9]. The results of extracting both SD and DD model are shown in Table 2 and Table 3 respectively, where the best values calculated using these algorithms are mentioned in bolt font.

Fig. 3. (Mu-FA) flowchart

Table 2. Comparison of results of various algorithms for the single-diode model Algorithm

Iph (A)

Isd (µA) Rs (Ω)

Rsh (Ω)

n

RMSE

MADE [9] CS [9] BHCS [12] LBSA [9] DE/BBO [9] CWOA [9] GWOCS [9] Proposed algorithm Mu-FA

0,760800 0,760480 0.76078 0,760760 0,760500 0,76077 0,760773 0,760776

0,32300 0,36015 0.32302 0,34618 0,32477 0,3239 0,32192 0,32302

53,7185 43,8423 53.71852 59,0978 55,2627 53,7987 53,6320 53,7181

1,4812 1,4929 1.4811 1,4881 1,4817 1,4812 1,4808 1,4812

9,86020E−04 2,01185E−03 9.8602E−04 1,01430E−03 9,99220E−04 9,86020E−04 9,86070E−04 9,86020E−04

0,036400 0,034920 0.03638 0,036200 0,036400 0,03636 0,036390 0,036377

A New Mutated-Firefly Algorithm for Parameters Extraction

557

Table 3. Comparison of results of various algorithms for the double-diode model Algorithm

Iph (A)

Isd1 (µA) Isd2 (µA) Rs (Ω) Rsh (Ω)

n1

n2

RMSE

MADE [9] CS-BBO [9] CWOA [9] ABC [2] ABSO [2] BMO [13] GWOCS [9] Proposed algorithm Mu-FA

0,760800 0,760780 0,760770 0,76080 0,7608 0,76078 0,76076 0,76078

0,739400 0,749350 0,241500 0,040700 0,267130 0,21110 0,53772 0.74936

1,9963 2 1,4565 1,4495 1,4651 1,4453 2,0000 1,9999

1,4505 1,4510 1,9899 1,4885 1,9815 1,9999 1,4588 1,3961

9.8261E−04 9.8249E−04 9,8272E−04 9,8610E−04 9,8344E−04 9.8262E − 04 9,8334E−04 9,8248E−04

0,224600 0,2260 0,600000 0,287400 0,3819 0,876880 0,248550 0.225974

0,0368 0,0367 0,0367 0,0364 0,0366 0,0368 0,0367 0,0367

55,4329 55,4854 55,2016 53,7804 54,6219 55,8081 54,73310 55,48544

This extraction approach is validated using the real measurement of I–V characteristics of a PV cell (RTC France silicon PV cell A 57 mm diameter), under 1000 W/m2 of irradiance at 33° C PV temperature cell, represented by IL and VL in Table 4 [13]. Table 4. Estimated values of current and individual absolute error (IAE) obtained by Mu-FA for the SD and DD model Item VL (V)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26

IL Measured (A)

−0.2057 0.7640 −0.1291 0.7620 −0.0588 0.7605 0.0057 0.7605 0.0646 0.7600 0.1185 0.7590 0.1678 0.7570 0.2132 0.7570 0.2545 0.7555 0.2924 0.7540 0.3269 0.7505 0.3585 0.7465 0.3873 0.7385 0.4137 0.7280 0.4373 0.7065 0.4590 0.6755 0.4784 0.6320 0.4960 0.5730 0.5119 0.4990 0.5265 0.4130 0.5398 0.3165 0.5521 0.2120 0.5633 0.1035 0.5736 −0.0100 0.5833 −0.1230 0.5900 −0.2100

I Estimated (A) Single Diode model

Double Diode model

IAE (IL) Single Diode model

Double Diode model

0.76180 0.76096 0.76018 0.75946 0.75880 0.75819 0.75759 0.75692 0.75604 0.75464 0.75215 0.74759 0.73949 0.72560 0.70396 0.67126 0.62634 0.56793 0.49678 0.41258 0.31828 0.21444 0.10518 −0.00666 −0.12542 −0.21146

0.76397 0.76260 0.76134 0.76018 0.75912 0.75813 0.75720 0.75626 0.75519 0.75373 0.75140 0.74729 0.73999 0.72723 0.70684 0.67520 0.63076 0.57200 0.49972 0.41374 0.31755 0.21212 0.10215 −0.00880 −0.12555 −0.20836

0.00220 0.00104 0.00032 0.00104 0.00120 0.00081 0.00059 0.00008 0.00054 0.00064 0.00165 0.00109 0.00099 0.00240 0.00254 0.00424 0.00566 0.00507 0.00222 0.00042 0.00178 0.00244 0.00168 0.00334 0.00242 0.00146

0.00003 0.00060 0.00084 0.00032 0.00088 0.00087 0.00020 0.00074 0.00031 0.00027 0.00090 0.00079 0.00149 0.00077 0.00034 0.00030 0.00124 0.00100 0.00072 0.00074 0.00105 0.00012 0.00135 0.00120 0.00255 0.00164

558

B. Aoufi and O. Hachana

The results of the proposed algorithm (Mu-FA) are shown in figures below. Figure 4 and Fig. 5 describe the evolution of the cost function RMSE for the SD model and DD model respectively. While, Fig. 6 and Fig. 7 demonstrate the measured and estimated I–V curves for the SD model and DD model respectively. Finally, Fig. 6 and Fig. 7 show the measured and estimated P-V curves for the SD model and DD model respectively.

Fig. 4. Evolution of RMSE for the SD model

Fig. 5. Evolution of RMSE for the DD model.

Fig. 6. I-V curves for the SD model.

A New Mutated-Firefly Algorithm for Parameters Extraction

559

Fig. 7. I-V curves for the DD model.

Fig. 8. P-V curves for the SD model.

Fig. 9. P-V curves for the DD model.

As can be seen from Fig. 4 and Fig. 5, the cost function RMSE reaches its minimum value when the number of iterations is at It = 8 for both (SD) and (DD) models. That is to say, that (Mu-FA) has a very high conversion speed compared with other metaheuristic methods that need a number of iterations much higher to reach the

560

B. Aoufi and O. Hachana

minimum value (more than 500 iterations in general). For the (SD) model, Fig. 6 and Fig. 8 show that the estimated I–V curve is matching the measured one and the P–V (Power-Voltage) curve is matching the estimated one respectively. The same observation is for the (DD) model as it is shown in Fig. 7 and Fig. 9 respectively. Finally, it is worth saying, that the population size needed for the (Mu-FA) algorithm is small, compared with other methods well-known in literature [9]. For example, for the (SD) model, the (Mu-FA) parameters are set as follows: Max Iteration = 10; Population size = 25; the results are as follows: RMSEmin = 9.9802E−4 corresponds to the 9th iteration, and the execution time T = 1.607 s. The algorithm is tested 10 times concluding to a standard deviation STD = 1.9295E−17.

5 Conclusion In this paper, a new hybrid optimization technique based on combination between FA and the mutation operator of the DE algorithm is proposed to identify the unknown parameters of the single-diode and the double-diode PV model. Depending on the results provided, the proposed algorithm (Mu-FA) shows the best performance in terms of robustness, conversion speed, which means that (Mu-FA) is a promising candidate approach for solar PV models parameters extraction. For future perspective, parameters extraction of photovoltaic modules is to be performed suing this proposed algorithm.

References 1. The People’s Democratic Republic of Algeria. Intended Nationally Determined Contribution INDC-Algeria September 3rd, 2015. https://www4.unfccc.int/sites/submissions/indc/ Submission%20Pages/submissions.aspx 2. Jamadi, M., Merrikh-Bayat, F., Bigdeli, M.: Very accurate parameter estimation of singleand double-diode solar cell models using a modified artificial bee colony algorithm. Int. J. Energ. Environ. Eng. 7(1), 13–25 (2016) 3. Soon, J.J., Low, K.S.: September. photovoltaic model identification using particle swarm optimization with inverse barrier constraint. IEEE Trans. Power Electron. 27(9), 3975–3983 (2012) 4. Dorigo, M., Maniezzo, V., Colorni, A.: February. ant system: optimization by a colony of cooperating agents. IEEE Trans. Syst. Man Cybern.-Part B Cybern. 26(1), 29–41 (1996) 5. Jadhav, H.T., Roy, R.: November: gbest-guided artificial bee colony algorithm for environmental/economic dispatch considering wind power. J. Expert Syst. Appl. 40(16), 6385–6399 (2013) 6. Digalakis, J.G., Margaritis, K.G.: An experimental study of benchmarking functions for genetic algorithms. Int. J. Comput. Math. 79(4), 403–416 (2002) 7. Storn, R., Price, K.: Minimizing the real functions of the ICEC 1996 contest by differential evolution. In: Proceedings of the IEEE Conference on Evolutionary Computation, pp. 842– 844 (1996) 8. Yang, X.S.: Firefly algorithms for multimodal optimization. In: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), LNCS, vol. 5792, pp. 169–178 (2009)

A New Mutated-Firefly Algorithm for Parameters Extraction

561

9. Long, W., Cai, S., Jiao, J., Xu, M., Wu, T.: A new hybrid algorithm based on grey wolf optimizer and cuckoo search for parameter extraction of solar photovoltaic models. Energ. Conver. Manag. 203, 112243 (2020) 10. Storn, R., Price, K.: Differential evolution - a simple and efficient heuristic for global optimization over continuous spaces. J. Glob. Optim. 11(4), 341–359 (1997) 11. Easwarakhanthan, T., Bottin, J., Bouhouch, I., Boutrit, C.: Nonlinear minimization algorithm for determining the solar cell parameters with microcomputers. Int. J. Solar Energ. 4(1), 1– 12 (1986) 12. Chen, X., Yu, K.: Hybridizing cuckoo search algorithm with biogeography-based optimization for estimating photovoltaic model parameters. J. Solar Energ. 180, 192–206 (2019) 13. Hachana, O., Hemsas, K.E., Tina, G.M., Ventura, C.: Comparison of different metaheuristic algorithms for parameter identification of photovoltaic cell/module. J. Renew. Sustain. Energy. 5 (2013)

A Survey on Cloud-Based Intelligent Transportation System Y. Khair(&), A. Dennai, and Y. Elmir Department of Computer Science, University of Tahri Mohammed, SGRE Laboratory, Bechar, Algeria [email protected], [email protected]

Abstract. Industrial development in the eighteenth century increased dependence on transportation, which led to the emergence of a group of traffic problems such as traffic congestion, increased travel time, as well as energy consumption, and these problems were solved by traditional methods, by creating new roads or expanding roads Existing congestion relief. However, the technological development that accompanied the twentieth century, in turn, provides a set of technologies to find appropriate solutions to these problems. Intelligent Transportation Systems (ITS) and Cloud computing are one of the most important computers and advanced communications technologies to control traffic flow to increase the productivity and safety of existing road infrastructure. The aim of this paper is to present promising solutions in the use of cloud computing for intelligent transportation systems to improve public transport results and discuss all aspects related to it (its concept, systems, and applications). Keywords: Intelligent transportation systems (ITSs)
Cloud computing
Transportation services
IaaS
PaaS
SaaS
Communication technologies (ICT)

1 Introduction The technological development that accompanied the end of the twentieth century and the beginning of the twenty-first century represented by the emergence of communication and information technologies, was mainly reflected in the form of life and the way in which the various activities are performed, leading to the emergence of a society of a new style that is increasingly dependent on digital knowledge and technologies, and performs various activities through means Virtualization instead of the usual means. The increasing number of urban residents and their increasing dependence on transportation has led to traffic problems in the urban structure that were not designed to accommodate this traffic census [1]. Previously, the approach to solving the congestion problem was to build more roads or expand them in order to increase their capacity while maintaining the same pattern in managing These facilities; however, this approach, even if it achieves the desired goal in the short term, but it constitutes in the long term an increased financial, operational and environmental burden [2]. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 562–572, 2021. https://doi.org/10.1007/978-3-030-63846-7_53

A Survey on Cloud-Based Intelligent Transportation System

563

So it was necessary to find smart solutions that aim to make the most of the unused potential capacity of the road by relying on modern technologies to reach more efficient traffic management. Hence intelligent transportation systems harnessed the latest technologies in Data Processing, Detection, Control, and Communications, that enable increasing the operational efficiency and capacity of the transportation system [3]. On the other hand, cloud computing has become a popular and effective technology where users can access any kind of computer resources such as shared resources, platforms, programs, and data in data centers distributed over a network such as the Internet [4]. Network computing is another common approach to managing resources from network servers and devices… etc. As access and dealing with it became geographically as a single computer with huge potential, and research conducted in recent years has shown that many service companies are moving towards cloud computing and restructuring their business in proportion to the cloud environment [5]. Hence, it was reasonable to apply cloud technology to solve the problems of intelligent transportation systems.

2 Intelligent Transportation System (ITS) 2.1

Intelligent Transportation Concept

Intelligent transportation systems (ITS) are defined as a set of integrated applications for sensors, computers, communication technologies, and electronics [6]. It aims to provide innovative services for traffic management and transportation [7]. To increase the efficiency of transportation systems and enhance traffic safety [2]. Intelligent transportation systems integrate information and communication technologies into existing transportation management systems that help avoid traffic congestion and accidents who increase with population growth and cause many undesirable effects such as long travel time, air pollution, and fuel consumption [8] and employ technologies to obtain information on the performance of transportation facilities, on-demand for transportation and mutual communication between vehicles themselves and between them and between roadside devices, as well as impending collision accidents [4]. These technologies provide smart cities by reducing the need for mobility, increasing passenger and cargo density in vehicles, and creating more efficient transportation networks [9] by offering cars and road infrastructure with intelligent systems that enhance connectivity. Intelligent transportation systems affect the performance of the transportation system with six main goals [10]: traffic Safety, Mobility, Efficiency, Productivity, Energy and Environment, User Satisfaction. Traffic safety is measured by changes in the accident rate or other measures such as vehicle speed, traffic collision, and traffic violations. Mobility improvements are measured by flight time or delays, as well as by flight cost and arrival time. Efficiency is created through a better ability to manage transportation facilities to accommodate additional demand, usually by increasing capacity or level of service within the existing transportation network or transportation systems. Productivity improvements achieve cost savings for transportation service

564

Y. Khair et al.

providers, travelers, and shipping companies. Benefits the field of energy and the environment by reducing fuel and reducing polluting emissions [11]. 2.2

Intelligent Transportation System Components

It consists of detection technologies that monitor the operations of the transportation system, and data processing that includes computer programs that process system performance information, as well as modern electronic technologies that provide information to people, and communication networks that flow data, as well as control tools [12]. (e.g., Fig. 1).

Fig. 1. The components of the intelligent transportation system

• Data collection methods: Devices that collect various data necessary for traffic flow, occupancy, and speed, and include parking cameras and public transport, as well as sensors • Data processing techniques: It is the software and hardware that work on processing data in a way that responds to the changes that occur in these systems. • Control and information transfer techniques: The techniques used to transform the results of data processing into reality, and it includes different types of traffic signs, dynamic message signals (DMS), and means of transmitting information to system users such as radio, mobile phone, and communication networks. 2.3

Intelligent Transportation System Subsystems

Most of the general terms used for the different subsystems of ITS in the United States have remained unchanged for several years. Somewhat similar subsystems have been adapted for use in Japan [13]. Intelligent transportation systems can be divided into five subsystems. (Table 1) provides a summary of the classification [12].

A Survey on Cloud-Based Intelligent Transportation System

565

Table 1. Classification ITS subsystems ITS subsystems Advanced traffic management systems

Applications services Driving aids Emergency events management Online cost recovery services Emergency and personal safety reports Advanced traveler information systems Information before the trip Information during the trip Route guidance Commercial vehicle operations system Electronic clearance of utility vehicles and freight automated road safety control Vehicle administrative operations Advanced public transportation systems Public transportation management Electronic payment Public on-demand transportation Vehicle control and safety systems Intelligent cruise controls Obstacle detection Automated driving

3 Cloud Computing Technology Cloud computing is defined as a technology that relies on transferring processing and private storage space, and the concept of cloud computing is defined as a server device that is accessed via the Internet. Thus, IT programs are transformed from products to services. [1] and the cloud computing infrastructure depends on advanced data centers that provide large storage spaces for users and also provide some programs as services to users. It depends on the capabilities provided by Web 2.0 technologies [14] Cloud computing, in terms of the services it provides, is divided into three main models, in the first type which is called “Infrastructure as a Service” (IaaS) which is the bottom layer of cloud computing, It provides computing infrastructure, and instead of buying servers, data center or network equipment spaces, customers buy these resources as a completely independent service. In the second type, “Platform as a Service” (PaaS), this type of cloud computing consists mainly of libraries, middleware, updates, and runtime tools that developers need in updating the software application as a service. The third type is known as “providing software as a service” (SaaS) [15], which is more concerned with the implementation of remote end-user applications, such as e-mail systems, customer relationship management applications [3]. It is worth noting that cloud computing can be divided in terms of deployment and application into two models, the Public and Private Cloud [9]. In the Public Cloud, services are sold to general Internet customers. Here, Amazon Elastic Compute Cloud (EC2) and Google App Engine are major public cloud service providers. The Private Cloud is known as the Interior Cloud, where it is a private property of a specific company or entities [9].

566

Y. Khair et al.

One of the IT requirements is to deal with different situations, including unexpected situations. In general, dealing with these situations highlights one of the forms of cloud computing intelligence, and its ability to feel sensitive to the surrounding environment, for example, Communications as a Service (CaaS) [16] that aims to encapsulate the capabilities of traditional communications and these communications may include VoIP and instant messaging. Moreover, it provides a network as a service (NaaS) [17], through network virtualization, which has become a vital role in modern cloud computing and acts as a bridge between the various elements. Cloud computing acquires particular importance when it represents one of the possible solutions in smart transportation systems, which require procedures and decisions to secure the minimum level of services and the quality of services [15].

4 Cloud-Based ITS Cloud computing provides more ability to manage resources from servers, network devices, etc. [18] so that dealing with them has become geographically as a single computer with huge capabilities through distributed virtualization techniques [3]. Hence, cloud computing technology has become an efficient technology, which makes the transportation system more efficient, faster, easier, and more reliable [9]. A. Fornaia et al. [2] provides a solution that allows clients to request a public vehicle for additional performance that stops on the main road, through high-level operations for both sides, from the client-side that gives users support for book transport vehicles, while server-side support is provided to deploy, implement, and monitor services on the Cloud. The authors provide a structured design Cloud infrastructure, and distributed architecture is suitable for assessing whether user requests can be accepted. The proposed software solution considers that requests that can accommodate available secondary paths are applicable, while also satisfying other pre-approved user requests. After that, drivers will be alarmed ahead of time to adjust their way. According to the workflow. Server-side components are independent of specific workflows and can implement services in a variety of ways to properly control the lifetime of the services. Moreover, a component is specially designed to plan the pre-requisite transportation service by modeling incoming requests and analyzing noise removal while identifying recurring trends. This is done by transferring the data to the wavelength field before giving it to the neural network component. Then, we are able to start operations, such as planning vehicle lanes and driver spells, and avoiding oversavings. S. Bitamet al. [9] he proposed a new vehicle cloud architecture called ITS-Cloud based on two sub-models: the Conventional cloud sub-model that is defined as a set of interconnected and paralleled and distributed computers that are considered as a complete computing system, and a Temporary (vehicular) cloud sub-model, which he presented as From a group of interconnected and/ or passenger vehicles initially placed in the automotive area as customers. In order to verify the effectiveness of the ITSCloud system, the authors suggested a load-balancing study using the Bees Life (BLA) algorithm, where a set of tests were performed on both types of clouds: Conventional cloud and ITS-Cloud, results obtained were obtained Access them using the

A Survey on Cloud-Based Intelligent Transportation System

567

Bees Life (BLA) algorithm applied to ITS-Cloud and compare to those accessed by the (BLA) only applied to the traditional cloud. The fully developed On-demand Bus system differs from the On-demand Bus system from the point of view of cloud computing technology K. Tsubouchi et al. [3]. The innovative On-demand Bus system has been developed as the public transportation service has adapted to the needs of car users, after reserving their places, and the car does not move if there is no reservation. The goal of the innovative bus system is to introduce the system in practice from a cost perspective. In order to build the system, there are four main components required: a scheduling system and a communication device in the car in addition to the presence of the reservation interface and the database P. Jaworskiet al. [5], in this study, designed a system of urban traffic control that relies on cloud computing, based on a component called the Intersection Control Service (ICS). The proposed system combines geographical addressing and cloud service detection mechanisms to request routing. Its main objectives are to manage traffic flow at the district level with the aim of increasing traffic productivity and increasing the safety of all traffic participants. Where vehicles are treated as cloud services, they are discovered and called using the cloud computing methodology. Different geographical addressing is used to focus all vehicles in the specified areas. The present study concluded the implementation of the system in its advisory form on public roads (in a limited test area), which allows realistic proof of reality without affecting the safety of participants in traffic. The vehicle can represent a set of next-generation sensors that reach the rich PaaS service platform. D. Bernstein et al. [1] concluded that the required capabilities resulting from such a cloud may exceed the capabilities of the current PaaS system. In this study, the authors suggest designing a PaaS system with mechanisms different from other existing PaaS systems that simulate millions of cars on the road. These requirements were taken and modified a typical PaaS platform architecture to accommodate, with the additional capabilities the typical PaaS platform architecture to accommodate, with additional capabilities. F. Sardis et al. [18]. In this study, the trend was to exploit the “Open” cloud, as this paper presented a new concept for Cloud-Based Mobile Media based on the cloud, considering that moving the user (mobile devices such as smartphones and tablets) from one geographical location to another in accessing these services from the local Cloud of their previous network will lead to increased congestion on the Internet and the transfer of a large amount of data, and thus the use of existing service models will lead to a deterioration in the quality of service and will not expand to cover the future needs of mobile media users. The authors believe that it is more efficient to move those services closer to that location. This will prevent the Internet backbone from facing high traffic loads due to multimedia streams and will provide service providers with a mechanism for allocating resources and automatic management for their services. By running these services on the local public cloud and are capable of populating other public Clouds to different geographical locations depending on service requirements and network status, it can provide a better solution to efficient network resource management while providing high QoS to clients. K. Ashokkumar et al. [4]. This study made contributions by proposing a standard and multi-layered transport information cloud platform. Simultaneously supports cloud

568

Y. Khair et al.

computing and IoT technologies. This platform supports mechanisms that enable it to collect and exchange knowledge between drivers. The authors suggested two models for changed data processing, a Naïve Bayes model, and a supplying Regression model. The goal of this platform, in the opinion of the authors, is to produce secure and customized time services for clients through clouds associated with a traditional cloud and vehicular cloud. For data exchange and resource sharing, heterogeneous devices related to the Internet of Things, networks, community technologies, and cloud-based services are combined on completely different layers. A. RAI et al. [14]. A generic infrastructure design has been developed to implement customized intelligent transportation systems in urban areas. The main idea of this structure is the interoperability between machine-to-machine technology and cloud computing that allows for seamless configuration and consumption of the service as well as speed in the deployment of new services by grouping different devices and access networks that may be owned by different actors such as telecom system providers and transportation service companies, And government organizations. The importance of the connection and communication between M2M and the cloud lies in the authors’ view on two aspects. It allows intelligent interactions between the consumer’s cloud consumer for many Internet of Things applications, and the other is the modeling of cloud applications in its economic aspect. The improvement of bus-based transportation operations in the city effectively helps save fuel by providing urban transportation services as efficiently as possible, M. Mathirajan et al. [15]. In this study [15], a Cloud-Based Decision Support System (C-DSS) was developed. The idea of (C-DSS) is based on a smart model that allows integrating the strategic problem in the warehouse site (to add new locations and remove existing sites) and allocating city buses to warehouses. For users to access this service, C-DSS-URTS was developed as a SaaS (Software as a Service) model that can be used in any browser. The study concludes with the proposed solution that enables an almost real decision to be obtained through the use of cloud computing technology with an easy-to-use graphical user interface based entirely on practical heuristic algorithms. In this paper, R. Zhang et al. [19] present a solution for intelligent traffic signal control, as they proposed a new algorithm based on deep learning. This algorithm helps mitigate performance loss and eliminate basic safety issues in Cloud-based traffic systems. The simulation results showed that the algorithm was able to withstand approximately 3 s of total delay without significant performance loss. Therefore, R. Zhang et al. [19] believes that the algorithm can be adopted in the real world and for all agent-based systems using remote computing resources. To summarize the various works cited above, we have conducted a comparative study, which will be described in detail below (Table 2): In (Table 3), we classified the studies according to the technical requirements of the intelligent transportation system in the formation of one of the functions based on cloud technology. From (Table 3), it is clear that cloud computing can process a large amount of traffic data and information of a different type that can be useful for extracting knowledge and adding intelligence to the transportation system.

A Survey on Cloud-Based Intelligent Transportation System

569

Table 2. Comparative study between related works

Table 3. Classification of the technical requirements ITS in the use of cloud technology. Cloud service models Technical Requirements Data collection Data processing Control and information dissemination

IaaS

PaaS SaaS

[4] [1] [2, 4] [5, 9, 14, 18]

[2, 3] [15]

5 Opportunities and Challenges The use of intelligent transportation systems contributes positively to solving mobility problems in general and addressing them to reach a balanced transportation system. Authors such as [1–3, 9, 20, 21] acknowledge the importance of implementing intelligent transportation systems, which are a reliable and useful tool in planning, monitoring, operating and managing transportation, and a factor that contributes to improving the quality of life, considering that transportation is an essential element in economic growth and development the society…

570

Y. Khair et al.

Currently, intelligent transportation systems are in place and used around the world, but high operational costs such as operating technical requirements and the cost of employing operators. In particular, it is a problem for governments and commercial companies to start and maintain this service. By using cloud computing, it will be able to take advantage of the extensive operating infrastructure, integrate its investments, services, and data centers, and use it in a completely different way and then reduce its costs. Cloud computing models are excellent platforms for data processing, acquisition, storage, and analysis of data. This technology was an ideal tool for processing large amounts of traffic data and monitoring the performance of network transportation systems. Thus improving transportation results such as road safety, travel reliability, and informed travel options. Given the foregoing, cloud computing is the main complement to the typical technological progress of intelligent transportation systems. However, as in any revolutionary technology, there are fundamental challenges that prevent widespread adoption. Cloud computing is no exception to this, perhaps the most prominent of which is the challenges: • Infrastructure: Like all other technologies, infrastructure is a key factor, although it is not about creating new infrastructure because cloud computing is heavily dependent on the Internet and networks, but in intelligent transportation systems access to services in real-time is important; therefore, the Internet needs broadband, and mobile devices via different geographical locations need major networks… etc. • Data management: One of the most important challenges is storage capacity. With the increasing number of cloud users and cloud service providers, it is necessary to address the issue of dealing with increasing amounts of data in an appropriate manner and to make the required data available to users to provide high-quality services at the right time. • Operation requirements: It is also one of the major challenges, as the leading companies in the ITS field face fear for their business and their suitability with cloud computing. • Data migration: The stage of data migration from the traditional form of intelligent transportation systems to those based on cloud computing is one of the most prominent challenges for users. Multiple data sources, which are usually of different types and representational forms, may lead to incompatibility with those in cloud computing. • Security: Despite the advantage of multiple data entry points in the cloud, it faces a major challenge in facing security threats regarding this data. Cloud computing is completely dependent on the service provider and the level of security it provides, such as information encryption, policy development, and procedures for accessing the cloud.

A Survey on Cloud-Based Intelligent Transportation System

571

6 Conclusions In this survey, we reviewed the contribution of cloud computing to intelligent transportation systems. We have briefly outlined the main concepts, applications, and methods used in both fields. Then we categorized possible solutions for intelligent computing systems based on cloud computing technology. Our literature survey shows that cloud computing technology will undoubtedly contribute to the development of intelligent transportation systems and make them more efficient, faster, easier, and reliable.

References 1. Bernstein, D., Vidovic, N., Modi, S.: A cloud PAAS for high scale, function, and velocity mobile applications-with reference application as the fully connected car. In: 2010 Fifth International Conference on Systems and Networks Communications, pp. 117–123 (2010) 2. Fornaia, A., Napoli, C., Tramontana, E.: Cloud services for on-demand vehicles management. Inf. Technol. Control 46, 484–498 (2017) 3. Tsubouchi, K., Yamato, H., Hiekata, K.: Innovative on-demand bus system in Japan. IET Intell. Transp. Syst. 4, 270–279 (2010) 4. Ashokkumar, K., Sam, B., Arshadprabhu, R.: Cloud based intelligent transport system. Proc. Comput. Sci. 50, 58–63 (2015) 5. Jaworski, P., Edwards, T., Moore, J., Burnham, K.: Cloud computing concept for intelligent transportation systems. In: 2011 14th International IEEE Conference on Intelligent Transportation Systems (ITSC), pp. 391–936 (2011) 6. Kaffash, S., Nguyen, A.T., Zhu, J.: Big data algorithms and applications in intelligent transportation system: a review and bibliometric analysis. Int. J. Product. Econo. 231, 107868 (2020) 7. Farahani, R.Z., Miandoabchi, E., Szeto, W.Y., Rashidi, H.: A review of urban transportation network design problems. Euro. J. Oper. Res. 229, 281–302 (2013) 8. Liebig, T., Piatkowski, N., Bockermann, C., Morik, K.: Dynamic route planning with realtime traffic predictions. Inf. Syst. 64, 258–265 (2017) 9. Bitam, S., Mellouk, A.: Its-cloud: Cloud computing for intelligent transportation system. In: 2012 IEEE Global Communications Conference (GLOBECOM), pp. 2054–2059 (2012) 10. Maccubbin, R.P., Staples, B.L., Kabir, F., Lowrance, C.F., Mercer, M.R., Philips, B.H., Gordon, S.R.: Intelligent transportation systems benefits, costs, deployment, and lessons learned: 2008 update, (2008) 11. Sobral, T., Galvão, T., Borges, J.: Visualization of urban mobility data from intelligent transportation systems. Sensors 19, 332 (2019) 12. Ezell, S.: Intelligent Transportation Systems/Ezell Stephen, ITIF. Inf. Technol. Innov. Found. 1–86 (2010) 13. Enoch, M., Ison, S., Laws, R., Zhang, L.: Evaluation study of demand responsive transport services in Wiltshire, Final Report, Wiltshire County Council, Trowbridge, Wiltshire (2006) 14. RAI, A., K.R, J.: Universal infrastructure of m2m enabled inter-cloud services for intelligent transportation system. Asian J. Pharm. Clin. Res. 239–243 (2017) 15. Mathirajan, M., Devadas, R., Ramanathan, R.: Transport analytics in action: a cloud-based decision support system for efficient city bus transportation. J. Inf. Optim. Sci. 1–46 (2020)

572

Y. Khair et al.

16. Hussein, M.K., Mousa, M.H., Alqarni, M.A.: A placement architecture for a container as a service (CaaS) in a cloud environment. J. Cloud Comput. 8, 7 (2019) 17. Boubendir, A., Bertin, E., Simoni, N.: On-demand dynamic network service deployment over NaaS architecture. In: NOMS 2016–2016 IEEE/IFIP Network Operations and Management Symposium, pp. 1023–1024 (2016) 18. Sardis, F., Mapp, G., Loo, J., Aiash, M., Vinel, A.: On the investigation of cloud-based mobile media environments with service-populating and QoS-aware mechanisms. IEEE Trans. Multi. 15, 769–777 (2013) 19. Zhang, R., Zhou, X., Tonguz, O.K.: Using AI for Mitigating the Impact of Network Delay in Cloud-based Intelligent Traffic Signal Control (2020) 20. Cai, W.H., Sun, T.T.: CTS: the new generation intelligent transportation system.In: 2011 Second International Conference on Innovations in Bio-inspired Computing and Applications, pp. 137–140 (2011) 21. Dong, Z., Xiaoguang, Y., Haode, L., Jing, T.: Internet-based advanced traveler information service: opportunities and challenges. In: 2010 International Conference on Optoelectronics and Image Processing, pp. 646–650 (2010)

Artificial Neural Network Based Solar Radiation Estimation of Algeria Southwest Cities D. Benatiallah1(&), K Bouchouicha2, A Benatiallah1, A. Harouz1, and B. Nasri1 1

2

Laboratory of Sustainable Development and Computer Science (LSDCS), Faculty of Sciences and, Technology University Ahmed Draia, 01000 Adrar, Algeria [email protected] Unité de Recherche en Energies Renouvelables en Milieu Saharien, URERMS, Centre de Développement des Energies Renouvelables, CDER, 01000 Adrar, Alegria

Abstract. Solar radiation estimation is the most integral part of design and performance of solar energy applications. Our paper aims to develop an artificial neural networks-based model for predicting the daily global solar radiation in three cities (Bechar, Naâma and Tindouf) in the south-west region of Algeria. Models’ inputs are: average temperature, wind speed, relative humidity, atmospheric pressure, extraterrestrial solar irradiation, sunshine duration. Three ANN multilayer architectures connection are used with the Levenberg-Marquardt algorithm for training. Efficiency of models was assessed using statistical tests including, correlation coefficient (R), root mean squared error (RMSE), mean bias error (MBE) and mean absolute percentage error (MAPE).The results during five years showed that, the Cascade-forward Neural Network (CFNN) and Feed-forward neural network (FFNN) models gives much better forecast of daily global solar radiation in the three Saharan cities. The models developed can be used for design and sizing solar energy systems, where radiation measuring stations are scarce in Algeria. Keywords: Artificial neural network
Solar radiation
Energy systems Cascade-forward
Feed-forward
Levenberg-Marquardt




1 Introduction Solar energy is becoming increasingly desirable in the 21st century, as the environmental problems caused by fossil fuel burning are becoming more serious. The shortage of fossil fuels is a worldwide long-term challenge that needs us striving for renewable and sustainable energy resources. Solar energy is highly desirable, since it is

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 573–583, 2021. https://doi.org/10.1007/978-3-030-63846-7_54

574

D. Benatiallah et al.

abundant in many parts of the planet. Solar energy can be considered the most suitable renewable source for meeting energy demand in urban as well as rural areas. Solar radiation data are indispensable for designing and assessing solar energy utilization technologies. Practically measured data is the most accurate but not always readily available, which is mainly due to the initial investment and maintenance cost of the measuring instruments and relevant recorders [1–3]. Solar system researchers developed several models and soft computing techniques for estimating solar radiation using different astronomical and meteorological parameters. In recent years, many artificial intelligence techniques have been used to predict solar irradiance, with the most successful being the ANFIS and ANN. Several review papers already exist in the pools of literature [4–7]. Kaushika et al. (2014), [8] applied ANN techniques to predict components of solar radiation in New Delhi (India), Muammer et al. (2012), [9] develop an artificial neural network model for estimating global solar radiation over 31 stations in Turkey. Kumar et al. (2015), [10] developed a variety of model ANNs for global solar radiation forecasting in Tamilnadu (India). The ANN models had better results than the other approaches in this study. Benatiallah et al. (2020), [11] used some geographical and meteorological parameters as data input into nine (9) ANNs models to predict global hourly solar radiation in Adrar city (Algeria). In general, the artificial neural network technique had better performance compared to other approaches and model parameters. The purpose of this work is to develop an artificial neural network (ANN) based model for predicting the daily global solar radiation from several input parameters in three cities in the south-west region of Algeria.

2 Methodology 2.1

Study Area and Climate Dataset

The study area is located in the southwest region of Algeria. This region known by very high daytime temperatures in summer, and low temperatures in winter [12]. This area is rich in potential solar radiation. In the southern region the insolation time over theses region reaches a maximum of 3900 h [13]. On a horizontal surface, the average solar energy obtained is 5 kWh/m2 over the majority of the national country (Fig. 1), or around 2263 kWh/m2/year for the south of the territory of Algeria [14].

Artificial Neural Network Based Solar Radiation Estimation

575

Fig. 1. The location of the study Cities

For our study, we collect a large amount of data and sufficient to make a representative database, which should be used for learning and testing the neural network. This database forms the neural network entry, and therefore it determines both the size of the network and the performance of the system. In this research, we used daily data for five (05) years (June 2015 – June 2020), data provided by the SODA (Simple Ocean Data Assimilation) database [15]. We worked with the meteorological and astronomical parameters (see Table 1) to predict the output of global solar radiation and as input: Average Temperature, Wind speed, Relative Humidity, Atmospheric pressure, and we calculates the parameters: extraterrestrial solar irradiation, sunshine duration of the day, by Eqs. (1–2) respectively: E0 ¼

24
p  g0 cosðhÞ cosðdÞ sinðwÞ þ sinðhÞ sinðdÞ w p 180   2 SD ¼ w 15

ð1Þ ð2Þ

With d (°) is declination of the sun; h (°) is latitude of city; (deg.) is the angle of solar height (°) and g0 (dimensionless) is the coefficient of extraterrestrial solar radiation.

576

D. Benatiallah et al. Table 1. Parameters used Parameters Average temperature Relative humidity Wind speed Atmospheric pressure Extraterrestrial solar irradiation Sunshine duration

2.2

Abbreviation Tavg RH WS AP E0 SD

Unit Category °C Meteorological % m/s Astronomical hPa Wh/m2 h

Artificial Neural Network for Estimation of Solar Radiation

Artificial neural networks (ANN) are non-linear models utilizing a structure that can describe arbitrary complex non-linear processes related to the inputs and outputs of any system by designing simple connections based on the net input function and activating function (see Fig. 2). Neuron is the basic unit inside an ANN. Neurons are linked by ties known as synapses, there is a weight factor associated with each synapse. The weighted sum of the input signals can be expressed in mathematical relation form as equation below [16]: vj ¼

Xn i

wij xi

ð3Þ

The results are then added to the transfer function as equation shown below.   y j ¼ u v j þ hj

ð4Þ

The Transfer function is used to transform the input signal to an output signal on the neural network node. A number of transfer functions are available like Tangent, LogSigmoid, linear, Sigmoid, Hyperbolic etc.… These function can be chosen based on nature of application. In this research, the Tan-Sigmoid activation function shown in Eq. (5) and the linear transfer seen in Eq. (6) are used respectively in the hidden layer and the output layer [2].   u vj ¼

2 1 1  e2vj   u vj ¼ cvj

ð5Þ ð6Þ

where c is the slope of the output of the system. There are serveral algorithms to train a network and adjust its weights. In this article, three types of neuronal connection architectures were adopted, namely FeedForward neural network (ANN1), Elman Neural Network (ANN2) and Cascadeforward Neural Network (ANN3) due to their accuracy in most case studies of solar radiation prediction and suitability for a variety of applications [17].

Artificial Neural Network Based Solar Radiation Estimation

577

The first and simplest form of artificial neural network invented was the feedforward neural network (FFNN). In this network, the information travels from the input nodes, through one or more hidden nodes, and to the output nodes in only one direction —forward. This network is without cycles or loops. The second is the Elman Neural Network (ENN) or feedforwardnet where each hidden layer feeds itself back through an additional collection of background nodes. In the third network, Cascade-forward (CFNN) networks are similar to FNNN but provide a link from the input and each previous layer to the layers that follow. Two- or more-layer CFNN can arbitrarily learn any finite relation between input and output. Table 4 in appendix shows the characteristics of those three types of neuron connection architecture. All datasets were standardized in the range [−1,1] to avoid the possibility of overfitting and to achieve higher accuracy of the Neural networks models. In addition, the dataset was randomly divided into two subsets to ensure the representativeness of the dataset, using 80% for training, and the remaining data for validating the model. We created and performed a computer programs for the three different ANN connection architecture using a script file written in MATLAB -R2014a on the three locations Bechar, Naâma, and Tindouf. Figure 2 describes the typical construction of an ANN multilayer model. This network model (6–5–1) structure with 6 variables (Tavg, RH, WS, AP, E0, SD) is presented in the input layer, 5 nodes in the hidden layer and one node (solar radiation) in the output layer. While training algorithms applied by Levenberg-Marquardt. Prior to training there is no proven methodology for selecting the correct network architecture [18]. In this analysis, different trials established the number of hidden neurons. The trial and error technique demonstrated that identical results are obtained when the number of neurons in the hidden layer is between 4 and 6 [19].

Fig. 2. General neural network’s architecture

2.3

Statistical Evaluation

The performance of the considered ANN models was assessed using measures widely used in evaluation scores [20], such as Root Mean Square Error (RMSE), Mean Bias Error (MBE), Mean Absolute Percentage Error (MAPE), and coefficient of correlation

578

D. Benatiallah et al.

value (R). The definitions of these measures are given in Table 2, where N represent the total number of data points, GAct and GSim are the actual values and the output solar radiation values of the estimation, respectively. Table 2. Equations of statistical scores Abbreviation Ideal value Equation rP ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi N 2 RMSE Zero ðGSim;i GAct;i Þ i¼1 RMSE ¼ N

(7)

N   P GSim;i  GAct;i

(8)

MBE

Zero

MAPE

Zero

R

One

MBE ¼ N1

i¼1

N P (9) GSim;i GAct;i MAPE ¼ 100 XAct;i N i¼1 PN  Act Þ (10) GAct;i G ðGSim;i G Simq Þðffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi i¼1 R ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi PN PN 2 2 ðGSim;i G Sim Þ  ðGAct;i G Act Þ i¼1 i¼1

3 Results and Discussion The ANN-based prediction is performed in three cities located in Southwest of Algeria, we established the neural artificial network computing models with five hidden layer by using Levenberg-Marquardt algorithms. There are three basic types of connection architecture for neurons; the Feed-Forward neural network (ANN1), Elman Neural Network (ANN2) and Cascade-forward Neural Network (ANN3) models. Tavg, RH, WS, AP, E0 and SD were used as input variables, using tangent sigmoid transfer function in hidden layer and linear transfer function in output layers, resulted in a very efficient model for estimating global solar radiation in Southwest region. The results of statistical evaluation of MAPE, MBE, RMSE, rRMSE and correlation coefficient of the ANNs models for training and testing data are reported in Table 3 and Fig. 3. In general, it can be observed that, the ANN1 model and the ANN3 model presents better estimation results in the three locations. In the city of Bechar (see Table 3), the Cascade-forward Neural Network (ANN3) model is the most accurate model in comparison with results of others models in the testing phase according to statistical performance (R = 0.9313, RMSE = 0.658 kWh/m2/day and MBE = 39.73 Wh/m2/day), followed by ANN1 (R = 0.9122, RMSE = 0.730 kWh/m2/day and MBE = −1.06 Wh/m2/day). In the second city (Naâma), the ANN1 using Feed-Forward neural network (FFNN) connection architectures, achieved the more accurate model than the others ANNs models according to score performance statistics (R = 0.9332, RMSE = 0.703 kWh/m2/day and MBE = −14.19 Wh/m2/day), followed by ANN3 (R = 0.9159, RMSE = 0.775 kWh/m2/day and MBE = −39.66 Wh/m2/day). In the third city (Tindouf), the ANN3 (Cascade-forward Neural Network) performed the best for all layer architectures models according to overall mean errors (R = 0.9157, RMSE = 0.673 kWh/m2/day and MBE = 43.84 Wh/m2/day), the ANN1 had a similar performance to the ANN3 (R = 0.9063, RMSE = 0.652 kWh/m2/day and MBE = 65.42 Wh/m2/day).

Artificial Neural Network Based Solar Radiation Estimation

579

Table 3. Statistical scores for each model during training and testing phases in Bechar, Naâma and Tindouf cities

From Fig. 3, when simulated between the estimated and actual data; it is evident that for the city of Bechar and Tindouf (city 1 and city 3), the MAPE were less than 8% for ANN3. The lowest value of MAPE for the city 2 (Naâma) registered in all models is related to the ANN1 (MAPE < 9%). As a result, rRMSE did not exceed the percentages of 11, 12 and 11%, for the city 1, 2 and 3, respectively for the best model.

Fig. 3. MAPE and rRMSE during testing phase

In addition, the Fig. 4 showed the scatter diagram of daily global solar radiation values predicted by the best artificial neural networks model in each city. We can observed that the majority of the data points are close to the centered slope line, where regression R was more than 0.91, indicating a good concordance between collected and predicted values in each city over the testing period. In all cases, the ANN3 model with Cascade-forward Neural Network connection architectures has excellent performance in Bechar and Tindouf cities than the ANN1 and ANN2 models for predicting global (G) solar radiation. Therefore the ANN1 with

580

D. Benatiallah et al.

Feed-Forward neural network connection architectures provided the best accuracy than the others ANNs models in Naâma location. Whereas ANN2 (Elman Neural Network) model performed the worst (see Table 3 and Fig. 3). Comparisons with other research models show that these models give similar or often better regular values compared to the suggested statistical score values.

Fig. 4. Scattering diagrams of daily global predicted by the best ANN architectures

Figure 5 show a comparison example of the estimation of the Cascade-forward Neural Network (ANN3) based models with actual value of Bechar city during study period. It can be observed that the Gsim (estimate values) of ANN3 give a good estimation with the actual values GAct and sometimes are very similar, during the sixty (60) months of study period in all sky conditions.

Artificial Neural Network Based Solar Radiation Estimation

581

Fig. 5. Comparison estimation of ANN3 (CFNN) based model - case study of Bechar city

4 Conclusion The present study examined the preliminary prediction of daily solar radiation in Algeria’s Southwest region. A comparative study was performed to select the best artificial neural networks (ANN) based model to predict global solar radiation under all sky condition, this study have been carried by using three connection architectures; Feed-Forward neural network, Elman and Cascade-forward Neural Networks. Various combinations of input parameters were used including the average temperature, wind speed, atmospheric pressure, relative humidity, extraterrestrial solar irradiation and sunshine duration. These models conducted at three locations: Bechar, Naâma and Tindouf using different statistical parameters by comparing the simulated and the actual values. This comparison study indicated that the Cascade-forward Neural Network (CFNN) based model is more suitable for predicting daily global solar radiation in the cities of Bechar and Tindouf. However, The Feed-Forward neural network (FFNN) connection architectures model gives much better forecast of solar radiation day by day in Naâma city, with a coefficient of correlation greater than 0.91 and MAPE (Mean Absolute Percentage Error) less than 9% for the two best models. The developed models can also be used to install solar-energy systems and make thermal condition assessments in building studies in Algeria’s Sahara semi-arid climate region and similar climatic conditions in other areas where data are available.

582

D. Benatiallah et al.

Appendix Table 4. Characteristic of the three architectures neuron connections Neural network Feed-Forward Neural Network (FFNN)

Cascade-forward Neural Network (CFNN)

Elman Neural Network (ENN)

Characteristic – Connections between the nodes do not form a cycle – Input layer consists of the neurons which get inputs and travel them on to the other layers – The output layer is the function predicted which depends on the type of model that you are constructing – Hidden layers contain a large number of neurons that apply transformations before moving through the inputs. – Similar to feed-forward networks – Have a relation to following layers from the input and each previous layer – This network of two or more layers can learn arbitrarily well any finite input-output relationship given enough hidden neurons – Have recurrent layer ties added with the tap delays – Use simplified derivative calculations – With one or additional hidden layers know how to learn any dynamic input-output relationship arbitrarily fine

.

References 1. Zhang, J., Zhao, L., Shuai, D., Weicong, X., Zhang, Y.: A critical review of the models used to estimate solar radiation. Renew. Sustain. Energ. Rev. 70, 314–329 (2017) 2. Bouchouicha, K., Bailek, N., Bellaoui, M., Oulimar, B.: Comparison of artificial intelligence and empirical models for energy production estimation of 20 MWp solar photovoltaic plant at the saharan medium of Algeria, Int. J. Energ. Sect. Manag. 195–203, Springer, Cham, (2020) 3. Benatiallah, D., Benatiallah, A., Bouchouicha, K., Nasri, B.: Estimation of clear sky global solar radiation in Algeria. AIMS Energ. 7, 710–727 (2019) 4. Yadav, A.K., Chandel, S.S.: Solar radiation prediction using artificial neural network techniques: a review. Renew. Sustain Energ. Rev. 33, 772–781 (2014) 5. Mellit, A., Benghanem, M., Bendekhis., M.: Artificial neural network model for prediction solar radiation data: application for sizing standalone photovoltaic power system. In: Power Engineering Society, General Meeting, USA, pp. 2187–2191. IEEE (2005) 6. Qazi, A., Fayaz, H., Wadi, A., et al.: The artificial neural network for solar radiation prediction and designing solar systems: a systematic literature review. J. Clean. Prod. 104, 1–12 (2015) 7. Yadav, A.K., Chandel, S.S.: Solar radiation prediction using artificial neural network techniques: a review. Renew. Sustain. Energ. Rev. 33, 72–81 (2014)

Artificial Neural Network Based Solar Radiation Estimation

583

8. Kaushika, N.D., Tomar, R., Kaushik, S.C.: Artificial neural network model based on interrelationship of direct, diffuse and global solar radiations. Solar Energ. 103, 327–342 (2014) 9. Muammer, O., Bilgili, M., Sahin, B.: Estimation of global solar radiation using ANN over Turkey. Expert Syst. Appl. 39, 5043–5051 (2012) 10. Kumar, R., Aggarwal, R.K., Sharma, J.D.: Comparison of regression and artificial neural network models for estimation of global solar radiations. Renew. Sustain. Energ. Rev. 52, 1294–1299 (2015) 11. Benatiallah, D., Benatiallah, A., Bouchouicha, K., Nasri, B.: Prédiction du rayonnement solaire horaire en utilisant les réseaux de neurone artificiel. Algerian J. Environ. Sci. Technol. 6, 1236–1245 (2020) 12. Benatiallah, D., Benatiallah, A., Bouchouicha, K., Hamouda, M., Nasri, B.: An empirical model for estimating solar radiation in the Algerian Sahara. Am. Inst. Phys. 7, 710–727 (2018) 13. Benatiallah, D., Bouchouicha, K., Benatiallah, A., Harrouz, A., Nasri, B.: Forecasting of solar radiation using an empirical model. Algerian J. Renew. Energ. Sustain. Develop. 1, 212–219 (2019) 14. Benatiallah, D., Benatiallah, A., Harouz, A., Bouchouicha, K.: Development and modeling of a geographic information system solar flux in adrar. Algeria, Int. J. Syst. Model. Simul. 1, 15–19 (2016) 15. SODA data. www.soda-pro.com/web-services#meteodata 16. Vassilis, Z., Antonopoulos, D., Papamichail, et al.: Solar radiation estimation methods using ANN and empirical models. Comput. Electron. Agricul. 160, 160–167 (2019) 17. Raza, M.Q., Mithulananthan, N., Summerfield, A.: Solar output power forecast using an ensemble framework with neural predictors and Bayesian adaptive combination. Sol Energ. 166, 26–41 (2018) 18. Yadav, A.K., Chandel, S.S.: Solar radiation prediction using artificial neural network techniques: a review. Renew. Sustain. Energ. Rev. 33, 772–781 (2014) 19. Jain, S.K., Nayak, P.C., Sudheer, K.P.: Models for estimating evapotranspiration using artificial neural networks and their physical interpretation. Hydrol. Process. 22, 2225–2234 (2008) 20. Stone, R.J.: Improved statistical procedure for the evaluation of solar radiation estimation models. Solar Energ. 89, 51–91 (1993)

ANN-Based Correction Model of Radiation and Temperature for Solar Energy Application in South of Algeria K. Bouchouicha1(&), N. Bailek2, M. Bellaoui1, B. Oulimar1, and D. Benatiallah3 1

2

Unité de Recherche en Energies renouvelables en Milieu Saharien, URERMS, Centre de Développement des Energies Renouvelables, CDER, 01000 Adrar, Algeria [email protected] Materials and Energy Research Group, SERL, Department of Matter Sciences, Faculty of Sciences and Technology, University Center of Tamanrasset, 10034 Tamanrasset, Algeria 3 Laboratory of Sustainable Development and Computer Science (LSDCS), Faculty of Sciences and Technology, University Ahmed Draia, 01000 Adrar, Algeria

Abstract. This study proposes a hybrid solar radiation prediction model combining a Numerical Weather Prediction (NWP) simulation and an Artificial Neural Networks model (ANN). Weather Research and Forecasting (WRF) model is used to simulate meteorological and solar radiation parameters for one calendar year from January to December 2016 using six-hourly interval 1°
1° NCEP FNL analysis data. The one calendar year results from the WRF model will be used as input data for ANN model to estimate the temperature and solar radiation, in a case study of a solar farm located in Adrar Province, Algeria. The results reveal that the use of the ANN model effectively improves the accuracy of solar and temperature prediction. The RMSE, MAE and Correlation were investigated for the model evaluated for hourly irradiation predictions. It was determined that the proposed hybrid WRF-ANN model performs better than WRF simulation values. Keywords: NWP simulation Temperature  Solar radiation

 WRF model  ANN model  Prediction 

1 Introduction With the increasingly severe energy crisis, the development and search of renewable energy has gradually become an important research goal in many countries. Among these renewable energy sources, solar energy is increasingly become one of the most important directions of energy development in an undergoing developing country. Much solar power is integrated into grid systems nowadays. However, having the characteristics of randomness, intermittence, and volatility, solar power may interfere with the reliability and stability of a grid system to optimize the operations of stand-alone and © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 584–591, 2021. https://doi.org/10.1007/978-3-030-63846-7_55

ANN-Based Correction Model of Radiation and Temperature

585

hybrid energy systems [1]. The accurate prediction of solar radiation and power allows the power dispatching department to optimize the integration and operation in current power grids, and in the planning of energy management [2]. Developing methods of prediction for solar radiation has been paid attention worldwide in recent years. several studies have been performed, physical, statistical and artificial intelligent models have been developed in the literature. Physical methods are more appropriate to perform better in medium- and long-term prediction based on atmospheric processes and physical laws [3, 4]. Statistical and artificial intelligent methods perform short and very-short solar irradiance predicting by training the relationship between input variables and outputs based on climatological time series [5–8]. Hybrid models integrate different category of prediction models, which generally have higher accuracy in short-term solar radiation prediction. The combination of Numerical Weather Prediction (NWP) and statistical or intelligent methods are particularly promising for approaching the variation of solar radiation modeling. These methods are applied the different weather variables to make changes in solar radiation [9, 10]. Nowadays, many NWP models are available for solar radiation prediction; e.g., the European Centre for Medium-Range Weather Forecasts (ECMWF) model, the fifthgeneration mesoscale model (MM5), and the Weather Research and Forecasting (WRF) model. All these NWP models allow day-ahead meteorological prediction, which is usually adopted in practice. This research aimed to developpe a hybrid solar radiation prediction model combining a Numerical Weather Prediction (NWP) simulation and an Artificial Neural Networks model (ANN). The objective of the present study is the forecast error reduction of the NWP simulation, exploiting the WRF meteorological parameter include global solar radiation, relative humidity, wind speed, and air temperature as an input to the Artificial Neural Networks model (ANN).

2 Methodology This study develops a multi-step solar radiation prediction model that combines numerical weather prediction simulation and an Artificial Neural Networks model. The model comprises three parts: numerical weather prediction based on Weather Research and Forecasting (WRF) model, temperature correction using neural network approach, and solar radiation prediction based on the same technique. (1) The WRF model is used to predict the meteorological parameters include global solar radiation, relative humidity, wind speed, and air temperature. (2) The proposed Artificial Neural Networks approach is used to improve the accuracy of the ambient temperature forecasted by WRF model (3) The meteorological parameters are used as the key inputs of the Artificial Neural Networks prediction model. The prediction results are evaluated using the mean absolute error (MAE) and root-mean-square error (RMSE) as presented in Sect. 2(c).

586

K. Bouchouicha et al.

a) WRF model Weather Research and Forecasting (WRF) [7] version 3.9 was used in this study. The simulations were done based on initial and lateral boundary conditions products from the National Centers for Environmental Prediction (NCEP) Global Forecasting System (GFS) global model at horizontal resolution of 0.25°. see website (https://rda.ucar.edu/ datasets/ds083.2/). The spatial resolution was six-hourly intervals with 00 UTC, datasets. Table 1. Present the dominant parameters and physical configuration of the WRF model, and Fig. 1. shows WRF model computational domain with topographic elevation contours. This domain extends from 18.55 °N to 46.45 °N and from 10.7 °W to 17.2° E. with 280
280 points for the post processing output [11, 12]. Table 1. Computational configuration of the WRF simulation Period start End Input data Domain Map projection Vertical layer Microphysics WRF Longwave radiation Shortwave radiation Surface layer Land surface Planetary boundary layer

1 January 2016 31 December 2017 NCEP FNL (6-hourly, 1°
1°) (10 km, 210
210 grids) Mercator 38 Single-Moment 5-class scheme RRTMG scheme RRTMG scheme MM5 scheme Noah land surface model Yonsei University scheme

b) Artificial Neural Networks The ANN model is designed in such a way that the output variables are estimated from input variables by the composition of basic connections based on the net input function and the activation function as shown in Fig. 2 [13]. First the weighted sum of the input signals can express in the form mathematical relationship as given below: vj ¼

n X

wij xi

ð1Þ

i

c) Statistical evaluatuion scores Several statistical score metrics have been employed in various literatures to estimate the forecasting performance. In this paper, we applied three essential metrics to estimate the forecasting performance of the models: mean bias (MBE), Mean Absolute Error (MAE), root mean square error (RMSE), correlation coefficient (R) [14].

ANN-Based Correction Model of Radiation and Temperature

587

Fig. 1. WRF model computational domain used in this study with topographic elevation contours in meter.

Fig. 2.

Basic neural network’s architecture with Multiple inputs neuron.

Mean Bias Error (MBE) or: indicates the average tendency of the model’s error. MBE ¼

1 XN ð f  oi Þ i¼1 i N

ð2Þ

Where N = number of cases; f i and oi the ith forecast and observation. Mean Absolute Error (MAE): It is the average of the absolute model’s error, if the MAE is equal to zero, the forecast is perfect and the value increases proportionally with the discrepancies between the forecast and the observation; it describes the typical magnitude for the forecast error. MAE ¼

N 1X jf  oi j N i¼1 i

ð3Þ

588

K. Bouchouicha et al.

Root Mean Square Error (RMSE): It measures the amplitude of the forecast error. vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi #ffi u " N u1 X RMSE ¼ t ð f  oi Þ 2 N i¼1 i

ð4Þ

Correlation coefficient (R): It describes the relationship between the forecast and the observation. R measures the strength and the direction of a linear relationship between two variables. The value of R varies between −1 and 1. The + and – signs are used for positive linear correlations and negative linear correlations, respectively.   PN   oÞ i¼1 f i  f ðoi   R ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi2 PN   2 PN ð oi   oÞ i¼1 f i  f i¼1

ð5Þ

3 Data This study is based on measurement radiation data collected at the ground station installed in the solar PV plan of the electricity and renewable energy company SKTM (Sharikat Kahraba wa Takat Moutajadida) a Sonelgaz (Algerian National Society for Electricity and Gas) group’s subsidiary in Kabertene, in Adrar site, in Southern Algeria. The site is located at 27°18 N latitude and 0°11 W longitudes. In this region, the solar radiation is an important climatic data to be taken into account as the daily average solar irradiance for a horizontal surface exceeding 5.5 kWh/m2/day [15, 16]. The site of measure is located at 27°55 N latitude and 0°19 W longitudes, the specifics informations of the site are given in Table 2. Table 2. Geographic and data records period of the studied station. Station Latitude (°N) Longitude (°E) Data series period Mean temperature (°C) Mean relative humidity (%) Mean wind speed (m/s) Mean air pressure (hPa) Mean GHI (W/m2)

KABERTENE (Adrar) 27.91 −0.32 2017–2018 31.03 19.0 5.78 978.74 799.91

4 Results In this section, analysis to determine the forecasting performance of the developed hybrid models are considered. First, we present the plotting of spatial distribution of surface pressure and 2-m air temperature parameters giving by the WRF model in

ANN-Based Correction Model of Radiation and Temperature

589

Fig. 3. And then, the statistical scores are engaged to test the forecasting performance of the developed model.

Fig. 3.

Exemple of WRF model outputs.

In this present study, the temperature at 2 m for five days obtaining from time and date homepage for location in Kabertene (Adrar) were used to analyze the difference between simulate and observed data. The results of the performance of 5 days comparison between observed and simulated temperature at 2 m of the WRF simulated data and the hybrid models (WRF + ANN), plotting of temperature in two seasons the first one in Fig. 4(a) from 10th to 14th January 2017, and the second in Fig. 4(b) from 5th to 9th Jun 2017. Time series of predicted temperature in comparison with measured irradiance which allows to visualize easily the forecast quality forecasting model. In Fig. 4a and b, as an example, a high forecast accuracy of the hybrid model and a low one in case of the WRF simulated data can be seen. The temperature at 2 m of WRF simulated data showed ±3 °C lower compared to weather station observed data in the mid-day time and ½v  ¼ ½ cosðhÞ > < a >   > > > > : vb ¼ ½ cosðhÞ

" sinðhÞ  " sinðhÞ 

Xa1 Xa2 Xb1

# #

Xb2

The above Eqs. (6) of va and vb expressions has a linear form given by:

ð6Þ

880

C. S. Elislam et al.

½Y  ¼ ½U ½K 

ð7Þ

The detailed algorithm steps are explained in the standard FF-RLS method in [9– 11], where Fig. 2 illustrates the real-time estimation scheme of the recursive least squares (RLS) method. From Eq. (7), the RLS algorithm could be applied to calculate instantaneously the vector of parameters [K]. Based on the values X 1a ; X 2a ; X 1b et X2b ; the sequence magnitudes v þ et v and the initial phases u þ , u may be calculated through the following relations: 8 þ va ¼ v þ cosð/ þ Þ ¼ 12 ðXa1 þ Xb1 Þ > > > > þ < vb ¼ v þ sinð/ þ Þ ¼ 12 ðXb1  Xa2 Þ þ  1 1 1 > v > a ¼ v cosð/ Þ ¼ 2 ðXa  Xb Þ > > :  vb ¼ v sinð/ þ Þ ¼  12 ðXb1 þ Xa2 Þ

ð8Þ

Where the positive and the negative sequence magnitudes are given by: qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 8 < v þ ¼ ðv þ Þ2 þ ðv þ Þ2 a b ffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : v ¼ ðv Þ2 þ ðv Þ2 a b

ð9Þ

The angles u+ and u− are defined by: (

vþ

/ þ ¼ cos1 ðvaþ Þ v b v

/ ¼ sin1 ð Þ

2.2

ð10Þ

DSC Algorithm

By delaying the obtained voltages along ab axes in Eq. (1) by a quantity of T/4, the following voltage equations could be derived as: 2

T 3
 va ðt  Þ þ  T T þ  6 4 7 ¼ v sinðxt þ /  4 Þ  v sinðxt þ /  4 Þ 4 5 þ  T v þ cosðxt þ /  T4 Þ þ v cosðxt þ /  T4 Þ vb ðt  Þ 4 By making some manipulations, Eq. (11) becomes:

ð11Þ

Extracting Methods of Positive and Negative Voltage Sequences

T 3
 va ðt  Þ v þ sinðxt þ / þ Þ  v sinðxt þ / Þ 6 4 7¼ 4 5 T v þ cosðxt þ / þ Þ þ v cosðxt þ / Þ vb ðt  Þ 4

881

2

ð12Þ

From Eqs. (1) and (3), the positive and negative sequences could be calculated by: 2

3

2 1 6 þ 7 6 vb ðtÞ 7 1 6 0 7 6 6 6 v ðtÞ 7 ¼ 2 4 1 4 a 5 0 v b ðtÞ vaþ ðtÞ

0 1 0 1

0 1 0 1

3 2 3 va ðtÞ 1 6 7 0 7 7
6 vb ðtÞ 7 1 5 4 va ðt  T4 Þ 5 vb ðt  T4 Þ 0

ð13Þ

Using Park’s transformation, the ab system could be transformed into dq reference frame as: "

"

vdþ ðtÞ vqþ ðtÞ

v d ðtÞ

v q ðtÞ

#

#




cosðhÞ ¼  sinðhÞ


cosðhÞ ¼  sinðhÞ

" þ # va ðtÞ sinðhÞ : þ cosðhÞ vb ðtÞ

ð14Þ

"  # va ðtÞ sinðhÞ :  cosðhÞ vb ðtÞ

ð15Þ

The corresponding schemes relative to extraction procedure base on the RLS algorithm and the DSC method are shown in Fig. 2 and Fig. 3 respectively.

Fig. 2. Block scheme for RLS algorithm

Fig. 3. Block scheme of DSC algorithm

882

C. S. Elislam et al.

3 Results and Discussion Several tests have been accomplished to evaluate the studied extraction algorithms using the Matlab/Simulink software. The RLS and the DSC algorithms are compared in term of which one have a short time response and in term of accuracy in extracting positive and negative sequences under unbalanced and distorted voltages. The first test concerns a sag voltage in two phases b and c is considered when the a-phase is taken healthy, the b-phase and the c-phase contain sag of 80% and 60% respectively. In the second test both a-phase and b-phase are taken healthy, whereas, the c-phase has sag of 50%. Both RLS and the DSC algorithms are examined under these conditions to evaluate their individual performances.

Fig. 4. The three-phase system under balanced conditions

Under balanced conditions, The three-phase voltages (Va,Vb,Vc) present the same amplitude and the voltage components along dq axis are constant as shown in Fig. 4. Figure 5 presents the source voltages under unbalanced conditions, it is seen that the voltages components along dq axes contain fluctuations due to the generated second order harmonic related to the negative sequence occurrence. From Fig. 6 and Fig. 9, it is clear that, the extracting procedure is successfully performed with the two methods with a slight droop at its amplitudes as an effect for the negative sequence. Figure 7 shows the extracted positive and negative voltages relative to the first test, one can notice that the RLS algorithm is more robust than the DSC algorithm. Indeed, the DSC algorithm has a higher transient time response with respect to the RLS algorithm. However, both algorithms extract the same amplitude for the two sequences. Figure 8 exhibits the obtained results relative to the second test where the dip of 50% at the c-phase occurs at t = 0.15 s and last until t = 0.25 s, at that time the voltage system regains its balance. The DSC algorithm presents more efficiency in extracting positive and negative voltage in comparison to the RLS performances. This fact is related to the long transient time response of the RLS algorithm to achieve the steady state. It can be said that this method lost its convergence when the system has changed from a balanced state to an unbalanced state as presented in Fig. 10.

Extracting Methods of Positive and Negative Voltage Sequences

883

Fig. 5. The unbalanced three-phase system Fig. 6. From top to bottom: Vd and Vq voltages (first case) using RLS, Vd and Vq voltages using DSC.

Fig. 7. From top to bottom: positive sequence using RLS, negative sequence using RLS positive sequence using DSC and negative sequence using DSC

Fig. 8. The unbalanced three-phase system (second case)

Fig. 9. From top to bottom: Vd and Vq voltages Fig. 10. From top to bottom: positive sequence using RLS, Vd and Vq voltages using DSC using RLS, positive sequence using DSC, negative sequence using RLS, and negative sequence using DSC

884

C. S. Elislam et al.

4 Conclusion In this work, a comparative study between two methods for extracting positive and negative sequences in real time is presented: the delay signal cancellation “DSC” and the method based on the RLS algorithm. Using the MATLAB/SIMULINK software, simulation results are carried out for the considered unbalanced three-phase systems. From this study, it can be concluded that, both of them remains insensitive to the harmonic oscillations. Furthermore, the time response of the method based on RLS algorithm is better than the DSC method. However, in the case when the system becomes balanced after a given asymmetry, the DSC is more accurate and effective to separate the positive and negative sequences with respect to the method which is based on the RLS algorithm.

References 1. Abad, G., et al.: Doubly Fed Induction Machine: Modeling and Control for Wind Energy Generation, vol. 85. Wiley, Hoboken (2011) 2. Xu, D., et al.: Advanced Control of Doubly Fed Induction Generator for Wind Power Systems. Wiley, Hoboken (2018) 3. Chittora, P., Singh, A., Singh, M.: Application of self-tuning filter for power quality improvement in three-phase-three-wire distorted grid system. In: 2017 7th International Conference on Power Systems (ICPS). IEEE (2017) 4. Tan, G., Sun, X.: Analysis of Tan-Sun coordinate transformation system for three-phase unbalanced power system. IEEE Trans. Power Electron. 33(6), 5386–5400 (2017) 5. Džafić, I., et al.: Three-phase power flow in distribution networks using Fortescue transformation. IEEE Trans. Power Syst. 28(2), 1027–1034 (2012) 6. Silva, L.H.S., et al.: A robust phase-locked loop against fundamental frequency deviations and harmonic distortions. Electr. Power Syst. Res. 163, 338–347 (2018) 7. Rocha-Osorio, C.M., et al.: Sequence component extraction based on recursive least squares for wind energy applications. J. Control Autom. Electr. Syst. 29(1), 110–118 (2018) 8. Wang, S., Etemadi, A., Doroslovački, M.: Adaptive cascaded Delayed Signal Cancellation PLL for three-phase grid under unbalanced and distorted condition. Electr. Power Syst. Res. 180, 106165 (2020) 9. Chelli, S.E., Nemmour, A.L., Ahmed, M.A., Boussaid, A., Khezzar, A.: An effective approach for real-time parameters estimation of doubly-fed induction machine using forgetting factor RLS algorithm. Eur. J. Electr. Eng. 22(2), 169–177 (2020) 10. Djadi, H., Yazid, Y.K., Menaa, M.: Parameters identification of a brushless doubly fed induction machine using pseudo-random binary signal excitation signal for recursive least squares method. IET Electr. Power Appl. 11(9) (2017) 11. Debbabi, F., Nemmour, A.L., Khezzar, A., Chelli, S.: An approved superiority of real-time induction machine parameter estimation operating in self-excited generating mode versus motoring mode using the linear RLS algorithm: Ideas & applications’. Int. J. Electr. Power Energy Syst. 118, 105725 (2020)

Extended Kalman Filter for the Estimation of the State of Charge of Lithium-Ion Batteries Mohamed Khalfaoui(&) and Aissa Hamlat Electrotechnical Engineering Laboratory, University Tahar Moulay of Saida, Saida, Algeria [email protected] Abstract. This paper proposes a dynamic model of accumulator able to predict the general behaviour of the battery of an electric vehicle during its operation with the proposed estimator of a state of charge applied to this model. Knowing the state of charge of the battery is an important key future in the management of the embedded systems such as electric vehicles. Since this information is not measurable directly, an estimator based on the nonlinear Kalman filter is proposed. This estimator is applied to the battery model and gives good results in the case of constant and variable currents. Keywords: State of charge (SOC) Lithium-ion battery


Battery model
Kalman filter


1 Introduction The accumulator has revolutionized the way of storing electrical energy [1, 2]. Its use is highly widespread and in full expansion. It allows having an electrical energy reserve, autonomous and mobile (i.e. cell phone, photovoltaic systems, space equipments, laptops and other devices for public or industrial use) [3]. In order to reduce the emission of greenhouse gases, which is now the major concern of humanity, these accumulators represent a power source in the new generation of electric vehicles that replace the conventional internal combustion systems. To use an accumulator battery efficiently, it is necessary to understand its operation, its dynamics and the parameters that may affect its performance. The problem of these accumulators is to maintain their duration of use as long as possible and optimize the use of their energy [4, 5]. In order to understand how the accumulators work, it is necessary to develop a model that can simulate their behavior. In the majority of systems involving an energy storage system, there is a power management system associated with the storage battery and this to ensure the efficiency of the use of energy provided by the battery [6, 7]. One of the main parameters of this management system is the state of charge of the battery. This paper is organized as follows: Sect. 2 presents a battery model that is modeled by a variable resistor in series with a voltage source. This model was mainly applied for lithium ion type accumulators, which are mostly used in electric vehicles (EVs). Section 3 concerns the state of charge estimation (SOC). It allows us to know the residual capacity of the battery in order to avoid the excessive charge and the deep discharge of the battery. This estimator is based on a nonlinear Kalman filter. Section 4 presents the results obtained and discussion. Section 5 presents the conclusion and remarks. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 885–893, 2021. https://doi.org/10.1007/978-3-030-63846-7_86

886

M. Khalfaoui and A. Hamlat

2 Battery Model The battery is modeled by a controlled voltage source in series with a variable resistor, as shown in the Fig. 1. This model assumes the same characteristics for charging and discharging the battery [8, 9]. The terminal voltage is calculated with a nonlinear equation based on the state of charge of the battery.

I b (t )

Ri

Vt (t )

Vc

Fig. 1. Equivalent circuit of the battery model

The controlled voltage source is described by the following equation Vc ¼ V0  K Q

Q Rt

Zt þ A expðB Ib dt

Ib dtÞ

ð1Þ

0

0

and the terminal voltage is expressed by the following equation [1]. vt ðtÞ ¼ V0  Ri Ib  K Q

Q Rt 0

Where vt ðtÞ: Terminal voltage of the accumulator v0 : Open circuit voltage Ri : Internal resistance of an accumulator Ib : Accumulator current K: Polarization voltage factor Q: Nominal capacity of the accumulator Rt Ib dt: Actual charge of the accumulator 0

A: Voltage factor 1=B: Load factor

Zt þ A expðB Ib dt

Ib dtÞ 0

ð2Þ

Extended Kalman Filter for the Estimation of the State of Charge

887

The proposed model in Eq. (2) represents a nonlinear voltage which depends only on the charge of the battery. This model gives precise results and presents the behavior of the battery (Fig. 2).

Fig. 2. Nonlinear battery model

3 Proposed Kalman Filter Battery State of Charge Estimation Filtering consists in estimating the state of a dynamic system, evolving over time, from observations or measurement generally noisy [10, 11]. Whether the system is linear or linear, discrete or continuous, there are several variations of the Kalman filter. In our case, the battery is a dynamic system where the state of charge is the only state variable to be estimated (Fig. 3).

Ib

Battery model

Vt

Kalman Filter Estimator

SOC

Fig. 3. The estimator principle

888

M. Khalfaoui and A. Hamlat

According to the proposed model, the voltage at the terminals of the battery is defined by the Eq. (2), and the state of charge can be defined by the following equation: Rt SOCðtÞ ¼ SOC0  Rt SOC ¼ 1 

Ib ðtÞdt

t0

Q

Ib ðsÞds

t0

ð3Þ

Q Q

¼

Rt

Ib ðtÞdt

t0

Q

Q 1 ¼ Rt SOCðtÞ Q  Ib ðtÞdt

ð4Þ ð5Þ

t0

where Zt Ib ðtÞdt ¼ Q½1  SOCðtÞ

ð6Þ

t0

In this model, the state variable to estimate is xk ¼ SOC and the output variable is the voltage at the terminals of the battery yk ¼ Vt . The state model is linear with a single state variable. The discrete model which is represented by Eq. (7) is deduced from the continuous model equation (3). xk þ 1 ¼ xx 

Dt ik Q

ð7Þ

The output model represents the voltage at the terminals of the battery, by replacing the Eqs. (5) and (6) in (2), we get: yk ¼ 

K þ A exp½BQð1  xk Þ þ V0  Ri ik xk

ð8Þ

The estimator of state of charge is realized using an extended Kalman filter, which has the following form: 

xk þ 1 ¼ f ðxk ; uk Þ þ wk yk ¼ hðxk ; uk Þ þ vk

ð9Þ

Extended Kalman Filter for the Estimation of the State of Charge

889

Where: f ð x k ; uk Þ ¼ x x  hðxk ; uk Þ ¼ V0  Ri ik 

Dt ik Q

K þ A exp½BQð1  xk Þ xk

ð10Þ ð11Þ

f ðxk ; uk Þ and hðxk ; uk Þ are nonlinear functions. Dt: represents the sampling period. In our case, f ðxk ; uk Þ is linear and hðxk ; uk Þ is non linear. To solve the problem of nonlinear estimation we must linearize the system. The linearization is done around a point of equilibrium or around a nominal trajectory. Thus, the linearization will be made around the estimated variable which is the estimated state of charge [12]. – Linearization of the model The linearization of the output model around the estimated state, using the development of Taylor series of the first order, is given as:  @hðxk ; uk Þ hðxk ; uk Þ  hð^xk ; uk Þ þ ðxk  ^xk Þ @xk xk ¼^xk

ð12Þ

In this new linear model, transition and observation matrices are defined as the following Jacobians  @f ðxk ; uk Þ @xk xk ¼^xk

ð13Þ

 @hðxk ; uk Þ Ck ¼ @xk xk ¼^xk

ð14Þ

Ak ¼

According to the battery model, we have:  @f ðxk ; uk Þ Ak ¼ ¼1 @xk xk ¼^xk

ð15Þ

And Ck ¼

 @f ðxk ; uk Þ K ¼ 2 þ ABQ exp½BQð1  ^xk Þ  @xk ^xk xk ¼^xk

ð16Þ

890

M. Khalfaoui and A. Hamlat

– Algorithm of the extended Kalman filter The Kalman filter algorithm has three distinct phases: initialization, prediction, and update [12]. • Initialisation ^x0 ¼ Eðx0 Þ

ð17Þ

  P0 ¼ E ð^x0  x0 Þð^x0  x0 ÞT

ð18Þ

• The prediction The prediction step is to use the estimated state of the previous instant to calculate an estimate of the actual state. The state prediction is given as ^xkjk1 ¼ f ð^xk1jk1 ; uk Þ

ð19Þ

The predicted estimate of the covariance of the estimation error is given by: Pkjk1 ¼ Ak Pk1jk1 ATk þ Qk

ð20Þ

• Update and correction In this step, the measures of the actual moment are used to correct the state predicted to obtain a more accurate estimate of the state. The gain of optimal Kalman filter is defined by the following expression: Kk ¼ Pkjk1 CkT ðCk Pkjk1 CkT þ Rk Þ1

Fig. 4. Structure of the extended Kalman filter

ð21Þ

Extended Kalman Filter for the Estimation of the State of Charge

891

Update of the estimated state:   ^xkjk ¼ ^xkjk1 þ Kk yk  hð^xkjk1 ; uk Þ

ð22Þ

Update of the covariance of the error (Fig. 4): Pkjk þ 1 ¼ ðI  Kk Ck ÞPkjk1

ð23Þ

4 Simulation Results The proposed estimator based on the Kalman filter has been applied to a single type of battery (lithium ions). We had applied a constant current and a variable current (sinusoidal signal). The noise of the process and the noise of measurement are considered constant with a spectral density (or covariance) Qk ¼ 106 and Rk ¼ 102 , respectively [3]. The sampling period T ¼ 1 s. The battery used has a capacity of 28 Ah and a nominal voltage of 12 V. The discharge current is 1.4 A. Estimated SOC vs. actual SOC is as shown below in Fig. 7 and 9 with different current profile. Figure 5 and 8, as shown the error decreases drastically, it proves that the estimation strategy provides a good estimate of the SOC (Fig. 6).

Fig. 5. Constant current (discharge and charge)

Fig. 6. Terminal voltage (discharge and charge)

892

M. Khalfaoui and A. Hamlat

Fig. 7. Estimated state of charge (discharge and charge)

Fig. 8. Variable current (sinusoidal)

Fig. 9. Estimated state of charge (discharge and charge)

5 Conclusion An interesting feature of this paper is the simulation of the battery’s dynamic behaviour with respect to current variation and the battery’s SOC. The estimate of the state of charge of the battery, which depends essentially on the battery model, is very useful for the energy management system. The knowledge of this information allows optimizing the energy and the power of the battery which will increase the duration of use of the battery. In this work, we have to propose an Extended Kalman Filter to estimate the state of charge of the battery, instead of using the conventional current integration method. The proposed estimator which has been applied in the case of a constant and variable current is based on the measurement of voltage and current, taking into account the noise of the model (process) and the noise of the measure.

Extended Kalman Filter for the Estimation of the State of Charge

893

References 1. Asghar, F., Talha, M., Kim, S.H., Ra, I.-H.: Simulation Study on Battery State of Charge Estimation Using Kalman Filter. J. Adv. Comput. Intell. Intell. Inf. 20, 861–866 (2016) 2. Zhou, Y., Wang, Y., Shi, G., Zhou, M.: Estimating method for lithium ion battery state of charge based on twin support vector regression. Int. J. Control Autom. 7, 413–420 (2014) 3. Hui, W.H.S.: New improved prediction algorithm for state of charge of battery. J. Electron. Meas. Instrum. 11, 003 (2010) 4. Chang, W.-Y.: The state of charge estimating methods for battery: a review. ISRN Appl. Math. 2013 (2013) 5. Yuan, S., Wu, H., Yin, C.: State of charge estimation using the extended Kalman filter for battery management systems based on the ARX battery model. Energies 6, 444–470 (2013) 6. Charkhgard, M., Zarif, M.H.: Design of adaptive H∞ filter for implementing on state-ofcharge estimation based on battery state-of-charge-varying modelling. IET Power Electron. 8, 1825–1833 (2015) 7. Zhu, G.Q., Wang, C., Feng, B.Y.: Simulation study on state of charge of battery for electric vehicle. Adv. Mater. Res, 1006–1009 (2014) 8. Gao, J., He, H.: Comparison of nonlinear filtering methods for estimating the state of charge of Li4Ti5O12 lithium-ion battery. Math. Prob. Eng. 2015 (2015) 9. Xu, L., Wang, J., Chen, Q.: Kalman filtering state of charge estimation for battery management system based on a stochastic fuzzy neural network battery model. Energy Convers. Manag. 53, 33–39 (2012) 10. Lee, S.-H., Park, M.-K.: Battery state of charge estimation considering the battery aging. J. IKEEE 18, 298–304 (2014) 11. Liao, Y., Huang, J.H., Zeng, Q.: A novel method for estimating state of charge of lithium ion battery packs. Adv. Mater. Res. 428–435 (2011) 12. Lee, Y.-S., Wang, W.-Y., Kuo, T.-Y.: Soft computing for battery state-of-charge (BSOC) estimation in battery string systems. IEEE Trans. Industr. Electron. 55, 229–239 (2008)

An Efficient Strategy for Power Quality Conditioner with Half-Bridge for High-Speed Railway Amira Chaib Ras(&), Ramdane Bouzerara, Hamza Bouzeria, Meriem Aissaoui, and Imen Mammeri Laboratory of Transportation Engineering and Environment (LITE), Transportation Engineering Department, Frères Mentouri University - Constantine 1, Constantine, Algeria [email protected], [email protected], [email protected]

Abstract. Power supply network and electric locomotive (train) are the most important components of railway electric traction. The latter is considered as a transportation mean that represents many advantages. However, its main disadvantage is the load variations (non-linear load) which cause power quality problems such as negative sequence current components (NSC), harmonics, reactive power consumption …etc. A study of a comprehensive and simplifier design based on new connected balance transformer Ynev with half-bridge railway power conditioner (HBRPC) is used to solve the power quality problems. The HBRPC adopts single-phase back-to-back converter with four power switches (IGBTs) and two capacitors. The proposed model is able to eliminate or minimize the impact of the above problems that can degrade the quality of energy in high-speed railway. The work has been carried out on MATLAB/SIMILINK in order to validate the performance of the model and the control block. Keywords: Power quality
High-speed railway HBRPC
Unfavorable sequence current


Balance transformer


1 Introduction The advancement made and the possibilities visualized in the different fields of electrical designing industry surmise a transient development in the vehicle segment, all the more especially in the railway electric traction. It was embraced in many countries because of its advantages such as lower air pollution, high efficiency and most recently because of CO2 generation reduction [1], which is more effective than diesel traction. Since its inception, the improvement of system performance, reliability and availability has been one of the most important concerns of designers and operators. In high-speed railway, the load is a single phase which affect the public electrical grid power quality because of various problems for example current and voltage unbalance in transmission network, harmonic, low frequency [2] and the most important problem is a negative sequence component (NSC) of current. Degradation of quality may lead to changes in behavior and performance or even destruction of equipment. Various effort and research have been © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 894–901, 2021. https://doi.org/10.1007/978-3-030-63846-7_87

An Efficient Strategy for Power Quality Conditioner

895

proposed to solve the power quality problems. Various balance transformers connections in rail transport for example: Scott, Leblanc, Woodbridge…etc. have been used to eliminate NSC [3–5]. It is a very important part in traction system design to choose a transformer since locomotive power is generally huge. But its complete elimination is impossible. Since this was not enough, another research is therefore necessary. Researchers are recently focusing on the compensation for harmonics, reactive power, load balancing and neutral current compensation [1]. FACTS technology opens up opportunities to control power and enhance the line capability usage. Many FACTS devices are developed shunt, series and hybrid. Power electronics conditioners (RPC, HBRPC, TW-RPC, HPQC…etc.) are a special technology of FACTS are used in high speed railway system [5–8] as illustrated in Fig. 1. This paper is divided as follows: Sect. 2 describes the components of system proposed. Section 3 presents the simulation results and analysis. Finally, Sect. 4 is a conclusion of this work.

Fig. 1. RPCs topologies used in high-speed railway compensation [8].

2 Component of System 2.1

Specially Balance Transformer Connection

The transformer used to adjust the voltage used by the electric locomotive. Different types of transformers are used in electric railway system to transformer three-phase to a two-phase supply, V-V transformer, Scott, Leblanc…etc. [4, 9]. In this paper, the Ynev balanced transformer is selected; it is equipped with a primary three-phase winding coupled as star with grounded neutral point and secondary winding coupled in open delta [10] transformer configuration and phasor diagram shown in Fig. 2. The voltage magnitude in transformer’s secondary side can be calculated in (1) and (2) [10]. jVaj ¼

pffiffiffi 3V(N22 þ N23 þ N2 N3 Þ1=3

ð1Þ

jVbj ¼

pffiffiffi 3V(N22 þ N23 þ N2 N3 Þ1=3

ð2Þ

Then, currents relationships are expressed in (3).

896

A. Chaib Ras et al.

Fig. 2. Phasor and connection diagram of Ynev transformer [10].

2

3 2 IA ðN2 þ N3Þ 4 IB 5 ¼ 4 N3 IC N2

3   N3 Ia 5 ðN2 þ N3Þ Ib N2

ð3Þ

Where; IA, IB, IC are the three-phase side current and Ia , Ib are the two-phase side currents. The current unbalance ration: e ¼

2.2

jIj jI þ j

ð4Þ

Load Model

In this paper, AC/DC/AC electric locomotive train is simulated, see Fig. 3. It is composed of static power converters for a transformation of the physical characteristics (voltage) before reaching the traction motor. The current supplied by the uncontrolled rectifier is converted to AC required by locomotive motors by the controlled inverter.

AC Power system

Traction load (train)

DC

AC

DC DC

Rectifier

AC

MAS

Two level inverter asynchronous machine

Fig. 3. Electric locomotive model.

An Efficient Strategy for Power Quality Conditioner

2.3

897

Simulation Model

The configuration of HBRPC is illustrated in Fig. 4, which consists of two half-bridge converters with two capacitors in series. The 230 kV three-phase high-voltage source is converted into 27.5 kV; the HBRPC is connected to Ynev transformer’s secondary side through an interface reactor and step-down transformer. Two trains in both sides of traction power supply substation are considered. Current of both load sections are denoted Ia, Ib while the compensation currents are Ica, I cb. Load current can be expressed as follow [10]: 

pffiffiffi P pffiffiffi Ia ðtÞ ¼ p2ffiffiffiI1 cosðxt þ h1 Þ þ 1 2I cosðxt þ hha Þ Ph¼2 pffiffiffi ha Ib ðtÞ ¼ 2I2 cosðxt þ h2 Þ þ 1 h¼2 2Ihb cos xt þ hhb

ð5Þ

Where Iha and Ihb are the hth harmonic order RMS currents of the a and b phases respectively. hha and hhb are the phases of hth harmonic order currents of the a and b phases. Compensation current can be calculated by subtracting the load currents from the desired currents: 

Ica ðtÞ ¼ Ia ðtÞ  Iref a ðtÞ Icb ðtÞ ¼ Ib ðtÞ  Iref b ðtÞ

ð6Þ

The compensating currents are related to the ideal current of Iref a and Iref b as: qffiffi Iref a ðtÞ ¼ 23Imp cosxt qffiffi : I ðtÞ ¼ 2I cosxt  2p ref b 3 mp 3 8
Dmax, the voltage gain drops quickly to zero. Experimental results show that the proposed boost converter circuit is capable to provide a constant DC output voltage with better overshoot reduction improving the performance of the system [13–15] (Figs. 12 and 13).

Fig. 10. Global block diagram of designed prototype of DC-DC Boost converter

Design and Simulation of a DC-DC Boost Converter

983

Fig. 11. Experimental platform of designed prototype of DC-DC Boost converter

Fig. 12. Experimental results of Boost converter operating at 50% duty cycle (a) Output and input voltage of Boost converter (b) Inductor current and control signal (c) Mosfet voltage and inductor current(d) Voltage and current of inductor (e) Output voltage and current of Boost converter

984

A. Hadj Dida et al.

Fig. 13. Experimental results of Boost converter operating at 75% duty cycle (a) Output voltage and current of inductor (b) Mosfet voltage and inductor current (c) Inductor current and control signal (d) Voltage and current of inductor

• For a Boost converter operating at 50% duty cycle: The output voltage and current are: Vo = 24 V and Io = 1 A. The input current: Iin = 1 A. The input power: Pin = Vin * Iin = 12 * 2.2 = 26.4 W. The output power: Po = Vo * Io = 24 * 1 = 24 W. g¼

Po 90 ¼ ¼ 90:9% Pin 26:4

• For a Boost converter operating at 75% duty cycle: The output voltage and current are: Vo = 37 V and Io = 1.52 A. The input current: Iin = 5.8 A. The input power: Pin = Vin * Iin = 12 * 5.8 = 69.6 W. The output power: Po = Vo * Io = 37 * 1.52 = 56.24 W. g¼

Po 56:24 ¼ ¼ 80:9% Pin 69:6

Design and Simulation of a DC-DC Boost Converter

985

6 Conclusion The paper attempted to contribute and design analysis of a DC-DC Boost converter control for power system embedded for space application. From the simulation results, it is found that the desired output voltages of Boost converter can be obtained by selecting proper values of inductor, capacitor and switching frequency and PID control coefficients [16, 17]. The proposed Boost converter with PID controller provides better output voltage regulation, overshoot reduction, thereby improving the performance of the system. The proposed electrical circuit of converter is simple, easy to understand and can be implemented with no additional components keeping lower-cost manufacturing with reduced size and weight. Simulation and experimental results were very satisfactory for further progression systems. This research work gives the opportunity to study and raise valuable knowledge in circuit designing and problem solving skills which has greatly enriched knowledge and understanding through the erudition route.

References 1. Seguier, G.: Power Electronic Converters: DC-DC Conversion. Springer, New York (1993) 2. Rashid, M.H.: Power Electronics, Circuits, Devices, and Applications, 3rd edn. Pearson Education, Inc., Prentice Hall (2004) 3. Uno, Md.M., Kukita, A.: 3-Port Converter Integrating a Boost Converter and Switched Capacitor Converter for a Single-Cell Battery Power System in a Small Satellite document is provided by JAXA 4. Abu-Qahouq, J., Batarseh, I.: Generalized analysis of soft-switching DC-DC converters. In: International Symposium on Circuits And Systems (ISCAS- IEEE); 28–31 May 2000 (2000) 5. Erickson, R.W., Maksimoni, D.: Fundamentals of Power Electronics. 2nd edn. Springer, New York (2001) 6. Kumar, J.S., Gajpal, T.: A Multi Input DC-DC Converter for Renewable Energy Applications (2016) 7. Li, W., Lv, X., Deng, Y., Liu, J., He, X.: A review of non-isolated high step-up DC/DC converters in renewable energy applications. In: Applied Power Electronics Conference and Exposition, 2009. APEC 2009. Twenty-Fourth Annual IEEE, pp. 364–369. IEEE (2009) 8. Ibrahim, O., Yahaya, N.Z., Saad, N.: Comparative studies of PID controller tuning methods on a DC-DC boost converter. In: 2016 6th International Conference on Intelligent and Advanced Systems (ICIAS), Kuala Lumpur, pp. 1–5 (2016) 9. Rathi, M.K., Ali, M.S.: Design and simulation of PID controller for power electronics converter circuits. Int. J. Innov. Emerg. Res. Eng. 3(2), 26–31 (2016) 10. Kim, I.H., Son, Y.I.: Regulation of a DC/DC boost converter under parametric uncertainty and input voltage variation using nested reduced-order PI observers. IEEE Trans. Ind. Electron. 64(1), 552–562 (2017) 11. Simoiu, M.S., Calofir, V., Arghira, N.: BOOST converter modelling as a subsystem of a photovoltaic panel control system. In: IEEE International Conference on Automation, Quality and Testing, Robotics (AQTR) (2020) 12. Erickson, R.W.: Dc–dc power converters. In: Wiley Encyclopedia of Electrical and Electronics Engineering, pp. 1–2 (2007)

986

A. Hadj Dida et al.

13. Miron, C., Olteanu, S., Christov, N., Popescu, D.: Advanced control system for photovoltaic power generation, storage and consumption. UPB Sci. Bull. Ser. C Electr. Eng. 81 (2019) 14. Adnan, M., Oninda, M., Nishat, M., Islam, N.: Design and simulation of a dc-dc boost converter with pid controller for enhanced performance. Int. J. Eng. Res. Technol. (IJERT) 6, 27–32 (2017) 15. Santoja, A., Barrado, A., Fernandez, C., Sanz, M., Raga, C., Lazaro, A.: High voltage gain DC-DC converter for micro and nanosatellite electric thrusters. In: 2013 Twenty-Eighth Annual IEEE Applied Power Electronics Conference and Exposition (APEC) (2013) 16. Forouzesh, M., Siwakoti, Y.P., Gorji, S.A., Blaabjerg, F.: Step-Up DC–DC converters: a comprehensive review of voltage-boosting techniques, topologies, and applications. IEEE Trans. Pow. Electron. 32(12), 9143 (2017) 17. Weinberg, A.K., Rueda Boldo, P.: A high power, high frequency, DC to DC converter for space applications. In: PESC 1992 Record. 23rd Annual IEEE Power Electronics Specialists Conference (n.d.)

Author Index

A Abadlia, I., 168 Abdel Karim, Ferouani, 958 Abdelfettah, Boussaid, 877 Abdelhamid, Slama, 15 Abdelkhalek, Othmane, 822 Abdelmoumene, Abdelkader, 487 Abdoune, F., 168 Abdulrazzak, Ibtihaj A., 649 Abri, Mehadji, 738 Adjerid, Chaouki, 635 Afoun, Laid, 635 Aggoun, Abderrezak, 457 Ahmed, Anas F., 649 Aidel, Salih, 697 Aissa, Oualid, 467 Aissani, Djamil, 782 Aissaoui, A., 592 Aissaoui, Meriem, 36, 294, 894 Aissat, A., 314 Allam, N., 114 Allam, Zehor, 946 Aloui, I., 592 Al-Shammari, Ahmed, 782 Amel, Abbadi, 79 Amraoui, M. A., 839 Anane, Zahira, 26 Aoufi, B., 551 Arif, Salem, 96, 252 Assem, H., 114 Aufy, Samah A., 135 Ayad, H., 727 Azib, T., 114 Azli, H., 478

B Babes, B., 105, 512 Bacetti, Abdelmoumen, 197 Badeche, M., 55 Bahi, T., 3 Bailek, N., 584 Bakria, K., 592 Bechka, M. L., 43 Beddar, A., 168 Bekhti, M., 975 Belahcene, B., 155 Belaidi, Hadjira, 487 Belgacem, A., 902 Belhaouas, N., 114, 592 Belkacem, Nassima, 782 Bellaoui, M., 584 Benabou, D., 727 Benahmed, A. Moumene, 946 Benahmed, Khelifa, 669 Benaissa, Mohamed, 738 Benalla, H., 222, 911 Benammar, Samir, 184, 856 Benatiallah, A, 573 Benatiallah, D., 573, 584 Benbouzid, Mohamed, 62 Benhanifia, Abdeldjalil, 747 Benidir, M., 36, 294 Bennabi, Souad, 847 Bennia, Ilyas, 302, 325 Benoudjafer, Cherif, 374 Benrabah, Abdeldjabar, 62 Bensaada, M., 975 Bensahal, Djamel, 831 Benslimane, S. M., 790 Bentaallah, Abderrahim, 336

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 M. Hatti (Ed.): ICAIRES 2020, LNNS 174, pp. 987–990, 2021. https://doi.org/10.1007/978-3-030-63846-7

988 Bentabet, Dougani, 689 Bentarzi, H., 766 Bentarzi, Hamid, 487 Benyamina, Fayçal, 62 Benyezza, Hamza, 447 Benziouche, Nihad, 697 Bessas, Aicha, 336 Bierk, Hussain M., 649 Bouafassa, A., 512 Bouafia, Abdelouahab, 351 Bouarfa, Hafida, 775 Boubakar, Hichem, 738 Boubchir, Larbi, 635 Bouchafaa, F., 114 Bouchahm, Y., 55 Bouchekara, Houssem R. E. H., 521, 911 Bouchouicha, K., 573, 584 Boudaoud, Chahrazad, 946 Boudechiche, Ghania, 467 Boudiaf, M., 902 Boudjella, Aissa, 425, 436 Boudjella, Manal Y., 425, 436 Boudjelthia, El Amin Kouadri, 811 Boughazi, Othmane, 374 Bouhedda, Mounir, 447, 457, 612, 625 Boukerma, Billal, 635 Boumadien, L., 930 Bouraiou, Ahmed, 88 Bousmaha, Bouchiba, 682, 822 Boutabba, T., 43, 281 Boutaghane, A., 105, 512 Bouzerara, Ramdane, 894 Bouzeria, Hamza, 36, 294, 894 C Chabane, Foued, 831 Chahtou, Amina, 811 Chaib Ras, Amira, 36, 294, 894 Chaoui, Abdelmajid, 351 Cheggou, R., 405 Chekired, Fathia, 811 Chellal, Arezki Abderrahim, 747 Chenine, Mossaab, 856 Chenni, Rachid, 391 Cherifi, Dalila, 635 D Dabou, Rachid, 88 Daili, Yacine, 302, 325, 868 Dennai, A., 562 Dhahbane, Djamel, 500

Author Index Djehaiche, Rania, 697 Djellab, S., 902 Djenadi, A., 178 Djeriri, Yousef, 336 Doghmane, Mohamed Z., 155, 184, 197, 243, 707, 856 Drid, S., 43, 281 E El Habib, Guedda, 958 Eladj, S., 155 El-Bayeh, Claude Ziad, 391 Elislam, Chelli Seif, 877 Elmir, Y., 562 F Faghihi, Yousef, 716 Fahsi, Mahmoud, 847 Ferroudji, F., 416 Fethia, Hamidia, 79 Fraoucene, H., 405 G Gacem, A., 210 Ghanemi, S., 71 Ghennam, Asma, 625 Ghouali, S., 727 Goudjil, Hadjer, 612 Gougui, A., 210 Guellil, MS., 727 Guettaf, Nacereddine, 26, 351 Guettaf, Seif El Islem, 26, 351 H Hachana, O., 551 Hadj Dida, A., 975 Hadj Youb, A., 178 Hadjrioua, F., 592 Hamida, M. A., 210 Hamid-Oudjana, Samir, 96 Hamied, A., 125 Hamlat, Aissa, 885 Hamouda, C., 405 Hamouda, N., 105, 512 Harouz, A., 573 Harrag, Abdelghani, 302, 325, 868 Hassaine, S., 270, 361 Hassaine, L., 168 Hassan, Heba Ahmed, 391 Hatti, Mustapha, 96 Hayi, Mohamed Yassine, 541

Author Index Hioual, Ouassila, 533 Hocini, Kenza, 756 Hossine, Guermit, 233 I Iloul, Zakaria, 635 Izeboudjen, N., 798 J Javaid, Muhammad S., 521 K Kahla, S., 105, 512 Kalkoul, S., 911 Karim, Ferouani Abdel, 966 Kassam, Allaeldin H., 135 Katia, Kouzi, 233 Kellil, N., 314 Kerrour, Fouad, 521 Kessai, Idir, 184 Khair, Y., 562 Khalfaoui, Mohamed, 885 Khattara, Abdelouahab, 252 Khedir, O., 727 Khelifa, Sadek, 184 Khemis R., A., 281 Khoucha, Farid, 62 Khoumeri, El Hadi, 405 Khoumeri, E., 405 Kidouche, Madjid, 155, 197, 243, 707 Klouche, B., 790 L Laguidi, Ahmed, 374 Lakhdara, A., 3 Larbes, C., 478 Lashab, Abderezak, 467 Lasmari, Adel, 391 Latreche, Yaaqoub, 521 Leila, Ghalmi, 920 Lemerini, Mostefa, 966 Lokmane, Nemmour Ahmed, 877 Louze, Lamri, 260 M Mahammed, Nadir, 847 Mammeri, Imen, 36, 294, 894 Mankour, M., 930 Medjati, Walid Yassine, 659 Megnafi, Hicham, 659, 747 Mehareb, Fatah, 811 Mellit, A., 125, 314 Melouah, Ahlem, 716

989 Mendil, Chafiaa, 243, 707 Mennad, Mebrouk, 336 Menzri, F., 43 Messaoud, Hamouda, 15 Messaoudi, Kamel, 260 Mihoub, Y., 270, 361 Miloudi, L., 178 Mohamed, Chafa, 260 Mohammed, Sahlaoui, 958 Moreau, S., 270, 361 Mosbah, Mustafa, 96, 252 Mostefaoui, Mohammed, 88 Mounir, Khiat, 15 Moussaoui, A. K., 3 Mraoui, A., 902 N Nabti, K., 911 Nait-Ali, Amine, 635 Nasri, B., 573 Necaibia, Ammar, 88 Nemmouchi, B., 222 Nemra, Abdelkrim, 500 Nouri, Hamou, 26 O Ouchani, Samir, 775 Oudjana, S. H., 252 Oudjouadj, Abderrahmane, 457 Oulimar, B., 584 R Rabhi, A., 125 Rabiai, Zakaria, 487 Rahmani, Mohamed, 831 Rais, M. S., 71 Raza, Ali, 62 Rebhi, Mhamed, 682, 822 Rebouh, Samia, 447, 457, 612, 625 Rehouma, F., 210 Rehouma, Y., 210 Rezgui, S. E., 222 Rouibah, N., 125 Roummani, K., 416 S Sahib, Khouloud, 716 Sahouane, Noredine, 88 Saidi, Ahmed, 669 Saidi, Karima, 533 Saihi, L., 416 Sakhi, Samir, 500 Sari-Hassoun, SE, 727

990 Sarra, Mustapha, 467 Seddiki, Nouredine, 669 Sekour, M., 930 Semchedine, Fouzi, 782 Settoul, Samir, 391 Sha’aban, Yusuf A., 521 Shahriar, Mohammad S., 521 Slim, Amel, 716 Slimani, Abdeljalil, 88 Soufi, Aicha, 946 Souhaila, Askri, 958, 966 Souhila, Bensmaine, 920 T Talbi, B., 105 Tir, Z., 210 Titri, S., 478, 798 Toumi, D., 270, 361

Author Index Toumi, Yassine, 447 Tsebia, M., 766 Y Yaichi, Mouaad, 682, 822 Yazid, Mohand, 756 Yedjour, Dounia, 604 Yedjour, H., 937 Yousfi, Ahmed, 831 Z Zahira, Chouiref, 541 Zellagui, Mohamed, 391 Zerrouki, Fahem, 775 Ziane, Abderrezzaq, 88 Zine, Rabie, 96, 252 Zouaidia, K., 71