Proceedings of International Conference on Computational Intelligence and Emerging Power System: ICCIPS 2021 (Algorithms for Intelligent Systems) 9811641021, 9789811641022

Citation preview

Algorithms for Intelligent Systems Series Editors: Jagdish Chand Bansal · Kusum Deep · Atulya K. Nagar

Ramesh C. Bansal Akka Zemmari K. G. Sharma Jyoti Gajrani Editors

Proceedings of International Conference on Computational Intelligence and Emerging Power System ICCIPS 2021

Algorithms for Intelligent Systems Series Editors Jagdish Chand Bansal, Department of Mathematics, South Asian University, New Delhi, Delhi, India Kusum Deep, Department of Mathematics, Indian Institute of Technology Roorkee, Roorkee, Uttarakhand, India Atulya K. Nagar, School of Mathematics, Computer Science and Engineering, Liverpool Hope University, Liverpool, UK

This book series publishes research on the analysis and development of algorithms for intelligent systems with their applications to various real world problems. It covers research related to autonomous agents, multi-agent systems, behavioral modeling, reinforcement learning, game theory, mechanism design, machine learning, meta-heuristic search, optimization, planning and scheduling, artificial neural networks, evolutionary computation, swarm intelligence and other algorithms for intelligent systems. The book series includes recent advancements, modification and applications of the artificial neural networks, evolutionary computation, swarm intelligence, artificial immune systems, fuzzy system, autonomous and multi agent systems, machine learning and other intelligent systems related areas. The material will be beneficial for the graduate students, post-graduate students as well as the researchers who want a broader view of advances in algorithms for intelligent systems. The contents will also be useful to the researchers from other fields who have no knowledge of the power of intelligent systems, e.g. the researchers in the field of bioinformatics, biochemists, mechanical and chemical engineers, economists, musicians and medical practitioners. The series publishes monographs, edited volumes, advanced textbooks and selected proceedings. All books published in the series are submitted for consideration in Web of Science.

More information about this series at https://link.springer.com/bookseries/16171

Ramesh C. Bansal · Akka Zemmari · K. G. Sharma · Jyoti Gajrani Editors

Proceedings of International Conference on Computational Intelligence and Emerging Power System ICCIPS 2021

Editors Ramesh C. Bansal Department of Electrical Engineering University of Sharjah Sharjah, United Arab Emirates K. G. Sharma Department of Electrical Engineering Engineering College Ajmer Ajmer, India

Akka Zemmari University of Bordeaux Talence Cedex, France Jyoti Gajrani Department of Computer Science and Engineering Engineering College Ajmer Ajmer, India

ISSN 2524-7565 ISSN 2524-7573 (electronic) Algorithms for Intelligent Systems ISBN 978-981-16-4102-2 ISBN 978-981-16-4103-9 (eBook) https://doi.org/10.1007/978-981-16-4103-9 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 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 Singapore Pte Ltd. The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore

Advisory Committee

Prof. Akka Zemmari, LaBRI, University of Bordeaux, France Dr. Akshay Kumar Rathore, Concordia University, Canada Dr. Baseen Khan, Hawassa University, Ethiopia Dr. Mahdi Khosravy, Osaka University, Japan Dr. Neeraj Gupta, Oakland University, USA Prof. R. C. Bansal, University of Sharjah, UAE Prof. Gaurav Bhatnagar, IIT Jodhpur Dr. Hari Prabhat Gupta, IIT BHU Prof. Manoj Singh Gaur, IIT Jammu Prof. Ranjan Choudhary, IIT Guwahati Prof. Virendra Singh, IIT Mumbai Dr. Harish Sahu, DRDO Delhi Dr. J. C. Bansal, South Asian University, New Delhi Dr. A. P. Mazumdar, MNIT Jaipur Dr. Meenakshi Tripathi, MNIT Jaipur Dr. Ram Niwas Mahia, NIT Hamirpur Dr. Rohit Bhakar, MNIT Jaipur Prof. S. G. Modani, Ex-Professor, MNIT Jaipur Dr. S. K. Vipparthi, MNIT Jaipur Dr. Ujjwal Kumar Kalla, MANIT Bhopal Prof. M. S. Sevda, Central Agriculture University, Sikkim Prof. Dhiren Patel, VJTI Mumbai Dr. Faruk Kazi, VJTI Mumbai Prof. R. N. Awale, VJTI Mumbai Prof. Akhil Ranjan Garg, JNVU Jodhpur Prof. K. L. Sharma, Ex-Professor, JNVU Jodhpur Prof. Rajesh Bhadada , JNVU Jodhpur Prof. Annapurna Bhargava, RTU Kota Prof. C. P. Gupta, RTU Kota Prof. D. K. Palwalia, RTU Kota Dr. Harish Sharma, RTU Kota v

vi

Advisory Committee

Prof. Ranjan Maheshwari, RTU Kota Dr. R. K. Bayal, RTU Kota Prof. S. C. Jain, RTU Kota Dr. R. R. Joshi, MPUAT Udaipur Dr. Vinod Kumar Yadav, MPUAT Udaipur Dr. Vinod P., CUST, Cochin Prof. P. M. Chawan, VJTI, Mumbai Prof. D. P. Kothari, SBJIT Nagpur Dr. Om Prakash Mahela, RVPN Dr. Rahul Garg, Department of Skill, Jaipur Dr. Rishi Pal Singh, GJUS&T Hisar Dr. Sanjay Jain, SPCGC Ajmer Prof. Bruhadeshwar Bezawada, MEC, Hyderabad

Technical Committee Gaurav Meena, CURAJ, India Arjun Choudhary, Sardar Patel University of Police, Security and Criminal Justice, Jodhpur Dr. Subhash Panwar, Engineering College Bikaner, Bikaner Vivek Prakash, Banasthali VidyaPeeth Dr. K. Narayanan, SASTRA, Tirumalaisamudram Dr. Ajay Kumar Bansal, Central University of Haryana Dr. Nand Kishor Meena, Aston University Vinay Kumar Jadoun, Manipal Institute of Technology, Jaipur Dr. Kailash Chand Sharma, Dr. B. R. Ambedkar National Institute of Technology, Jalandhar Vikas Sihag, Sardar Patel University of Police, Security & Criminal Justice, Jodhpur Dr. Jyoti Grover, MNIT Jaipur, Jaipur Dr. Subhash Panwar, Engineering College Bikaner, Bikaner Dr. Anil Kumar Dubey, ABES Engineering College, Ghaziabad, Uttar Pradesh

Organizing Committee Chief Patrons Dr. Subhas Garg Minister, Department of Technical Education, GoR and President BoG, ECS Ajmer Smt. Aparna Arora (IAS) Secretary, Department of Technical Education, GoR

Advisory Committee

Patrons Prof. N. C. Shivaprakash Chairman BoG, ECS Ajmer and Professor, IISC Bangaluru Dr. Rekha Mehra Principal, Engineering College Ajmer

General Chairs Dr. Krishan Gopal Sharma Dr. Jyoti Gajrani Dr. Adarsh Mangal

Organizing Secretaries Vinesh Kumar Jain Nand Kishore Gupta

vii

Preface

The International Conference on Computational Intelligence and Emerging Power System (ICCIPS 2021) is the first International Conference that has been organized with the publication support of Springer. It is jointly organized by the Department of Electrical Engineering and Department of Computer Science of Engineering College Ajmer under TEQIP III. The objective of the conference was to bring together people with common ideas and to discuss issues related to artificial intelligence, machine learning, IoT, intelligent algorithms, swarm optimizations, big data, and their applications in power system, energy optimization, power optimization, power system, smart grid and renewable energy system, etc. The aim was to involve graduate students, research scholars, faculty, and industry persons to present their research work, theories, and ideas. The conference has been planned in three tracks, viz, Track-1 Computational Intelligence, Track-2 Emerging Power System, and Track-3 Renewable Energy System. On behalf of Engineering College Ajmer, we are pleased to welcome all the readers of the Proceedings of the International Conference on Computational Intelligence and Emerging Power System (ICCIPS 2021). This conference has provided an environment to conduct intellectual discussions and exchange ideas that are instrumental in shaping the future of artificial intelligence. The conference got an overwhelming response with 150 institutes participating from India and other countries like Portugal, Algeria, Ethiopia, etc. A total of 185 papers were received and peerreviewed by reviewers from different reputed institutions across India. A total of 30 papers were selected for presentation at the conference. All selected papers were registered and presented during March 9–10, 2021. We are thankful to Engineering College Ajmer for giving us this opportunity to organize this conference under the TEQIP III project. Sharjah, United Arab Emirates Talence Cedex, France Ajmer, India Ajmer, India

Ramesh C. Bansal Akka Zemmari K. G. Sharma Jyoti Gajrani

ix

Contents

1

Reading Gauges Using Computer Vision . . . . . . . . . . . . . . . . . . . . . . . . Dalvi Ameya, Ketkar Prathmesh, Dani Chinmay, Mundada Kapil, Gujarathi Mohnish, and Iyer Anand

2

Efficient Classification for Age and Gender of Unconstrained Face Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Guramritpal Singh Saggu, Keshav Gupta, and Palvinder Singh Mann

1

13

3

Prediction of Road Accidents Using Data Mining Techniques . . . . . . Fahim Ahmed Shakil, Sayed Muddashir Hossain, Rifat Hossain, and Sifat Momen

25

4

Eye Gaze-Based Student Readability Analysis . . . . . . . . . . . . . . . . . . . Niranjan Patil, Rahulkumar Das, Komal Dhusia, Varsha Sanap, and Vivek Kumar Singh

37

5

Enhancement of Life Time of Sensor Nodes in Wireless Sensor Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hemant Kumar Vijayvergia and Uma Shankar Modani

6

7

8

Social Distancing Through Image Processing, Video Analysis, and CNN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Abdul Bari, Sharfuddin Waseem Mohammed, Sameena, and Sharanya Deep Learning for Video Based Human Activity Recognition: Review and Recent Developments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sonika Jindal, Monika Sachdeva, and Alok Kumar Singh Kushwaha Machine Learning Methods for Protein Function Prediction . . . . . . Saurabh Biswas and Yasha Hasija

49

59

71 85

xi

xii

9

Contents

Intelligent Monitoring and Control of Vertical Farms for Metropolitan Areas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mihir Khara, Devashri Naik, Dhruvik Shah, and Sudha Gupta

99

10 A Brief Review on Parameters of Link Stability in MANETs . . . . . . 109 Kapila Pareek and Sumegh Tharewal 11 Combined LSTM GRU Model for Prediction of Congestion State in QUIC Protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 Sujeet Singh Bhadouria and Shashikant Gupta 12 Prediction of Autism Spectrum Disorder Using Feature Selection and Machine Learning Algorithms . . . . . . . . . . . . . . . . . . . . . 133 Mousumi Bala, Ayesha Aziz Prova, and Mohammad Hanif Ali 13 A Novel Hybrid Approach for Improving the Accuracy of Recommender Systems Using Weights . . . . . . . . . . . . . . . . . . . . . . . . 149 Dhiraj Khurana and Sunita Dhingra 14 A Novel Approach for Hand Gesture Recognition . . . . . . . . . . . . . . . . 159 Neha Kulshrestha, Satyanarayan Tazi, Uma Shankar Modani, and Manish Gupta 15 Android Malware Analysis Using Machine Learning Classifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 Sakshi Jain, Tarul Khandelwal, Yash Jain, and Jyoti Gajrani 16 Performance Enhancement of Renewable ADN by Strategical Placement of Dispatchable DER . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181 Pushpendra Singh and S. K. Bishnoi 17 Restructuring of Transmission Network to Cater Load Demand in Northern Parts of Rajasthan Using Renewable Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193 Rahul Choudhary, Om Prakash Mahela, Surendra Singh, Krishan Gopal Sharma, and Akhil Ranjan Garg 18 Performance Analysis of Multiobjective Particle Swarm Optimization Based Optimal Power Flow . . . . . . . . . . . . . . . . . . . . . . . . 213 Vineeta Chauhan and Jaydeep Chakravorty 19 Solar Power Predictions in Stochastics Framework . . . . . . . . . . . . . . . 223 Nandkishor Gupta, K. G. Sharma, A. Mangal, K. C. Sharma, and R. A. Gupta 20 Economic Analysis of Rural Distribution System with DER and Energy Storage System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235 Prashant Kumar and Vinod Kumar

Contents

xiii

21 An Intelligent Technique to Mitigate the Transient Effect on Circuit Breaker Due to the Occurrence of Various Types of Faults . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245 Shripati Vyas, R. R. Joshi, and Vinod Kumar 22 Power Quality Improvement by Using STATCOM for DFIG-Based Wind Energy Conversion System . . . . . . . . . . . . . . . . 253 Megha Vyas, Monika Vardia, Vinod Kumar, Shripati Vyas, and Yashwant Joshi 23 Optimization of Standalone Microgrid’s Operation Considering Battery Degradation Cost . . . . . . . . . . . . . . . . . . . . . . . . . . 267 Rekha Swami and Sunil Kumar Gupta 24 Investigation and Assessment of TCAD-Based Modeled and Simulated PPV/PCBM Bulk Heterojunction Solar Cells . . . . . . 279 Mohammad Asif Iqbal and Virendra Sangtani 25 Torque Ripple Reduction of PMSM Based Electric Vehicle . . . . . . . . 289 Ashish Kumar Panda, Giribabu Dyanamina, and Rishi Kumar Singh 26 Design and Implementation of Soft Computing-Based Robust PID Controller for CSTR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301 Rupali R. Gawde, Sharad P. Jadhav, and Bhawana A. Garg 27 A Double-SOGI-Based Power Quality Improvement for a Weak-Grid-Connected PV System . . . . . . . . . . . . . . . . . . . . . . . . . 319 Sanjay Kumar Peeploda, Tripurari Nath Gupta, and Mahiraj Singh Rawat 28 Multi-machine Power System Stabilizer Design Using Grey Wolf Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331 Ravi Kant Sharma, Dhanraj Chitara, Shashi Raj, K. R. Niazi, and Anil Swarnkar 29 Stability Enhancement of Grid Connected AC Microgrid in Modern Power Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345 Vivek Ranjan, Amit Arora, and Mahendra Bhadu 30 Performance Analysis of P&O and FLC Method of MPPT for PV Module Based on Five-Parameter Model . . . . . . . . . . . . . . . . . . 357 Prakash Bahrani and Naveen Jain Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371

About the Editors

Prof. Ramesh C. Bansal has more than 25 years of diversified experience of research, scholarship of teaching and learning, accreditation, industrial, and academic leadership in several countries. Currently, he is Professor in the Department of Electrical Engineering at University of Sharjah. Previously, he was Professor and Group Head (Power) in the ECE Department at University of Pretoria (UP), South Africa. Prior to his appointment at UP, he was employed by the University of Queensland, Australia; University of the South Pacific, Fiji; BITS Pilani, India; and Civil Construction Wing, All India Radio. He has significant experience of collaborating with industry and government organisations. He has made a significant contribution to the development and delivery of BS and ME programmes for utilities. He has extensive experience in the design and delivery of CPD programmes for professional engineers. He has carried out research and consultancy and attracted significant funding from industry and government organisations. He has published over 325 journal articles, presented papers at conferences, books, and chapters in books. He has Google citations of over 11000 and h-index of 50. He has supervised 25 Ph.D., 4 postdocs, and currently supervising 5 Ph.D. students. His diversified research interests are in the areas of renewable energy (wind, PV, microgrid), power systems, and smart grid. He is Editor/Associate Editor of several highly regarded journals including IEEE Systems Journal, IET Renewable Power Generation, and Technology and Economics of Smart Grids and

xv

xvi

About the Editors

Sustainable Energy. He is Fellow and Chartered Engineer IET-UK, Fellow Institution of Engineers (India), and Senior Member of IEEE-USA. Dr. Prof. Akka Zemmari is an associate professor at LaBRI, University of Bordeaux—CNRS. He received his Ph.D. degree and his HDR in Computer Science from Université Bordeaux in 2000 and 2009, respectively. His research areas deal with (1) the design, the analysis, and simulation of distributed algorithms, (2) the static/dynamic analysis of programs with application to malware detection, and (3) machine and deep learning with application to security aspects and to image analysis. Currently, he is serving as the head of Distributed Algorithms research group in LaBRI, the computer science research laboratory of the University of Bordeaux. He is PI/Co-PI of six research projects which also includes Joint Indo-French projects. He has acted as a referee for the many international journals and international conferences. He has published over 30 research articles in reputed journals. He has published over 50 research articles in reputed conferences. He has published a book on Deep Learning in Mining of Visual Content with Springer Briefs in Computer Science ISBN 978-3-030-34376-7. Dr. K. G. Sharma is an associate professor and the head of Electrical Engineering (EE) at Govt. Engineering College Ajmer, Rajasthan. He is contributing in the engineering profession since 20 years. He received the B. Tech. degree in EE from CTAE, Udaipur, M.Tech. in Power System from MNIT, Jaipur, and Ph.D. degree in EE from RTU, Kota. He published 25 national and international research papers, presented papers at conferences, authored 01 book, received education excellence award, and supervised more than 10 PG students. He is qualified as certified energy auditor of BEE. He is PI/Co-PI of three CRS projects under TEQIP-III. He has got two patents. His diversified research interests are in the areas of power system stability, renewable energy sources, spectral analysis, power system dynamics and control. Dr. Sharma is a fellow of the Institution of Engineers India, Life Member of Indian Society for Technical Education (ISTE), Member of ISHRAE and ISLE, Member of Soft Computing Research Society. He is a

About the Editors

xvii

member of various prestigious boards, DRC, inspection committee of university and AICTE. He has delivered several expert talks. He is a reviewer of many repute Journals of Research along with several international conferences. He has organized FDPs/STC/workshops along with one international conference. Dr. Jyoti Gajrani is an assistant professor and the head of the Department of Computer Science and Engineering at Engineering College Ajmer (Rajasthan). She completed B.Tech. in Computer Engineering from Mody Institute of Engineering and Technology, Lakshmangarh, Rajasthan, M.Tech. in Computer Engineering from the Indian Institute of Technology, Bombay, and Ph.D. in Computer Science and Engineering from Malaviya National Institute of Technology, Jaipur. She has 15 years of teaching and research experience. She has successfully supervised 05 M.Tech. scholars and 05 are currently working. Her research interests include the area of security and privacy, computational intelligence, operating system, and system software. She has received special reorganization by Dr. Nitin Deep Blaggan, ExSP Ajmer for development of Ajmer Police Traffic App. She is Co-PI in project titled “Performance Measurement of Security Algorithms on IoT Devices” under competitive research scheme by Rajasthan Technical University (TEQIP-III) with grant of 2.4 lakh. She has published around 40 research papers in reputed international journals and conferences including SCI and Scopus indexed papers. She is the author of one book on Telecommunication Engineering Fundamentals. She has delivered several expert talks. She is reviewer of IETE Journal of Research (Taylor and Fransis) and ACM DTRAP along with several international conferences ICCIS 2020, ICDLAIR2019, INDICON 2019, SIN 2017, SIN 2018. She has organized more than 30 FDPs, STCs hackathons, and international conferences.

Chapter 1

Reading Gauges Using Computer Vision Dalvi Ameya, Ketkar Prathmesh, Dani Chinmay, Mundada Kapil, Gujarathi Mohnish, and Iyer Anand

1 Introduction In the process industry like Oil and Gas sector, chemical plants and operations industry, analogue pointer-type gauges are prominently used. For such gauges, readings can be obtained only by visual inspections which require a technician to travel to the gauge location and log the current reading which can be further used in different operations. This method is time consuming, unsafe and is susceptible to human errors. An alternate approach to reading analogue gauges is to capture an image or record a video and use computer vision techniques. Using Open CV Image processing tool in Python Environment and some functions help in reading the present value [3]. That value will be directly stored in the database, so that it can be viewed in future anytime. Also, the graphic user interface (GUI) is created, so that operator in the control room can easily see the gauge and respective reading.

D. Ameya · K. Prathmesh · D. Chinmay · M. Kapil · G. . Mohnish (B) Instrumentation and Control Department, Vishwakarma Institute of Technology, Pune, India e-mail: [email protected] D. Ameya e-mail: [email protected] K. Prathmesh e-mail: [email protected] D. Chinmay e-mail: [email protected] M. Kapil e-mail: [email protected] I. Anand Global head, Fieper project, Pune, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 R. C. Bansal et al. (eds.), Proceedings of International Conference on Computational Intelligence and Emerging Power System, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-16-4103-9_1

1

2

D. Ameya et al.

2 Literature Review In today’s world of automation, the majority of the work is done by machines and robots. The use of advanced technologies like Image processing, the Internet of Things, Artificial Intelligence, etc. is increasing nowadays in Process Industries also. This part focuses on the approaches used by some publication papers that have an aim to build similar models. Depending upon the best suitable approaches, we also suggest our approach for the model. These papers majorly make use of technologies like pyzbar and OpenCV [4]. This section tells us about what were the methodologies and approaches that were used by the authors, whose papers were reviewed. Section of Media Technology, Aalborg University, Denmark 2EnviDan, Aalborg, Denmark, has presented a paper on r automated recognition and translation of pointer movement in analogue circular gauges. The proposed method processes an input video frame-wise in a module-based manner. Noise is minimized in each image using a bilateral filter before a Gaussian mean adaptive threshold is applied to segment objects. Subsequently, the objects are described by a set of proposed features and classified using probability distributions estimated using Expectation Maximization. The pointer is classified by the Mahalanobis distance and the angle of the pointer is determined using PCA. The output is a low pass filtered digital time series based on the temporal estimations of the pointer angle. https://core.ac.uk/download/pdf/304 612162.pdf. Jianbo Song and Lei Zhang, aimed at the problem of automatic interpretation of analogue instrument’s dial indicating data, making the automatic interpretation solution, analysing the relevant techniques and solving the key technical problems, so as to provide a reference for realization of automatic calibration for an analogue meter.

3 Methodology In the proposed approach, Python as the programming language is used and OpenCV and Pyzbar Libraries are the main tools. The following methodology has been used to read gauges of different types: • • • •

Circular Gauge with Pointer pivot in the centre. Circular Gauge with Pointer pivot just below centre. Rectangular Gauge with Pointer pivot at a corner. Circular Gauge with Pointer of a different colour.

1 Reading Gauges Using Computer Vision

3

3.1 Database Creation Excel is used to create the database. The database will have the following information of every gauge that it has to read: • • • • • • • • • • •

Gauge shape (round, square). Gauge dimension in mm. Pointer Colour. Pointer pivot position (coordinates). Pointer length. Pointer tail length. Scale Minimum. Minimum scale position angle. Scale maximum. Maximum Scale position angle. Units.

The above information of a gauge will be required to get the gauge reading. Records of a particular gauge are accessed using the QR code.

3.2 QR Code Reading QR codes are often used to contain web address information and links. In this particular case, QR codes are used to detect a gauge in the process industry and to access information about that particular gauge in the database. The QR code will be placed near the gauge and will be read using the Camera. The database record to which the QR code will point has the necessary information required to read gauges. This simplifies the process of detecting gauge in the industry environment and also saves the processing time (see Fig. 1).

3.3 Reading Gauge and Storing Value The following steps are used to get the gauge reading from the image: • Coordinates of the needle pivot and the length of the needle are extracted from the database. • For locating the gauge needle, radial lines are drawn from the gauge centre to the pixels on the circumference and the line with the darkest average pixel value will be the needle [5]. • The needle angle is calculated from the lowest point on the circle circumference and the angle is converted to appropriate reading [6].

4

D. Ameya et al.

Fig. 1 Finding patterns in QR code

Fig. 2 Radial lines identified as pointer location

• The gauge reading will be directly stored in the database, for inspection purpose (see Figs. 2 and 3).

3.4 Testing For testing the algorithm, images of the gauge dial and needle are considered (Fig. 4).

1 Reading Gauges Using Computer Vision

5

Fig. 3 Sample radial lines

Fig. 4 Gauge dial and gauge needle

These images are used to create synthetic images for testing the algorithm by rotating the needle on the gauge dial to get different gauge readings. For this purpose, the following steps are performed [7]: • Determining angle corresponding to the minimum scale reading on the gauge and set needle at that particular angle. • Iterate over the range between the minimum scale angle and the maximum scale angle. • Create the image with the labels in gauge units instead of degrees and the following is the result (see Fig. 5).

6

D. Ameya et al.

Fig. 5 Synthetic Images for testing the proposed methodology

3.5 GUI A simple GUI is designed for importing images and then running the program with the imported image as an input. The imported image is displayed in the GUI. The image is pre-processed and the steps for reading gauges are carried out as explained above. The final result is displayed in the same. The final image result is displayed in the GUI with the gauge reading (see Fig. 6).

Fig. 6 GUI

1 Reading Gauges Using Computer Vision

7

Fig. 7 GUI with final results

4 Results 4.1 GUI A GUI is designed to view the image as well as the results on the interface. The required image is selected using the ‘Get Gauge Image’ button and then the program runs by clicking on the ‘Run’. Then the gauge reading appears in the ‘Result’ box (see Fig. 7).

4.2 Database The gauge parameters such as gauge location, shape, pointer radius and centre coordinates are stored in the database given below. Once the program is executed, the present value of the gauge is then stored in the database (see Fig. 8).

4.3 Testing The gauges used for testing are of different shapes, sizes and different pointer colours. (a)

(b)

Test 1: The input image is a pressure gauge with a range of 0–15psi. The shape of the gauge is round and the pointer colour is black. The output image shows the gauge reading that is 1.92 psi (see Fig. 9). Test 2: The input image is a voltmeter with a range of 0–300 V. The shape of the gauge is rectangular with a black pointer. The output image shows the reading as 0 V (see Fig. 10).

8

D. Ameya et al.

Fig. 8 Database snapshot

Fig. 9 Result of Test 1

(c)

(d)

(e)

(f)

Test 3: The gauge is a voltmeter having a rectangular shape with a range of 0–300 V. The peculiarity of the gauge is that it has a pointer of red colour. The algorithm is able to detect the pointer successfully and the final value is 39.56 V (see Fig. 11). Test 4: The input image is the voltmeter and the ammeter of a single-phase submersible pump captured by a mobile camera in natural conditions. The voltmeter range is 0–300 v and the ammeter range is 0–30 A (see Fig. 12). Test 5: The input image is a power gauge in KW ranging from 0–100 KW. It has rectangular shape with a black colour pointer. The gauge reading comes out to be 0.82 KW as the pointer is very close to 0 KW (see Fig. 13). Test 6: The input image is a voltmeter with a range of −10–10 V. The shape of the gauge is round and the pointer colour is black (see Fig. 14).

1 Reading Gauges Using Computer Vision

Fig. 10 Result of Test 2

Fig. 11 Result of Test 3

Actual image Ammeter reading(4.05A) Voltmeter reading(3.98V) Fig. 12 Result of Test 4

9

10

D. Ameya et al.

Fig. 13 Result of Test 5

Fig. 14 Result of Test 6

(g)

Test 7: The input image is a speedometer with a range of 0–160kmph. The glass of the speedometer is covered with fog and the image is slightly blurred. The algorithm is not able to detect the pointer location due to fog and it is giving a false location (see Fig. 15).

5 Conclusion The proposed approach is successfully able to decode QR codes to get information from the database and that information is successfully used to read gauges using Computer vision tools in Python. • The proposed approach aims to read analogue gauges using computer vision tools. • The proposed approach yields successful results for gauges of various shapes, different pointer positions and different pointer colours. • The proposed approach is designed in a way to effectively use the computational power of the robot and not overuse it.

1 Reading Gauges Using Computer Vision

Input

11

Output

Fig. 15 Result of Test 7

• For better accuracy and results, the information in the database has to be correct and the input images need to be clear. • For better results, the position of the barcode and the gauge should be fixed. • It is observed that the approach doesn’t work for images with fog on them. • The proposed approach can be deployed in a test environment and will be suitable for use for purposes stated in the project.

6 Future Scope • A more robust approach can be used to read gauges with fog. • The database can be periodically checked and updated for getting errorless results. • The proposed approach can be modified to work on videos as well.

References 1. Chi J, Liu L, Liu J, Jiang Z, Zhang G (2015) Machine vision based automatic detection method of indicating values of a pointer gauge. Math. Problems Eng. 2015:19 pages. Article ID 283629. https://doi.org/10.1155/2015/283629 2. (2019) Open physics, vol. 17(1), pp 86–92, eISSN 2391–5471. https://scholar.google.co.in/ scholar?q=Open+Physics,+Volume+17,+Issue+1,+Pages+86%E2%80%9392,+ISSN+23915471&hl=en&as_sdt=0&as_vis=1&oi=scholart 3. Gellaboina MK, Swaminathan G, Venkoparao V (Jun 2013) Analog dial gauge reader for handheld devices. https://ieeexplore.ieee.org/abstract/document/6566539 4. Sablatnig R, Kropatsch W (Nov 1994) Automatic reading of analog display instruments. IEEE Xplore. https://cvl.tuwien.ac.at/wp-content/uploads/2014/12/icpr94.pdf 5. Tian E, Zhang H, Hanafiah MM (2019) A pointer location algorithm for computer vision based automatic reading recognition of pointer gauges. https://www.degruyter.com/document/. https:// doi.org/10.1515/phys-2019-0010/html

12

D. Ameya et al.

6. Ye X, Xie D, Tao S (May 2013) Automatic value identification of pointer-type pressure gauge based on machine vision. Explor J Comput 8(5). https://www.researchgate.net/publication/272 798714_Automatic_Value_Identification_of_Pointer-Type_Pressure_Gauge_Based_on_Mach ine_Vision 7. oci-labs/deep-gauge (GitHub). https://github.com/oci-labs/deep-gauge

Chapter 2

Efficient Classification for Age and Gender of Unconstrained Face Images Guramritpal Singh Saggu, Keshav Gupta, and Palvinder Singh Mann

1 Introduction Classification of age and gender is part of the facial analysis which has been becoming popular in the computer vision community. In the wild means that the classification or the analysis is done on the images that are not taken in the controlled environment so, they tend to mimic the real world pretty well. A highly accurate age and gender classification system can provide various applications in the fields involving human-computer interaction, entertainment and cosmetology. Moreover, it would assist in more critical fields like surveillance, forensic, etc. [1]. Despite the benefits this classification can provide, the ability to automatically classify age and gender with accuracy and reliability from images of faces is a long way from meeting the necessities of enterprise applications. Introductory methods involved the usage of statistical models and features designed by experts for classification of age and gender [2–4] the achieved good scores on the dataset which contained images taken in control environment, i.e. noise free images with relatively less changing background. But their performance proved to be below expectations for “in the wild” images like present OIU-Adience benchmark dataset [5] which we will use for training and validation of our approach. Recent methods that have used Convolutional Neural Network to classify age and gender [6, 7] have been able to perform better on in the wild images but they still suffer from problems like having low accuracy for age classification or low classification speed because of increasing depth in the latest neural networks. These problems make them unsuitable for commercial real-world applications. G. S. Saggu (B) · K. Gupta Indian Institute of Information Technology and Management, Gwalior, India P. S. Mann DAV Institute of Engineering and Technology, Jalandhar, India

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 R. C. Bansal et al. (eds.), Proceedings of International Conference on Computational Intelligence and Emerging Power System, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-16-4103-9_2

13

14

G. S. Saggu et al.

In this paper, we attempt to improve on the convolution neural network techniques and cover-up for the main disadvantages faced by them, i.e. low accuracy on age classification and slow speed. Our main contribution through the proposed method is to provide a method and a model which is accurate, efficient and generalizes well, all while requiring low resources for training and easy to fine-tune for other more specific age and gender classification.

2 Related Works Before proceeding towards describing our proposed method, we will briefly review the literature and previously used methods for age and gender classification and also provide an overview of the other related topics.

2.1 Age and Gender Classification Now, we will briefly review the previous methods that have been proposed before and work related to our methodology, we will also briefly review literature related to the task. The early methods are based on controlled imaging environments and using manual and handcrafted features of the face. There were very few studies which focused on age and gender classification in the unconstrained real-world environments. In [8], the authors developed an age estimation method that helped determine the ratios of facial features among various dimensions using the geometric features. This method had drawbacks and was not successful in properly distinguishing the junior and senior adults while it worked well enough for infants and adults. In [9], the authors used both the texture and geometric features for the task of prediction, using an active appearance model. Thus, this method doesn’t prove to be suitable for constrained conditions of imaging as they have a lot of variations. In [4], the authors leveraged manually designed features for this task and various other works as well [10, 11]. In [12], Dileep et al. proposed a convolutional neural network and an approach 3-sigma based on control limits for classifying a person’s age into various classes. Most of the approaches discussed here work only on images taken in controlled environment and is unable to achieve decent enough results on the images taken in uncontrolled environment which are often encountered in real-world and practical applications.

2.2 Deep Convolutional Neural Network In recent times, most of the research work for classification of age and gender involves the use of convolutional neural networks (CNNs). As CNN has good feature extrac-

2 Efficient Classification for Age and Gender of Unconstrained Face Images

15

tion techniques thus its has the ability to classify age and gender based on the face images [13–15]. Deep CNN has been widely adopted in recent times due to the large availability of the data and the computation capabilities required to build the CNN models. CNN models are very powerful to learn discriminative and compact features from faces if the amount of data available is sufficient enough. Real-world face images classification model for age was developed by [7], composed of both convolutional and fully connected dense layer which are three and two in quantity. The issues of misalignment in the unconstrained images were handled using oversampling and centre-crop methods. In [16], the authors developed an multitask CNN system which was able to learn complex structures and weights to perform classification on age, ethnicity and gender. Qawaqneh et al. [17] proposed a deep VGG-Face pre-trained CNN methodology for age estimation tasks. CNN models trained for this purpose have eight convolutional layers with three fully connected dense layers. In [6], the authors proposed a residual network of the residual network (RoR) for the task of age and gender estimation. In [47], the authors proposed an end-to-end multitask learning framework which solved the task of age estimation and gender recognition using a single CNN. Ranjan et al. [18] developed a CNN for classification of age using a single image of the face. This solution used a face alignment preprocessing before feeding the images to CNN. The focal loss function was used in CNN models developed by authors in [19]. In [20], the authors performed classification using hybrid CNN structure, as it used CNN to extract features but extreme learning machines to perform classification. Similarly authors in [21] also used CNN and an extreme learning machine (ELM) and this model showed improved performance. In this research work mentioned in Sects. 2.1 and 2.2, only a few experiments have been carried out in the unconstrained imaging conditions while some of these work well on images taken in controlled environment. However, the image in uncontrollable conditions is also known as inthe-wild environment is still a challenge to classify them due to variations in lighting, alignment, viewpoint, etc. This requires an effective and robust method so that the application of these models can make largely possible in the real-world environment. Here, we address these various issues and developed a CNN-based framework which is more effective and generalize for these conditions.

3 Methodology In this section, we will discuss the methodology and describe the architecture of the proposed method’s model, the training technique and other various aspects. The motivation behind the decisions is taken in the process. The methodology primarily consists of using efficientnet architecture as base on your convolutional neural network and adjusting it to the domain by iterating on a large face dataset and fine-tuning for the OIU-Adience dataset as shown in Fig. 1 which shows the pipeline for training the proposed method. It includes making changes to VGGFace2 dataset followed by domain change of neural networks and fine-tuning for the task at hand.

16

G. S. Saggu et al.

Fig. 1 The pipeline for training model for age and gender classification

Image Preprocessing: Image preprocessing phase is important to effectively classify age and gender under the real-world unconstrained conditions. As images in such settings are non-frontal and unaligned most of the time with a lot of differences in the parameters such as lighting, background and other conditions. Thus, these images required to be detected, cropped and aligned before using them as input for our classification models. Face Detection: Image preprocessing starts with the detection of the face. In this phase, detection helps to find the face in the input images. In our methodology, we have used fast deep convolutional face detection in the wild exploiting hard sample mining [22]. In this approach, images are passed through two different networks, outputs where combined and evaluated. Both the networks are capable of detecting faces in the input images under unconstrained pose variations and dire occlusion. After this, the best output one with highest confidence is selected and passed on to the next phase. Cropping and Face Alignment: After we are done with face detection, we perform the cropping and face alignment. To align the faces to a single reference coordinate frame, we used the state-of-the-art method proposed in [5]. This novel technique is based on iterative estimation of the uncertainties of the facial feature localization’s which provide robust facial alignment. They used an Iterative re-weighted least squares approach which obtained transformation until convergence. Affine transformation is performed and each iteration provided considerable improvements in performance, thus optimal results were obtained. Then, these images are cropped along the rectangular border of aligned face. CNN Architecture: The task of age and gender classification from face images is a rather complex one, i.e. simple feature extraction for classification would rather not perform that well. For extracting complex features, we need deep networks and training a deep network on relatively small dataset such as OIU-Adience would result in overfitting. So we use the concept of transfer learning with some changes. Transfer learning allows one to use networks trained for one task and fine-tune for the current task. More the current task is close or similar to the task for which the model was made, better the model is expected to perform on the current task. First, we have to choose a network architecture which can work effectively for this task. After

2 Efficient Classification for Age and Gender of Unconstrained Face Images

17

Fig. 2 Architecture of a single convolutional block of EfficientNet-B0

exploring various architecture, we decided to go with the efficientnet with pre-trained weights of imagenet dataset, single block architecture of efficientnet-b0 is shown in Fig. 2. EfficientNet was preferred for this task because it is optimized for accuracy and efficiency which both are the main focus of our method. The efficientnet was created using the concept of compound scaling, which claims to balance the increase of all the important aspects of the network, i.e. depth, width and image resolution to provide best increased accuracy for the resources consumed. Moreover, the baseline model on which compound scaling is applied is made using neural architect search with the objective of optimizing accuracy and computation. In the pre-trained network, we remove the output layer and replace it by two fully connected layers, the final layer is the output layer which consists of two units.

18

G. S. Saggu et al.

Probability of each class label is given by softmax and its equation is given as follows: exp(xi ) Softmax(xi ) =  (1) j exp(x j ) where xi denotes the vector of class scores of the ith element. Binary cross entropy loss is used to measure the performance of the binary classification task is calculated as follows: − (y log( p) + (1 − y) log(1 − p)) (2) and in case of multiclass classification, multiclass cross entropy loss is used calculated as follows: M  yo,c log( po,c ) (3) − c=1

where y is binary indicator if c class label is correctly classified for observation o and p is predicted observation o is of class c. Training Details: Since our efficientnet model was trained on the imagenet dataset, the domain of the model is general and quite different from the domain of the task, we decided to do what we call as domain change of the model. We took a large dataset of face images VGGFace2 and mapped the current annotations to the gender of the person in the image. We then tuned the weights of the entire model by iterating it over multiple times. We believe that this would allow models to extract better features from the facial images. By doing this, we are trying to bring the domain of the model closer to the domain of the task. Following this the model is trained separately for classification of age and gender. Age Classification: For age classification, we again start by taking the model tuned for better extracting the facial futures. But since this time the output of age classification is different from that of the model, we will remove the old output layer and add a new one according to the task of age classification. First, we only train the last layer on the OIU-Adience benchmark and then the last few layers are fine tuned on the dataset. Fine tuning here means that we will not change the weights of the layers present in the beginning but only the layers present at the end. Gender Classification: For classification of gender, we begin by taking the model tuned for better extracting facial features. Since the model was previously tuned with gender classification dataset only, we need not scrap the output layer of the model. We will simply fine-tune the model on the adience dataset so that it can work better on noisy images.

2 Efficient Classification for Age and Gender of Unconstrained Face Images

19

4 Results and Experiments This section is used to describe the various datasets being used and present the results, analysis of the experiments performed, evaluate the performance of the proposed model and compare it with other techniques used in the past.

4.1 Datasets VGGFace2 [23], Adience Benchmark [5] and MORPH-II [24] dataset have been used in the experiments of the proposed method where VGGFace2 is used to change the domain of efficientnet network trained on imagenet to better learn the face features. Proposed method has been fine tuned and tested in the OIU-Adience and validated on MORPH-II dataset. A concise overview of the datasets such as size, subjects, age and its range are given in Table 1. Also, the illustration of a few images of each dataset has been shown in Fig. 3. Now, we will look into the description of each dataset. VGGFace2: VGGFace2 is a large scale dataset used for face recognition. Images for the dataset have been taken for Google Image Search and thus possess a lot of variations in terms of lighting, pose, age, gender, occupation and much more. All faces are unfiltered and are captured “in the wild” conditions with variations.

Table 1 Datasets used in the experiments with their details Dataset Size Subjects Age VGGFace2 [23] 3,310,000 OIU-Adience [5] 26,580 Morph-II [24] 55,134

9131 2284 13,618

– Group Actual

Range – 0–60+ 16–77

Fig. 3 Sample faces from OIU-Adience dataset and VGGFace2 dataset respectively

20

G. S. Saggu et al.

OIU-Adience: OIU-Adience dataset is an aggregation of real life and unfiltered conditions face images. It provides all the features that can be anticipated from the dataset collected “in the wild” conditions. These are the faces which were uploaded to the cloud with any filter. They were uploaded to Flickr website therefore adience images possess large variations in noise, display, pose among various other variations. OIU-Adience collection has more than twenty five thousand face images with more than two thousand different subjects, with eight age range. More details of the dataset are in Table 1. The dataset has been divided into five folds and both the age and gender models has been cross validated on each of the five folds. Morph-II: Morph-II is the publicly available dataset which is frequently used for classification of age and gender. This dataset has more than 13,000 subjects with 55,134 images. Morph-II has ages ranging from 16 to 77 years with more than 4600 images of male and more than 8000 females. We have classified the age range of this dataset into groups 16–20, 25–32, 38–43, 48–53 and 60+ so that model proposed can be validated on this dataset. Experimental Results and Analysis: We will present the results of the proposed model. For OIU-Adience, the 5 fold cross-validation results along with comparison with other results is provided. We only use Morph-II for testing, so results for the same are also provided. For gender classification, we use accuracy as a metric but for age classifcation we use two metrics accuracy and one-off accuracy. Accuracy: Accuracy or exact accuracy is defined as the percentage of predictions that are correct, which would translate to percentage of age and gender group predicted correctly. It would be mathematically defined as Accuracy =

No. of accurate prediction Total number of predictions

(4)

One-Off Accuracy: This evaluation metric is defined as off by one neighbouring group and measures predicted label of the class is the same as that of the ground truth or lies in the two neighbouring classes. This metric is used in the presentation of the results of the age classification only. Age Classification For age classification, the performance of the proposed method has been evaluated using two metrics which extract accuracy and one-off accuracy. The model has been trained to classify the face image into one of eight age groups. Age classification has been trained and validated on Adience and tested on MorphII. Our proposed network when evaluated on the Adience dataset obtains an exact accuracy of 89.50% and one-off accuracy of 94.79% on cross validation on five folds. These results are remarkably better than the previously reported state-of-theart results and show significant improvement of nearly 6%. On Morph-II dataset, we receive the accuracy of 88.23% indicating good generalizability of the model. The plot of exact accuracy of training and validation of OIU-Adience dataset has been shown in Fig. 4b for age classification. Table 2 shows the classification results of the proposed model as compared to other works in the OIU-Adience benchmark dataset.

2 Efficient Classification for Age and Gender of Unconstrained Face Images

21

Fig. 4 Plots of accuracy for the age and gender classification Table 2 Age classification cross-validation results on OIU-Adience benchmark dataset Method Exact accuracy (%) One-Off (%) SVM-dropout [5] Multiple CNNs [25] DAPP [26] RoR-34 [6] RoR-152 [6] CNN [27] Proposed method

45.1 58.5 62.2 66.91 67.34 83.1 89.50

79.5 – – 97.49 97.51 93.8 94.79

Gender Classification The proposed method has also been used for gender classification on the OIU-Adience dataset and Morph-II. Gender classification has been trained and validated on Adience and tested on Morph-II. For gender classification, we have trained our network on two labels, male and female and reported the performance of the validation set on exact accuracy. As shown in Fig. 4, we have achieved an exact accuracy of 94.12% on five-fold cross validation which is very much comparable to the previous state-of-the-art results. On Morph-II dataset, we receive the accuracy of 92.57% without even fine-tuning on the dataset. Table 3 shows the classification results of the proposed model as compared to other works in the OIU-Adience benchmark dataset. Other results in the table also use five-fold cross validation. The graph of accuracy of training and validation set is shown in Fig. 4a. As the accuracy graphs indicate the number of epochs, it takes the said accuracy is only 20 and 30 for gender and age classification, respectively, which is considerably less than normally used so training or adjusting the dataset according to particular condition takes significantly less time. This also indicates that the proposed method is very efficient in training and learning new patterns. For training, we use RMSProp optimizer with the learning rate in the order of 0.001. The learning rate keeps decreasing throughout the course of training.

22

G. S. Saggu et al.

Table 3 Gender classification cross-validation results on OIU-Adience benchmark dataset Method Exact accuracy (%) SVM-dropout [5] 4c2f-CNN [28] ResNets-34 [6] RoR-34 [6] CNN [27] Method

79.3 86.8 93.17 93.24 96.2 94.12

Though our method might not have achieved state-of-the-art performance in gender classification, it performs significantly better than previous models in age classification. So we see this as an good trade-off. Since most of the earlier works used accuracy as their metric, we had to use the same as otherwise we’d have nothing to compare our results against. Moreover age detection was posed as classification problem, and it limited the metrics to the ones used for classification.

5 Conclusion and Future Work We posed the problem of age and gender detection as a classical classification problem and used the standard tools, functions and training methodology available for the same to solve our task. The proposed method involves taking a pre-trained model on imagenet, which is good at extracting basic image features and building on top of them and changing its domain to facial images by training it on large facial dataset like VGGFace2. Finally, the model is fine tuned on the OIU-Adience benchmark which contains unfiltered images, so that model can get accustomed to noise. We also rely on a pre-processing algorithm that provides us with aligned and cropped facial images. Finally, the model’s performance is tested on both OIU-Adience and morph-II dataset and we achieve significant improvement in the results of age classification compared to previous state of the art and for gender classification the results are comparative to the state of the art as well. Validation of the method and model is checked using K-fold cross validation. For future we will consider using model which is pre-trained on the facial dataset rather than general purpose dataset, and we will also consider including other miscellaneous techniques in order to boost the robustness and accuracy of the method.

2 Efficient Classification for Age and Gender of Unconstrained Face Images

23

References 1. Rothe R, Timofte R, Van Gool L (2018) Deep expectation of real and apparent age from a single image without facial landmarks. Int J Comput Vision 126(2–4):144–157 2. Ma Z, Leijon A (2011) Bayesian estimation of beta mixture models with variational inference. IEEE Trans Pattern Anal Mach Intell 33(11), 2160–2173 3. Ma Z, Teschendorff AE, Leijon A, Qiao Y, Zhang H, Guo J (2014) Variational bayesian matrix factorization for bounded support data. IEEE Trans Pattern Anal Mach Intell 37(4):876–889 4. Gao F, Ai H (2009) Face age classification on consumer images with gabor feature and fuzzy lda method. In: International conference on biometrics, pp 132–141. Springer 5. Eidinger E, Enbar R, Hassner T (2014) Age and gender estimation of unfiltered faces. IEEE Trans Inform Forensics Secur 9(12):2170–2179 6. Zhang K, Gao C, Guo L, Sun M, Yuan X, Han TX, Zhao Z, Li B (2017) Age group and gender estimation in the wild with deep ror architecture. IEEE Access 5:22492–22503 7. Levi G, Hassner T (2015) Age and gender classification using convolutional neural networks. In: Proceedings of the IEEE conference on computer vision and pattern recognition workshops, pp 34–42 8. Kwon YH, da Vitoria Lobo N (1999) Age classification from facial images. Comput Vision Image Underst 74(1):1–21 9. Lanitis A, Draganova C, Christodoulou C (2004) Comparing different classifiers for automatic age estimation. IEEE Trans Syst Man Cybern Part B (Cybernetics) 34(1):621–628 10. Gunay A, Nabiyev VV (2008) Automatic age classification with lbp. In: 2008 23rd international symposium on computer and information sciences, pp 1–4. IEEE 11. Guo G, Mu G, Fu Y, Huang TS (2009) Human age estimation using bio-inspired features. In: 2009 IEEE conference on computer vision and pattern recognition, pp 112–119. IEEE 12. Dileep MR, Danti A (2018) Human age and gender prediction based on neural networks and three sigma control limits. Appl Artif Intell 32(3), 281–292 13. Szegedy C, Liu W, Jia Y, Sermanet P, Reed S, Anguelov D, Erhan D, Vanhoucke V, Rabinovich A (2015) Going deeper with convolutions. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 1–9 14. He K, Zhang X, Ren S, Sun J (2016) Deep residual learning for image recognition. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 770–778 15. Lin M, Chen Q, Yan S (2013) Network in network. arXiv preprint arXiv:1312.4400 16. Yi D, Lei Z, Li SZ (2014) Age estimation by multi-scale convolutional network. In: Asian conference on computer vision, pp 144–158. Springer 17. Qawaqneh Z, Mallouh AA, Barkana BD (2017) Deep convolutional neural network for age estimation based on vgg-face model. arXiv preprint arXiv:1709.01664 18. Ranjan R, Sankaranarayanan S, Castillo CD, Chellappa R (2017) An all-in-one convolutional neural network for face analysis. In: 2017 12th IEEE international conference on automatic face & gesture recognition (FG 2017), pp 17–24. IEEE 19. Liu W, Chen L, Chen Y et al (2018) Age classification using convolutional neural networks with the multi-class focal loss. In: IOP conference series: materials science and engineering, vol 428 20. Duan M, Li K, Yang C, Li K (2018) A hybrid deep learning cnn-elm for age and gender classification. Neurocomputing 275:448–461 21. Duan M, Li K, Li K (2017) An ensemble cnn2elm for age estimation. IEEE Trans Inform Forensics Secur 13(3), 758–772 22. Danai Triantafyllidou, Paraskevi Nousi, and Anastasios Tefas. Fast deep convolutional face detection in the wild exploiting hard sample mining. Big data research, 11:65–76, 2018 23. Cao Q, Shen L, Xie W, Parkhi OM, Zisserman A (2018) Vggface2: a dataset for recognising faces across pose and age. In: 2018 13th IEEE international conference on automatic face & gesture recognition (FG 2018), pp 67–74. IEEE

24

G. S. Saggu et al.

24. Ricanek K, Tesafaye T (2006) Morph: a longitudinal image database of normal adult ageprogression. In: 7th international conference on automatic face and gesture recognition (FGR06), pp 341–345. IEEE 25. Anand A, Labati RD, Genovese A, Muñoz E, Piuri V, Scotti F (2017) Age estimation based on face images and pre-trained convolutional neural networks. In: 2017 IEEE symposium series on computational intelligence (SSCI), pp 1–7. IEEE 26. Md Tauhid Bin Iqbal, Mohammad Shoyaib, Byungyong Ryu, Mohammad Abdullah-AlWadud, and Oksam Chae. Directional age-primitive pattern (dapp) for human age group recognition and age estimation. IEEE Transactions on Information Forensics and Security, 12(11), 2505–2517, 2017 27. Agbo-Ajala O, Viriri S (2020) Deeply learned classifiers for age and gender predictions of unfiltered faces. The Scientific World J 2020 28. Ekmekji A (2016) Convolutional neural networks for age and gender classification. Stanford University

Chapter 3

Prediction of Road Accidents Using Data Mining Techniques Fahim Ahmed Shakil, Sayed Muddashir Hossain, Rifat Hossain, and Sifat Momen

1 Introduction Road accident is among the leading causes of deaths around the world [9]. According to Global status report on road safety 2015, the leading cause of deaths among young people aged around 15–29 is traffic accident [9]. On the same report, it is also stated that around 1.24 million people faces death each year due to traffic accidents. These statistics surely draw the attention for preventive measurements. There are many attempts going on to take measurements to reduce traffic accident. Machine learning and data mining techniques in this field is emerging as a promising solution. There are many factors that may cause a traffic accident—such as drunk driving, crossing speed limit and breaking traffic laws [11]. To discern the factors that leads to traffic accidents, many traffic accident prediction models have been developed. Previously, there has been work of traffic data visualization of Dhaka city [10]. Machine learning is one of the prevalent technologies that can be used to ascertain the reasons that lead to traffic accidents by analyzing data regarding traffic accident. Machine learning models can also be used to predict the severity of the casualty in a traffic accident. By using the predicted casualty severity numbers, authorities can take preparations for emergency medical and other supports for severely injured even F. A. Shakil · S. M. Hossain (B) · R. Hossain · S. Momen Department of Electrical and Computer Engineering, North South University, Plot-15, Block-B, Bashundhara, Dhaka, Bangladesh e-mail: [email protected] F. A. Shakil e-mail: [email protected] R. Hossain e-mail: [email protected] S. Momen e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 R. C. Bansal et al. (eds.), Proceedings of International Conference on Computational Intelligence and Emerging Power System, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-16-4103-9_3

25

26

F. A. Shakil et al.

before the occurrence of an accident. Insurance agencies can also use the prediction of casualty severity in traffic accidents to set premiums for their customers. Again, machine learning models can be used to find out the dominant features that influence the predictability of severity of injury of traffic accidents too. Using of machine learning in this field to take preventive measures is very promising. Even though many organizations that works to ensure traffic safety have implemented various methods to predict the causes behind traffic accidents and severity of casualties, there are still many ambiguous aspects that may lead to traffic accidents and casualty severity. Thus, the scope of this research is to evaluate the prominent aspects that determines the casualty severity of a traffic accident. In this study, we have processed road accidents data of Great Britain from 1979 to 2015 [2] by applying various machine learning models. The objective of the study is to figure out the important aspects that affect the casualty severity of a traffic accident by analyzing the efficiency of various machine learning models. The performances of different machine learning models are evaluated after removing irrelevant aspects that do not influence the casualty severity of traffic accidents. Features that influence the prediction of casualty severity of traffic accidents are valuable information that can be used to reduce the impacts of traffic accidents. The paper is outlined as follows: Sect. 2 reviews related articles on casualty severity prediction due to traffic accidents using machine learning models. Section 3 discusses the methodology of the research and data and features used. Section 4 analyzes the experiments and results. In Sect. 5, conclusion and limitation are discussed.

2 Literature Review A research to predict the severity of traffic accident by using an artificial neural network (ANN), genetic algorithm (GA), and pattern search (PS) was conducted by Kunt [6]. They used a dataset that was derived from the total number of crashes that occurred on the Tehran–Ghom freeway in 2007 [6]. It was found that ANN with MLP architecture performed better than GA and PS models to predict traffic injury severity. In a study conducted in 2008, it was identified that feed forward back propagation (FFBP) networks such as multilayer perceptron (MLP) models generate best results to predict the casualty severity [7]. The authors determined 25 influential independent features in the prediction of crash severity by fatality-injury crash percent in urban highways. In the study, relationship between crash severity in urban highways, and traffic variables (traffic volume, flow speed, human factors, road, vehicle and weather conditions) were found [7]. To analyze the influential factors that affects the severity of injury in traffic accident, a non-parametric tree-based (CART) model was used in a study conducted by Chang [3]. The authors used 20 predictor variables with injury severity level as class attribute and found out collision type, contributing circumstance and driver/vehicle action as the features that influence the outcome of injury severity in traffic accidents.

3 Prediction of Road Accidents Using Data Mining Techniques

27

In the study, the CART model predicted the casualty severity for the learning data with 90.3% accuracy and testing data with 91.7% accuracy. To classify traffic accidents by casualty severity, Juan de Ona conducted a study where Bayesian networks (BNs) were used [8]. The authors used accident data of rural highways of Spain where 18 features were used to develop three BNs and identified— accident type, driver age, lighting and number of injuries; as the influential features to classify traffic accidents by casualty severity. Comparison between popular ANN paradigms to predict casualty severity level was analyzed by Mohamed A. Abdel-Aty in a study conducted in 2004 [1]. In the study, the authors applied ANNs to predict casualty severity of drivers involved in traffic accidents and compared the neural network models with calibrated ordered probit model. The accuracies for testing classification were 73.5, 70.6 and 61.7% for MLP, fuzzy ARTMAP and ordered probit model respectively [1]. This demonstrates that neural networks typically performs better in predicting traffic accidents severity. In a study conducted in 2005 [4] Miao Chong and his colleagues compared four machine learning paradigms namely Neural networks trained using mixed learning methods, supporting vector machines, decision trees and a combined mixed model incorporating decision trees and neural networks. Their experimental findings showed that the hybrid decision-making tree-neural network approach among the machine learning paradigms had outperformed the individual methods. Their research showed that the hybrid method performed better than neural network, decision trees and support vector machines for the non-invalid injury, the incapacitating injury, and the fatal injury classes. The hybrid approach accomplished better than neural network for the no injury and potential injury levels. The no injury and potential injury classes could best be specifically modeled by decision trees [4]. So we can see that, the reviewed literature on casualty severity prediction due to traffic accident shows that many studies have been conducted so far to figure out the severity of injury in a traffic accident. Different methods and models have been used to predict the casualty severity and some of them performed better than the others. Even though many research have been conducted regarding the prediction of casualty severity not many machine learning models have been used. Also the influential features that affects the severity of casualty in a traffic accident is still not completely figured out and varies depending on the dataset and models used. Therefore, the scope of this study is to apply different machine learning models to predict casualty severity and figure out the dominant features that influence the prediction of casualty severity.

3 Methodology In this research several machine learning methods were used to predict casualty severity of traffic accidents. In the following subsections our work methodology is discussed in details. This section presents details of data pre-processing, selected features and machine learning models utilized in the study. See Fig. 1 for the overview of our methodology and workflow.

28

F. A. Shakil et al.

Fig. 1 Proposed framework for traffic accident casualty severity classification

In brief, first we have collected the data. Then we cleaned the data for several steps. Cleaning consisted of getting rid of some attributes and defining the missing values so that it fit our machine learning tool. Then we ran the machine learning algorithms to get the models and from those we got the predicted results. The detailed work process and methodology has been described in the following subsections.

3.1 Data Capturing and Pre-processing The dataset named—“Road Accidents Incidence: Road Accidents Data Great Britain 1979–2015” used in this study, is obtained form Kaggle [2]. It contains many data from numerous road accidents occurred in United Kingdom from 1979 to 2015 including casualty severity and many other attributes like age of the driver, weather condition, road condition, vehicle details, etc. There are in total 70 attributes and 285K instances in the dataset. A detailed list of the attributes can be found at Kaggle dataset [2]. Information pre-processing is a significant step for the preparation of the dataset, and it deals with all kind of commotion. For example, managing missing qualities, discretization of factors, normalizing certain factors, etc. The goal of this step is to clean the data for applying machine learning algorithms and also determining the influential classes. For this we have used WEKA as a tool [5]. We have used WEKA as a data mining tool and also for modeling some machine learning models. It was used for filtering and clearing the data and making it convenient for our study. It has many different ways of clustering, visualizing, and classifying the data. Also, it is very well equipped with some filtering facilities which were very useful for our study.

3 Prediction of Road Accidents Using Data Mining Techniques

29

Details Of Pre-processing The data we got at first had many noises. Some attribute fields did not have finite values. Meaning that the values of those fields varied in wide ranges and for making the study precise we needed the values to be around some band ranges. Therefore, we removed those features. The removed features were accident index, age band of driver, age band of casualty, location easting osgr, location northing osgr, longitude, latitude, date, 1st road number, 2nd road number, lsoa of accident location, casualty reference, age band of casualty, number of vehicles and number of casualties. Also some field values were edited to fit in as finite values. For time attribute, we made a band of times and placed those band values for feature time. The bands for feature time are 05–10 = a, 11–14 = b, 15–18 = c, 19–23 = d, 00–04 = e. Then, we removed unnecessary attributes and apply band ranges for selected attributes. We ran some classifiers and found out that still there are many undefined missing values. According to WEKA syntax the missing values must be replaced with ‘?’ symbol. So again we cleaned our data. And that was the final dataset we used for our study. The final dataset contains 55 attributes and 285K instances.

3.2 Feature Selection and Learning Models Feature selection is an important process that can directly influence the performance of models. In this procedure, dominant features are figured out which lessens the dimensionality of input dataset by getting rid of insignificant attributes. As a result, it enhances the learning process and increases the prediction accuracy. Therefore, key attributes are identified in feature selection process to enhance the performance of used models. To identify the key features, we have used “Select Attribute” tool of Weka and used three different evaluation method—Correlation-based Feature Selection (CFS) subset evaluator, Information gain ranking filter, and Correlation ranking filter. CFS subset evaluator identifies subset of attributes that are highly correlated with the class attribute while having low intercorrelation are preferred. Information gain attribute evaluator identifies the worth of an attribute by measuring the information gain with respect to the class. And correlation ranking filter orders the attributes from higher to lower relation with class attribute (Table 1). After applying the aforementioned methods where ‘casualty-severity’ is selected as class attribute, the following features are identified as key attributes. From the table, we see that some features are common in all the three applied method—pedestrian-road-maintenance-worker, accident-severity, car-passenger, bus-or-coach-passenger. Five features sets have been selected after applying the methods where features set ‘D’ is selected by combining the common features found in other three selected features sets.

30

F. A. Shakil et al.

Table 1 The description of selected feature sets Dataset FS method A

CFS subset evaluator

B

Information gain ranking filter

C

Correlation ranking filter

D

Common features

E

Complete dataset

Selected features Vehicle-manoeuvre, accident-severity, car-passenger, bus-or-coach-passenger, pedestrian-road-maintenance-worker Casualty-type, casualty-class, pedestrian-location, pedestrian-movement, sex-of-casualty, bus-or-coach-passenger, pedestrian-road-maintenance-worker, car-passenger, age-of-casualty, casualty-home-area-type, casualty-imd-decile, accident-severity Pedestrian-road-maintenance-worker, bus-or-coach-passenger, pedestrian-location, pedestrian-movement, car-passenger, sex-of-casualty, casualty-class, casualty-home-area-type, casualty-type, casualty-imd-decile, age-of-casualty, accident-severity, No-of-Vehiclesinvolved-unique-to-accident-index Pedestrian-road-maintenance-worker, accident-severity, car-passenger, bus-or-coach-passenger Full features in Table 1

Learning Models We used the following machine learning classifiers to train the data: – – – –

ZeroR OneR Decision Tree Algorithm: J48 Naïve Bayes

The simplest of all classifiers is ZeroR. It does not have rule as the name suggests. It only helps to cluster and more clearly visualize the data, and to get an overall image of the main class. While there is no power of predictability in ZeroR, it is useful as a benchmark for other classification methods to evaluate a baseline score. OneR, meaning “one rule,” is a simple but concise characterization calculation which produces in the details one standard for each indicator, at which point the standard with the smallest absolute error is chosen as its “one guideline.” To render an indicator standard, we build a recurrence table for each indicator against the goal. It has appeared that OneR produces leads just a little less accurate than the best in class grouping calculations, thus making decisions that are easy for people to understand.

3 Prediction of Road Accidents Using Data Mining Techniques

31

Grouping is the way to build a class model from a lot of records which contain class marks. Preference Tree Calculation is to discover how the vector features carry on for various occasions. In addition, the classes for the recently produced examples are found on the basis of the training. This estimate provides expectations for goal variable expectation. The simple transmission of the information is essentially rational with the aid of tree order calculation. J48 reflects an ID3 extension. J48’s extra highlights reflect missing values, pruning trees of preference, unceasing price worth extents, standards induction, and so forth. The J48 is an open-source Java execution of the C4.5 equation in the WEKA knowledge mining apparatus. The WEKA appliance furnishes numerous tree pruning-related alternatives. In the case of uncertainty for fitting pruning, this can be used as a precision tool. The grouping is done recursively in various formulas until each and every leaf is unadulterated, i.e., the grouping of the information should be as immaculate as required under the circumstances. It produces the guidelines from this calculation from which basic personality of that knowledge is generated. The goal is to logically speculate a tree of choice until it acquires harmony of adaptability and accuracy. Naïve Bayes is a binary (two-class) classification algorithm and multi-class classification problem. When represented using binary or categorical input values, the technique is easiest to understand. It’s called Naive Bayes or fool Bayes as it simplifies the estimation of the probability for each hypothesis to make their estimation tractable. Rather than trying to measure the values of each type of attributes P(c|x) =

P(x|c)P(c) P(x)

(1)

they are considered to be conditionally independent, and measured as and so on. This is a very strong assumption that, in real data, is most impossible, i.e., the attributes are not interacting. The method nonetheless performs remarkably well on data where this presumption does not hold. This approach is pretty much perfect for machine learning models.

4 Experimental Results and Analysis The experiments were performed on platform WEKA 3.8.3. All models were tested using ten-fold Cross Validation [5] and all tests were performed on a computer with 8 GB RAM and 4-core processing power of 3.40 GHz. Evaluation metrics used to measure model efficiency include predictive precision, recall, consistency, and a harmonic mean (F1). Accuracy is a measure of how effective the model is in predicting performance. Accuracy =

(T P + T N ) (T P + F P + F N + T N )

(2)

32

F. A. Shakil et al.

Table 2 The prediction accuracy matrix Dataset Algorithm Accuracy A

B

C

D

E

ZeroR OneR Decision Tree (J48) Naive Bayes ZeroR OneR Decision Tree (J48) Naive Bayes ZeroR OneR Decision Tree (J48) Naive Bayes ZeroR OneR Decision Tree (J48) Naive Bayes ZeroR OneR Decision Tree (J48) Naive Bayes

Precision

Recall

F1

0.3333 0.9591 0.9613

NA 0.969 0.966

0.333 0.959 0.961

NA 0.962 0.963

0.9592 0.3333 0.9591 0.9613

0.967 NA 0.969 0.966

0.959 0.333 0.959 0.961

0.961 NA 0.962 0.963

0.9582 0.3333 0.9591 0.9628

0.959 NA 0.969 0.966

0.958 0.333 0.959 0.963

0.958 NA 0.962 0.964

0.9587 0.3333 0.9591 0.9608

0.959 NA 0.969 0.967

0.959 0.333 0.959 0.961

0.959 NA 0.962 0.963

0.9591 0.3333 0.9591 0.9674

0.969 NA 0.969 0.968

0.959 0.333 0.959 0.967

0.962 NA 0.962 0.967

0.9531

0.954

0.953

0.954

Precision is a measure for positive prediction. Pr ecision =

TP (T P + F P)

(3)

Recall is a measure of correctly expected positive responses to all positive class observations. Recall =

TP (T P + F N )

(4)

F1 Score is the weighted average of Precision and Recall. F1Scor e =

2 × (Recall × Pr ecision) (Recall + Pr ecision)

(5)

3 Prediction of Road Accidents Using Data Mining Techniques Table 3 The confusion matrix Slight Actual class

159466 2721 40

33

Serious

Fatal

2443 19187 100

300 226 1589

From Table 2, the results obtained after applying the machine learning models on different features sets can be seen. It can clearly be seen that prediction accuracy on distinctive set of features (A-E) are almost same and varies very little. By taking of the average of prediction accuracy for features set: A-E; the highest prediction accuracy is found for features set A. Also, Decision Tree (J48) has the best accuracy for all the distinct features set, though prediction accuracy by oneR and Naive Bayes are very close too. On Table 3, the confusion matrix for decision tree (J48) on test set for features set E is provided. As the purpose of this study is to figure out the dominant features for class attribute “casualty-severity”, Table 2 was generated to compare the performances of different features sets for the applied machine learning algorithms. As, it we have seen that features set D performs the best, the features selected in set A are the influential features for predicting “casualty-severity”. Therefore, the influential features are vehicle-manoeuvre, accident-severity, car-passenger, bus-or-coachpassenger, pedestrian-road-maintenance-worker. Figure 2 illustrates the receiver-operating-characteristic (ROC) curve and precision-recall (PR) curve for class value- Slight, Serious and Fatal; for all the applied models zeroR, oneR, Naive Bayes, and decision tree (J48) on feature set ‘E’ are provided. By analyzing the curves, we can determine that the models performed well on the test data. From Fig. 3, we can see the performance of the models on both train and test data for feature set ‘E’ and it can clearly be seen that other than zeroR model all the other models performed well on both the train and test data.

5 Conclusion In this study, we wanted to build machine learning models which will be able to predict the severity of road accidents. Here, we did so based on available dataset and in this paper we are only presenting our findings and results. We only hope our study will be helpful for further research related to this field and also be applicable in reallife situations to prevent and take measurements to reduce of stop road accidents. This is one of the leading cause of death all over the world. This factor alone at the first place made us interested working in this field. But due to lack of dataset and also many missing data of course our study has its shortcomings. But we only hope this will be of any help to save all those life those are lost in countless road accidents every moment.

34

F. A. Shakil et al.

Fig. 2 ROC curves and PR curves

Fig. 3 Comparison of model performances on train and test dataset

References 1. Abdel-Aty MA, Abdelwahab HT (2004) Predicting injury severity levels in traffic crashes: a modeling comparison. J Trans Eng 130(2), 204–210 2. Babbar A (2017) Road accidents incidence; road accidents data great britain, 1979–2015. https://www.kaggle.com/akshay4/road-accidents-incidence 3. Chang LY, Wang HW (2006) Analysis of traffic injury severity: an application of non-parametric classification tree techniques. Accid Anal Prev 38(5), 1019–1027

3 Prediction of Road Accidents Using Data Mining Techniques

35

4. Chong M, Abraham A, Paprzycki M (2005) Traffic accident analysis using machine learning paradigms. Informatica 29(1) 5. Hall M, Frank E, Holmes G, Pfahringer B, Reutemann P, Witten IH (2009) The WEKA data mining software: an update. SIGKDD Explor 11(1):10–18 6. Kunt MM, Aghayan I, Noii N (2011) Prediction for traffic accident severity: comparing the artificial neural network, genetic algorithm, combined genetic algorithm and pattern search methods. Transport 26(4):353–366 7. Moghaddam FR, Afandizadeh S, Ziyadi M (2011) Prediction of accident severity using artificial neural networks. Int J Civil Eng 9(1):41 8. de Oña J, Mujalli RO, Calvo FJ (2011) Analysis of traffic accident injury severity on spanish rural highways using bayesian networks. Accid Anal Prev 43(1):402–411 9. Organization WH (2015) Global status report on road safety 2015. World Health Organization 10. Quadir F, Ameen MFA, Momen S (2014) Visualization and queuing analysis of spatio-temporal traffic data. In: 2014 17th international conference on computer and information technology (ICCIT), pp 223–228 11. Causes of road accidents, http://jhtransport.gov.in/causes-of-road-accidents.html. Accessed December 2020

Chapter 4

Eye Gaze-Based Student Readability Analysis Niranjan Patil, Rahulkumar Das, Komal Dhusia, Varsha Sanap, and Vivek Kumar Singh

1 Introduction Student performance evaluation is a crucial part of the educational system. The foremost goal is to boost the abilities of the students. Also, it is necessary to seek the interests of students in activities and the focus with which they pursue the tasks assigned to them. This can be made attainable with the use of emerging IT technologies. This work focuses on the utilization of technologies such as Data Mining, Machine Learning, and Image Recognition. These technologies will facilitate in analyzing the readability and understanding of students and provide feedback on their activities. The clustering algorithms are used to map the iris data collected and represent them on the screen. The data coordinates obtained can be imported into a CSV file and then fed to the data mining algorithms. Python supports eye tracking through multiple libraries like OpenCV, PyGaze, Eye, PIL, etc. The obtained coordinates can be plotted to highlight the most and least viewed areas. Eye gaze coordinates are collected to track the eye gaze moments. The camera will track the coordinates of the iris and it will give X-coordinate and Y-coordinate. These coordinates will be used to automate the mouse movements in the information retrieval module. There are many different approaches described in the literature that specialize in eye tracking. The methods described will be used as a way to develop an eye-tracking system that has the best accuracy, best performance, and is a low-cost system. There are many approaches to develop an eye-tracking system that achieves the highest accuracy, best performance, and lowest cost.

N. Patil · R. Das · K. Dhusia · V. Sanap (B) · V. K. Singh Ramrao Adik Institute of Technology, Mumbai, India e-mail: [email protected] V. K. Singh e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 R. C. Bansal et al. (eds.), Proceedings of International Conference on Computational Intelligence and Emerging Power System, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-16-4103-9_4

37

38

N. Patil et al.

The next section sheds light upon the existing systems studied to determine the scope of the problem and determine a suitable approach for the given problem statement.

2 Literature Survey Many researchers studied the eye gaze technique by developing their methods. In few papers, Wang and Sung [1], Fiequeira et al. [2] calculated the gaze point using the line of sight and the orientation of eyes. The designed system provided robust and quick solutions due to the nature and simplicity of the features used in the algorithms. The technique used by Wang and Sung [1] is one-circle and two-circle algorithm with an error of 0.3% and 10%, respectively. Figueira et al. [2] developed various synthetic eye images using libraries. The images generated were used to detect the position of the iris and, in turn, helping with iris detection. The research work done by Asteriadis et al. [3], Meier et al. [4], Yang et al. [5] has led to several analytical studies in understanding student behavior and motivation and engagement to learn. They have utilized the user feedback from the e-learning platforms which presents the data from the behavioral state of the user in the context of reading an electronic document. They have used various data like facial movements, hand movements, the performance of students in tests based on which they have developed their algorithms that are used to suggest the user more personalized courses that the user may be interested to enroll. Asteriadis et al. [3] calculated the level of interest with the help of a neuro-fuzzy network. Similarly, Meier et al. [4] developed a data mining approach to predict the personalized course suggestions to users and grade prediction. Yang et al. [5] developed two algorithms named FTSNN in which the learning from the quiz was evaluated and the IFTSNN behavior factor was also taken into consideration for the evaluation. To extract the image of the eye, Nguyen [6] made use of the Gaussian process. Nguyen [6] has used open CVs to extract the images of the eye. It made use of a new training model to detect and track the eye movements, and then after extracting the cropped image of the eye, it is deployed to train the Gaussian functions for gaze estimation. The major drawback of this system is that the user has to keep the head in a stable position in front of the camera even after the training process is completed. The method has low adaptability and accuracy and has a common fault which does not consider the head posture. Some of the researches have highlighted eye-tracking problems and to solve those issues, Cheung et al. [7], Wang et al. [8] have proposed new methods in their paper using which eye tracking can be done in a convenient way. Cheung et al. [7] used a web camera to track the face and then extract the eye region and then the iris center and eye corner are located combining intensity energy and edge strength to form an eye vector. To reduce the error, sinusoidal head method (SHM) is used to simulate the 3D head shape and propose it to the POSIT algorithm. On the other hand, Cheung et al. [7] used eye tribe eye tracker glasses for tracking the iris movement and used

4 Eye Gaze-Based Student Readability Analysis

39

a dedicated software Unity v5.1.1 for mapping the coordinates on the screen. Wang et al. [8] discovered that the existing eye trackers require an explicit calibration to get the eye parameters. To simplify the calibration process, it was performed without any active participants or even the knowledge of the participants. A regression-based deep convolutional neural network was used to learn image feature to predict the eye gaze. The results which are collected from the experiments of Cheung et al. [7] and Wang et al. [8] show the reduced time of computation and an easy approach for eye tracking. Koles and Hercegfi [9] have used an electromagnetic head tracker and a headmounted eye tracker to record the readings in a virtual CAVE environment. The participants were made to sit in a physical space measuring length and width. Participants were seated behind 2.5 m from the front wall. Before and after the task was performed, they were asked to look at four points placed in front of them one by one, and heat maps were generated. The analysis of the eye-tracking data was done in Tobii Studio and to calculate the accuracy of the heat map, each heat map was loaded into the software to calculate the distance between the target and the largest fixation point. The results depicted that some participants had perfect hit before and after the experiment, while some others had to add offset. If compared between infra ray light gaze tracking systems and visible light tracking systems as suggested by Kao et al. [10] it is observed that visible light tracking system provides new applications for consumers. The visible light tracking system suffers from difficulties of various illumination conditions and which leads to a problem of low accuracy in detecting the location of the pupil. By improving the ellipse matching algorithm, the average angular error is 65%

Fan

MQ135

CO2

0 Spike at lag 1, then 0; + ve spike if ϕ1 > 0 and alternating in sign, starting on –ve and − ve spike if ϕ1 < 0 side, if ϕ1 < 0

AR ( p)

Exponential decay or damped sine wave. The exact pattern depends on the signs and sizes of ϕ1 ,..., ϕ p

Spikes at lags 1 to p, then zero

MA (1)

Spike at lag 1, then 0; + ve spike if ψ1 < 0 and − ve spike if ψ1 > 0

Exponential decay: on + ve side if ψ1 < 0 and alternating in sign, starting on + ve side, if ϕ1 < 0

MA (q)

Spikes at lags 1 to q, then zero

Exponential decay or damped sine wave. The exact pattern depends on the signs and sizes of ψ1 ,..., ψq

19 Solar Power Predictions in Stochastics Framework

231

3 Solar Power Uncertainty Modeling and Results For uncertainty modeling using ARIMA model, first, the order of suitable ARIMA model on the basis of collected historical data is determined. Order of AR and MA terms are obtained by the observation of ACF and PACF plot. These plots are shown in Figs. 1 and 2, respectively. From these figures, it is observed that ACF plot is sinusoidal and PACF for first two legs are out of bounds. Thus, ARIMA (3,0,0) is suitable for modeling solar power uncertainty. The estimated values of AR coefficients are 1.18, -0.309, and -0.032. The values of constant term and variance of white noise are 97.86 and 0.43, respectively. 1 0.8 0.6

ACF

0.4 0.2 0 -0.2 -0.4 -0.6 -0.8

0

10

20

30

40

50

60

70

80

90

100

60

70

80

90

100

Lag

Fig. 1 ACF plot of sample solar power data

1 0.8 0.6

PACF

0.4 0.2 0 -0.2 -0.4 -0.6 -0.8

0

10

20

30

40

50

Lag

Fig. 2 PACF plot of sample solar power data

232

N. Gupta et al.

Power Output (kW)

200

150

100

50

0

1

2

3

4

5

6

7

8

9

10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

Time (h)

Fig. 3 Generated 1000 solar power scenarios (green) along with mean scenario (red) and forecasted solar power (black)

To solve MG generation scheduling problem using proposed stochastic optimization based methodology, 1000 solar power scenarios are generated. Generated scenarios are reduced to 25 scenarios using backward reduction algorithm. Generated 1000 solar power scenarios along with mean scenario (red) and forecasted solar power (black) is shown in the Figure. Figure 4 shows the reduced 25 solar power scenarios along with mean scenario (red) and forecasted solar power (black). When spread of uncertain parameter is not very high, a mean solar power scenario can be used to solve MG generation scheduling problem. Practically, spread is there in solar power uncertainty, thus reduced solar power scenarios can be used for solving MG generation scheduling problem in stochastic optimization framework (Figs. 3 and 4).

4 Conclusion This paper discusses the solar power predictions in stochastic framework. The scenarios of solar power are generated through ARIMA model. Further, basics of ARIMA model with its features have been discussed. Here, firstly, solar power scenarios are generated and reduced to a significant number (25) using backward and reduction method.

19 Solar Power Predictions in Stochastics Framework

233

Power Output (kW)

200

150

100

50

0

1

2

3

4

5

6

7

8

9

10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

Time (h)

Fig. 4 Reduced 25 solar power scenarios (green) along with mean scenario (red) and forecasted solar power (black)

Acknowledgements The author acknowledges the AICTE-NPIU for giving financial aid under CRS Project 1-5748344986.

References 1. Jacobson MZ, Delucchi MA (2011) Providing all global energy with wind, water, and solar power, Part I: Technologies, energy resources, quantities and areas of infrastructure, and materials. Energy Policy 39(3):1154–1169 2. Wang MQ, Gooi HB (2011) Spinning reserve estimation in microgrids. IEEE Trans Power Syst 26(3):1164–1174 3. Ahn S-J et al (2013) Power scheduling of distributed generators for economic and stable operation of a microgrid. IEEE Trans. Power Syst 4(1):398–405. 4. Liao G-C (2012) Solve environmental economic dispatch of Smart MicroGrid containing distributed generation system–using chaotic quantum genetic algorithm. Int J Elect Power Energy Syst 43(1):779–787 5. Gupta RA, Gupta NK (2015) A robust optimization based approach for microgrid operation in deregulated environment. Energy Convers Manag 93(2015):121–131 6. Gupta RA, Gupta NK (2014) Robust microgrid operation considering renewable power uncertainties. In: 2014 eighteenth national power systems conference (NPSC). IEEE 7. Gupta NK (2014) Generation scheduling at PCC in grid connected microgrid. In: Recent advances and innovations in engineering (ICRAIE), 2014. IEEE 8. Saber AY, Venayagamoorthy GK (2012) Resource scheduling under uncertainty in a smart grid with renewables and plug-in vehicles. Syst J IEEE 6(1):103–109 9. Doherty R, O’Malley M (2005) A new approach to quantify reserve demand in systems with significant installed wind capacity. IEEE Trans Power Syst 20(2):587–595

Chapter 20

Economic Analysis of Rural Distribution System with DER and Energy Storage System Prashant Kumar and Vinod Kumar

1 Introduction Variation and reliability of electricity supply, integration of distributed energy resources, and elimination of transmission losses and greenhouse gas emissions are supported by electrical distribution systems. Electrical distribution systems can generate, transmit, distribute, store electricity, and fulfill electricity consumption. The Indian electrical distribution systems operational strategies are calculated by minimizing the overall cost of investment and operation and maintenance costs. There are portable, distributed power generators, batteries, and heat storage facilities in the Indian electrical systems. To assess the investment and optimum operating strategies of electrical distribution networks, economic and optimization studies have been carried out. Furthermore, there is an economic analysis of the electrical distribution systems. To assess optimum expenditures in power generators and battery devices, economic factors are measured, including the payback time, current value, and net profits of rural electricity distribution networks. The built environment is the biggest area that consumes energy and consumes about 33% of the total demand of the world. Consequently, laws and legislation embrace favorable benefits by stimulating dispersed energy infrastructure to reduce power consumption and CO2 pollution. Photovoltaic electricity production is now the greatest source of energy generation and can achieve 22% of the whole generation of world up to 2050. New trends show that when grid parity and policy targets are achieved, a growing total of countries may quickly limit their feed-in tariff compensation. Alternative methods for ensuring a successful operation from distributed energy resources are now being applied. However, due to the regular and seasonal variations in solar radiation, the availability of solar energy is unpredictable. During P. Kumar (B) · V. Kumar Department of Electrical Engineering, College of Engineering and Technology, MPUAT, Udaipur, Rajasthan, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 R. C. Bansal et al. (eds.), Proceedings of International Conference on Computational Intelligence and Emerging Power System, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-16-4103-9_20

235

236

P. Kumar and V. Kumar

the evening and morning hours, residential demand profiles show peaks, while PV power peaks are generated generally around midday. As the quantity of unreliable electricity sources is raised, energy storage devices may be helpful as they boost the dispatch ability of clean energy technology. It is possible to implement different kinds of energy storage devices like batteries for power storing and water storage tanks as heat storage. Today, the most popular type of household storage is home energy storage (HES), in which the energy is stored inside the house. However, a new study shows that additional advantages may be provided through Community Energy Storage (CES), like other control uses (e.g. maximum shaving) and cost-effectivity. Consequently, CES draws concentration and has recently been adopted in the built environment as an appraisal scale. The idea of a rural electrical distribution system where both the sources of localized storing facilities and distributed energy are managed with Energy Management System (EMS) that gives multiple uses, one of the technologies relevant to HES and CES. Furthermore, the incorporation of moving loads like battery-operated vehicle will give energy systems mobility. Due to more aggressive climate change initiatives, the idea of all-electric distributed generation has recently been launched to supplement fossil-fuel-based electricity systems. The costs and environmental effects of such development, however, are quietly great (e.g. batteries). Subsequently, it is vital to assess additional energy networks on economic viability and environmental efficiency. To determine the rates of renewable energy networks for various storage sizes, multiple studies and optimization models are suggested (i.e. HES and CES). In addition, a CES method was suggested to control the distribution of available energy in the storage facility, in which residential allocate the CES system with allowing utilization of the EMS. In addition, an algorithm was proposed to evaluate the specific advantages of CES using the benefits of dynamic energy markets. In addition, HES and CES were compared in terms of households. An optimization model was developed by the authors to evaluate the technical and cost-effective viability of HES along with CES by various battery sizing technologies. Their work focuses on the conservation of energy-storing systems with distributed energy resources and neglects the efficiency of HES and CES. In addition, the techno-economic advantages of the energy storage system and distributed energy resources were analyzed as a result of the neighborhood’s scale to conduct demand load shifting along distributed energy resources time shifting [1]. Furthermore, an optimization technique was suggested and an economic study of the DER and energy storage system was presented when participating in various energy markets.

2 Energy Management System Energy Management system manages the electricity regulation and requirement of the distribution network. Moreover, for the management and planning of network strategies and components, various information control rounds between techniques

20 Economic Analysis of Rural Distribution System …

237

are shared. By maximizing the usage of self-generated electricity, the Energy Management Device minimizes the volume of electricity use by the consumer [2]. This is accomplished using the excess electricity by the immediate use of electricity from distributed energy resources, by the discharging of the batteries. An optimization algorithm does this.

3 System Description Figure 1 presents a distribution system integrated along with a distributed energy system and an energy storage system. Investment by distribution system load aggregator in the general electricity markets to buy electrical energy to service the customers of the distribution system. For now, it is often believed that the load aggregator manages the distributed energy resources and energy storage system built into the distribution system served. Distributed generation of electricity may be regulated by reducing the usable output of distributed energy resources [3]. Electricity price is believed to be dictated and the load is decided by the consumers, which is inflexible to the price. The purpose of the load aggregator is to provide a stable power supply to the customers while minimizing the cost of electricity in the power markets. The aims of this study are to suggest new methods of activity to increase dependability and wealth and to provide a systematic structure for measuring reliability along the economy [4, 5].

Fig. 1 Flowchart

238

P. Kumar and V. Kumar

Fig. 2 Presentation of a rural distribution system linked with distributed energy resources and energy storage system

At present, a large number of distribution networks are functioning on the radial system, this research aims to explore a radial distribution network along with advanced distributed energy sources and energy storage networks. A radial distribution grid of interconnected distributed energy sources and energy storage is presented in Fig. 2 [6].

3.1 Objective Reducing the cost ofenergy buying from the period I to i + n. Min u(i). p(i) + i+n k=i+1 u f (K ). p f (K ) Where u f (K )= Energy purchased in future period K. p f (K )= cost of energy for future period K. The first  part u(i). p(i) is the energy purchasing cost of the current period i. The second part i+n k=i+1 u k (k). pk (k) is the predicted total energy purchasing cost of the following periods from i + 1 to i + n.u(i) and uf (k) are the decision variables to be solved [7].

3.2 Constraints 3.2.1

Energy Storage System Operation Constraints

0 ≤ c(K) ≤ cmax . 0 ≤ d(K) ≤ dmax .

20 Economic Analysis of Rural Distribution System …

239

SoCmin ≤ SoC(K) ≤ SoCmax . Where C(K) = Charged storage device for period K. cmax = full charging capacity of storing device. D(K) = Discharged storage device for period K. dmax = full discharging capacity of storing device. SoCmin = least condition of charge level of energy storing device. SoC(K) = condition of charge stage at the last of period K. SoCmax = Maximum stage of charge capacity of storing device. K = I, I + 1,…….I + n. The operation of charging and discharging is to be resolved by keeping the maximum charging and discharging rates constant. We have taken the results for 1 h duration in this research, and the charging energy is equal to c(K) multiplied by 1 h. For ease, c(K) is taken interchangeably as charging rate and energy charged in 1 h. d(K) is employed in a similar means.

3.2.2

Available Distributed Energy Resources Constraints

0 ≤ r(K) ≤ rmax(K). 0 ≤ rf (K) ≤ rf.max (K). Where r(K) = consumed distributed resources in period K. rmax (K) = Available distributed resources in period K. rf (K) = Forecasted distributed resources utilization with period K. rf.max (K) = Forecasted accessible distributed resources with period K. K = I + 1,…….I + n. The utilized renewable energy is equal to or less than the available renewable energy. Extra energy not utilized is dumped in ways such as adjusting the wind turbines’ blade pitch, so wind turbines do not generate the maximum power they can in that period. Utilized renewable energy for the current period and future period are to be solved [8].

3.2.3

Power Balance Constraints

u(I) + r(I) = l(I) + c(I)-ηdd(I). uf (K) + rf (K) = lf (K) + c(K)-ηd d(K). Here, K = I + 1,…….I + n. Although loads, existing renewable sources, and rates are dictated by predicting methods in future periods, constraints are given for the optimization issue. The result of this optimization issue provides an optimum working plan for the process of electrical storage system charging/discharging, purchasing electricity, and use of renewable energy from time I to I + n.

240

P. Kumar and V. Kumar

4 Economic Assessment Framework The proposed method id for the economic analysis in rural distribution systems with distributed energy resources and energy storage system framework. During operation, the energy storage device often functions as a generation that provides the power to the load, and for a short time it acts like a load during the charging period. Current energy storage system stage of charge capacity at a state in time is determined by the preceding processes [9]. Over the current cycle, the consumption of electricity from the energy storage device is determined by both its existing amount of charge status and its expected potential use. Because of these unusual sequential features of the energy storing device, its effect on the economy of the system is better, in which its basic operational techniques are combined.

4.1 Distribution System Economy Analysis The economic indexes are used for annual utility purchasing rates and consumer outage costs. Based on the actual operation, hourly electricity purchasing expenses are determined. Then, the amount of hourly energy costs per year is the annual expense of purchasing energy. In this article, during operation, we concentrate on the cost. The development costs of the distributed energy resources and the energy storage system are, therefore, not contained in the economic index but should be contained if needed [10]. The investments should be addressed if the optimum capacity of the distributed energy resources and the size of the energy storage facility are to be overcome (Table 1). The cost of customer interruption is the cost of harm to consumers incurred by the interruption of the electricity supply. The routine operations of consumers in the distribution system could be disturbed and bear such interruption costs while a power failure happens. Customers are divided into seven industries due to the scope of their activities: large-scale consumer, manufacturing, commercial, farming, domestic, government and administrative, office, and building. To measure the expense of consumer disturbance, postal surveys have been performed [11, 12]. Table 1 Energy storage system parameters Capacity (MWh)

Power limit (MW)

Capacity (MWh)

Power limit (MW)

Capacity (MWh)

Power limit (MW)

5

1

10

1

15

1

5

2

10

2

15

2

5

2.5

10

2.5

15

2.5

5

5

10

5

15

5

241

Price

20 Economic Analysis of Rural Distribution System …

Time/h Fig. 3 Electricity price curve

5 Case Studies An efficient rural distribution system integrated with distributed energy resources and electrical storage system, as seen in Fig. 3 is being studied. The external grid and the rural distribution system are connected by a step-down transformer. The network topology and the capacity to manage the required power injection may evaluate the integration node. It is assumed that this node will handle the injection of power. It was also possible to select other relevant nodes. If the converter has a malfunction or the external grid fails to provide adequate electricity owing to an interruption, it is not feasible to provide electricity to the rural distribution system [13, 14].

6 Result The findings show the economic progress brought on by the incorporation of distributed energy resources and electricity storage systems and the planned operational strategies. They also offer information on how the capability of the electricity storage system, the power limit, and the ability of distributed energy resources impact the economy. These findings may also be useful in evaluating the appropriate capability of the electrical storage system, the power limit, and the distributed capacity of energy resources to achieve the optimal degree of efficiency and economic benefit (Table 2). Results show that improving the size of the electric storage system, power limit, or capacity of distributed energy resources will both increase stability, conserve energy costs, and reduce the cost of consumer disruption. The influence that each aspect has on the economy, though depends on the situation.

242

P. Kumar and V. Kumar

Table 2 Economy indices of system with an electrical storage system and distributed energy resources Electrical energy storage Power (MW)

Distributed energy resources capacity (MW)

Energy cost

Energy capacity (MWh) 5

1

1

5

1

4

5

2.5

5

2.5

10

Customer interruption High commercial

High residential

3.112

1.752

0.614

2.437

1.343

0.469

1

3.058

1.716

0.603

4

2.377

1.319

0.463

1

1

3.083

1.672

0.585

10

1

4

2.403

1.265

0.441

10

5

1

2.916

1.566

0.550

10

5

4

2.246

1.193

0.417

Energy Purchasing Cost (Millions/Year)

Figure 4 shows the cost of purchasing energy with the capability of 5 and 15MWh energy storage devices. The electricity expense of energy storage devices with a capacity of 15 MWh and a power limit of 1 MW is larger than that of energy storage devices with a capacity of 5 MWh and a power limit of 2 MW. This phenomenon indicates the value of the proper matching of the capability and power limit of energy storage systems to obtain the required economic benefits. Customer disruption costs for a large commercial mixing system with a capacity of 15 MWh for energy storage systems as seen in Fig. 5. By growing the capacity cap

Distributed Energy Resources Capacity (MW) Fig. 4 Energy purchasing cost when of energy storage systems capacity at 5 and 15 MWh

243

Electricity Purchasing Cost (Millions/Year)

20 Economic Analysis of Rural Distribution System …

Distributed Energy Resources Capacity (MW) Fig. 5 System customer interruption cost for high commercial mix system when fixing of energy storage systems capacity at 15 MWh

of energy storage systems from 1to 2 MW, there is a reduction in disruption costs. The decrease, however, is rather restricted as the power limit rises above 2 MW. This outcome indicates the impact of nonlinear and saturation by using energy storage systems and distributed energy resources to increase the economy of the system.

7 Conclusion This paper estimates the economic study of the rural distribution system integration with energy storage devices and distributed energy resources. The proposed utility grid operation strategy related to the development of energy storage devices and distributed energy resources strategy for the operation of electricity supply is being introduced. The proposed economic evaluation process is used to evaluate the effect on the economy of proposed operating policies, energy storage technologies, and the incorporation of distributed energy resources. The findings of case studies illustrate the advantages of the proposed operation strategies and also provide insights into how the ability of energy storage devices, the power limit, and the distributed energy resources impact the economy.

244

P. Kumar and V. Kumar

References 1. Chen Y-H, Lu S-Y, Chang Y-R, Lee T-T, Hu M-C (2013) Economic analysis and optimal energy management models for microgrid systems: a case study in Taiwan. Appl Energy 103:145–154. https://doi.org/10.1016/j.apenergy.2012.09.023 2. Kumar P, Mathew L, Shimi SL, Singh P (2016) Need of ICT for sustainable development of power sector proceedings of international conference on ICT for sustainable development, 607–614. https://doi.org/10.1007/978-981-10-0129-1_63 3. Kumar P, Kumar V (2020) Energy storage options for enhancing the reliability of Power system in the presence of Renewable Energy Sources.In: 2020 second international conference on inventive research in computing applications (ICIRCA). https://doi.org/10.1109/icirca48905. 2020.9183349 4. Sharma H, Mishra S (2019) Techno-economic analysis of solar grid-based virtual power plant in Indian power sector: a case study. Int Trans Electric Energy Syst. https://doi.org/10.1002/ 2050-7038.12177 5. Bellekom S, Arentsen M, Van Gorkum K (2016) Prosumption and the distribution and supply of electricity. Energy Sustain Soc 6(1):1–17. https://doi.org/10.1186/s13705-016-0087-7 6. Calvillo CF, Villar J, Martín F (2016) Optimal planning and operation of aggregated distributed energy resources with market participation. Appl Energy 182:340–357. https://doi.org/10.1016/ j.apenergy.2016.08.117 7. Olivella-rosell P, Bullich-massagué E, Aragüés-peñalba M, Sumper A (2017) Optimization problem for meeting distribution system operator requests in local flexibility markets with distributed energy resources. Appl Energy 210:881–895. https://doi.org/10.1016/j.apenergy. 2017.08.136 8. Ahmad S, Naeem M, Ahmad A (2019) Low complexity approach for energy management in residential buildings. Int Trans Electr Energy Syst 29(1):1–19. https://doi.org/10.1002/etep. 268031 9. Hassan AS, Cipcigan L, Jenkins N (2017) Impact of optimised distributed energy resources on local grid constraints. Energy 142:878–895. https://doi.org/10.1016/j.energy.2017.10.07432 10. Ahmad J, Imran M, Khalid A et al (2018) Techno economic analysis of a wind-photovoltaicbiomass hybrid renewable energy system for rural electrification: a case study of Kallar Kahar. Energy 148:208–234. https://doi.org/10.1016/j.energy.2018.01.133 11. Talwariya A, Pushpendra S, Kolhe M (2019) A stepwise power tariff model with game theory based on Monte-Carlo simulation and its applications for household, agricultural, commercial and industrial consumers. Int J Electrical Power and Energy Syst 111:14–24 12. Gill A, Yadav SK, Singh P (2019) Biogeography based optimization technique for optimal siting and sizing of distributed generation system in a distribution system. Int J Eng Appl Manag Sci Paradig (IJEAM) 54(2). ISSN: 2320-6608. (May 2019) 13. Gill A, Yadav SK, Singh P (2019) Optimal siting and sizing of distributed generation system in radial distribution network using particle swarm optimization technique. J Emerg Technol Innov Res (JETIR) 6(5). ISSN: 2349-5162. (May 2019) 14. Gill A, Yadav SK, Singh P (2019) An adaptive scheme for Optimal siting of Distributed Generation system in a distribution network. Int J Recent Technol Eng (IJRTE) 8(1). ISSN: 2277-3878. (May 2019)

Chapter 21

An Intelligent Technique to Mitigate the Transient Effect on Circuit Breaker Due to the Occurrence of Various Types of Faults Shripati Vyas, R. R. Joshi, and Vinod Kumar

1 Introduction Usually, a power system operates in a balanced condition but in any abnormality, the system can get unbalanced. If there is any failure in its insulation or two or more phases working at different voltages come into contact then there is the possibility of any fault. There are several reasons for fault occurrence like high winds, snow, surges, etc. There may be some other causes like dropping of trees on line, accidents with supporting structures and lower insulation may cause faults, etc. There is a greater possibility of short circuits due to defective insulation caused by aging or overloading of conductors. Hence, during power system analysis the fault analysis is of crucial importance. There is a need for protective relays for protection from greater short circuit currents while disconnecting the faulty part. At the time of selection of circuit breakers and relays, we should approximate the degree of currents that will flow under faulty conditions that are the part of the fault study. The selection of phase relays is done by 3L-G fault data and the ground relays are selected on the basis of L-G faults. There is very less chance of 3L-G faults in the system, most of the occurred faults are of 2L-G and L-G. In our investigation, we have included the asymmetrical faults because their effect is most severe in system, while the symmetrical faults are very uncommon in nature. The power system fault can be divided into mainly two parts that are symmetrical and asymmetrical. On the occasion of symmetrical faults, the fault current is equal for all three phases and the network is in a balanced condition while in the unsymmetrical conditions the fault current magnitudes are dissimilar for every phase. The various types of faults are

S. Vyas · R. R. Joshi (B) · V. Kumar Department of Electrical Engineering, College of Technology and Engineering / MPUAT / Research Scholler, Udaipur, Rajasthan 313001, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 R. C. Bansal et al. (eds.), Proceedings of International Conference on Computational Intelligence and Emerging Power System, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-16-4103-9_21

245

246

1. 2. 3.

S. Vyas et al.

L-G fault (Single-line-ground fault), Double-line-ground fault, Three-line-ground fault.

These faults are also called short circuit faults which are common on transmission lines. Such faults take place due to different factors like insulation failure of equipment caused by lightning and switching surges are coming in contact and also when a foreign object came in contact with bare power lines [6]. The foreign objects may be falling off trees on line or birds shorting outlines. In this paper, we are using a MATLAB Simulink. MATLAB stands for Matrix Laboratory. MATLAB is a highperformance language for technical computing. It has many advantages compare to convectional computer language for solving a technical problem, this software is available since from 1984 and is considered as a standard tool. MATLAB can be used for graphical and programmatically analysis, MATLAB has many toolboxes and we are using Simulink for graphical modelling.

2 Modelling and Simulation In this model, we are using three-phase sources of 13.8 kV, 50 Hz is connected to the series RLC load with an impedance of the load in ohms, and it is connected through transmission lines. Three transmission lines are used of 50 km long, while one of the transmission lines is parallel to the two series transmission lines: here two-way supply is given to the RLC series load through transmission lines since our main aim is to create a fault in one of the transmission lines during that time the supply is not interrupted to the load. To isolate that unhealthy part, we are using a circuit breaker for switching operation that is, open and close the circuit breaker, the circuit breaker operation is opposite to that ideal switch, that is, if an ideal switch is an open circuit breaker that remains in the closed condition and vice versa. The operation is similar to double-line-ground fault, three-line-ground fault but while double-lineground fault two lines of the transmission line is short-circuited and three lines of the transmission line is short-circuited for three-line-ground fault. The obtained results can be calculated theoretically by using the formula.

2.1 Single-Line-Ground Fault There are three phases that are phase-a, phase-b, and phase-c, during single-lineground fault it is assumed that phase-a shorted to the ground directly. Where (Va = 0, Ib = 0, Ic = 0) and the fault current is If = 3(Ea / Z1 + Z2 + Z3).

21 An Intelligent Technique to Mitigate the Transient Effect on Circuit Breaker …

247

2.2 Double-Line-Ground Fault The fault takes place in phases b and c, this phases shorted ground directly. Then the condition is (Vb = 0, Vc = 0 and Ia = 0) and the fault current is If = - 3 Ia1 (Z2/ Z1 + Z2).

2.3 Three-Line-Ground Fault Here three phases are shorted ground directly. The condition is Ia + Ib + Ic = 0 and Va = Vb = Vc , the fault current is If = Ea/ Z. Figure 1 presents a model which is designed in MATLAB toolbox, the network is having a voltage generation source of 13.8 kV, 50 Hz, having pre-fault from 1 to 5 cycle at CB1. The system has three circuit breakers that are installed in the transmission line. The fault is created near CB1 and then circuit breaker voltages and currents are investigated.

3 ANN-Based Fault Detection Neural network architecture is employed to detect the fault at sending end, with the intention that the fault can be early detected and the faulty part can be removed. The designing of the fault detector is done by following stages.

Fig. 1 Proposed model for fault analysis

248 Table 1 Training of conductor fault

S. Vyas et al. Type of fault

A-G, AB-G, ABC-G

Fault location in Km

5,15,20,30,35,40,45

Inception angle of fault

0–360 (°)

Fig. 2 Complete arrangement of the ANN to detect the fault

3.1 Foundation of Training Data Set The foundation of training data has the required details to adjust input patterns with more real output patterns. The training speed and efficiency of the neural network are affected by the training data set. There is a mixture of several fault situations in the training data set of an ANN. Preparation of training data sets is done by simulating various faults in the network. For obtaining several training samples the fault phase and locations are changed. Table 1 presents the categorization and location task (Fig. 2).

4 Results and Discussion The fault study of the transient effect on circuit breaker has three cases which are described in this section, in these sections L-G, 2L-G, and 3L-G faults are applied on the transmission line. Case I- In this case, the voltage and current waveform of faulted circuit breaker and fault breaker are shown the location of the fault is near to sending end. The fault type is a single line to ground fault (Fig. 3). Case II- In this case, the voltage and current waveform of faulted circuit breaker and fault breaker are shown the location of the fault is near to sending end. The fault type is double line to ground fault (Fig. 4). Case III- In this case, the voltage and current waveform of faulted circuit breaker and fault breaker are shown the location of the fault is near to sending end. The fault type is triple line to ground fault (Fig. 5).

21 An Intelligent Technique to Mitigate the Transient Effect on Circuit Breaker …

249

(a) Voltage waveform

(b) Current waveform Fig. 3 a and b shows the voltage and current waveform of faulted CB and FB, respectively, near to sending end under the single line to ground fault

250

S. Vyas et al.

(a)Voltage wavefrom

(b) Current waveform Fig. 4 a and b shows the voltage and current waveform of faulted CB and FB, respectively, near to sending end under the double line to ground fault

21 An Intelligent Technique to Mitigate the Transient Effect on Circuit Breaker …

251

(a)Voltage waveform

(b) Current waveform Fig. 5 a and b shows the voltage and current waveform of faulted CB and FB, respectively, near to sending end under the triple line to ground fault

252

S. Vyas et al.

Table 2 Required output of ANN-based fault detector Type of fault

Phase

A

B

C

G

Single phase to ground

A-G

1

0

0

1

B-G

0

1

0

1

C-G

0

0

1

1

AB-G

1

1

0

1

BC-G

0

1

1

1

CA-G

1

0

1

1

ABC-G

1

1

1

1

Double phase to ground

Triple phase to ground

The testing of ANN-based fault detector was done by data sets from sending end with 860 fault patterns. The system was investigated and authenticated by testing several faults at several locations. (Lf 0–45 km) and inception angles (i = 0–360°). The operational time ANN-based fault detector can be computed as follows: Operation time = pattern number required for detection x sampling time (Table 2).

5 Conclusions The simulation study of various types of faults at several locations is done using MATLAB Sim-power system and Simulink toolbox is done and the effect of faults on a circuit breaker is also presented. An artificial neural network-based fault detector is designed and investigated. The test results show that by employing ANN-based fault detecting approach we can get early detection of fault at the circuit breaker and it may be saved from damage by monitoring the fault in advance and its location.

References 1. Shakuntla B, Yadav M, Kumar N (2019), MATLAB simulation-based study of various types of faults occurring in the transmission lines. Int J Eng Res Technol (IJERT) 08 2. Pooja P, Preethi KR, Chetan HR, Nandish BM (2016) Three phase transmission line fault analysis using MATLAB Simulink. Int J Sci Eng Dev Res 1:376–378 3. Qiu W (2011) Simulation Study of Single Line-to-Ground Faults on Rural Teed Distribution Lines, and Computing Technologies in Agriculture IV. Springer, Berlin Heidelberg, pp 521–524 4. Tharani V, Nandhini M, Sundar R, Nithiyananthan K (2016) MATLAB based simulations model for three phase power system network 5. Srivastava S, Reshu K, Singh S (2014) Mathematical calculation and MATLAB programming of symmetrical and asymmetrical faults in power system, pp 184–187 6. Boora Dr (2019) MATLAB simulation based study of various types of faults occurring in the transmission lines. Int J Eng Res 7. Gill P (2016) Electrical power equipment maintenance & testing. CRC Press

Chapter 22

Power Quality Improvement by Using STATCOM for DFIG-Based Wind Energy Conversion System Megha Vyas, Monika Vardia, Vinod Kumar, Shripati Vyas, and Yashwant Joshi

1 Introduction In the present time, the sources like solar, wind, hydro and tidal energies are majorly used as renewable sources, and use of these sources is growing rapidly due to less cost, and their clean and sustainable nature [1, 2]. In early time, the wind farms were installed with constant speed induction generators and turbines [3, 4], but due to fixed speed of these, the power efficiency is very low because wind speed changes whole day. These days, the use of intelligent wind generators with a changeable speed are increased sufficiently for enhancing the efficiency. That is why the DFIGs are employed a lot because of their variable speed working, better control of reactive and active power and using moderately rated power converter [5, 6]. The power quality is a serious issue; variations have a negative effect on stability and PQ [7]. Furthermore, the grid-connected wind farms have flickers, voltage sag and current harmonics. Due to the PQ problems, automatic reset, data errors and damage of equipment are more severe. The sag in voltages is a main issue which arises because of 3 L-G fault or by running big motors, and it can cause sudden tripping of equipment, mal-operation and shutdown of system [8, 9].

2 Doubly Fed Induction Generator (DFIG) A DFIG system has an AC/DC-DC/AC, IGBT-dependent PWM converter and a wound rotor induction generator as shown in Fig. 1 [10]. In DFIG system, there is an utmost extraction of energy with the minimum wind speeds by improving turbine M. Vyas · M. Vardia · V. Kumar (B) · S. Vyas · Y. Joshi Department of Electrical Engineering, College of Technology and Engineering / MPUAT / Research Scholler, Udaipur, Rajasthan 313001, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 R. C. Bansal et al. (eds.), Proceedings of International Conference on Computational Intelligence and Emerging Power System, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-16-4103-9_22

253

254

M. Vyas et al. Transformer DFIG

Ulity Grid

Drive train gearbox

RSC Wind Turbine

AC

Line Filter

DC link capacitor GSC C

DC

DC

AC

Fig. 1 Grid-connected DFIG-based wind power system

speeds and reducing mechanical pressures. The DFIGs have some more benefits like generating or absorbing reactive power by power electronic converter, and it reduces the requirement of capacitor banks in presence of squirrel cage induction generator [11].

3 DFIG Modelling and Control The modelling of DFIG is shown in Fig. 1. Controlling of the DFIG is done in a rotating d-q reference frame, along d-axis aligned by the stator flux vector, which is presented in Fig. 2. The control of stator reactive and active powers is done by changing the voltage and current of the rotor. Hence, current and voltage of rotor require to be decomposed within the stator reactive and active power [12].

3.1 Dynamic Modelling The stator’s electrical and mechanical power output can be calculated as: Fig. 2 Vector presentation of stator flux-familiarized control

22 Power Quality Improvement by Using STATCOM …

255

Pt = Tk ∗ wt

(1)

Pz = Tqk ∗ wz

(2)

For a loss less generator, the mechanical equation is: Jdwt = Tk − Tqk dt

(3)

In steady-state at fixed speed for a loss less generator Tk = Tqk and Pk = Pz + Pt

(4)

Pt = Pk − Pz = Tk wt − Tqk wz = −sPz

(5)

Here is the generator slip, which is usually much less than 1, and so, Pz is only a small part of Ps. 1. Voltage Equations: a. Stator Voltage Equations Vqs = pλqs + wλqs + rs iqs

(6)

Vds = pλds − wλqs + rs iqs

(7)

b. Rotor Voltage Equations Vqr = pλqr + (ω − ωr )λds + rs iqr

(8)

Vdr = pλdr − (ω − ωr )λqr + rs idr

(9)

c. Power Equations  3 Vds Ids + Vqs Iqs 2

(10)

 3 Vqs Ids + Vds + Ids 2

(11)

 p Vds Iqs − Vqs Ids 4

(12)

Ps = Qs = d. Torque Equation

Tg = −3

256

M. Vyas et al.

2. Flux Linkage Equations: a. Stator Flux Equations λqs = (Lis + Lm )iqs + Lm Iqr

(13)

λds = (Lis + Lm )ids + Lm Idr

(14)

λqr = (Lir + Lm )iqr + Lm Iq

(15)

λdr = (Lir + Lm )idr + Lm Ids

(16)

Rotor Flux Equation:

In wind generation, a quantity of the accessible power is converted to mechanical power by the rotor blades working on the wind turbines rotor shaft [13]. For computation of mechanical power’s steady-state from a wind turbine, the supposed Cp (λ, β) curve may be employed. Figure 3 shows the power coefficient and tip speed ratio curves. Figure 4 presents DFIG d-q equivalent circuit. The voltages, currents and fluxes may be presented as with stator flux-oriented control [14]. λdqs = Ls idqs + Lpq idqr

(17)

λdqr = Lr idqr + Lm idqs

(18)

vdqr = Rs idqs +

Fig. 3 Cp (power coefficient) v/s tip speed ratio, λ

d λdqt + jωe λdqt dt

(19)

22 Power Quality Improvement by Using STATCOM …

257

Fig. 4 Equivalent circuit of DFIG [6]

Vdqr = Rr idqr +

d λdqr + j(ωe − ωr )λdqr dt

(20)

Here: Lm —(Magnetizing Inductance). Ls —(Stator self-inductance). Lr —(Rotor self-inductance). λdqs —(Stator d-q axis flux linkage). λdqr —(Rotor d-q axis flux linkage). idqr , idqs —(Stator and Rotor d-q axis current). Stator flux vector phase angle can be computed as;  λsdqs =

s s (vdqs − Rs idqs )dt

θe = tan−1

λsqs λsds

(21) (22)

Here, the superscript ‘s’ represents quantities within stationary reference frame.

4 STATCOM STATCOMs are of two types which are voltage source converter (VSC) and current source converter (CSC). Usually, the CSC methodology is less effective than VSC, that is why VSCs are employed in STATCOM. In large capacity wind farms, larger voltage rating is preferable, hence this research uses voltage source STATCOM like a dynamic reactive power compensator [15–17]. The basic for VSC-STATCOM linked with an ac source is presented in Fig. 5, in which Vs indicates the ac voltage source, Vsi indicates controlled ac output voltage of STATCOM and interfacing resistance and reactance as Rs and Xs.

258

M. Vyas et al.

Fig. 5 STATCOM model

Here, we are supposing the power transmission’s positive path from the grid to STATCOM, if Xs > > Rs, the active power P and reactive power Q to the STATCOM. Equations (23) and (24) evidently presents that the active power P and reactive power Q may be forced by varying the output voltage Vsi and the phase difference α between Vs and Vsi. P=

Vs Vsi sinα Xs

  Vsi Vs cosα − Vsi Q= Xs

(23)

(24)

The direction of active power can be obtained by polarity of α, and reactive power direction may be obtained by amplitude subtraction of Vs and Vsi.

5 Power Quality Improvement A good power quality enhances the system operation efficiency, while poor power quality may harm the equipment operation and further the whole power system. A poor power quality has various issues like voltage fluctuation, voltage sag and swell, long term voltage interruption, noise, waveform distortion, power frequency variations, harmonics, voltage spike, transient and flicker that causes power quality problems and can be traced to a specific type of electrical disturbance. Out of the referenced issues, the problem of power ripple has become the most severe issue in advancement of wind generation system, because such ripple results in a poor power

22 Power Quality Improvement by Using STATCOM …

259

quality that affects the system reliability and leads to malfunctioning of the system. The various power quality problems are shown in Fig. 6. The block diagram shown in Fig. 7 proposes an intelligent controller-based wind energy conversion system which consists of two back-to-back converters. The neural network, fuzzy logic and neuro-fuzzy controllers are one of the forms of artificial intelligence techniques that can be applied for controlling of these two converters. Further, the use of fast reactive power compensators can improve the power system stability, and hence maximum power transfers through the electric system. An intelligent method of control strategy using PI controller and neuro-fuzzy system may be used in order to achieve smooth grid integration, energy savings, reactive power compensation and effective real power flow.

3% 8%

29%

voltage sag

voltage swell

60%

transients

Interrupon

Fig. 6 Power quality problems

ANFIS/ FLC

Wind Turbine

Gear Box

DFIG

Transformer

AC/DC

DC/AC

Controller

Fig. 7 Power quality Improvement of controller-based wind energy conversion system

Grid

260

M. Vyas et al.

Fig. 8 Conventional Pi controller

6 Control Strategy We are employing an intelligent controlling strategy for improving the performance of STATCOM with FLC. The advantage over conventional controller is FLC can be used without any mathematical modelling [16]. Furthermore, FLC is more efficient to handle nonlinearity and is further vigorous in comparison to conventional PI controller that too enhances the efficiency of STATCOM. The employed control technique is presented below.

6.1 Conventional PI Controller Here, the controlling of STATCOM is done by conventional PI controller as presented in Fig. 8. We assumed gain values of P and I as KP = 0.1 and Ki = 2 with trial and error technique. The sag voltage is evaluated with reference voltage, and the error is evaluated with PI controller, and then, output is renovated to 3 phases via unit vector generation. Further, it is given to PWM generator to offer gate pulses to series so that it can insert the needed voltage for the reduction of voltage sag [17].

6.2 Fuzzy Logic Controller The use of FLC is best over other operations of fuzzy set theory; the major advantage is that it utilizes linguistic variables over numerical variables. It uses human potential to understand the system’s behaviour and hence, built up on quality control system. It employs basic method to reach a certain end depending on indistinct, indefinite, unclear, noisy or absent input data [18–20]. The FLC in a simplified way is shown as in Fig. 9.

22 Power Quality Improvement by Using STATCOM …

261

Fig. 9 Fuzzy logic controller

• The input data is altered into suitable linguistics by a fuzzification interface. • The study of data base within the necessary linguistic descriptions and control law set is done with knowledge base. • The collection of fuzzy control act from the data of the control rules and the linguistic variable explanations is done by a Decision-Making Logic. • A Defuzzification interface that surrenders a non-fuzzy control act from an inferred fuzzy control act.

7 Simulation Model In the presented model, wind turbines of 9 MW capacities are employed, which utilizes the DFIGs. The wind turbines are simulated with grid of 50 Hz and rotors are simulated to the 50 Hz grid, and the rotors are obsessed by a changeable-pitch wind turbine. Usually, the DFIG speed is kept somewhat more than the synchronous speed for generating the power. The variation in speed occurs around 1pu at condition of no load and in full load, 1.006. There is also a protection system to monitor the speed, voltage and current. The remaining reactive power needed for maintaining the bus voltage near 1pu by a 10-Mvar STATCOM. The mechanical power (turbine speed) presented for wind speeds range from 5 m/s to 13 m/s. The investigation on transient stability is done by wind turbine model and the STATCOM model along large simulation times (Fig. 10).

262

M. Vyas et al.

Fig. 10 Wind turbine model

8 Simulation Results To confirm the achievability of the planned control scheme, MATLAB simulations were done. In the study, the employed DFIG is of 2 MW capacity and the wind speed is steady at 10 m/s. To create the unbalancing condition in voltage, the value of phase-a voltage was reduced by 20%. In the starting, the system was running in balanced conditions, after 1.5 s, the grid voltage was varied, and further, the balanced condition was achieved in 3 s. The grid current and voltage with no compensation are presented in Fig. 11a and b. Here, major ripple factors in the rotor torque, speed, stator reactive and active power are presented in Figs. 12 and 13. The fault happens at 0.2 s, voltage sag appears about 30%; at this moment, reactive power produced by STATCOM is -0.27 pu, whereas there is about zero power in conventional circuit. Further, the peak capacitive power produced by a STATCOM reduces linearly along voltage reduce (constant current). Availing enhanced power in fault condition is the main benefit of the STATCOM. There is severe pressure on generator shaft and gearbox due to fluctuations in torque and rotor speeds. For mitigating the above referenced issue, the reactive power ripple control technique is employed.

22 Power Quality Improvement by Using STATCOM …

263

(a)

(b) Fig. 11 a Grid Current. b Grid voltage

9 Conclusion In this research, the techniques for enhancement of voltage and current quality are presented for grid-linked DFIG network. Several PQ issues like sag in voltage and current harmonics are investigated in MATLAB environment. The FLC-based STATCOM is presented as a solution for PQ improvement, and the results are also evaluated for conventional PI and FLC-based controller. By investigating, we can

264

M. Vyas et al.

(a)

(b) Fig. 12 a Active power performance. b Reactive power performance

conclude that the conventional controller is capable to reduce voltage sag but not the load current harmonics efficiently, while the presented FLC-based STATCOM controller reduces the voltage sag and along with load current harmonics efficiently by maintaining the THD in standard limits. Hence, the FLC-based controller was proved efficient for enhancing PQ in a grid-connected wind power network.

22 Power Quality Improvement by Using STATCOM …

265

(a)

(b) Fig. 13 a Generator torque. b Constant and Impulsive Wind speed condition

References 1. Xu D, Li R, Liu Y, Lang Y (2009) Reactive power analysis and control of doubly fed induction generator wind farm. In: IEEE, 13th European conference on power electronics and applications, pp 1–10 2. Renewable Energy Akshay Urja (2013) Newsletter of the ministry of new and renewable energy, Government of India, p 6 3. Xu L, Cartwright P (2006) Direct active and reactive power control of DFIG for wind energy generation. IEEE Trans Energy Convers 21:750–758 4. Erlich HW, Feltes C (2008) Dynamic behavior of DFIG-based wind turbine during grid faults. IEEJ Trans 128:396

266

M. Vyas et al.

5. Molinas M, Suul JA, Undeland T (2007) A simple method for analytical evaluation of LVRT in wind energy for induction generators with STATCOM or SVC. In: IEEE power electronics and applications, European conference, pp 1–10 6. Hasan AD, Michalke G (2007) Fault ride-through capability of DFIG wind turbines. Renew Energy 32:1594–1610 7. Santos-Martins D, Arnaltes S, Rodriguez Amenedo JL (2008) Reactive power capability of doubly fed asynchronous generators. Electric Power Syst Res 78:1837–1840 8. Cai G, Sun Q, Liu C, Li P (2011) A new control strategy to improve voltage stability of the power system containing large-scale wind power plants. In IEEE, 4th international conference on electric utility deregulation and restructuring and power technologies (DRPT), Weihai, Shandong, pp 1276–128 9. Santos S, Le HT (2007) Fundamental time-domain wind turbine models for wind power studies. Renew Energy 32:2436–2452 10. Chowdhury BH, Chellapilia S (2006) Doubly-fed induction generator control for variable speed wind power generation. Electric Power Syst Res 76:786–800 11. Karim-Davijani H, Sheikjoleslami A, Livani H, Karimi-Davijani M (2009) Fuzzy logic control of doubly fed induction generator wind turbine. World Appl Sci J 6:499–508 12. Sun T, Chen Z, Blaabjerg F (2005) Transient stability of DFIG wind turbines at an external short circuit fault. Wind Energy J 8:345–360 13. Zou Y, Elbuluk M, Sozer Y (2010) A complete modelling and simulation of induction generator wind power systems. In: Proceedings of IEEE industry applications society annual meeting, pp. 1–8 14. Santacana E, Rackliffe G, Tang L, Feng X (2010) Getting smart. IEEE Power Energy Mag. (April 2010) 15. Tapia A, Tapia G, Ostolaza JX, Saenz JR (2003) Modelling and Control of a wind turbine driven doubly fed induction generator. IEEE Trans Energy Convers 18:194–204 16. Adamczyk A, Teodorescu R, Mukerjee RN, Rodriguez P (2010) Overview of facts devices for wind powerplants directly connected to the transmission network. In: IEEE international symposium on industrial electronics (ISIE), pp 3742–3748 17. Zhiand D, Xu L (2007) Direct power control of DFIG with constant switching frequency and improved transient performance. IEEE Trans Energy Convers 22 18. Poure GAP, Saadate S (2009) Variable speed DFIG wind energy system for power generation and harmonic current mitigation. Renew Energy 34:1545–1553 19. Phanand V-T, Lee H-H (2012) Performance enhancement of stand-alone DFIG systems with control of rotor and load side converters using resonant controllers. IEEE Trans Ind Appl 48 20. Hosseini SMH, Olamaee J, Samadzadeh H (2011) Power oscillations damping by Static Var Compensator using an Adaptive Neuro-Fuzzy controller. In: Proceedings of 7th international conference on electrical and electronics engineering, pp 80–84

Chapter 23

Optimization of Standalone Microgrid’s Operation Considering Battery Degradation Cost Rekha Swami and Sunil Kumar Gupta

1 Introduction To supply power to isolated areas from the main utility grid involves high capital costs due to the cost of building transmission and distribution infrastructure. An isolated microgrid consisting of both dispatchable sources like diesel generators and non-dispatchable renewable distributed generators can be an alternative to this. In an isolated microgrid, balance between generation and demand is necessary, otherwise systems’ voltage and frequency will deviate and the system may collapse [1–3]. To keep the frequency of the system within limit either load is curtailed (when generation is less than load) or renewable power generation is curtailed (when generation is higher than load) [4]. Both cases affect the reliability and economy of the microgrid. One effective solution to perform the task of matching generation with the demand is the use of a battery storage system (BSS) [5]. BSS stores energy during the offpeak time and releases it during peak times, thus reducing overall system cost [6]. However, these storage systems have high capital cost and low lifecycle which limits their application in microgrid [7]. Depending upon the use, battery performance degrades over years and its capacity decreases. Degradation of BSS is majorly due to calendrical aging and cyclic aging [8, 9]. Calendrical aging occurs when BSS is not in use and is affected by the cell temperature and potential of the battery, i.e., SOC of the battery. Cyclic aging results from the use of BSS and it is dependent upon the depth of discharge and cycle numbers. Neglecting cyclic aging of BSS in microgrid expansion problems may result in an incorrect economic assessment as the BSS replacement might be needed before the project finishes. To estimate the lifecycle of BSS different methods are proposed in R. Swami (B) · S. K. Gupta Department of Electrical Engineering, Poornima University, Jaipur, India S. K. Gupta e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 R. C. Bansal et al. (eds.), Proceedings of International Conference on Computational Intelligence and Emerging Power System, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-16-4103-9_23

267

268

R. Swami and S. K. Gupta

[10–13]. However, the BSS manufacturers normally proffer the relationship between lifecycle and depth of discharge in the form of a curve. For lead-acid batteries, this relationship is in linear form while for lithium-ion batteries it is exponential. In general, BSS lifecycle decreases with an increase in depth of discharge [10]. A depreciation cost model for lithium batteries based on cycle life is developed along with the implementation of a practical charge/discharge strategy for battery management in [14]. Reference [15] presented dual objective optimization model for the standalone microgrid to minimize the generation cost and enhance the life of the lead-acid battery. In [16], a stochastic optimization model is presented for minimizing the operation cost of grid-connected microgrid for a single day considering battery degradation cost. A multi-objective model for island microgrids having solar, wind, diesel, storage systems for power generation and seawater desalination load is presented in [17]. A linear model for degradation cost of the battery for sizing multicarrier microgrid considering economic and environmental aspects is developed in [18]. In [19] a novel scheduling algorithm is proposed for minimizing the operating cost of remote microgrids and extending battery life. An optimization model for community microgrid is developed for minimizing electricity bills considering battery degradation cost along with the cost for charging/discharging cycles [20]. A day-ahead energy management strategy for a community microgrid for minimizing operating cost considering the cyclic cost for the battery is proposed in [21]. It also demonstrated that uncertainty in renewable generations and load has no effect on the energy scheduling of BSS but is affected by the real prices. The technical and economic performance of lead-acid as well as Li-ion batteries is analyzed under the rural and urban scenarios in [22]. In the majority of the research work, lead-acid batteries were used as energy storage for microgrid applications. Although these batteries are more than 100 years old, matured and have lower capital cost but has short life, therefore, they need replacement after use of 4–5 years during the whole project lifetime, which affects the project lifetime cost [14–19]. In the considered model, lithium-ion batteries are chosen because it has high energy density, longer cycle life, lower maintenance, specific energy, smaller size, and weight as compared to lead-acid batteries. Because of these characteristics, they are gaining popularity to be used in power system applications [19–30]. The contributions of the presented work are the following: 1. 2. 3.

100% utilization of renewable energy. Battery degradation cost due to cyclic degradation is considered in order to show the real operation cost of microgrid. No load curtailment.

Figure 1 shows the schematic diagram of the standalone microgrid that is studied in this paper. The power sources considered are wind turbine, solar photovoltaic, and diesel generator with a battery storage system.

23 Optimization of Standalone Microgrid’s Operation …

269

Fig. 1 Standalone microgrid model

G BSS WT

Load PV

2 Optimization Model of Standalone Microgrid 2.1 Objective Function For the considered standalone microgrid model (Fig. 1) optimization problem is formulated as the minimization of the operating cost of the microgrid for a period of 24-h including the battery degradation cost. 24     F = ag pg2 (t) + bg pg (t) + cg + Cdeg . pk (t).t k

(1)

t=1

where ag , bg , cg —Diesel generators’ cost coefficients ($/K w2 h, $/Kwh, $). pg (t)—Diesel generator power output at time t (Kw). Cdeg —Battery degradation cost ($/Kwh).

2.2 Battery Degradation Cost Model Battery capacity decreases depending upon use and operating conditions, this is known as battery aging. Battery is said to be dead when its capacity has reached 20% of its initial capacity and then the battery needs to be replaced. Therefore, it is necessary to add battery degradation cost into the operating cost to show the real operating cost (Fig. 2). The total number of charge/discharge cycles represents the battery life cycle before its capacity decreases to 80% of the initial capacity (under standard conditions). In

270

R. Swami and S. K. Gupta

Fig. 2 Battery retaining capacity versus cycle life [14]

literature [24] for different DOD, the capacity retention rate of li-ion batteries with an increase in cycle life has been tested. When battery capacity drops to 80% then cycle life is pointed out and given as below in Table 1. To show the relationship between lifecycle and DOD of LiFePO4 battery, a 3parameter function is applied that is popularly used in engineering applications [14]. N = R/(D O D − d)m

(2)

where N is the lifecycle of the battery and m, d and R are fitting parameters (Fig. 3). Table 1 Experimental test data for battery

Fig. 3 Fitted curve for life cycle and DOD [14]

LiFePO4 battery

DOD

Cycle

80

700

50

1440

20

2900

10

6300

23 Optimization of Standalone Microgrid’s Operation …

271

Using the data given above, least square method is used to fit the lifecycle function. The fitted curve is shown below and parameters c, d, and m are found as 612.313, − 0.0223, and 1.0953. After calculating lifecycle using Eq. (2), the degradation cost of the battery ($/KWh) is obtained by dividing battery investment cost from battery lifecycle. Cdeg =

Cinv cost Battery life

(3)

2.3 Constraints 2.3.1

Power Supply Balance Pg + PW T + PP V + PB = PL

2.3.2

Diesel Generator Limits pgmin ≤ p g ≤ pgmax

2.3.3 (a)

Battery state of charge at every instant is given by (6)

Pc (t)—Battery charging power at each instant (Kw). Pd (t)—Battery charging power at each instant (Kw). S OC(t)—State of charge of battery (Kwh). ηc ,ηd —Charging/Discharging efficiency. Generally to enhance the battery life battery SOC is set to a minimum limit that will restrict the economic benefits of battery operating below the minimum limit. In this paper, no lower limit of SOC is considered. 0 ≤ S OC(t) ≤ S OC max (t)

(c)

(5)

Battery Constraints

S OC(t) = S OC(t − 1) + Pc (t)ηc − Pd (t)/ηd

(b)

(4)

(7)

To guarantee the secure operation for the next day, SOC at the end and starting must be equal. S OC(0) = S OC T

(8)

272

R. Swami and S. K. Gupta

3 Case Study and Results For the case study, a standalone microgrid on Dongfushan Island in China is considered [15]. WT and PV output power based on actually measured solar irradiation, temperature and wind speed, and load demand of Dongfushan island are given in Fig. 4 (Table 2). Battery Parameters: Battery storing capacity is 90 KWh, Investment cost of battery $600/KWh, SOC(0) = S OC T = 0.5 p.u., S OC max = 1 p.u., ηC and ηd = 0.95. The objective function is optimized with and without considering degradation cost of battery under two scenarios, when there is: 1. 2.

Short of renewable power generation, Excess renewable power generation. Results obtained for each scenario are discussed below.

Scenario 1 In this case, renewable power generation is limited. Microgrid operating cost without including battery degradation cost is found as $545.234. In this case battery is charged during time interval 1, 2, 3, 4, 5, 14, 21, 22, 23, 24 and discharged during 7, 8, 9, 10, 11, 12, 15, 16, 17. For the rest of the period, battery is in idle condition. Battery SOCs remain within limits (Figs. 5 and 6). Now battery degradation cost is added to the operating cost and different weights are assigned to each cost so that microgrid real operating cost is obtained. On taking w1 = 0.9 and w2 = 0.1, the operating cost is reduced by 8.84%. With this small proportion of battery degradation cost addition, the discharge amount of the battery is decreased from 67.105 to 61 KW. With further increase in W2, the discharge amount decreases to 45.16 KW and the number of charge-discharge cycles of the battery reduces (as shown in Fig. 7), which will improve the battery life. The obtained results in table form are given below (Table 3). The use of generators and battery depends upon these weights. In case of w1 = 1 (battery life was not considered, i.e., w2 = 0), battery gives high power output, therefore the life of the battery reduces and needs replacement. With an increase in w2 , battery cost is given the importance that results in reduced battery consumption. Since in a standalone microgrid, more priority is given to minimize the fuel cost so w1 is weighted higher than w2 . Scenario 2 The operating cost of microgrid with no battery degradation cost is found as $255.214. Battery is charged in time interval 3, 4, 12, 13, 14, 15, 20, 22, 23, and 24 and discharged in 6, 7, 8, 9, 10, 11, 16, 17, 18, and 19. Battery power output is shown in Fig. 8. Now like scenario 1, battery degradation cost of the battery is added to the operating cost of the diesel generator and different weights are assigned to each cost. On taking w1 = 0.9 and w2 = 0.1, the operating cost is reduced by 6.63%. With this small proportion of battery degradation cost addition, the total discharge amount

23 Optimization of Standalone Microgrid’s Operation …

273

Fig. 4 a Power output in case of short renewable generation, b Power output in case of excess renewable generation, c Load demand of standalone microgrid [15]

Wind PV

(a)

Wind PV

(b)

(c) Table 2 Diesel generator unit characteristics ag ($/K W 2 h)

bg ($/KWh)

cg ($)

Pimin (KW)

Pimax (KW)

0.003

0.29

8

5

50

274 Fig. 5 Battery output power under short renewable generation

R. Swami and S. K. Gupta 30 20 10 0 1

3

5

7

9

11 13 15 17 19 21

23

-10 -20 -30

Fig. 6 Battery SOC under short renewable generation

100 80 60 SOC 40

Load

20 0 1 3 5 7 9 11 13 15 17 19 21 23 Fig. 7 Decrease in no of charge and discharge cycle of battery with w2 = 0.2

20 10 0 1

3

5

7

9

11 13 15 17 19 21 2 3

-10 -20

Table 3 Optimization results using Li-ion battery in case of short of renewable generation Operating cost ($) Without battery degradation cost

544.697

With battery degradation cost

w1 = 0.9 & w2 = 0.1

w1 = 0.8 & w2 = 0.2

w1 = 0.7 & w2 = 0.3

496.545

445.537

394.310

23 Optimization of Standalone Microgrid’s Operation … Fig. 8 Battery output power under excess renewable generation

275

30 20 10 0 -10

1

3

5

7

9

11 13 15 17 19 21

23

-20 -30 -40

Table 4 Optimization results for using Li-ion battery in case of excess renewable generation Operating cost ($) Without battery degradation cost With battery degradation cost

255.214 w1 = 0.9 and w2 = 0.1 w1 = 0.8 and w2 = 0.2 w1 = 0.7 and w2 = 0.3 238.291

221.332

204.374

of the battery is decreased from 112.86 to 108.3 KW. With further increase in W2, the battery throughput decreases that will improve the battery life. The optimization results in the case of excess renewable generation are given in Table 4.

4 Conclusion In the presented work, a battery degradation cost model is developed for Li-ion batteries. For the case study, Dongfushan island of China is considered to investigate the economic performance of Li-ion battery in the microgrid operation. The objective function considered is taken as minimization of the operating cost of standalone microgrid and battery degradation cost is included to give the real operating cost of the microgrid. The case study is performed for two scenarios under short and excess renewable generation. Used weighted sum approach to optimize the total operation cost in GAMS software with the help of CONOPT solver. From the dayahead optimization results, it was found that the addition of battery degradation cost even in small proportion reduces the microgrid operation cost by 8.84 and 6.63% under short and excess renewable generation scenarios, respectively, compared to the case of neglecting battery degradation cost in the microgrid operation cost.

276

R. Swami and S. K. Gupta

References 1. Parhizi S, Lotfi H, Khodaei A, Bahramirad S (2015) State of the art in research on microgrids: a review. IEEE Access 3:1–36. https://doi.org/10.1109/ACCESS.2015.2443119 2. Khodaei A, Shahidehpour M (2013) Microgrid-based co-optimization of generation and transmission planning in power systems. IEEE Trans Power Syst 28(2):1582–1590 3. Khodaei A (2014) Resiliency-oriented microgrid optimal scheduling. IEEE Trans Smart Grid 5(4):1584–1591 4. Elrayyah A, Sozer Y, Elbuluk M (2015) Microgrid-connected PV based sources: a novel autonomous control method for maintaining maximum power. IEEE Ind Appl Mag 21(2):19–29 5. Khodaei A (2014) Microgrid optimal scheduling with multi-period islanding constraints. IEEE Trans Power Syst 29(3):1383–1392 6. Alharbi A, Bhattacharya K (2014) Optimal sizing of battery energy storage systems for microgrids. IEEE electrical power & energy conference, Canada, pp 275–280. https://doi.org/10. 1109/EPEC.2014.44 7. Alsaidan I, Khodaei A, Gao W (2016) Distributed energy storage sizing for microgrid applications. IEEE PES transmission and distribution conference, Dallas. https://doi.org/10.1109/ TDC.2016.7519904 8. Weitzel T, Schneider M, Glock CH, Löber F, Rinderknecht S (2018) Operating a storageaugmented hybrid microgrid considering battery aging costs. J Clean Prod 9. Karanasios E, Ampatzis M, Nguyen PH, Kling WL, van Zwam A (2014) A model for the estimation of the cost of use of Li-Ion batteries in residential storage applications integrated with PV panels. In: 49th international universities power engineering conference (UPEC), pp 1–6 10. Bocca A, Sassone A, Shin D, Macci A (2015) An equation-based battery cycle life model for various batteries chemistries. IEEE Int. conference on very large scale integration (VLSI-SOC). Korea (South), pp 57–62. https://doi.org/10.1109/VLSI-SoC.2015.7314392 11. Seiger HN (1981) Effect of depth of discharge on cycle life of near-term batteries. Paper presented at the 16th Intersociety Energy Conversion Engineering Conference, pp 102–110 12. Burke AF (1995) Cycle life considerations for batteries in electric and hybrid vehicles. SAE Technical Paper, Warrendale, PA, SAE Technical Paper 951951 13. Thaller LH (1983) Expected cycle life vs. depth of discharge relationships of well-behaved single cells and cell strings. J Electrochem Soc 130(5): 986–990 14. Zhang Z, Wang J, Wanget X (2015) An improved charging/discharging strategy of lithium batteries considering depreciation cost in dayahead microgrid scheduling. Energy Convers Manage 105:75–84 15. Zhao Bo, Zhang X, Chen J (2013) operation optimization of standalone microgrids considering lifetime characteristics of battery energy storage systems. IEEE Trans Sustain Energy 4:934– 943 16. Wtencongu Su, Wang J, Roh J (2013) Stochastic energy scheduling in microgrids with intermittent renewable energy resources. IEEE Trans Smart Grid 5:1875–1883 17. Tang Q, Liu N, Zhang J (2014) Optimal operation method for microgrid withwind/PV/diesel generator/battery and desalination. J Appl Mathe 2014(ID857541). https://doi.org/10.1155/ 2014/857541 18. Cardoso G, Brouhard T, DeForest N, Wang D, Heleno M, Kotzur L (2018) Battery aging in multi- energy microgrid design using mixed integer linear programming. Appl Energy 231:1059–1069 19. Chalise S (2014) Energy Management of remote microgrids considering battery lifetime. Electr J 29:1–10 20. Alamgir Hossain Md, Pota HR, Squartini S, Zaman F (2019) energy management of community microgrids consi dering degradation cost of battery. J Energy Storage 22:257–269 21. Hossain HA, Pota HR, Squartini S, Zaman F, Guerrero JM (2019) Energy scheduling of community microgrid with battery cost using particle swarm optimization. Appl Energy 254, 113723

23 Optimization of Standalone Microgrid’s Operation …

277

22. Dhundhara S, Verma YP, Williams A (2018) Techno-economic analysis of the lithium-ion and lead-acid battery in microgrid systems. Energy Convers Manage 177:122–142 23. Lai CS, Locatelli G, Pimm A, Tao Y, Lia X, Laia LL (2019) A financial model for lithium-ion storage in a photovoltaic and biogas energy system. Appl Energy 251, 113179 24. Wanga J, Liua P, Hicks-Garnera J, Sherman E, Soukiazian S, Verbrugge M (2011) Cycle-life model for graphite-LiFePO4 cells. J Power Sources 19(8):3942–3948. https://doi.org/10.1016/ j.jpowsour.2010.11.134 25. Lorestani A, Gharehpetian GB, Nazari MH (2019) Optimal sizing and techno-economic analysis of energy- and cost- efficient standalone multi-carrier microgrid. Energy. https://doi.org/ 10.1016/j.energy.2019.04.152 26. Kabir MN, Mishra Y, Ledwich Z, Xu Z, Bansal RC (2015) Coordinated operation algorithm using reactive power and integrated battery storage in distribution system. In The handbook of clean energy systems. Johan Wiley & Sons, vol 5, no 28, pp 2891–2902 27. Adefarati T, Bansal RC, Naidoo R, Potgieter S, Rizzo R, Padmanaban S (2020) Optimization of PV-Wind-battery storage microgrid system utilizing a genetic algorithm. IET-Renew Power Gener 14(19):4053–4062 28. Lupangu C, Justo JJ, Bansal RC (2020) Model predictive for reactive power scheduling control strategy for PVĂbattery hybrid system in competitive energy market. IEEE Syst J 14(3):4071– 4078 29. Mbungu NT, Bansal RC, Naidoo R (2017) Optimisation of grid connected hybrid PV-windbattery system using model predictive control design. IET-Renew Power Gener 11(14):1573– 1584 30. Bansal RC, Zobaa AF (eds) (2021) Handbook of renewable energy technology and systems. World Scientific Publisher, UK

Chapter 24

Investigation and Assessment of TCAD-Based Modeled and Simulated PPV/PCBM Bulk Heterojunction Solar Cells Mohammad Asif Iqbal and Virendra Sangtani

1 Introduction In a typical organic PV cell operation, there are four fundamental physical processes, which are described by schematic depiction in Fig. 1. Step 1: Light Absorption When light is captivated in donor material, robustly bound, electron–hole pairs are created. Step 2: Exciton Diffusion The photo created through excitons is firmly coulomb bound because of low dielectric steady in polymer materials. Consequently, electrically unbiased excitons can just move by dispersion so as to separate into an electron-gap pair at interface of donor– acceptor particles (as shown in Fig. 1b). Step 3: Excitons Dissociation The excitons separate just at vivaciously positive donor–acceptor interfaces, when vitality gain is bigger than excitons restricting vitality. An electron move (or charge move) happens; separating a gap staying on natural (appeared in Fig. 1c). After separation, the created pair is still coulombically limited, this is known as polaron pair.

M. A. Iqbal (B) · V. Sangtani Department of Electrical Engineering, Vivekananda Institute of Technology, Jaipur, India V. Sangtani e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 R. C. Bansal et al. (eds.), Proceedings of International Conference on Computational Intelligence and Emerging Power System, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-16-4103-9_24

279

280

M. A. Iqbal and V. Sangtani

Fig. 1 Schematic Processes During Operation of Organic PV Cells under 1 Sun Illumination

Step 4: Charge Transport The polaron pair is dissociated into free electron and hole with built-in electric field provided by two different electrodes. As electrons and holes are transported to respective electrodes, driven by electric field. Carriers in respective phases are moved by a hopping transport process, however, there is several loss mechanisms involved during operation of PV cells. The field of organic PV cells (often called as organicfullerene PV cells) has taken a huge splurge with discovery of conducting organics [4]. Although conducting organics are highly conjugated organics, having backbones of continuous sp2 hybridized carbon centers, which have a large overlap between pz orbital’s providing large delocalized states. As described in the above section, organic PV cells employ conducting organics as electron donor materials and fullerenes as electron acceptor materials. Organic solar cells are much improved after research in the field of variation in molecular structure, now the efficiency improved from 1 to over 9%. Introduction of transport layers (hole & electron transport), tuning of active layer morphologies, employment of various device structures, etc. [12]. In this regard, various conducting organics were evolved, (i) Firstly, (poly phennylane vinylene)s (PPV), MEH-PPV, MDMO-PPV were introduced. Power conversion efficiencies up to 3% were achieved by employing PPV-based organics as donors and (PC61BM) as acceptor materials [1]. (ii) Secondly, smaller-band-gap organics (thiophenes) were investigated. P3HT with PCBM as acceptor material demonstrated PCE up to 5% [2]. (iii) Lately, polycarbazole-based organics and some patented absorber materials have demonstrated world record efficiencies over 9%. Although, many investigations were made during this period, no suitable materials were found to replace highly efficient fullerene derivatives (PC61BM, PC71BM, ICMA, ICBA) as acceptor materials in organic PV cells. Some of photovoltaic properties with various conducting organics are presented. In present time inorganic material; for example, silicon is driving route in PV cell innovation. Effectiveness of organic devices can be reached up to over 15%; theoretical organic PV cells have become the most approachable to collect energy source in future. Higher Fabrication cost and extremely hard to create has become its disadvantages (Figs. 2 and 3).

24 Investigation and Assessment of TCAD …

281

Fig. 2 Schematic depiction of carrier transport in inorganic and organic semiconductors [2]

Fig. 3 Peak power and fill factor in a polymer (Donor) PV cell

2 Modeling and Simulation of PPV/PCBM-Based PV Cell 2.1 Selection of Material When a photon is absorbed by an active material, it forms excited-state excitons, an electron–hole pair found in the local environment. These excitons can be transmitted through matter that diffuses into life-limited matter, allowing the distance it can travel before it collapses. Usually, it is only 10 to 20 nm in size. This means that the optimal size structure in the active material must also lie in the order of the material, which is referred to as bulk heterojunction or macrophage separation. The sequence of donor, polymer (donor), and acceptor is located in several domains, in which holes and electrons can propagate. All types of polymer solar cells (donor solar cells) are constructed with double-layer geometry [8]. Therefore, in the manufacture of solar cells, a layer consists of the polymer (donor) donor and the acceptor, which also means that the exciton have to travel a long distance. The exciton migrate to

282

M. A. Iqbal and V. Sangtani

Fig. 4 Schematic of energy levels, process electron donating, and electron accepting materials [6]

Fig. 5 Proposed PV cells with electron–hole flow [9]

the phase boundary between the donor and acceptor material, where charges can be separated, electrons can migrate to the acceptor material, and holes in the polymer (donor) material can migrate and pass through the two through the electrode kind of material [11]. Charge carriers need to move to multiple electrodes instead of moving randomly to generate electricity. An internal electric field is creates by using different electrode materials with different energy levels and also by using an intermediate layer, which preferably transports electrons or hole charge carriers in the correct direction of electrode (Figs. 4 and 5). PCMB is a fullerene material and used as an electron acceptor. It is a material with good optical and electrical properties, which is a primary requirement of a semiconductor. It is useful to make N region in PV cell. It is also known as C60 bucky ball. In the process of organic solar cell designing, it is desirable to choose PPV in association with PCBM [5]. PPV is a polymer material and used as an electron donor. It is a material with good optical and electrical properties, which is a primary requirement of a semiconductor, as it is useful to make P region in PV cell [3]. Poisson’s equation and singlet exciton continuity equation is expressed as: ds = G ph − KNRS.EXCITONS · S - RD np + R L np dt

(1)

24 Investigation and Assessment of TCAD …

283

where S: singlet concentration, r L : Langevin recombination rate constant. Gph: Photogeneration rate. E: Local electric field. Er: Relative permittivity. T: Temperature. A.SINGLET and S.BINDING: user specified parameters. KNRS.EXCITON: non-radioactive singlet decay rate. J1 : first order Bessel function. S: singlet concentration. U(x): Net generation rate. Dn,p: carrier diffusion coefficients. Exciton dissociation is given by the equation: √ S.B I N D I N G JI (2 −2b) 3r L exp(− ) √ R D np = 4π A.S I N G L E T kT −2b b=

q 3 | E| 8π ξr ξo k 2 T 2

(2)

(3)

The thicknesses used to play out these estimations are 120 nm thick PPV/PCBM BHJ PV cells.

2.2 Simulation Performance Poisson equation can be written as follows: q ∂2 ∀(x) = [n(x) − p(x)] 2 ∂x ε

(4)

The current continuity equation can be expressed as: ∂ J p (x) = −qU (x) ∂x

(5)

Jn = −qnμn

∂ ∂ ∀ + q Dn n ∂x ∂x

(6)

J p = −qnμ p

∂ ∂ ∀ − q Dp p ∂x ∂x

(7)

Dn. p = μn. p Vt

(8)

284

M. A. Iqbal and V. Sangtani

Fig. 6 Structure of a Bulk Heterojunction Polymer (donor) PV Cell [7]

Vt = k B T lq

(9)

The average net generation rate can be written as: < u >=

Jn (L) − Jn (0) 1 L ∫ U (x)d x = L 0 qL

(10)

2.3 Device Structure With respect to photo generation, creators recommended a consistent photo generation rate all through the gadget of 2.7 × 1021 cm−3 s−1 . The last worth disregards front reflections or light going through cells. In light of recommended boundary esteems, simulation have been carried out. Figure 7 shows the correlation of simulation with introduced trial results [10]. Albeit, impede is a decent match, rest of bend shifts a considerable amount from test. Because of absence of information on different boundaries and models, for example. Absorbed light generates abundant excitons, which diffuses and separate to shape electron opening sets (Figs. 6, 7 and 8, Table 1).

3 Simulation Results and Discussion Power conversion efficiency of a PV cell is dependent upon short-circuit current density (ISC ), open circuit voltage (VOC ), and fill factor (FF) (Table 2). The simulation numerical models for current–voltage (I-V) curve of polymer (donor)/fullerene bulk heterojunction (PPV/PCBM) PV cells have been done and compared both numerical models for current–voltage (I-V) characteristics of polymer (donor)/fullerene bulk heterojunction (PPV/PSBM) PV cells (Table 3).

24 Investigation and Assessment of TCAD …

285

Fig. 7 Langevin recombination at short circuit

Fig. 8 VI and VP characteristics for PPV/PSBM PV cell

4 Conclusion Allowing for PCBM in designing of Polymer solar cell instead of P3HT, the cell fabrication process is not only simpler and cheaper. PPV/PCBM led to a maximum attainable efficiency of 13.33% which was only 5% in the previous researches. With the help of TCAD, software extraction of the optical and electrical parameters before fabrication is possible with ease.

286

M. A. Iqbal and V. Sangtani

Table 1 Optimizer value of DeckBuild Iteration Materialknrt.exciton Materiala. Materials.binding Solve b1 singlet

IV Sensitivity (%)

1

1.50e + 06

1.3

0.35

1

err = 46

–

2

1.5474e + 06

1.3

0.35

1

err = 47

69%

3

1.5e + 06

1.34111

0.35

1

err = 50

2.7e + 02%

4

1.5e + 06

1.3

0.361068

1

err = 64

1.2e + 03%

5

1.5e + 06

1.3

0.35

1.03162

err = 44

−1.4e + 02%

6

1 6448e + 06

1.32324

0.321206

0.762386

err = 25

–

7

1.78231e + 08

1.34459

0.296568

0.5

err = 31

–

8

1.76051e + 06

1.34494

0.300619

0.566989

err = 25

–

9

1.76525e + 08

1.34494

0.300619

0.566969

err = 25

–

10

1.82694e + 06

1.32167

0.28484

0.587647

err = 1.7

–

Table 2 Results obtained by (PPV/PCBM) PV cells I (A/m2)

Va

Average dissociation rate (%)

Loss (%)

Short circuit

0

28.0

61.0

7.0

Maximum power

0.653

20.2

51.5

24.9

Open circuit

0.846

0

47.4

97.8

Table 3 Analysis of results obtained from TCAD Parameters/materials

Voc (V)

Jsc (mA/cm2 )

Vmax

Imax

FF (%)

η (%)

PPV/PCBM

0.846

28

0.66

20.2

0.56

13.33

24 Investigation and Assessment of TCAD …

287

References 1. Iqbal MA, Gupta SK (2020) Comparative analysis between numerical simulation of PPV/PCBM and InGaN based solar cells. Mater Today. https://doi.org/10.1016/j.matpr.2020. 05.519 2. Iqbal MA, Gupta SK (2020) TCAD based simulation and performance optimization of PPV/PCBM and Perovskite PV cells. Int J Comput Digital Syst 10:2–7 3. Iqbal MA (2014) Analysis & comparison of various control strategy of hybrid power generation, vol 2, no Iconce. IEEE, pp 184–189. https://doi.org/10.1109/ICONCE.2014.6808717 4. Iqbal MA, Dwivedi ADD (2019) Modelling & efficiency analysis of ingap/gaas single junction PV cells with BSF. Int J Eng Adv Technol 8(6):623–627. https://doi.org/10.35940/ijeat.F8081. 088619 5. Iqbal MA, Dwivedi ADD (2019) Efficiency improvement approach of InGaN based PV cell by investigating different optical & electrical properties. SSRN Electron J 1(1):1–9. https://doi. org/10.2139/ssrn.3355989 6. Iqbal MA, Dwivedi ADD (2019) TCAD based simulation & performance optimization of InxGa(1-X)N based PV cell. Glob J Res Eng F Electr Electr Eng 19(4):27–33 7. Iqbal MA, Srivastava G (2019) A review on field programmable gate arrays control based photovoltaic energy management. Iconic Res Eng J 1(9):350–355 8. Iqbal MA, Srivastava G (2018) A review on donor material of bulk heterojunction Polymer(Donor) PV cells. Iconic Res Eng J 1(9):30–33 9. Iqbal MA, Pachori S (2018) Solar energy programs for rural electrification: experiences & lessons from South Asian countries. Int J Trend Sci Res Dev 2(3):860–872. https://doi.org/10. 31142/ijtsrd11049 10. Iqbal MA (2018) Investigations of indium gallium nitride properties for enhancement of performance of PV cells. Int J Creat Res Thoughts 6(1):605–608 11. Iqbal MA, Dwivedi ADD (2017) A comparative study of microgrid load frequency control techniques with incorporation of renewable energy. Published in International Conference on recent innovation and trends in engineering, technology and research (ICRITETR-2017). pp 160–165. ISBN: 978–93–5291–761–7 12. Iqbal MA, Sharma S (2014) Multi-agent model of hybrid energy system. Am Int J Res Sci Technol Eng Math 5(2):164–168

Chapter 25

Torque Ripple Reduction of PMSM Based Electric Vehicle Ashish Kumar Panda, Giribabu Dyanamina, and Rishi Kumar Singh

1 Introduction In the control mechanism of motor drive systems, the inverter transmits the signal and the motor act as a demodulator [1]. 2 level inverters have harmonics and high dv/dt in ac output waveform because the instantaneous error is more. Multilevel inverters are extensively used in motor drive applications due to their superior result compared to two level inverters. Multilevel inverters are emerging technology for high and medium voltage applications because of their ability to generate high quality output waveforms with a low switching frequency. Multilevel inverter topologies mainly classified as cascaded H-bridges inverters, diode clamped Multilevel inverters, and flying capacitors [2]. Diode clamped inverters are widely used with ac drives due to its simple structure. Multilevel inverters reduce the harmonics and voltage stress on switches [3]. In sinusoidal pulse width modulation, the time duration for starting vector and ending vector in a switching cycle are not equal whereas in SVM these vectors are equal duration. In a complete cycle modulation range is from 0 to 1 for sinusoidal pulse width modulation. In SVM it is 0–0.866. So, In SVM, 15% more voltage magnitude is obtained from the same DC bus voltage, and its harmonics performance is also better. So, SVM is widely used over sine-pulse width modulation. SVM is most promising switching algorithms proposed in the literature for multilevel inverter because it offers a great flexibility in optimizing switching pattern design, and it is also well suited for digital implementation. Space vector PWM has recently grown as a very popular PWM method for voltage fed converter ac drives because it offers the advantages of improved PWM quality and extended voltage range in under modulation region. The SVM technique, average voltage produce by reference vector should be the same as average voltage produce by inverter over one switching period. A. K. Panda (B) · G. Dyanamina · R. K. Singh Electrical Department, Maulana Azad NIT, Bhopal, MP, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 R. C. Bansal et al. (eds.), Proceedings of International Conference on Computational Intelligence and Emerging Power System, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-16-4103-9_25

289

290

A. K. Panda et al.

15 percent output voltage is increased by SVM and also reduce the switching number the same as that carrier frequency of the PWM method [4]. Generally, permanent magnet motor and induction motor are most widely used in an electric vehicle application. Induction motor has low efficiency and poor power factor. At low-speed operation induction motor experiences poor performance characteristics and low starting torque. To improve motor performances PMSM has been extensively used in electric vehicle, transport application for motor drives due to its high power density, better reliability, and high torque at low speeds [5, 6]. Various speed control techniques were proposed for the PMSM drives including scalar control [7], vector control [8], and direct torque control [9]. The vector control method is dependent on motor parameters and the requirement of coordinate transformation and current loop, which make the model more complicate [10]. To overcome the disadvantage of vector control a simplified method of direct torque control is proposed. In the closed-loop, conventional DTC receives the torque and flux as the control signal, calculates the sector number, and then selects the voltage vector for VSI, based on a look-up table [11]. Due to the use of flux comparator and torque comparator in the conventional DTC, the waveforms have ripple. SVM-DTC technique is used to overcome the above disadvantages [12]. Conventional DTC and SVM-DTC of PMSM with flux observer based on rotor position is simulated in this paper.

2 Mathematical Model of PMSM System Figure 1 shows that the 3-phase rotating vectors a, b, c can be transformed into 2-phase rotating vectors such as α, β using parks transformation.

Fig. 1 Vector diagram for coordinate transformation

25 Torque Ripple Reduction of PMSM Based Electric Vehicle

291

The vector model equations of stator flux and stator voltage of PMSM are 

Vsα Vsβ



⎛

⎞⎛ ⎞ 1 1 Van − − 2 ⎜ 1 √2 2√ ⎟ ⎝ Vbn ⎠ = ⎝ 3⎠ 3 0 3 − Vcn 2 2

(1)

i sα = i a 1 i sβ = √ (i a + 2i b ) 3

ψsα = (Vsα − Rs i sα )dt

ψsβ = (Vsβ − Rs i sβ )dt

(2)

(3)

Electromagnetic torque equation Te =

3 n p (ψsα i sβ − ψsβ i sα ) 2

(4)

where Vsα , Vsβ are the stator voltage in α-axis and β-axis; isα , isβ are the stator current in α-axis and β-axis;  sα ,  sβ are the stator flux linkage in α-axis and β-axis; np is the motor pole pair; and Rs is the stator resistance.

3 PMSM and VSI Fed Conventional DTC In the DTC of the PMSM drive, inverter terminal voltage and current are used to estimate the flux and torque. Then estimated results are compared with the reference values. Error is passed through torque and flux comparator. The voltage lookup Table 1, which is based on Fig. 2a, gets the signal from the comparator and produces a suitable switching state for the voltage source inverter [13] which is shown in Fig. 2a. The stator flux magnitude and angle are used to identify the sector number which is shown in Fig. 3. The amplitude and phase of the stator flux equation Table 1 Sector one vector selection −1

Flux comparator

1

Torque comparator

1

0

−1

1

0

−1

Vector selection

V2

V0

V6

V3

V7

V5

292

A. K. Panda et al.

Fig. 2 a 2 level VSI b variation of stator flux in the first sector

Fig. 3 Block model of conventional DTC

|ψs | =



2 + ψ2 ψsα sβ   ψsα θ = arcsin ψsβ

(5)

4 PMSM and Multilevel Inverter Fed SVM-DTC 4.1 Space Vector Modulation for 2 Level VSI 2 level voltage source inverter has 2 states in the output either +Vdc /2 or −Vdc /2. So, SVM has 8 (23 ) possible vectors and 6 sectors. Each sector is enclosed by a 3 vector. Three nearest vector includes two non-zero vectors and one zero vector according

25 Torque Ripple Reduction of PMSM Based Electric Vehicle

293

Fig. 4 2 level space vector diagram

to the position of motor rotating reference vector in a different sector. Here starting and ending vectors are zero vectors, which helps for switching transition from one vector to another. The on-time duration for each switch for 2 level VSI is calculated by applying the volt-sec balance principle with the three nearest vectors which is shown in Fig. 4. Suppose the reference vector is in sector 1, by applying volt second balance along alpha axis Vr e f cos(θ )Ts = V1 T1 + V2 cos(600 )Ts

(6)

Vr e f sin(θ )Ts = V2 sin(600 )Ts

(7)



sin 600 − θ T1 = Ts m n sin 600 sin(θ ) T2 = Ts m n sin 600 T0 = Ts − (T1 + T2 )

(8)

along beta axis

Putting V2 in Eq. (7)

V

where m n = 2 Vr e f . 3 dc Similarly switching time in six sectors is calculated as per the Eqs. 6, 7, 8 and generate pulse for inverter switches using MATLAB.

294

A. K. Panda et al.

Fig. 5 Rotating reference vector (a) In 6 Sector with (b) angle

Table 2 Switching time duration for sector one

Sector

Switches Switch 1

1

T1 + T2 +

Switch 3 T0 2

T2 +

T0 2

Switch 5 T0 2

4.2 Space Vector Modulation for 3 Level NPCI 3 level NPCI has a more dense space vector over 2 level VSI which is shown in Fig. 6b. So, it helps to reduce the instantaneous error and harmonic and give more sinusoidal voltage. To minimize the switching always small vectors are used for the

Fig. 6 a Circuit diagram for 3 level NPCI b 3 level space vector diagram

25 Torque Ripple Reduction of PMSM Based Electric Vehicle

295

starting and ending vector in a switching cycle Ts . By manipulating the switching time for the starting and ending vector, torque ripple and current ripple can reduce and also minimize the losses in NPCI. Suppose the resultant vector is in region 2 and the volt second balance equation is Vr e f Ts = V1 ta + V7 tb + V2 tc

(9)

After putting the value of V1 , V2 , V7 and solving the complex form of the equation, the on-time duration is indicated in Fig. 6b 4 ta = Ts − √ m n Ts sin(θ ) 3 π  4 tb = √ m n Ts sin + θ − Ts 3 3 π  4 tc = Ts − √ m n Ts sin −θ 3 3

(10)

V

where m n = 2 Vr e f . 3 dc Similarly, the time duration for each switch for six sectors is designed as per the Eqs. 9 and 10 and generate pulses for switches using MATLAB. Fig. 7 3 level SVM output for 1a, 1b, 1c inverter switches

Table 3 On-Time duration for upper switches in region 2 Reg

Time Switch 1a

2

tc 4

+

ta 4

+

tb 2

Switch2a

Switch1b

Switch2b

Switch 1c

Switch 2c

Ts 2

tc 2

Ts 2

0

tc 2

−

ta 4

+

ta 4

296

A. K. Panda et al.

Fig. 8 Block diagram model of SVM-DTC using multilevel inverter

4.3 Performance of PMSM Using SVM-DTC SVM-DTC does not use the comparator; it reduces the computation time. The switching patterns of SVM are used for controlling the inverter of PMSM instead of the voltage lookup table [14]. Here the estimated value of flux and torque from inverter terminal voltage and current is compared with the reference value. The error is passed to SVM, and the reference voltage from SVM is calculated as polar coordinates Vref (Vsα_ref , Vsβ_ref ) which finally decide the gate pulses for NPCI. To avoid non-linearity between the increment angle  θ and torque, a PI controller is used which is shown in Fig. 8. The reference vector equations for SVM-DTC

  Vα_r e f = ψs∗  cos(θ +  θ ) − |ψs | cos(θ ) /Ts + Rs i sα

  Vβ_r e f = ψ ∗ sin(θ +  θ ) − |ψs |sin(θ ) /Ts + Rs i sβ

(11)

s

Where magnitude and angle of the reference vector, Vs_r e f =



2 2 Vsα_r e f + Vsβ_r e f   Vsα_r e f θ = arctan Vsβ_r e f

(12)

25 Torque Ripple Reduction of PMSM Based Electric Vehicle

297

5 Simulation Results and Discussion The simulation model of conventional DTCand SVM-DTC using 2-L VSI and 3-L NPCI is simulated in MATLAB Simulink software. Comparison study carried out in this helps to identify a good control technique for PMSM drive application for electric vehicle. PMSM parameters are as pole pair (np ) = 2, stator resistance (Rs ) = 1.07 , flux = 0.3 wb, direct axis inductance = 9.9 mH, inertia (J) = 0.0018 kg.m2 . Torque reference is set as 2–3 Nm at 0.5 s and machine rated speed is 300 rad/sec with a constant stator flux magnitude of 0.3 wb shows the MATLAB simulation result of conventional DTC and 3 level NPCI fed SVMDTC. Figure 10 shows the stator current of PMSM using conventional DTC and 3level NPCI fed SVMDTC. Figures. 9b and 10b indicates that using NPC Inverter fed SVM significantly reduces the torque ripple and current ripple. Figure 11 ensures that after changing the electromagnetic torque at 0.5 s machine speed is maintained at 300 rad/sec. Figures 12 and 13 shows the line voltage and THD of 2 level VSI and 3 level NPC inverter with DC-link voltage as 400 V and switching frequency as 2 kHz. Figures (12b and 13b) shows the THD of 2 level VSI and 3 level NPCI with (7.12)% and (5.84)%. Figure 13b shows that THD

Fig. 9 Torque comparison of a conventional VSI fed DTC b SVM NPCI fed DTC

Fig. 10 Current comparison of a conventional VSI fed DTC b SVM NPCI fed DTC

298

Fig. 11 Speed result a conventional VSI fed DTC b SVM NPCI fed DTC

Fig. 12. 2 level Inverter a Line to line voltage and b THD

Fig. 13. 3 level Inverter a Line to line voltage and b THD

A. K. Panda et al.

25 Torque Ripple Reduction of PMSM Based Electric Vehicle

299

is improved by using NPC inverter in place of VSI for SVM direct torque control of PMSM.

6 Conclusion A new nearest three space vector algorithm for 2 level VSI and 3 level NPCI is developed in this article for the direct torque control of a PMSM drive. The proposed DTC using NPCI fed PMSM improves the total harmonics distortion (THD) in the output voltage and decreases the torque and current ripples at the steady-state as well as transient state. SVM ensures that one switching transition occurs from one vector to another. So it is simple and reduces the computational burden on the system. MATLAB simulation result ensures that the SVM-DTC with NPCI improves the performance of the PMSM drive system.

References 1. Jacob B, Baiju MR (2015) A new space vector modulation scheme for multilevel inverters which directly vector quantize the reference space vector. IEEE Trans Indust Electr 62(1):88–95 2. Panda A, Dyanamina G, Singh RK (2021) MATLAB simulation of space vector pulse width modulation for 3-level NPC Inverter and 2-level Inverter. In: 2021 International conference on sustainable energy and future electric transportation (SEFET). Hyderabad, India, pp 1–5 3. Deng Y, Harley RG (2015) Space-vector versus nearest-level pulse width modulation for multilevel converters. IEEE Trans Power Electron 30(6):2962–2974 4. Dyanamina G, Srivastava SP, Pathak MK (2012) Rotor flux-based MRAS for three level inverter fed induction motor drive using fuzzy logic controller. Int J Power Electron 4:463–478 5. Feng G, Qi W, Zhang B, Li C (2011) Analysis and comparison of three-phase variable frequency PMSM with single-phase induction motor in household appliances. In 2011 International Conference on Electrical Machines and Systems. Beijing, pp 1–5 6. Xu Y, Fan Y, Zhong Y (2012) An improved direct torque control method of PMSM based on backstepping control. In: Proceedings of the 7th international power electronics and motion control conference. Harbin, pp 2362–2366 7. Dementyev YN, Kojain NV, Bragin AD, Udut LS (2015) Control system with sinusoidal PWM three-phase inverter with a frequency scalar control of induction motor. In: International Siberian conference on control and communications (SIBCON). Omsk, pp 1–6 8. Mishra A, Makwana JA, Agarwal P, Srivastava SP (2012) Modeling and implementation of vector control for PM synchronous motor drive. IEEE-International conference on advances in engineering, science and management (ICAESM–2012). Nagapattinam, Tamil Nadu, pp 582–585 9. Zhong L, Rahman MF, Hu WY, Lim KW, Rahman MA (1999) A direct torque controller for permanent magnet synchronous motor drives. IEEE Trans Energy Convers 14(3):637–642 10. Dyanamina G, Kumar A (2016) Performance improvement of grid connected DFIG fed by three level diode clamped MLI using vector control. In: 2016 IEEE region 10 conference (TENCON). Singapore, pp 560–565 11. Malla SG, Rao MHL, Malla JMR, Sabat RR, Dadi J, Das MM (2013) SVM-DTC Permanent magnet synchronous motor driven electric vehicle with bidirectional converter. In: 2013 International mutli-conference on automation, computing, communication, control and compressed sensing (iMac4s). Kottayam, pp 742–747

300

A. K. Panda et al.

12. Tang L, Rahman MF (2001) A new direct torque control strategy for flux and torque ripple reduction for induction motors drive by using space vector modulation. In: 2001 IEEE 32nd annual power electronics specialists conference (IEEE Cat. No.01CH37230). Vancouver, BC, pp 1440–1445 13. Wang S, Li C, Che C, Xu D (2018) Direct torque control for 2L-VSI PMSM using switching instant table. IEEE Trans Indust Electron 65(12):9410–9420 14. Li B, Wang C (2016) Comparative analysis on PMSM control system based on SPWM and SVPWM. In: 2016 Chinese control and decision conference (CCDC). Yinchuan, pp 5071–5075

Chapter 26

Design and Implementation of Soft Computing-Based Robust PID Controller for CSTR Rupali R. Gawde, Sharad P. Jadhav, and Bhawana A. Garg

1 Introduction CSTR is a unit where highly exothermic reactions take place and final product is obtained by maintaining a specific temperature. This can be challenging, so various control approaches have been developed with improved versions and applied by changing the operating conditions. PID is one of them and commonly used in process industry for many years. Conventional controllers provide satisfactory results but achieving appropriate control considering variations in operating point and environmental conditions is not possible. Hence, researchers have put enormous efforts to develop controller with intelligent features of GA, PSO, ABC, TLBO fuzzy logic, neural networks to auto-tune parameters of PID and achieve desired results. These algorithms are known as nature-inspired or population-based optimization algorithm. They are employed with population sizes, uncertainties, probabilities, iterations which help find the desired solution for a problem [1]. In optimization techniques, the initial population is a set of candidate solution which creates a new population based on the fitness. The resultant is an optimal solution with efficient performance. The cycle repeats until its true condition is reached. These methods can be categorized in artificial intelligence as Swarm Intelligence and Evolutionary Algorithms [1]. TLBO was first introduced by Professor R. Venkata Rao in 2011. It is a self-learning nature inspired algorithm which uses population size and iterations in a problem for process control. These parameters are also called as design variables for TLBO method. TLBO works on a practical example of impact of teacher on students in a class. The students are considered as population, subjects offered to them are the design variables and the students result is the fitness score

R. R. Gawde (B) · S. P. Jadhav · B. A. Garg Ramrao Adik Institute of Technology, Nerul, Navi Mumbai, Maharashtra, India URL: http://www.dypatil.edu/mumbai/rait/ © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 R. C. Bansal et al. (eds.), Proceedings of International Conference on Computational Intelligence and Emerging Power System, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-16-4103-9_26

301

302

R. R. Gawde et al.

for the given problem. The best solution is formulated which is the optimal value of objective function and is called as teacher [2]. The aim of this research work is to control the temperature of the reactor to control the process satisfactorily. The proposed method improves the transient response, further improving the time response parameters for TLBO–PID with comparative analysis with existing methods like Z–N, GA, PSO, ABC using MATLAB. In simulation results, it is observed that TLBO-based PID controller efficiently reaches the setpoint with less control effort. Robustness analyses show that TLBO–PID performs better in comparison to open loop, Z–N–PID, GA–PID, PSO–PID, ABC–PID despite of uncertainties introduced in the system, proving the controller is robust.

2 Mathematical Model of CSTR CSTR process model helps us understand the dynamic characteristics of the system and design a proper control scheme. Figure 1 shows a chemical reactor in which exothermic chemical reactions takes place. The temperature is regulated by a coolant stream which flows in the jacket in the reactor [1]. The chemical reaction by Arrhenius law is a formula for the temperature dependence of reaction rates. Arrhenius equation gives the impact of rate constant of a chemical reaction on the temperature, an empirical constant and other constants which is illustrated by Eq. 1  −r A = k(T )C A = ko ex p

 −E CA RT

(1)

where r A is the reaction rate of component A, A is the pre-exponential factor constant, k is the rate constant, C A is the concentration of substance A, T is the absolute

cAin

Fig. 1 Chemical reactor

Tin

Q

Mixer

Tjkt

Warming up

V

cA

Product ( Outgoing )

QT

cA

26 Design and Implementation of Soft Computing …

303

temperature in kelvins, E is the activation energy, R is the universal gas constant and ko is the frequency factor. The mass–energy balance is given in Eqs. 2 and 3  Q dC A = C A,in − C A − k(T )C A dt V

(2)

  Q dT hr U Ar  = T A,in − T − k(T )C A − T − T jkt dt V ρC P VρC P

(3)

Energy balance equation for jacket can be expressed as in Eq. 4   dT jkt ρw  U Ar  P = Q cw + Tcw − T jkt + T − T jkt dt Mo Cwater Mo Cwater Mo

(4)

where t is the time, Q is the volumetric flow rate, C is the concentration, T is the temperature, T jkt is the jacket temperature, C P is the specific heat capacities, M O is the overall effective mass of heating/cooling system, V is the reactor’s Volume, Ar is the area, C water is the heat capacity of water, Tcwcw is the temperature of cooling water, P is power input, U is the heat transfer coefficient, hr is the heat of reaction and ρ is the density. A linear model can be obtained based on C A , T and C A,in , T in and can be used to design and describe the relationship between output temperature and the jacket temperature of CSTR. Transfer function of CSTR is given in Eq. 5 2.293 s + 9.172 T (s) = 2 T jkt (s) s + 10.29 s + 25.17

(5)

3 PID Controller Design and Tuning for CSTR PID is tuned using the GA, PSO, ABC, TLBO soft-tuning methods. The output temperature is controlled by PID controller after the input temperature is applied. Difference in the output and the set point is the error which is indicated using MSE and ITAE methods. Figure 2 shows the schematic diagram of a closed-loop CSTR System. PID gains are set and calculated to define the controller response. This is computed by the application of GA, PSO, ABC and TLBO tuning methods. The output of PID controller is given in Eq. 6 t u(t) = K p e(t) + K i

e(t)dt + K d 0

d e(t) dt

(6)

304

R. R. Gawde et al. MSE =

1t 2 te (t)dt t∫ 0

GA/PSO/ABC/TLBO

t

ITAE = ∫ t e(t) dt 0

Kp

Ki Kd

Feed (q, CAo, To ) qC ,Tco

r(t)

e(t)

t d u(t) = K p e(t) + Ki ∫ e(t)dt + K d e(t) dt 0

qc , Tc

Coolant

c(t) u(t) CA , T ,q

Fig. 2 Schematic diagram of closed loop CSTR system

Table 1 Z–N Rule Table

Control type

kp

ki

kd

P

0.50ku

–

–

PI

0.45ku

1.2kp/tu

–

PID

0.60ku

2kp/tu

kpTu/8

where Kp is the proportional gain, Ki is the integral gain, Kd is the derivative gain, t is the time. These parameters of PID need to be set to achieve the optimal output using tuning methods like GA, PSO, ABC and TLBO.

3.1 Ziegler–Nichols-Based PID Ziegler and Nichols employed a procedure to set the values for PID gain: Kp, Ki, Kd. Two parameters, namely, period Tu and gain margin Ku are derived from measurements. ki and kd values are set to zero, Kp is increased until it reaches gain Ku, further the output of the loop starts oscillating. Ku and Tu are associated with values which help set the gains according to the following tuning rule Table 1 [17].

3.2 Genetic Algorithm-Based PID GA is derived from Darwin’s Theory of Survival of Fitness which is based on natural selection. This process is divided into five phases, namely, initial population, fitness function, selection, crossover, mutation. Individual is selected to be parents from the current population depending on the fitness score. Further, crossover and mutation is carried out to children for the next generation. Optimal solution is obtained after a number of iterations, else the cycle repeats [1].

26 Design and Implementation of Soft Computing …

305

3.3 Particle Swarm Optimization-Based PID PSO is inspired by behaviour of flocking of the birds or schooling of the fish that uses the velocity and position to update its particles. The information from the previous particle is used in general back propagation to provide progress in the global optimum point. Selection of the particles is made from the population then the swarm flies around in space in search of food. They fly by their own experience as well as flying experience of other swarms in space. Each swarm follows up with their neighbour to find the best solution which is called personal best or pbest. The best value obtained by the particle from the neighbourhood is called global best or gbest. These values help to evaluate the target error criteria [18].

3.4 Artificial Bee Colony-Based PID ABC is based on the movement of honey bee colonies which can be recorded in three phases: Employed bees, Onlooker bees and Scout bees [19]. Employed bees go to food source and shares information about food source by dancing inside the hive. It discovers a food source in each iteration and evaluates and calculates its nectar amount which is the fitness value. Onlooker bees hold on the dancing area to decide to choose a food source depending on the information collected by employed bees which is based on good quality food. They observe the waggle dance and calculate the profitability. When food is exploited, employed bees become scout bees. Scout bees search for new food source in neighbourhood near their hive.

3.5 Proposed Teaching and Learning-Based Optimization-Based PID Teaching and Learning-Based Optimization is based on the impact of a teacher on the learners in a class. There are two phases: Teacher and Learner. Teacher phase simulates learning through teacher by putting efforts to increase the mean result of the class and bring learners level up in terms of knowledge. Learner phase simulates learning and gain knowledge through interaction amongst themselves to increase their knowledge from the teacher and through interaction between themselves by group discussions, presentations, etc. A learner learns new things if the other learner has more knowledge. The process control through a smart controller can be explained by steps in a sequential order. Figure 3 shows flow chart of CSTR control using intelligent controllers. The objective of the problem is to increase the efficiency and improve the performance. At first, population is created through set of individuals then featured by parameters. The fitness function determines fitness level and is used to reduce the

306

R. R. Gawde et al.

Create initial random population

1t

te 2 (t)dt t0 t ITAE = t e(t)dt 0 MSE =

Evaluate the process using objective function

t d ( ) =u tKpe t( ) + Ki e(t )dt + Kd e(t) dt 0

Select fitness r(t)

e(t)

c(t) u(t)

Yes Termination criteria satisfaction

Best

K KK p

i

d

No

GA * Reproduction * Crossover * Mutation PSO * Aacceleration constant * Inertia weight factor * Velocity matrix ABC * Employed bee * Onlooker bee * Scout stage TLBO * Initialization * Teacher Phase * Learner Phase

Fig. 3 Flow chart of CSTR control using Intelligent controllers

error using MSE and ITAE. Each individual solution like the chromosome, swarm, bee and teacher of the population is decided by objective function which will correspond to the three PID gains to regulate the controller. Hence, the performance of each particle can be evaluated [1]. In TLBO algorithm, learners are the population and the different subjects are the design variables. The mean result of all subjects in the class is the fitness value and solution to the problem [20]. PID gains are determined by the fitness function as given in Eq. 7

26 Design and Implementation of Soft Computing …

307

t |(Tr e f − Tkp,ki,kd )|dt

Min f (kp, ki, kd) =

(7)

0

where kp, ki, kd are the gains of PID, Tref is the set-point value of temperature and TKp;Ki;Kd is the actual temperature of CSTR. The purpose of ITAE is to settle the output at set-point of the system at the earliest possible. Higher overshoot for TKp;Ki;Kd can be minimized by reducing the upper bound value of Kp, increasing the settling time [1]. The optimization algorithms are used to tune PID parameters (Kp, Ki, Kd) using the model offline. Performance index shows the system performance of the designed PID [20]. Hence, the algorithm terminates if population converges thus providing a set of solutions.

4 TLBO Algorithm The steps involved in the working of TLBO algorithm can be summed up as follows: (1) (2) (3)

First, initialize grades (Xold,i) randomly for iteration i of n students which is the population size. Evaluate the objective function. Select the objective function and compute Diffi by using Eq. 8; X new,i = X old,i + Di f f i

(4)

Compute Xnew,i for n students and iteration i by using Eq. 9; Di f f i = ri (X T eacher − TF Mi )

(5) (6)

(9)

Compare Xold,i with Xnew,i. The better one goes to next step. For i = 1: Pn, randomly select two learners X i and X j , where i = j. Then new solution is computed for every student by using Eq. 10;  X new,i =

(7) (8)

(8)

X old,i + ri (X i − X j )if X i > X j X old,i + ri (X j − X i )if X j > X i

(10)

Accept Xnew whichever gives a better function value, goes to next step. Evaluate the objective function for n students and check if the stop criteria is satisfied. Else, go back to Step (3). Figure 4 shows the steps involved in system identification.

308

R. R. Gawde et al.

Fig. 4 Flowchart of TLBO

5 Comparison of TLBO with Other Intelligent Optimization Techniques The following are the difference between TLBO and the different algorithms like GA, PSO, ABC: (1)

Like GA, PSO and ABC; TLBO is a population-based optimization tool which carries out operations in each iteration to find optimal solution.

26 Design and Implementation of Soft Computing …

(2)

309

These algorithms require parameters that affect the performance of the system. GA requires the crossover, mutation and selection method; PSO requires learning, weight and velocity; and ABC requires the limit value. TLBO does not require any algorithm parameters to be tuned. Like in PSO, TLBO uses the best solution to change the existing solution which increases the convergence rate. Population is not divided in TLBO like in ABC. GA uses selection, crossover and mutation and ABC uses employed, onlooker and scout bees; similarly, TLBO uses two different phases, the teacher phase and the learner phase. TLBO implements greediness to accept a good solution like in ABC.

(3) (4) (5)

(6)

6 Simulation Results This section describes the time response and frequency response analysis results obtained by simulation.

6.1 Time Response Analysis The performance of the proposed TLBO-based PID controller is tested using MATLAB by the following: – – – – –

Design Variables for Graphical Analysis Comparative Performance Analysis Error Indicators Control Effort Robustness Analysis.

6.1.1

Design Variables for Graphical Analysis

Parameters of CSTR are CA = 0.114 lbmol/ft raise to 3, TJ = 100 °F, T = 40 °F. Initial parameters or design variables for graphical analysis are as shown in Tables 2, 3, 4, and 5. These parameters are used in the algorithm for computing. Table 6 shows the PID controller design parameters obtained by various tuning methods. Table 2 Design variables of GA–PID

No. of population

50

No. of iteration

100

Crossover operator

Arithmetic

No. of variables to be optimization

03

310

R. R. Gawde et al.

Table 3 Design variables of PSO–PID Particle size

30

No. of iteration

100

Weight Function

Arithmetic

Acceleration constants C1 and C2

0.012

Dimension of the search-space

03

Table 4 Design variables of ABC–PID No. of colony size (Np)

20

No. of food sources (S = Np/2)

10

No. of iteration

100

Table 5 Design variables of TLBO–PID Population size

50

No. of iteration

30

Table 6 PID design variables of GA–PID, PSO–PID, ABC–PID and TLBO–PID Parameters

Z–N–PID

Kp

5

9.992

Ki

50

Kd

0.5

6.1.2

GA–PID

PSO–PID

ABC–PID

TLBO–PID

10

10

23.8872

99.9787

85.764

71.8911

150

0.0115

0.0101

0.01

0.01

Comparative Performance Analysis

The performance has been analysed for system with open loop, Z–N–PID, GA–PID, PSO–PID and ABC–PID and TLBO–PID, Fig. 5 illustrates the time response graphs. The initial temperature is taken as 40 °F and the set-point temperature is 100 °F. Table 7 depicts that the open loop system does not reach to the set-point temperature, while conventional Z–N controller reaches to desired value in 0.7580 s. The response is improved with the implementation of smart controllers. Z–N–PID controller shows stability at 0.75 s; GA–PID, PSO–PID, ABC–PID and TLBO–PID shows steady state at 0.5 s. The response gets better with the addition of the Z–N– PID, GA–PID, PSO–PID and ABC–PID. The settling time for TLBO–PID is 0.0732 s which is the comparatively the least, thus improving the system performance.

26 Design and Implementation of Soft Computing …

311

120 Open Loop ZN−PID GA−PID PSO−PID ABC−PID TLBO−PID

Amplitude

100

80

60

40

20

0 0

0.5

1

Time

1.5

Fig. 5 Comparative time response analysis of CSTR performance using different control strategies

Table 7 Comparison of different methods with time response parameters Response parameters

Without controller

Z–N–PID

GA–PID

PSO–PID

ABC–PID

TLBO-PID

Rise time (s)

0.3489

0.2314

0.0753

0.0817

0.0905

0.0411

Overshoot%

0.0000

5.5544

4.4651

2.7155

0.9648

0.0016

Peak time (s)

1.5000

0.4790

0.1820

0.2050

0.2520

0.2330

Settling time (s)

0.6203

0.7580

0.3050

0.2812

0.1429

0.0732

6.1.3

Error Indicators

The error indices, namely, integral mean square error and ITAE is measured to minimize the error to obtain the best results. Table 8 shows the values of error obtained for different controllers. TLBO-based PID controller has the least value of error index amongst the other controllers. Error index in Z–N–PID is 1.7801 and in TLBO–PID is 0.0343. The error is reduced to zero after 0.5 s. It is observed that TLBO–PID gives comparatively minimum error than the other methods. Table 8 Comparison of different methods with error indicators Error

Z–N–PID

GA–PID

PSO–PID

ABC–PID

TLBO–PID

MSE

152.8729

134.0125

137.5606

142.0142

62.8431

ITAE

1.7801

0.2932

0.2732

0.2278

0.0343

312

R. R. Gawde et al.

Table 9 Control efforts for various tuning methods

6.1.4

PID tuning methods

1-Norm

2-Norm

∞-Norm

Z–N–PID

8.88e+03

479.0222

46.5875

GA–PID

4.36e+03

448.5006

97.4308

PSO–PID

4.33e+03

454.399

97.7365

ABC–PID

4.31e+03

461.6961

97.7584

TLBO–PID

1.88e+03

307.1277

97.7584

Control Effort

Minimizing the effort of the controller is important in control theory. It is the measure of consumption of energy of the controller. Therefore, reducing the control efforts increases the functional safety of the controlled process. It is understood as controller output. The index shows the output–input signal energy effectiveness of the controller [22]. The commonly used measures of error are the norms such as the 1-, 2- and ∞norm. Table 9 indicates the values of 1-Norm, 2-Norm and ∞-Norm for all the controller tuning techniques. It shows that the TLBO-tuned PID controller shows good results for 1-Norm and 2- Norm with values of 1.88e+03 and 307.1277 and 8.88e+03 and 479.0222 for Z–N method, respectively. This shows that TLBO-based PID controller is more efficient and requires very less effort to settle to the set-point value.

6.1.5

Robustness Analysis

Robustness analysis tests the ability of the system to respond to uncertainties if induced in the parameters. It is an operational measure to analyse the flexibility of the system. This will lead to useful future decision choice while designing and modelling. Irrespective of variations in input, output shall vary within allowed tolerance, this guarantees the maximum value of the sensitivity function [21]. The uncertainties in CSTR system are observed by a deviation of ±10% in the values of the coefficients of TF. The response in Figs. 6 and 7 show that the variation in time response parameters is minimal. Table 10 shows that the variation of +10% records settling time of 0.7108 s for Z–N–PID and 0.1266 s for TLBO–PID. For variation of −10% the settling time for Z–N–PID is 1.1729 s and for TLBO–PID it is 0.2148 s. The peak overshoot for Z–N–PID is 8.3055% and for TLBO–PID, it is 2.9561%. Hence, TLBO–PID method gives better performance with no overshoot and faster settling time when compared with open loop, Z–N–PID, GA–PID, PSO–PID, ABC– PID. The time response parameters values are within the tolerable limits. Hence, it can be concluded that the proposed TLBO–PID is robust and can perform adequately under the critical changes in the system parameters thus keeping the system stable.

26 Design and Implementation of Soft Computing …

313

120 Open Loop Conventional GA PSO ABC TLBO

Amplitude

100

80

60

40

20

0 0

0.5

Time

1

1.5

Fig. 6 Comparative frequency response analysis of CSTR performance using different control strategies for +10% 120 Open Loop Conventional GA PSO ABC TLBO

Amplitude

100

80

60

40

20

0 0

0.5

Time

1

1.5

Fig. 7 Comparative frequency response analysis of CSTR performance using different control strategies for −10%

6.2 Frequency Response Analysis To examine the stability of the proposed control system, Bode plot has been used. The frequency of the input signal is changed and output is sinusoid of the similar

314

R. R. Gawde et al.

Table 10 System response for ±10% change in system parameters Response parameters

% Change

Rise time (s)

Settling time (s)

Overshoot (%)

Without controller

+10

0.2023

0.3400

0.3276

0.6450

−10

0.4040

0.6367

0.2854

1.0210

Z–N controller

+10

0.2263

0.7108

2.6697

0.4790

−10

0.2300

1.1729

8.3055

0.4790

GA–PID

+10

0.0527

0.0964

0.0000

1.5000

−10

0.0901

0.3819

9.8829

0.2080

+10

0.0573

0.1408

0.0000

1.5000

−10

0.0971

0.3981

7.6484

0.2240

ABC–PID

+10

0.0645

0.2334

0.0000

1.5000

−10

0.1062

0.4018

5.2099

0.2460

TLBO–PID

+10

0.0244

0.1266

0.0000

1.5000

−10

0.0534

0.2148

2.9561

0.1430

PSO–PID

Peak time (s)

Phase (deg)

Magnitude (dB)

frequency but with a different magnitude and phase. Bode plot provides exact values for process with time delays. The graph of frequency response of CSTR performance with different control techniques is as shown in Fig. 8. It indicates that the phase never passes −180z , so the gain margin is ∞ db for all the methods. An infinite gain margin in the graph Bode Diagram

10 5 0 −5 −10 −15 −20 −25 −30 −35 −40 0

ZN−PID GA−PID PSO−PID ABC−PID TLBO−PID

−30 −60 −90 −1 10

10

0

10

1

10

2

10

3

10

4

10

5

Frequency (rad/s)

Fig. 8 Comparative frequency response analysis of CSTR performance using different control strategies

26 Design and Implementation of Soft Computing …

315

Table 11 Comparison of different methods with frequency response parameters Response parameters

Z–N–PID

GA–PID

PSO–PID

ABC–PID

TLBO–PID

Phase margin (deg)

166

156

163

178

−180

Delay (s)

0.778

0.244

0.36

2.73

∞

At frequency (rad/s)

3.71

11.1

7.89

1.13

0

Closed-loop stable

Yes

Yes

Yes

Yes

Yes

Gain margin

∞

means that the system will never become unstable anyhow in future, even if the gain is kept on increasing. Or the margin to reach the verge of instability is infinity. Table 11 shows the comparison of different methods with frequency response parameters. Hence, it can be said that the proposed TLBO–PID provides the highphase margin and hence provides better stable operating area as compared to other algorithms.

7 Conclusion An adaptive mechanism to control the reactor temperature has been proposed. Tuned control parameters are obtained to maintain the output temperature of CSTR at given set temperature. Time response specification prove that the TLBO–PID provides better response when compared to open loop, Z–N–PID, PSO–PID and ABC–PID. The dynamic behaviour of the system is enhanced using a mathematical formulation by the error indicators MSE/ITAE. Control efforts show that TLBO-based PID controller is more efficient and requires less effort to reach set-point value. Robustness analysis show that the TLBO–PID is robust and can perform satisfactorily despite of variation in the system parameters. Stability analysis using bode plot proves that the system is stable. TLBO never claimed to be the best algorithm amongst all optimization algorithms available. There may not be any such best algorithm as it all depends on the application and suitable algorithm for specific reason. TLBO has proved to be capable to solve the problem. In case an algorithm is found having drawbacks, then efforts are put to find ways to overcome and strengthen the algorithm. It can be said that TLBO is simple to apply, it has no algorithm-specific parameters, and it provides the optimum results in a comparatively less evaluations. It has established itself and has set itself distinct amongst all the algorithms. Researchers are encouraged to make improvements so that the algorithm will become more powerful with improved performance.

316

R. R. Gawde et al.

References 1. Goud H, Swarnkar P (2019) Investigations on metaheuristic algorithm for designing adaptive PID controller for continuous stirred tank reactor. Mapan 34(1):113–119 2. Rao RV, Savsani VJ, Vakharia DP (2011) Teaching–learning-based optimization: a novel method for constrained mechanical design optimization problems. Comput Aided Des 43(3):303–315 3. Kuniel V et al (2018) Survey on basic control schemes for continuous stirred tank system. In: 2018 international conference on control, power, communication and computing technologies (ICCPCCT). IEEE 4. Russo LP, Bequette BW (1993) CSTR performance limitations due to cooling jacket dynamics. In: Dynamics and control of chemical reactors, distillation columns and batch processes, Pergamon, pp 149–154 5. Ponnusamy TKL, Kirubagaran R, Atmanand MA (2013) Design and optimization of temperature controller for high pressure rated modified CSTR system. In: 2013 IEEE international conference on signal processing, computing and control (ISPCC), Solan, pp 1–6. https://doi. org/10.1109/ISPCC.2013.6663462 6. Hambali N, Masngut A, Ishak AA, Janin Z (2014) Process controllability for flow control system using Ziegler-Nichols (ZN), Cohen-Coon (CC) and Chien-Hrones-Reswick (CHR) tuning methods. In: 2014 IEEE international conference on smart instrumentation, measurement and applications (ICSIMA), Kuala Lumpur, pp 1–6. https://doi.org/10.1109/ICSIMA. 2014.7047432 7. Holland JH (1984) Genetic algorithms and adaptation. In: Adaptive control of III-defined systems, pp 317–333 8. Shi Y (2004) Particle swarm optimization. IEEE Connect 2(1):8–13 9. Karaboga D, Basturk B (2007) Artificial Bee Colony (ABC) optimization algorithm for solving constrained optimization problems. In: Foundations of fuzzy logic and soft computing, pp 789–798 10. Singh A, Sharma V (2013) Concentration control of CSTR through fractional order PID controller by using soft techniques. In: 2013 fourth international conference on computing, communications and networking technologies (ICC-CNT). IEEE 11. Sahed OA, Kara K, Benyoucef A (2015) Artificial bee colony-based predictive control for non-linear systems. Trans Inst Meas Control 37(6):780–792 12. Wahab A, Nadhir M, Nefti-Meziani S, Atyabi A (2015) A comprehensive review of swarm optimization algorithms. PloS One 10.5:e0122827 13. Sahu BK et al (2015) Teaching–learning based optimization algorithm based fuzzy-PID controller for automatic generation control of multi-area power system. Appl Soft Compu 27:240–249 14. Sahu BK et al (2016) A novel hybrid LUS–TLBO optimized fuzzy-PID controller for load frequency control of multi-source power system. Int J Electr Power Energy Syst 74:58–69 15. Chatterjee S, Mukherjee V (2016) PID controller for automatic voltage regulator using teaching–learning based optimization technique. Int J Electr Power Energy Syst 77:418–429 16. Aljaifi T et al (2019) Applying genetic algorithm to optimize the PID controller parameters for an effective automatic voltage regulator. Commun Comput Appl Math 1(2) 17. Sen R et al (2015) Comparison between three tuning methods of PID control for high precision positioning stage. MAPAN 30(1):65–70 18. Koyuncu H, Ceylan R (2019) A PSO based approach: scout particle swarm algorithm for continuous global optimization problems. J Comput Des Eng 6(2):129–142 19. Ali RS, Aldair AA, Almousawi AK (2014) Design an optimal PID controller using artificial bee colony and genetic algorithm for autonomous mobile robot. Int J Comput Appl 100(16):8–16 20. Kumar KS, Samuel RH (2015) Teaching learning based optimization

26 Design and Implementation of Soft Computing …

317

21. Jumani TA et al (2020) Jaya optimization algorithm for transient response and stability enhancement of a fractional-order PID based automatic voltage regulator system. Alex Eng J 22. Bartys M, Hryniewicki B (2019) The trade-off between the controller effort and control quality on example of an electro-pneumatic final control element. In: Actuators, vol 8, no. 1. Multidisciplinary Digital Publishing Institute

Chapter 27

A Double-SOGI-Based Power Quality Improvement for a Weak-Grid-Connected PV System Sanjay Kumar Peeploda, Tripurari Nath Gupta, and Mahiraj Singh Rawat

1 Introduction In the past few decades, the utilization of RES (Renewable Energy Sources) expanding at a quicker rate. Petroleum derivatives have been for some time utilized for outfitting energy, yet they won’t keep going forever, as conventional sources will disappear after some decades [1]. In this way, the best substitute approach is to utilize the regular assets. In India, the utilization of RES, for example, Wind energy generation, Solar Photovoltaic generation, Hydropower generation systems, etc. became more popular in past few decades especially, the solar photovoltaic generation system because of appropriate climate conditions [2, 3]. The perfect energy age is a significant concern in view of the persistently increment of carbon content in climate, so to get freed out of this issue renewable energy sources are gotten extremely well known [2]. There are many MPP tracking techniques to extract optimal power from solar PV arrays such as InC (Incremental Conductance), Hill Climbing or P&O (Perturb and Observe), Reference cell, Voltage sampling, etc. techniques are used. The rate of accuracy, feasibility, and response of the InC technique is much better than other MPPT techniques like Perturb and Observe [5]. The biggest advantage of InC MPPT is that it finds out the perturbation direction in an exact way for the operating array voltage [3–5]. InC technique proves to be the most accurate and widely used technique. The InC MPPT technique is utilized to estimate reference voltage and that voltage is then utilized to evaluate the duty cycle of the boost converter. In this work, a double-SOGI-based control technique is applied to extract power frequency components from the grid current to make the system more efficient and robust. Nonlinear loads connected at CPI (Common Point of Interconnection) S. K. Peeploda · T. N. Gupta (B) · M. S. Rawat (B) National Institute of Technology, Uttarakhand, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 R. C. Bansal et al. (eds.), Proceedings of International Conference on Computational Intelligence and Emerging Power System, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-16-4103-9_27

319

320

S. K. Peeploda et al.

demands reactive power which leads to harmonics and distortion in the grid current and the same will reflect on the power factor of the system [6, 7]. To get rid of the above-mentioned problem, PLL-less double-SOGI-based control algorithm is applied in this study. The PI controller is utilized to get the converter loss component of the current in the system. At CPI, to maintain a constant voltage during abnormal grid conditions, a DC link voltage control is applied. DC link control also helps in reducing harmonics content in injected current to the grid and to reduce switching losses of VSC [7].

2 System Topology In this work, the system consists of a solar PV array, boost converter (LB, SB, Diode), DC link capacitor (Cdc), voltage source converter (VSC), interfacing inductor (Lf), ripple filter (Rf, Cf), nonlinear load, and the three-phase weak AC grid having impedance (Za, Zb, Zc) as shown in Fig. 1. The weak AC grid is considered in this system which is contaminated by lower order harmonics. Both converters are connected through the DC link of VSC. Incremental conductance MPP tracking technique is utilized to extract optimal power from the solar PV array. The output MPP voltage is utilized for switching to the boost converter. The PV array is designed by series and parallel connected units according to requirement at the load side. Interfacing inductor is used to eliminate ripples in the injected grid current. The higher order harmonics get introduced in injected current at CPI due to high-frequency switching. To eliminate higher order harmonics from the injected grid current at CPI, a ripple filter is used [6, 7].

Fig. 1 Schematic of system under study

27 A Double-SOGI-Based Power Quality Improvement …

321

3 Control Algorithm In this work, a weak AC grid is considered which contains lower order harmonics such as fifth and seventh harmonics by which THD of the grid voltage has become 9.43%. Due to the distortion in the grid voltage harmonics also gets introduced in the grid current which makes the grid current distorted. In this control algorithm, a double-SOGI filter is used to filtered out harmonics existing in the grid voltage to get proper reference waveform of the grid current.

3.1 MPPT Technique To achieve the optimal power from the solar PV array, in this system InC technique is used. The rate of accuracy, feasibility, and response of the InC (Incremental conductance) technique is much better than other MPPT techniques. The biggest advantage of the InC MPP technique is that it finds out the perturbation direction in an exact way for the operating PV array voltage [4, 5]. The InC is an accurate and widely used technique. Equations (1–4) shown represents how the InC tracks the maximum power point, PP V = V P V ∗ I P V

(1)

  I P V PP V = I P V + VP V V P V V P V

(2)

At maximum power point, PP V =0 V P V

(3)

I P V IPV =− V P V VP V

(4)

The ΔV PV and ΔI PV can be determined by a sampled value taken from solar PV array by the help of Eqs. (5)–(6) V P V = V P V (n) − V P V (n − 1)

(5)

I P V = I P V (n) − I P V (n − 1)

(6)

where I PV , V pv , PPV are PV array current, voltage, and power, respectively.

322

S. K. Peeploda et al.

Fig. 2 Structure of double-SOGI filter

3.2 Structure of Double-SOGI In second-order generalized integrator, two outputs are obtained, one is an in-phase fundamental component and another one is a quadrature component [7]. In Fig. 2, a schematic of a double-SOGI is shown, in which two SOGI are in cascade connection [7, 8]. The fundamental component of the first SOGI is given to the second SOGI filter as input which helps to get a better waveform of the fundamental component of the distorted signal. The THD of signal gets reduced in double-SOGI as compared to single SOGI used, where ξ represents the damping ratio.

3.3 Grid-Interfaced Voltage Source Converter The VSC in this work performs several functions such as harmonics mitigation of the grid current, feeding power to grid, maintaining power factor at unity, and reactive power compensation [7, 8]. The VSC current indirectly controls the above-mentioned functions. The switching losses of VSC mainly depend on the DC link voltage as shown in Fig. 3a. So, by controlling the DC link voltage, it is possible to minimize the switching losses of VSC. The switching given to the boost converter is achieved by comparing the carrier signal with a difference of V PV and V PVRef as shown in Fig. 4. The phase voltage is determined by sensed line voltage and the synchronization signal [9] is calculated by using phase voltages. So, the amplitude of the voltage at CPI can be estimated as   2 (7) Vz = ∗ (Vsa2 + Vsb2 + Vsc2 ) 3 The PV array voltage and current are utilized to estimate photovoltaic feedforward term (PVFF), which helps in achieving reference grid current (I gRef ) as shown in Fig. 3a. The feed-forward term represents the amplitude of the grid current for lossless ideal system and at no-load at the common point of interconnection (CPI) [9, 10]. PVFF term used for a better dynamic response during the variation in CPI voltage and during a sudden change in the solar PV power. The PVFF term is calculated by the following equation:

27 A Double-SOGI-Based Power Quality Improvement …

323

(a)

(b) Fig. 3 a Control algorithm of VSC b Estimation of magnitude of CPI voltage and unit vectors

Fig. 4 Switching control of boost converter

IPV f f =

2 PP V ∗ 3 Vz

(8)

The set of unit vectors as shown in Fig. 3b are calculated to synchronize the output current of VSC [11]. The unit vectors help in estimating the value of reference grid current (I gRef ). That reference value (I gRef ) gets equated with sensed grid current, the resultant will pass through the current hysteresis controller to get switching pulses for VSC [12, 13]. The unit vectors are calculated with the help of the following equations:

324

S. K. Peeploda et al.

za =

Vsa Vz

(9)

zb =

Vsb Vz

(10)

zc =

Vsc Vz

(11)

where V z is the amplitude of CPI voltage which contains information of all three phases, za , zb , zc are the unit vectors of three phases, respectively. I pvff is PV feed-forward term, Ppv PV array Power, V sa , V sb , V sc are the phase voltages of three-phase grid, respectively. The reference grid current evaluated by I g multiplied with unit vector [12, 13], where I g is Ig = Iloss − I P V f f

(12)

where I loss is the current loss component and I PVff is achieved by calculating feedforward term.

4 Simulation Results Figures 5, 6, 7, 8, 9, and 10 shows the simulation results under varying conditions. The PV array considered is approximately 15 KW at 1000 W/m2 and the connected load

Fig. 5 Performance under the change in irradiance (1000 W/m2 to 800 W/m2 ), variations in (a). Irradiance (W/m2 ), Vgrid (V), Igrid (A), Iload (A), (b). Irradiance (W/m2 ), VPV (V), IPV (A), PPV (W)

27 A Double-SOGI-Based Power Quality Improvement …

325

Fig. 6 Performance under the change in irradiance (800 W/m2 –1000 W/m2 ), variations in (a). Irradiance (W/m2 ), Vgrid (V), Igrid (A), Iload (A), (b). Irradiance (W/m2 ), VPV (V), IPV (A), PPV (W)

is 6.5 KW. The switching frequency selected is 10 kHz. The results are categorized into two portions, one is taken as a change in irradiance and another one is a change in load. The simulation results are taken under non-ideal conditions. Non-linear load is connected at CPI which makes the grid injected currently distorted. A weak AC grid is taken for simulation which is contaminated by lower order harmonics, which makes THD of the grid voltage 9.43% as shown in Fig. 7a. The double-SOGI filter is used to filter out harmonics from the grid voltage to get proper reference current.

4.1 Performance Under Change in Solar Irradiance (1000 W/m2 –800 W/m2 ) Figure 5 shows performance under sudden variation in the solar insolation from 1000 W/m2 to 800 W/m2 in presence of load at CPI. The system is working in a steady state at 1000 W/m2 till t = 1 s. After t = 1 s, the irradiance was reduced to 800 W/m2 . By change in solar insolation, solar PV power gets reduced due to which grid current also decreases. The connected load current will be the same as before. After change also, the injected grid current has a sinusoidal waveform, whereas the load withdraws a non-sinusoid current. The grid current THD is 1.87% at 1000 W/m2 and 2.81% at 800 W/m2 , which are within IEEE-519 standard as shown in Fig. 7b, c. The PV array parameters also get stable in very little time during the change of irradiance as shown in Fig. 5b.

326

S. K. Peeploda et al.

Fig. 7 THD analysis of, a Grid voltage, b Grid current at irradiance 1000 W/m2 , c Grid current at irradiance 800 W/m2

27 A Double-SOGI-Based Power Quality Improvement …

327

Fig. 8 Performance at irradiance 0 W/m2 in presence of load at CPI, variations in (a). Irradiance (W/m2 ), Vgrid (V), Igrid (A), Iload (A), (b). Irradiance (W/m2 ), VPV (V), IPV (A), PPV (W)

Fig. 9 Performance at Irradiance 1000 W/m2 without load, variations in (a). Irradiance (W/m2 ), Vgrid (V), Igrid (A), Iload (A), (b). Irradiance (W/m2 ), VPV (V), IPV (A), PPV (W)

4.2 Performance Under Change in Solar Irradiance (800–1000 W/m2 ) Figure 6 shows the performance under a change in the solar PV irradiance (800– 1000 W/m2 ) in presence of load at CPI. After t = 1 s, the irradiance changes to 1000 W/m2 . After a change in irradiance, the injected grid current will increase, and load current will remain the same as before. The grid current will be sinusoidal after a change in irradiance also even when VSC current is non-sinusoidal that presents the satisfactory work done by VSC control. The THD observed is 2.81% at 800 W/m2 and 1.87% at 1000 W/m2 of the injected grid current, which satisfied the IEEE-519

328

S. K. Peeploda et al.

Fig. 10 THD analysis of, a Grid current at irradiance 0 W/m2 , b Grid current at irradiance 1000 W/m2 without load at CPI

standard as shown in Fig. 7b, c which shows the system works in a better way under dynamic conditions also.

4.3 Performance at 0 W/m2 Solar Irradiance (DSTATCOM Mode) Figure 8 shows the performance at 0 W/m2 . At no irradiance, the grid will feed power to the load. In this condition also, distorted grid voltage will supply sinusoidal grid current which shows the robustness of the double-SOGI based control technique. The PV parameters do not show much variation during DSTATCOM mode also as shown in Fig. 8. The THD of the grid current is observed as 3.12% as shown in Fig. 10a, which satisfies the criteria stated in IEEE-519 standard.

4.4 Performance Under No-Load Condition Figure 9 shows the performance under the no-load condition at 1000 W/m2 . At noload, whole solar PV power will be delivered to the AC grid which results in increase in the grid current. The THD of injected grid current was observed as 1.58% as

27 A Double-SOGI-Based Power Quality Improvement …

329

shown in Fig. 10b, which satisfies the criteria stated in IEEE-519 standard. The PV parameters also get intact at the removal of load and do not show much variations during a change in load.

5 Conclusion By taking reference of above simulation results, it is proved that the system shown in this paper performs satisfactorily. The system works under different loading and change in irradiance conditions, which shows the robustness of the system and gives better results under the abnormal grid condition. The power factor is also the main concern while feeding to the grid and in this system, it is maintained close to the unity power factor. The system has performed multifunctional tasks for power quality improvement during feeding power to the grid-like harmonics mitigation, power factor correction, reactive power compensation under different operating conditions. The THD of the injected grid current is observed by simulation results is less than 5% in all conditions which satisfy the criteria stated in the IEEE-519 standard.

References 1. Gupta TN, Murshid S, Singh B (2018) Single-phase grid interfaced WEGS using frequency adaptive notch filter for power quality improvement. In: International conference on power electronics, intelligent control and energy systems (ICPEICES), IEEE, pp 630–635, Delhi, India 2. Patel N, Gupta N, Babu BC (2020) Photovoltaic system operation as DSTATCOM for power quality improvement employing active current control. IET Gener Transm Distrib 14:3518– 3529 3. Deo S, Jain C, Singh B (2015) A PLL-Less scheme for single-phase grid interfaced load compensating solar PV generation system. IEEE Trans Indus Inform 11(3):692–699 4. Jain C, Singh B (2014) A SOGI-FLL based control algorithm for single phase grid interfaced multifunctional SPV under non-ideal distribution system. In: 2014 Annual IEEE India conference (INDICON), IEEE, pp 1–6 5. Khadidija S, Mountassar M, M’hamed B (2017) Comparative study of incremental conductance and perturb & observe MPPT methods for photovoltaic system. In: International conference on green energy conversion systems (GECS), IEEE, pp 1–6, Hammamet 6. Gupta TN, Murshid S, Singh B (2018) Modified notch filter based control for wind energy generation system integrated to single-phase weak grid. In: Uttar Pradesh section international conference on electrical, electronics and computer engineering (UPCON), IEEE, pp 1–6, Gorakhpur 7. Gupta S, Verma A, Singh B, Garg R, Singh A (2020) Power quality improvement of PVWPGS based grid interactive microgrid using MSWL-CLMS control. IET Energy Syst Integr 2:362–372 8. Jain C, Singh B (2017) An adjustable DC link voltage-based control of multifunctional grid interfaced solar PV system. J Emerg Selec Topics Power Electron IEEE 5(2):651–660 9. Singh B, Jain C (2017) A decoupled adaptive noise detection based control approach for a grid supportive SPV system. Trans Indus Appl IEEE 53(5):4894–4902

330

S. K. Peeploda et al.

10. Gupta TN, Murshid S, Singh B (2018) Power quality improvement of single-phase grid connected hybrid solar pv and wind system. In: 8th power India international conference (PIICON), IEEE, pp 1–6, Kurukshetra 11. Gupta TN, Murshid S, Singh B (2018) Single-phase grid interfaced hybrid solar PV and wind system using STF-FLL for power quality improvement. In: 8th India international conference on power electronics (IICPE), IEEE, pp 1–6, Jaipur, India 12. Singh Y, Singh B, Mishra S (2021) Photovoltaic-battery powered grid connected system using multi-structural adaptive circular noise estimation control. IET Power Electron 14:397–411 13. Nirmal Mukundan C, Jayaprakash P (2020) DSOGI with proportional resonance controlled CHB inverter based two-stage exalted photovoltaic integration in power system with power quality enhancement. IET Renew Power Gener 14:3126–3137

Chapter 28

Multi-machine Power System Stabilizer Design Using Grey Wolf Optimization Ravi Kant Sharma, Dhanraj Chitara, Shashi Raj, K. R. Niazi, and Anil Swarnkar

1 Introduction In recent years, the problem of low frequency oscillations in interconnected power system is observed due to insufficient damping available. These small magnitudes low frequency oscillations may sustain and raises, causes system separation and affect the power transfer capability if adequate damping is not provided. These changes in the system may be due to tripping of line or generator, variations in load, etc. In multi-machine power system, these oscillations are broadly classified in to two types [1]. The oscillations among the same area generators are known as local area modes while any generator in one area oscillates with respect to other area generator are known as inter-area modes of oscillations. In case of interconnected complex power system, the problem of inter-area oscillations is more dangerous. It causes the complete system collapse due to cascading failure of power system components [2]. Power System Stabilizers (PSSs) are frequently employed in the industry to damp out such oscillations of the system and enhance the damping performance and the small-signal stability [1]. A proper selection of PSS parameters results in satisfactory performance during the system disturbance. For a lot of decades, the Conventional PSSs (CPSS) is one of the most effective damping controllers [2, 3]. The designing of PSS has been focused on lot of papers in the literature. In earlier stage, the CPSS design based on linearity principal of control theory [4], modern control approach [5], digital control method [6], etc. are used for small-signal stability enhancement in the multi-machine power system. The linearized model of complex R. K. Sharma (B) · D. Chitara · S. Raj Department of Electrical Engineering, SKIT, Jaipur, Rajastan, India D. Chitara e-mail: [email protected] K. R. Niazi · A. Swarnkar Department of Electrical Engineering, MNIT, Jaipur, Rajastan, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 R. C. Bansal et al. (eds.), Proceedings of International Conference on Computational Intelligence and Emerging Power System, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-16-4103-9_28

331

332

R. K. Sharma et al.

power system has been efficiently used to analyse low frequency oscillations and the tuning of CPSS for specific operating conditions of the system. As the operating condition changes under wide range, the CPSS design results are not satisfactory [2, 3]. Other type of designing procedure of PSS uses the optimization techniques. In this process, the main objective is to simultaneously control the real part of eigenvalues and its damping ratio. According to the literature, the optimization techniques can be classified in to two main categories: bio-inspired techniques and mathematical programming. Earlier, many bio-inspired meta-heuristic techniques, e. g., Genetic Algorithm (GA) [7– 9], simulated annealing [10], particle swarm optimization [11], tabu search [12], harmony search optimization [13], bat algorithm [14], rule based bacteria foraging [15], cuckoo search optimization [16–19], ant colony optimization [20], bee colony optimization [21], etc. have been tested to design the PSS parameters using eigenvalue based objective function for controlling damping factor and/or damping ratio of standard IEEE based multi-machine test systems. In order to take more benefits of meta-heuristic techniques, hybrid techniques have been also proposed [22, 23]. Although, bio-inspired optimization techniques required more run time, slow convergence, depending on the size of the system under case study but they are off-line and free from mathematical modelling of large system. In this paper, a novel bio-inspired meta-heuristic technique Grey Wolf Optimization (GWO) algorithm is explored for tuning the PSS parameters by minimizing a multi-objective optimization problem. The tuning strategy involved simultaneously improving the real part of eigenvalue and its damping ratio to a selected D-shape stable zone in the s-plane. To depict the effectiveness, this algorithm is tested on 3-machine, 9-bus Western System Coordinating Council Power System (WSCCPS) [24] for various operating conditions under critical scenarios of disturbances and non-linear time-domain simulations, eigenvalues analysis and performance indices using Power System Analysis Toolbox (PSAT) [25], and results are compared with that of obtained by HSO, PSO and GA. The PSS parameters designed using HSO, PSO, GA and GWO are named as HSOPSS, PSOPSS, GAPSS and GWOPSS, respectively [26–28]. Also, the result shows that the designed GWOPSS provide guarantees for robust damping characteristics and superior overall dynamic stability to that of HSOPSS, PSOPSS and GAPSS to mitigate the low frequency oscillations.

2 Problem Formulation 2.1 Power System Model and Structure of PSS A complex power system model is generally represented by a set of nonlinear differential equations. In the designing problem of PSS, a linearized incremental model about a particular operating point is normally used [5–7]. The PSSs are feedback

28 Multi-machine Power System Stabilizer Design …

333

controllers and stabilizes the generator excitation such that electrical torque component, i. e,. damping torque comes in phase with rotor speed [18–20]. The structure of PSS is given as follows:  Ui = K i

sTw 1 + sTw



 (1 + sT1i ) (1 + sT3i ) wi (1 + sT2i ) (1 + sT4i )

(1)

where Δwi and ΔU i are deviation in speed of ith machine in pu from synchronous speed and output signal of PSS. This transfer function comprises washout filter as high-pass filter to eliminate the steady state error in the terminal voltage, dynamic controlling damping gain and a lag-lead phase compensator for decreasing the error between electrical torque and excitation. The time constant T w , T 2i , and T 4i are selected as chosen value while damping controlling gain K i, and compensator time constants T 1i and T 3i values are to be calculated for the tuning the PSS [18–20].

2.2 Test Case Study on 3-Machine, 9-Bus WSCCPS In this paper, the three-machine, nine-bus WSCCPS is selected for case study and its layout is shown in Fig. 1 and system data details are given in [24]. Three loading cases, i.e., normal loading as Case-1, light loading as Case-2 and heavy loading as Case-3 are selected and depicted in Table 1 [29].

Fig. 1 Layout of 3-machine, 9-bus WSCCPS

334

R. K. Sharma et al.

Table 1 Three operating cases of WSCCPS Generators

Case-1 P

Case-2 Q

Case-3

P

Q

P

Q

(p. u.) G1

0.71

0.62

0.96

0.22

3.57

1.81

G2

1.63

0.06

1.00

– 0.19

2.20

0.71

G3

0.85

0.45

– 0.26

1.35

0.43

– 0.10

Load A

1.25

0.50

0.70

0.35

2.00

0.90

B

0.90

0.30

0.50

0.30

1.80

0.60

C

1.00

0.35

0.60

0.20

1.60

0.65

Load at G1

1.00

0.35

0.60

0.20

1.60

0.65

Table 2 depicts the eigenvalues and damping ratio of WSCCPS with No-PSS for unstable and/or lightly damped modes with three operating cases only. Table 2 illustrates that the WSCC system becomes highly unstable for Case-3 with one-pair of eigenvalues modes lie in unstable zone of the s-plane and have negative damping. Moreover, it is noticed that unstable modes and/or lightly damped modes have high participation as compared to other frequency modes. Therefore, participation factor values indicate that the corresponding generators G2 and G3 are most favourable locations for installing PSSs [29]. Table 2 Eigenvalues and damping ratio comparison with No-PSS and with designed HSOPSS, PSOPSS, GAPSS and GWOPSS for three-loading cases Case-1 With No-PSS With GAPSS With PSOPSS With HSOPSS

– 0.110 ± j 8.588, 0.012

Case-2

Case-3

– 0.637 ± j 8.515, 0.074 0.158 ± j 8.372, – 0.018

– 0.653 ± j 13.023, 0.050 – 1.274 ± j 12.752, 0.099 – 0.308 ± j 12.896, 0.024 – 1.778 ± j 8.323, 0.209

– 1.659 ± j 7.724, 0.210 – 0.961 ± j 7.148, 0.133

– 1.887 ± j 7.160, 0.254

– 2.811 ± j 7.480, 0.351 – 1.930 ± j 8.508, 0.221

– 1.212 ± j 7.549, 0.158 – 1.614 ± j 7.563, 0.208

– 0.768 ± j 7.381, 0.103

– 2.007 ± j 14.393, 0.138 – 2.669 ± j 14.041, 0.351 – 1.570 ± j 14.157, 0.110 – 1.466 ± j 6.856, 0.209

– 1.876 ± j 6.935, 0.261 – 0.982 ± j 6.791, 0.143

– 2.278 ± j 17.457, 0.129 – 2.918 ± j 16.950, 0.169 – 1.956 ± j 17.143, 0.113

With – 2.008 ± j 7.363, 0.263 – 2.235 ± j 7.598, 0.282 – 1.561 ± j 7.244, 0.210 GWOPSS – 2.619 ± j 17.189, 0.150 – 3.351 ± j 17.010, 0.193 – 1.944 ± j 17.526, 0.110

28 Multi-machine Power System Stabilizer Design …

335

2.3 Objective Function To guarantee stability for unstable modes and to attain the relative stability for lightly damped modes, the six parameters of the PSS may be picked to minimize an eigenvalue based multi-objective function comprises simultaneously control of real-part of eigenvalues and damping ratio as given by: J=

np  

(σ0 − σi, j )2 +

j=1 σi, j ≥σ0

np  

(ξ0 − ξi, j )2

(2)

j=1 ξi, j ≤ξ0

where np is the number of chosen operating cases for tuning the PSS, and σ i,j and ξ i,j are the real-part and the damping ratio of the ith eigenvalue of the jth operating case, respectively. In this case study, the chosen values of σ 0 and ξ 0 are –0.5 and 0.1, respectively [29]. This will move the lightly and/or unstable damped eigenvalues to a chosen D-shape zone in the left-half of s-plane for considered operating cases. To reduce the computation burden, the value of T w is fixed as 10 s, T 2i and T 4i are reserved constant at numerical values of 0.1 s. The values of planned parameters K, T 1 and T 3 are put in the range of [1–100], [0.01–1] and [0.01–1], respectively [29]. Minimize J subject to K imin ≤ K i ≤ K imax

(3)

T1imin ≤ T1i ≤ T1imax

(4)

T3imin ≤ T3i ≤ T3imax

(5)

3 Grey Wolf Optimization The GWO is based on natural behaviour of wolf pack. Grey wolf are the vertex predators in the food chain and spend life in the groups. They have maintained a strict social behaviour and categorized into four types as: Alpha, Beta, Delta, and Omega. The main three steps of hunting are searching for prey, encircling prey and attacking prey [30]. The main merit of GWO is that it has few parameters and no derivative information needed in the initial search. GWO has a special capability of making the correct combination between exploration and exploitation on the time of searching that undergoes towards favourable convergence. GWO algorithm shows the leadership hierarchy and hunting scheme of grey wolves. Grey wolves always live in a pack of approximately 5–11 wolves. Social hierarchy is the main aspect of their pack. According to Munro et al. [31] three steps of hunting followed by the

336

R. K. Sharma et al.

wolves are (i) Chasing and approaching the prey (ii) Encircling the prey and (iii) attacking the prey. After evaluating the social behaviour of grey wolves, Mirjalili et al. [30] modelled the social behaviour and hunting process to design GWO. The mathematical execution of GWO technique is given as follows: The encircling approach of wolves around the prey is mathematically modelled by introducing the following equation as. X t+1 = X p,t − μd

(6)

d = |cX p.t − Xt |

(7)

μ = 2 · a · r1 −b

(8)

c = 2 · r2

(9)

where d, X t , X t + 1 and X p,t denotes difference vector expressed by Eq. (7), the position of the wolf at tth iteration, the position of the wolf at (t + 1)th iteration, the position of the prey at tth iteration, respectively. μ and c are constant coefficient vectors, and b is a linearly reducing vector from 2 to 0 over iterations, expressed as:  b = 2 − 2.

t maximum no. o f iteration

 (10)

and r 1 , r 2 are random vectors that uniformly distributed in the range 0 and 1 [30, 31]. Accordingly by going through these approximations each wolf can update their position by: X 1 = X α −μα · dα

(11)

X 2 = X β −μβ · dβ

(12)

X 3 = X δ −μδ · dδ

(13)

X t+1 = (X 1 + X 2 + X 3 )/3

(14)

where X α , X β , X δ shows the approximated position of the α, β, and δ solutions (wolves) with the help of Eq. (6).

28 Multi-machine Power System Stabilizer Design …

337

3.1 Exploration and Exploitation in Attacking Prey It is self-evident that when the movement of the prey is stopped, wolf will kill him and complete the hunting mechanism. In this process when |μ| > 1 or c > 1, then the wolf will search the whole search space for prey’s (optima) availability and search space will be exploited by the wolves if |μ| < 1 or c < 1. When t approaches maximum number of iterations, then by Eq. (8), b → 0 and μ → 0 in this case coefficient c will be responsible for exploration. By these way grey wolves completes the process of hunting by repeating the encircling and hunting steps as discussed above. Implementation of GWO algorithm 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12.

Set population size n, parameter a, coefficient vector μ, b, c and iterations. Produce random initial population of X i (t) (i = 1, 2, …, n) (each including of PSS parameters via. K i , T 1i and T 3i for each generator). Evaluate fitness F i = f(x i ) using the objective function f(x) = J for each population. Protect a best solution of X α , X β , and X δ and assign as first, second and third ranking respectively. Update each search agent population using Eqs. (11) Update the parameters a = 2 in the range of 2 to 0. Update the coefficient c and b using Eqs. (9) and (10) respectively. Evaluate the fitness function F j = f(x j ) Check the condition if (F j > F i ) then replace the solution x i with x j and F i with F j . Otherwise update the X α , X β , and Xδ . Preserve the best solutions with better fitness value. Repeat the steps 5 to 11 for achieving convergence of result till the termination condition satisfied.

4 PSS Design and Simulations Results In this section, the superiority of GWO algorithm is used for tuning of six PSS parameters for two generators of WSCCPS and performance is compared with GA, PSO and HSO based PSSs. The GWO is capable to determine the optimal parameters for which objective function J becomes zero shows that two pair of lightly and/or unstable damped eigenvalues modes are moved to a particular D-shape stable zone in the s-plane. The optimized six parameters evaluated by HSOPSS, PSOPSS, GAPSS and GWOPSSs for two generators are illustrated in Table 3. The eigenvalues and damping ratio with designed HSOPSS, PSOPSS, GAPSS and GWOPSS for three loading cases are determined by PSAT [25] and are illustrated in Table 2. Table 2 shows that the GWOPSSs move the eigenvalues to a chosen Dshape stable zone in the s-plane with desired real-part and damping ratio as compared to obtained by HSOPSS, PSOPSS, GAPSS. Hence, designed GWOPSS suggest

338

R. K. Sharma et al.

Table 3 Optimized designed parameters of HSOPSS, PSOPSS, GAPSS [29] and GWOPSS T1

T3

With GAPSSs [20]

Generators G2

1.000

0.464

0.060

G3

1.000

0.61

0.670

With PSOPSSs

G2

1.000

1.000

0.156

G3

1.000

0.400

0.060

G2

1.770

1.000

0.133

G3

1.810

0.060

0.714

G2

13.7402

0.248

0.060

G3

1.000

0.313

0.630

With HSOPSSs With GWOPSSs

K

improved overall dynamic stability and damping characteristics of the WSCCPS as compared to same obtained using HSOPSS, PSOPSS and GAPSS. To study the success of designed GWOPSS in terms of speed deviations, a 3phaseshort circuit, six-cycle fault occur at 1 s on bus 7 at the end of line 5–7 for heavy loading case is selected. The non-linear simulations of WSCCPS are performed with GWOPSS for critical disturbance of operating heavy loading only. The generator speed deviations Δw12 , Δw23 and Δw31 under severe disturbance of Case-3 for the system without PSS are illustrated in Fig. 2a and comparison of same speed deviations with designed HSOPSS, PSOPSS, GAPSS and GWOPSSs for same severe disturbance of Case-3 are shown in Fig. 2b–d respectively.

Fig. 2 Generator speed deviations a with No-PSS and b Δw12 c Δw23 d Δw31 with GAPSS, PSOPSS, HSOPSS and GWOPSS for severe disturbance of operating Case-3

28 Multi-machine Power System Stabilizer Design …

(a)

339

(b)

Fig. 3 Performance indices bar-chart: a IAE b ITAE for operating cases 1–3 of WSCCPS

From Fig. 2a, it is realized that in all speed deviation plots, the system with No-PSS is not competent to die-out these small amplitude low frequency oscillations because they are growing in magnitude with time and finally all generators loose the synchronism. Moreover, the Δw12 is most critical speed deviation than others. From Fig. 2b–d, it is obvious that with designed GWOPSS, all oscillations are quickly settle down and reaches in steady state speedily as compared to that of obtained by HSOPSS, PSOPSS, GAPSS for critical disturbance of heavy loading case. The GWOPSS give the impression to provide superior damping characteristics of designed set of PSS parameters. This illustrates the superiority of GWO technique to obtain desired set of PSS parameters. In addition to non-linear simulation results, the superior success of designed GWOPSS is realized by plotting bar charts of performance indices: Integral of Time Multiplied Absolute Value of Error (ITAE) and Integral of Absolute Error (IAE) with other PSSs for same disturbance of three cases and are illustrated in Fig. 3a, b, respectively. The figure depicts that values of both indices with GWOPSS are as low as for severe disturbance of operating cases 1–3 as compared to the same obtained by HSOPSS, PSOPSS, and GAPSS to settle low frequency local modes of oscillations. These comparisons evidently depict that GWOPSS offer comparatively superior damping performance than that of other designed PSSs.

5 Robustness of Designed GWOPSSs To examine robustness performance of previously designed GWOPSS for WSCCPS, three unnoticed operating cases 4–6 are considered and are shown in Table 4. To examine the strength of designed GWOPSS, time-domain simulation results, eigenvalue analysis and performance indices are calculated for the same severe disturbances under unnoticed cases 4–6. The comparison of real-part of eigenvalues and their damping ratio with No-PSS and with designed HSOPSS, PSOPSS, GAPSS for unnoticed cases 4–6 are illustrated in Table 5.

340

R. K. Sharma et al.

Table 4 Unnoticed operating cases of WSCCPS Generators

Case-4 P

Case-5

Case-6

Q

P

Q

P

Q

(p. u.) G1

0.33

1.12

1.09

0.79

1.41

0.59

G2

2.00

0.57

2.45

0.57

2.60

0.38

G3

1.50

0.38

1.27

0.21

1.2

0.02

A

1.50

0.90

1.90

0.75

2.00

0.60

B

1.20

0.80

1.30

0.45

1.50

0.30

C

1.00

0.50

1.50

0.50

1.60

0.20

Load

Table 5 Eigenvalues and damping ratio comparison with No-PSS and with designed HSOPSS, PSOPSS, GAPSS and GWOPSS for three unnoticed operating cases Case-4

Case-5

Case-6 0.604 ± j 8.375, – 0.072

With No-PSS

0.341 ± j 8.339, – 0.040

0.465 ± j 8.357, – 0.055

– 0.109 ± j 12.803, 0.0085

– 0.250 ± j 12.931, 0.019 – 0.233 ± j 12.981, 0.018

With GAPSSs

– 0.766 ± j 7.225, 0.105

– 1.228 ± j 8.052, 0.150

– 1.829 ± j 8.273, 0.215

– 1.327 ± j 7.440, 0.175

– 0.664 ± j 7.530, 0.087

– 0.557 ± j 7.442, 0.074

– 0.465 ± j 7.442, 0.062

– 1.565 ± j 13.977, 0.111

– 1.587 ± j 14.234, 0.110

– 1.495 ± j 14.387, 0.103

– 0.939 ± j 6.922, 0.134

– 0.828 ± j 6.835, 0.120

– 0.746 ± j 6.827, 0.108

– 2.038 ± j 17.156, 0.118

– 1.974 ± j 17.283, 0.113

– 1.816 ± j 17.429, 0.103

– 2.008 ± j 7.363, 0.263

– 2.235 ± j 7.598, 0.282

– 1.561 ± j 7.244, 0.210

With PSOPSSs With HSOPSSs With GWOPSSs

– 0.746 ± j 8.283, 0.089 – 1.692 ± j 7.092, 0.232

– 2.619 ± j 17.189, 0.150 – 3.351 ± j 17.101, 0.193 – 1.944 ± j 17.526, 0.110

Table 5 depicts that the GWOPSS move the lightly and/or unstable eigenvalues to a specific D-shape stable zone in the s-plane with improved damping factor and damping ratio as compared to the same obtained by HSOPSS, PSOPSS, GAPSS for unnoticed cases 4–6. Moreover, it is realized that only HSOPSS and GWOPSS satisfy the particular criterion for the real-part of eigenvalues and its damping ratio for tuning of PSS parameters. Hence, the designed GWOPSS are optimal and robust due to their superior damping characteristics even for unnoticed operating cases 4–6 as compared to the same obtained by other designed PSSs. In order to inspect the robustness performance of earlier designed GWOPSS in terms of generator speed deviations Δw12 , Δw23 and Δw31 , previously critical disturbance is performed on severe unnoticed Case-6 for the system with No-PSS and are illustrated in Fig. 4a. The comparison of speed deviations Δw12 , Δw23 and Δw31 with HSOPSS, PSOPSS, GAPSS and GWOPSS for same disturbance are illustrated in Fig. 4 b–d, respectively.

28 Multi-machine Power System Stabilizer Design …

341

Fig. 4 Generator speed deviations a with No-PSS and b Δw12 c Δw23 d Δw31 with HSOPSS, PSOPSS, GAPSS and GWOPSSs for severe disturbance of operating Case-6

Figure 4a illustrates that the generator speed responses with No-PSS is not efficient to damp-out the oscillations because they are growing in amplitude with time and finally all generators loose the synchronism. From Fig. 4b–d, it may be experienced that the designed GWOPSS for severe disturbance of unnoticed Case-6 shows enhanced damping performance because low frequency oscillations are die out speedily as compared to the same obtained by other designed PSSs. In addition to simulation results, the efficacy and robustness performance of designed GWOPSS are also experienced by plotting bar charts of ITAE and IAE for the same disturbance of unnoticed cases 4–6 and are illustrated in Fig. 5a, b, respectively. From Fig. 5, it is experienced that the both indices values for GWOPSS are low as far as and for GAPSS are greatest, for severe disturbance of unnoticed cases

Fig. 5 Performance indices bar-chart: a IAE b ITAE for unnoticed operating cases 4–6 of WSCCPS

342

R. K. Sharma et al.

4–6. It emphasis that the designed GWOPSS are most appropriate as compared to other designed PSSs to die-out the local-modes of oscillations with superior dynamic stability and damping characteristics not only for selected loading cases 1–3 but also for unnoticed operating cases 4–5 under same critical disturbances.

6 Conclusions This paper presents exploration of a new meta-heuristic technique GWO for designing of optimal and robust PSS parameters of the three-machines, nine-bus WSCCPS under a wide range of operational cases. The planned approach includes an eigenvalue-based multi-objective function to move lightly and/or unstable modes to a specific D-shape stable zone in the s-plane with improved small-signal stability. The non-linear simulation results, eigenvalue analysis and performance indices have clarified that with GWOPSS, the WSCCPS rapidly damped out the low frequency oscillations, even for the severe disturbance under considered operating cases as well as for unnoticed operating cases as compared to other designed PSSs. This depicts that GWOPSS are more robust and optimal than other designed PSSs. The present work can also be extended to apply for tuning the PSS parameters for large test systems, e. g., 10-machine, 39-bus New England power system or 16-machine, 68bus New England extended system for enhancement of damping performance and dynamic stability.

References 1. Kundur P (1994) Power system stability and control. In: Balu NJ, Lauby MG 4.2 2. Kundur P et al (2004) Definition and classification of power system stability IEEE/CIGRE joint task force on stability terms and definitions. IEEE Trans Power Syst 19.3:1387–1401 3. Demello FP, Concordia C (1969) Concepts of synchronous machine stability as affected by excitation control. IEEE Trans Power Appar Syst 88(4):316–329 4. Bollinger K et al (1975) Power stabilizer design using root locus methods. IEEE Trans Power Appar Syst 94(5):1484–1488 5. Shrikant RP, Sen I (2000) Robust pole placement stabilizer design using linear matrix inequalities. IEEE Trans Power Syst 15.1:313–319 6. DeMello FP et al (1982) A power system stabilizer design using digital control. IEEE Trans Power Appar Syst 8:2860–2868 7. Abdel-Magid YL, Dawoud MM (1996) Tuning of power system stabilizers using genetic algorithms. Electric Power Syst Res 39(2):137–143 8. Bomfim D, Antonio LB, Taranto GN, Falcao DM (2000) Simultaneous tuning of power system damping controllers using genetic algorithms. IEEE Trans Power Syst 15.1:163–169 9. Alkhatib H, Duveau J (2013) Dynamic genetic algorithms for robust design of multimachine power system stabilizers. Int J Electr Power Energy Syst 45(1):242–251 10. Abido MA (2000) Robust design of multimachine power system stabilizers using simulated annealing. IEEE Trans Energy Convers 15(3):297–304 11. Abido MA (2002) Optimal design of power-system stabilizers using particle swarm optimization. IEEE Trans Energy Convers 17(3):406–413

28 Multi-machine Power System Stabilizer Design …

343

12. Abido MA (1999) A novel approach to conventional power system stabilizer design using tabu search. Int J Electr Power Energy Syst 21(6):443–454 13. Hameed KA, Palani S (2014) Robust design of power system stabilizer using harmony search algorithm. Automatika 55.2: 162–169. 14. Ali ES (2014) Optimization of power system stabilizers using BAT search algorithm. Int J Electr Power Energy Syst 61:683–690 15. Mishra S, Tripathy M, Nanda J (2007) Multi-machine power system stabilizer design by rule based bacteria foraging. Electric Power Syst Res 77(12):1595–1607 16. Abd Elazim SM, Ali ES (2016) Optimal power system stabilizers design via cuckoo search algorithm. Int J Electric Power Energy Syst 75:99–107 17. Chitara D et al (2015) Optimal tuning of multimachine power system stabilizer using cuckoo search algorithm. IFAC-Papers On Line 48.30:143–148 18. Chitara D et al Robust tuning of multimachine power system Stabilizer via Cuckoo search optimization algorithm. In: 2016 IEEE 6th international conference on power systems (ICPS). IEEE 19. Chitara D et al (2018) Cuckoo search optimization algorithm for designing of a multimachine power system stabilizer. IEEE Trans Indus Appl 54.4:3056–3065 20. Linda MM, Kesavan Nair N (2012) Optimal design of multi-machine power system stabilizer using robust ant colony optimization technique. Trans Instit Measure Control 34.7:829–840 21. Eke ¯I, Taplamacıo˘glu MC, Lee KY (2015) Robust tuning of power system stabilizer by using orthogonal learning artificial bee colony. IFAC-PapersOnLine 48.30:149–154 22. Das TK, Venayagamoorthy GK, Aliyu UO (2008) Bio-inspired algorithms for the design of multiple optimal power system stabilizers: SPPSO and BFA. IEEE Trans Indus Appl 44.5:1445– 1457 23. Abd-Elazim SM, Ali ES (2013) A hybrid particle swarm optimization and bacterial foraging for optimal power system stabilizers design. Int J Electr Power Energy Syst 46:334–341 24. Anderson PM, Fouad AAA (2003) Institute of Electrical, and Electronics Engineers. Power Syst Control Stab 25. Milano F (2010) Power system analysis toolbox manual-documentation for PSAT version 2.1. 6. Univ Coll. Dublin Dublin Irel, Tech Rep 26. Chopra R, Joshi D, Bansal RC (2009) Analysis delta-omega and fuzzy logic power system stabilizer performances under several operating conditions. J Renew Sustain Energy 1(3):1–11 27. Chitara D, Swarnkar A, Gupta N, Niazi KR, Bansal RC (2015) Optimal tuning of multi-machine power system stabilizer using cuckoo search algorithm. In: 9th IFAC symposium on control of power and energy systems, Indian Institute of Technology. Delhi, India 28. Bansal RC, Zobaa AF (eds) (2021) Handbook of renewable energy technology and systems. World Scientific Publisher 29. Chitara D, Meena NK, Yang J, Niazi KR, Swarnkar A, Gupta N, Vega-Fuentes E (2019) Smallsignal stability enhancement of multi-machine power system using Cuckoo and harmony search optimization techniques. International conference on applied energy, Västerås, Sweden 30. Seyedali M (2014) Mirjalili Seyed Mohammad, and Lewis Andrew: Grey wolf optimizer. Adv Eng Softw 69:46–61 31. Munro L (2012) The animal rights movement in theory and practice: a review of the sociological literature. Sociol Compass 6(2):166–181

Chapter 29

Stability Enhancement of Grid Connected AC Microgrid in Modern Power Systems Vivek Ranjan, Amit Arora, and Mahendra Bhadu

1 Introduction The gradual depletion of conventional energy source due to the limited stock of these natural resources and continuously increasing demand of electricity consumption due to industrialization and increasing living standard of human worldwide introduced the Microgrid (MG) concept. The energy policy set by the government promote to construct renewable energy based small scale power plant, which has added advantage such as environmentally friendly and economically suitability, is known as Microgrid (MG) [1]. The renewable energy sources (RESs) based small scale grid is smoothly integrate into the main grid, and this fact is responsible for the development of the Microgrid concept. The Microgrid (MG) is define as a localized group of controlled power producing unit that consist of distributed energy resources (DERs), loads, storage devices, and controllers [2]. The DERs used in Microgrid are based on both renewable and non-renewable energy resources. The rapid development of renewable energy technologies and continuously increases the cost of non-renewable based energy sources increases the opportunities to developed Microgrid, which is based on renewable energy resources [3]. The renewable energy based power sources are pollution free and abundant in nature. Among in renewable sources, wind and solar are considered to be more popular of their reduced cost and technological advancements, variable speed wind turbine with doubly fed induction generator (DFIG), are used for wind energy extraction due to its advantages such as reduced in converter rating, less acoustic noise, highly energy efficient, and lower power loss [4, 5]. In other side, there has been increasing power generation through solar energy conversion system. Microgrid has two operating modes; one is grid connected mode and another is islanded mode [6]. One of the major advantages of Microgrid is that it has an ability V. Ranjan · A. Arora (B) · M. Bhadu Electrical Engineering Department, Engineering College Bikaner, Bikaner, Rajasthan, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 R. C. Bansal et al. (eds.), Proceedings of International Conference on Computational Intelligence and Emerging Power System, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-16-4103-9_29

345

346

V. Ranjan et al.

to sell or purchase power to/from main grid. This feature makes it an intellectual distribution network which satisfy the modern changeable load demand. Voltage and frequency stability of the network is maintained by balancing the supply and demand of power in a network. The Microgrids have less inertia as compare to the conventional power plant, so the instability problem in the Microgrid is more compare to the other. The major challenge faces by the Microgrid that include such as voltage and frequency deviation during high integration of intermittent RESs. The necessity to integration of RESs increases day by day due to the environmental and economical concern [2, 3]. The Microgrid control can be achieved by the proper coordination of three hierarchical control structure such as primary, secondary, and tertiary. There is need to improve the stability of AC Microgrid, and this can be achieved by designing an appropriate and robust controller and the result of this controller design the system is insensitive to all types of disturbances and change in system parameters [7]. The installation of power system stabilizers (PSSs) on power producing units will dampen the electromechanical oscillation and by doing this the transient stability of the network will enhance. In order to achieve the desired performance, the various types of power system stabilizers such as conventional PSS, Multi-band PSS (MB-PSS4B), and fuzzy logic-based PSS (FLPSSs) are installed on the synchronous generator of diesel and hydro farm. The PSS is more effective countermeasure tool against undamped oscillatory mode in a network [8, 9]. The rest of the paper is described as follows: Sect. 2, the brief concept regarding to the AC Microgrid structure design and its mathematical modelling is discuss. Section 3 reviews the controlling of AC Microgrid by different controllers with tuned parameters. In Sect. 4, the discussion of simulation result validates the controller performance. Finally, the paper concluded in Sect. 5, end by the references.

2 Modeling of AC Microgrid Components In this paper, there are four different generating units such as Diesel, Hydro, Solar, and DFIG used to make a small power producing grid, known as AC Microgrid. The grid connected AC Microgrid whose single line diagram is shown in Fig. 1. This AC MG is directly connected to a host grid which is 20 kV, 50 Hz, and 1000 MVA capacity at the point of common coupling. A distribution transformer whose capacity of 1200 MVA is used to connect the AC MG with the main grid, and B-G bus is selected as swing bus. The PSS provide supplementary stabilizing signal to the rotor excitation system of Diesel and Hydro farm which is used in the AC microgrid. The generalized systematic arrangement of PSS, which is connected to the synchronous generator is shown in Fig. 2. Using multiport input and single output switch for the connection of the different PSS by appropriate switching and at a given time only one PSS is working [7, 8]. Figure 3 shows Equivalent circuit of a solar cell, which consist of current sources, diode, and resistors as specified in [10]. In this diagram, the PV cell is represented by

29 Stability Enhancement of Grid Connected …

347

Fig. 1 Systematic representation of AC Microgrid model by single line diagram

an equivalent current source and the energy generated by the PV farm is depending on that surface area (A) of the PV cell, solar irradiance data, and solar panel efficiency (η). The solar cell is represented by the electrical model is shown in Fig. 3 Its I–V characteristic is expressed by the following equation. Figure 4 shows Wind turbine system based on DFIG. Power output from the wind turbine can be represented as follow:  q (V +I RS )  V + IR s I = I L − I0 e AK T − 1 − RS H

(1)

Pm = 0.5ρ AC p (λ, β)Vw3 ,

(2)

348

V. Ranjan et al.

Fig. 2 Basic arrangement of synchronous generator with PSS [8]

Fig. 3 Equivalent circuit of a solar cell

where ρ is the air density, A is the rotor swept area, C p (λ, β) is the power coefficient function which is determine by tip speed ratio λ and pitch angle β, and Vw is the wind speed [11]. The mechanical power and the stator electric power output are computed as follows: Pm = Tm ωr Ps = Tem ωs . For a lossless generator the mechanical equation is as follows:

(3)

29 Stability Enhancement of Grid Connected …

349

Fig. 4 Wind turbine system based on DFIG

J

dωr = Tm − Tem . dt

(4)

In steady-state at fixed speed for a lossless generator, Tm = Tem , and Pm = Ps + Pr

(5)

. It follows that: Pr = Pm − Ps = Tm ωr − Temωs = −Tm

ωs − ωr ωs = −sTm ωs = −s Ps , ωs

(6)

where s is defined as the slip of the generator s=

ωs − ωr ωs

(7)

3 Controllers of the AC Microgrid The AC Microgrid system faces many stability issues such as voltage and frequency deviation, rotor angle deviation, and active power oscillation at the abnormal condition such as fault. And this stability issue is more at the faulty condition due to the low inertia of the Microgrid. So, in order to stabilize all the oscillation and provide damping, PSS is introduced which is works as a supplementary controller device; whose main function is to modulate the excitation system of synchronous generator by providing additional damping signal to the excitation system to dampen the rotor oscillating mode. The installation of various PSS on the Microgrid reduces the oscillation problem and make the system stable.

350

V. Ranjan et al.

Fig. 5 Conventional PSS

3.1 Conventional PSS (CPSS) Figure 5 shows the block diagram of classical lead-lag power system stabilizer. It consists of gain block, washout circuit, phase compensation block, and limiter. The input signal for this PSS is either rotor speed deviation ( ω) or accelerating power Pa = (Pm – Pe). The output signal (Vstab ) from the PSS is fed into the excitation system, and it acts as an additional stabilizing input signal to the excitation system. The transfer function of CPSS is given as follows: V P SS = K P

(sTW ) (1 + sT1 )(1 + sT3 ) y (1 + sTW ) (1 + sT2) (1 + ST4 )

(8)

where y is the input signal (Pa or  ω) and T1 , T2 , T3 , and T4 are the time constant. TW = washout time constant. K P = gain of PSS [9].

3.2 Multi-band PSS (MB-PSS4B) MB-PSS has ability to damp out the rotor oscillation by proving adequate damping torque over wide range of oscillating mode. This is achieved by proper functioning of three tunable bands used in MB-PSS for low, intermediate, and high frequency modes of oscillation. When the rotor oscillation is very slow, i.e., global modes (0.05–0.1 Hz) take care by the low band of MB-PSS. The inter modes (0.1–0.8 Hz) take care by intermediate band and local, and intra-plant modes (0.8–4 Hz) take care by high band. MB-PSS work is accomplished by three basic stages.

3.3 Fuzzy logic control power system stabilizer (FLC-PSS) Fuzzy logic is designed based on human thinking and degrees of truth. The element of the fuzzy sets which is used in fuzzy logic contains degrees of memberships. This can be said in another way, every element is linked with multiple fuzzy sets. The

29 Stability Enhancement of Grid Connected …

351

range of membership value in fuzzy sets are from 0 to 1. The membership function takes crisp variable in input and convert it into a fuzzy variable. Figure 6 shows the block diagram of fuzzy controller which include Fuzzifier block, rule base and Inference engine, and the defuzzifier block. The role of the fuzzy inference process is to connect the membership functions with fuzzy rules, which is used to derive the output of fuzzy (Table 1). The given rule table uses two input signal, speed deviation ( ω) and acceleration power (Pa ) and single stabilizing output (Vstab ) so in this fuzzy controller 49 decision rules (7 × 7 member shi p f unction) are designed by using these input and output [12, 13]. For these inputs and output uses seven linguistic variables, which are given as: Big Negative (NB), Medium Negative (NM), Small Negative (NS), Zero (ZE), Small Positive (PS), Medium Positive (PM), and Big Positive (PB). The input and output membership function for fuzzy are shown in the Figs. 7, 8 and 9.

Rules

Crisp output

Crisp input Fuzzifier

Defuzzifier

Inference Fuzzy input sets

Fuzzy output sets

Fig. 6 Basic block diagram of fuzzy controller

Table 1 Rule table for seven membership function Speed deviation (ω)

Acceleration power ((Pa )) NB

NM

NS

ZE

PS

PM

PB

NB

NB

NB

NB

NB

NM

NS

ZE

NM

NB

NB

NM

NM

NS

ZE

PS

NS

NB

NM

NS

NS

ZE

PS

PM

ZE

NM

NS

ZE

ZE

PM

PB

PB

PS

NM

NS

PS

PM

PM

PB

PB

PM

NS

ZE

PS

PM

PM

PB

PB

PB

ZE

ZE

PM

PB

PB

PB

PB

352

V. Ranjan et al. NB

-0.015

NM

-0.010

NS

-0.005

ZE

0

PS

PM

PB

0.005

0.010

0.015

Fig. 7 Input membership function for speed deviation (ω)

NB

NM

-1

-0.8

-0.6

NS

-0.4

-0.2

ZE

0

PS

PM

0.2

0.4

PB

0.6

0.8

1

Fig. 8 Input membership function for acceleration power (Pa )

NB

NM

NS

-0.1

-0.08 -0.06 -0.04 -0.02

ZE

0

PS

0.02

PM

0.04

0.06

PB

0.08

0.1

Fig. 9 Output membership function plot for stabilizing signal (Vstab )

4 Results and discussion For result and discussion purpose five different cases has been considered. There are different types of controllers are used in this test model. And a three-phase fault occurred at a time t = 2 s in the grid. The duration of the fault is (9/60) sec. The performance of different controllers has been investigated and compare with each other after inserting a fault in the grid. Figure 10 shows diesel generator active power in MW and comparatively results are presented, and FLPSS gives better results.

29 Stability Enhancement of Grid Connected …

353

ω

Fig. 10 Diesel Generator active power contribution to grid (MW)

ω

Fig. 11 Hydro Generator active power contribution to grid (MW)

Hydro generator active power and Grid active power shown in Fig. 11 and Fig. 12 respectively. Diesel generator and hydro generator rotor speed deviation (pu) shown in Fig. 13 and Fig. 14 respectively. Diesel generator and hydro generator rotor speed (pu) shown in Fig. 15 and Fig. 16 respectively. It can be observed that the FLPSS and MB-PSS give the improved dynamic response over the conventional power system stabilizers ( ω-PSS,  Pa -PSS). The system with FLPSS seems to be the best at given operating point.

5 Conclusion The enhancement of the system stability of grid connected AC Microgrid (MG) examines by this research paper. The AC Microgrid comprised of three renewable based DG units such as hydro, solar photovoltaic (PV) array, wind energy using doubly

354

V. Ranjan et al.

ω

Fig. 12 Grid active power (MW)

ω

Fig. 13 Diesel Generator rotor speed deviation (pu)

ω

Fig. 14 Hydro generator rotor speed deviation (pu)

29 Stability Enhancement of Grid Connected …

355

Fig. 15 Diesel generator rotor speed Wm (pu)

Fig. 16 Hydro generator rotor speed Wm (pu)

fed induction generator (DFIG), and one additional emergency diesel generator for the backup supply to provides reliable and secure power supply if the power unbalanced occur due to the disturbances. The implementation of various PSSs such as conventional PSSs, Multi-band PSS, and fuzzy logic PSS on the synchronous generator of diesel and hydro farm of Microgrid have been investigated and compared under disturbances and its effectiveness to improve the dynamics stability of grid connected AC Microgrid are examined. The obtained result show that the FLPSS is the most suitable stabilizer for the stabilization of the system oscillation compare to the MB-PSS and CPSS and it also proves that the MB-PSS is another best choice then after FLPSS. The installation of FLPSS on synchronous generator decreases the overshoot of frequency oscillation with lesser settling time compare to the other used PSSs. This proves the supremacy of FLPSS to dampen the rotor frequency oscillation over the MB-PSS and CPSS.

356

V. Ranjan et al.

References 1. Mohammed A, Refaat SS, Bayhan S, Abu-Rub H (2019) AC microgrid control and management strategies: evaluation and review. IEEE Power Electron Mag 6(2):18–31. https://doi.org/10. 1109/MPEL.2019.2910292 2. Fu Q et al (2012) Microgrid generation capacity design with renewables and energy storage addressing power quality and surety. IEEE Trans Smart Grid 3(4):2019–2027. https://doi.org/ 10.1109/TSG.2012.2223245 3. Hossain A, Roy H, Hossain J, Blaabjerg F (2019) Electrical power and energy systems evolution of microgrids with converter-interfaced generations: challenges and opportunities. Electr Power Energy Syst 109(January):160–186. https://doi.org/10.1016/j.ijepes.2019.01.038 4. Li W, Chao P, Liang X, Ma J, Xu D, Jin X (2018) A practical equivalent method for DFIG wind farms. IEEE Trans Sustain Energy 9(2):610–620 5. Adefarati T, Bansal RC, John Justo J (2017) Techno-economic analysis of a PV–wind–battery– diesel standalone power system in a remote area. J Eng 2017(13): 740–744 6. Hou X et al (2018) Distributed hierarchical control of AC microgrid operating in gridconnected, islanded and their transition modes. IEEE Access 6:77388–77401. https://doi.org/ 10.1109/ACCESS.2018.2882678 7. Rathor B, Bhadu M, Bishnoi SK (2018) Modern controller techniques of improve stability of AC microgrid. In: 2018 5th international conference on signal processing and integrated networks (SPIN), Noida, pp 592–596, https://doi.org/10.1109/SPIN.2018.8474249 8. Bhadu M, Rathor B, Bishnoi SK (2017) Modern control techniques of AC microgrid. In: 2017 international conference on computing and communication technologies for smart nation (IC3TSN), Gurgaon, pp 1–6, https://doi.org/10.1109/IC3TSN.2017.8284440 9. Agrawal V, Rathor B, Bhadu M, Bishnoi SK (2018) Discrete time mode PSS controller techniques to improve stability of AC Microgrid. In: 2018 8th IEEE India international conference on Power Electronics (IICPE), pp 1–5 10. Ramos JA, Zamora I, Campayo JJ (2010) Modeling of photovoltaic module. In: International conference on renewable energies and power quality (ICREPQ’10) Granada, Spain, 23–25 March 2010 11. Miller NW, Sanchez-Gasca JJ, Price WW, Delmerico RW (2003) Dynamic modeling of GE 1.5 and 3.6 mw wind turbine-generators for stability simulations. GE Power Systems Energy Consulting, IEEE WTG Modeling Panel 12. Lee CC (1990) Fuzzy logic in control systems: fuzzy logic controller, part I and II. IEEE Trans Syst Man Cybern 20(2):404–435 13. Li HX, Gatland HB (1995) Enhanced methods of fuzzy logic control. In: IEEE Trans Syst Man Cybern 331–336 14. Dou CX, Liu B (2013) Multi-agent based hierarchical hybrid control for smart microgrid. IEEE Trans Smart Grid 4(2):771–778. https://doi.org/10.1109/TSG.2012.2230197 15. Olivares DE et al (2014) Trends in microgrid control. IEEE Trans Smart Grid 5(4):1905–1919. https://doi.org/10.1109/TSG.2013.2295514 16. Kundur P (1994) Power system stability and control. McGraw-Hill, NewYork, NY 17. Kamwa I, Gérin-Lajoie L (2000) State-space identification-towards MIMOModels for modal analysis and optimization of bulk power systems. IEEE Trans Power Syst 15(1): 326–335 18. Grondin R, Kamwa I, Trudel G, Gérin-Lajoie L, Taborda J (2003) Modeling and closed-loop validation of a new PSS concept, The Multi-Band PSS. In: Presented at the 2003 IEEE/PES general meeting, panel session on new PSS technologies, Toronto, Canada 19. IEEE Standard 421.5 (1992) IEEE Recommended Practice for Excitation Systems Models for Power System Stability Studies, August 1992 20. Klein M, Rogers GJ, Moorty S, Kundur P (1992) Analytical investigation of factors influencing power system stabilizers performance. IEEE Trans Energy Conv 7(3):382–390

Chapter 30

Performance Analysis of P&O and FLC Method of MPPT for PV Module Based on Five-Parameter Model Prakash Bahrani and Naveen Jain

1 Introduction With the increased demand for electrical energy, reduction in losses of generation and harvesting maximum energy out of installed capacity is the need of time. A lot of continuous research and development work in the field of renewable energy sources (RES) are reported in the past few decades. Common RES are wind turbines, solar PV system, hydroelectric generation, geothermal generation, etc. Among all, RES Solar PV (photovoltaic) system has become the most reliable, green, and effective way of electricity generation in Distributed Generation (DG) [1, 2]. The concept of DC microgrid (MG) consisting RES is a cost-effective integration of DG in the existing power system. The PV module manufacturers provide limited details about the ratings and output of the module, which of course are obtained through Standard Test Conditions (STC) and at standard operating conditions. Due to nondeterministic behavior of solar insolation, the limited operating conditions are not suitable for performance analysis under dynamic weather conditions [3]. Hence this encourages the researchers to devise some tools to incorporate all possible dynamic conditions for the PV system. Maximum Power Point Tracking (MPPT) is the technique associated with wind turbines and solar PV system (but not limited to) which primarily is the technique to extract maximum power out of the DG system in all possible operating conditions. In other words for variable power source, it is an application that leads the path to observe the maximum power available out of the source [4–6]. Solar PV system exists in different configurations like Stand-alone system, Gridconnected system, Grid connected with load system, Grid-connected hybrid system, etc. In different configurations, various parameters are involved in relations of different parts of the Solar PV and load arrangements like electrical loads, some P. Bahrani (B) · N. Jain Department of EE, CTAE, MPUAT, Udaipur, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 R. C. Bansal et al. (eds.), Proceedings of International Conference on Computational Intelligence and Emerging Power System, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-16-4103-9_30

357

358

P. Bahrani and N. Jain

Fig. 1 Solar PV cell equivalent model a Single diode, b Double diode, c Three diode

converter arrangements, battery banks, or external grid. All these arrangements are mandatory as per the requirement of the system and these arrangements address various problems also, but MPPT mainly addresses the efficiency of power output from the solar cell considering the solar cell input as solar radiations and electrical characteristics of the load. PV system is used to absorb solar energy from solar insolation and convert it into electrical energy and generation mainly depends upon climatic conditions (irradiance and ambient temperature), load resistance, and insolation angle. Due to the intermittent nature of solar energy dynamic operating conditions are faced by the solar cell and hence the requirement of some MPPT techniques arises to harness the maximum output of the panel. The PV cell can be represented by either single diode, double diode, or three diode models as shown in Fig. 1. In general, single-diode model is sufficient to represent a PV cell equivalent circuit. In this paper, a designed example of PV model addressing five parameters, namely Photo current I ph , diode reverse saturation current I 0 , PV cell series resistance Rs , PV cell shunt resistance Rsh , and output current I under varying conditions are considered for analysis [3]. This model takes input parameters as provided by the manufacturer and the analysis of panel characteristics is done using semi-empirical equations. Two different MPPT techniques, namely P&O and FLC methods are compared to understand the performance analysis of the PV cell under dynamic conditions. This paper is organized as follows; Sect. 2 presents the modeling of PV module suitable for dynamic operating conditions and basic principle of MPPT along with P&O and FLC methods of MPPT. Analysis of simulation results along with a discussion on results is presented in Sect. 3. Section 4 concludes the study.

2 Solar PV System Modeling For the practical use of the PV cell of any type, the parameters on which the performance of the cell depends must be analyzed through an electrical equivalent circuit for getting the proper relation between and the current and voltage (I-V characteristics). Using the equivalent circuit output equation of the module is characterized. As

30 Performance Analysis of P&O and FLC Method …

359

the efficiency of the module depends upon the aging factor also, which is caused due to influence of temperature changes, intense radiation, and high humidity condition in the field, therefore the electrical equivalent circuit must consist of the effect of temperature on various parameters to present the different characteristics in different operating conditions. In this paper, a five-parameter extraction method is utilized and modeled in MATLAB/Simulink environment representing a single diode model. To make the PV generation efficient, cost-effective, less complex operation and hardware implementable, various PV modules/array and operating techniques are designed and developed. The primary goal is to obtain the maximum power output of the solar module and the process of getting maximum power output is known as Maximum Power Point Tracking system (MPPT). Modeling of PV cell and MPPT are discussed in the following subsections.

2.1 Modeling of PV Cell Using Five-Parameter Extraction Model PV panel manufacturing companies give some operational parameters based on Standard Operating Conditions (SOC) such as current and voltage at maximum power (I mp & V mp ), short-circuit current and open-circuit voltage of panel (I sc & V oc ), and open-circuit voltage’s and short-circuit current’s temperature coefficients (KV oc & KI sc ). The SOC are 1000 W/m2 irradiance and 25 °C cell temperature which generally is not the operating conditions in actuality. An equivalent circuit diagram for PV cells in a single diode model is represented in Fig. 2. To increase the efficiency and feasibility the solar PV cells, they are assembled in form of series and parallel configuration to develop PV module (and further array), which serves the purpose of delivering high power demand. Thus the mathematical modeling of the PV module is required to be done. The output equation of the single diode model is as under  V +I Rs  V + I R  s ( ) (1) I = I P V − I0 e nVT − 1 − Rsh

Fig. 2 Five-parameter equivalent circuit diagram for PV cell

360

P. Bahrani and N. Jain

where I PV —Output current of PV module, A I 0 —Diode saturation current, A RS —Series resistance,  RSh —Shunt resistance,  V T —Thermal voltage, V V —Output voltage of PV array, V I—Output current of PV array, A n—Fill factor (ideal = 1).

The expression of thermal voltage V T is given by VT =

KT q

(2)

Diode current is expressed as   I D = Io eq V d/n K T − 1

(3)

Output current is expressed as I = I pv − I D − Ish

(4)

Shunt resistance current is expressed as Ish =

V + I Rs Rsh

(5)

where q Charge of Electron (1.60217646 × 10−19 C), K Boltzmann constant (1.3806503 × 10−23 J/K). Photo or PV cell current is expressed as    V ph + Rsh Ish −1 I pv = I − Io exp n

(6)

Reverse saturation current is expressed as Io = n p I ph

   KV −1 − n p Ir s ex p ns

where nS —No. of cells connected in series, nP —No. of cells connected in parallel (Fig. 3).

(7)

30 Performance Analysis of P&O and FLC Method …

361

Fig. 3 MATLAB/Simulink model for PV module using five-parameter extraction

2.2 Modeling of MPPT Methods The primary goal of PV module modeling is to obtain the maximum power output and the process of getting maximum power output is known as the Maximum Power Point Tracking system (MPPT). The role of different MPPT techniques depends upon efficiency required, requirement of sensors, speed of convergence, complexity of system and hardware implementation, etc. The main objective of any MPPT technique is to obtain the resulting output. Various MPPT methods based on the variable control strategy are categorized as below: • • • • •

Tracking with constant parameters methods, Measurement and comparison of parameters methods, Trial and error methods, Mathematical calculation based methods, Intelligent prediction methods.

Various MPPT methods based on convergence rate, stability, and dynamic response during variable operating conditions are also categorized for the specific application area. MPPT is broadly categorized into two categories as follows. • Conventional approach: Conventional categories are such that Perturb & Observation (P&O), Incremental Conductance (INC), Hill Climbing (HC), Curve Fitting (CF), Sliding Mode Control (SMC), etc. • Soft computing approach: The soft computing category are such that Fuzzy Logic Control (FLC), Artificial Neural Network (ANN), Biological Swarm Optimization like Ant Colony Optimization (ACO), Particle Swarm Optimization

362

P. Bahrani and N. Jain

MPPT Algorithm Measured V and I

S

Duty Cycle Control

I0

I1

PV Array

V1

L-C Filter

L o a d

Fig. 4 Block diagram of MPPT application

(PSO), Grey Wolf Optimization (GWO), Firefly Algorithm (FA), Ant Lion Optimization (ALO), Multi-verse Optimization (MVO) This section presents modeling and development of the FLC method of MPPT and a basic model of P&O from MATLAB library is taken for comparative analysis of the five-parameter model of PV module. The output of the PV system is intermittent in nature as the input parameters are dynamic, but the load connected to the output of the DC converter must be supplied with the maximum power available at the time. The converter which is used to enhance the voltage level of PV output can control this power output during the dynamic conditions by controlling the duty ratio (D) (Fig. 4). For the PV cell, the output power is the function of irradiance and temperature. The output power of PV cell is expressed as PP V (t) = F(V P V (t), I P V (t), γ (t))

(8)

where PPV (t) is the output power of PV system. V PV (t) is the output voltage of PV system. I PV (t) is the output current of PV system. Ƴ(t) all parameters depending upon climatic conditions. t any time instant. The basic concepts of MPP are presented in the equations below. d PP V =0 d VP V

when V P V = Vmp

(9)

d PP V >0 d VP V

when V P V < Vmp

(10)

30 Performance Analysis of P&O and FLC Method …

d PP V < 0 when V P V > Vmp d VP V

363

(11)

• P&O MPPT method P&O is the most common and effective technique to track the maximum power from the solar PV system. The operation of P&O is basically described as increasing or decreasing the reference voltage value or duty ratio of the converter and observes its effect on the value of the power output from the system. Now if the value of the kth power P(k) is higher than (k − 1)th value P(k − 1) then it keeps the same direction and if not then it reverses the direction of change. The change is given by perturbation and its effect is given by observation. The duty cycle of the DC-DC converter is continuously varied with a change of load condition and source output to match the peak point of power till the maximum power is obtained (FIg. 5). • FLC MPPT method In the PV system where the system parameters are highly nonlinear, FLC provides a robust system as compared to conventional methods. FLC has basic three elements as below. • Fuzzification and Rule base, • Inference engine, • De-fuzzification. The inputs to the FLC-based MPPT are error (E) and change in error (dE) at some time instant k. The expression for the E and dE is given as below. E(k) =

PP V (k) − PP V (k − 1) V P V (k) − V P V (k − 1)

d E(k) = E(k) − E(k − 1)

(12) (13)

PPV and V PV are power and voltage of the PV system, respectively. After calculation of E and dE, it has been converted to the linguistic variable and the output of the controller is the change in the duty cycle is the change in the duty ratio ΔD of the DC converter. The linguistic variable or the membership function of the FLC is as shown in Fig. 6. The fuzzy inference engine consists of the rule base and implication sub-blocks. Input parameters as fuzzified in the above figure are now fed to the fuzzy inference engine and the rule base is applied. The inference method utilized in the present work is Mamdani which is Max–Min fuzzy combination. The Table 1 shows the fuzzy inference. The method of de-fuzzification used is the center of gravity, and the center of gravity for the PV system is given as follows (Table 2).

364

P. Bahrani and N. Jain

Fig. 5 P&O algorithm for MPPT n

D=

μ(D j ) − (D j )

j=1 n

j=1

. (D j )

(14)

30 Performance Analysis of P&O and FLC Method …

365

NB

NS

ZE

PS

PB

b

a

0

a

b

Fig. 6 Membership function of FLC MPPT

Table 1 Fuzzy inference for MPPT E

dE NB

NS

ZF

PS

PB

NB

ZE

ZE

NB

NB

NB

NS

ZE

ZE

NS

NS

NS

ZE

NS

ZE

ZE

ZE

PS

PS

PS

PS

PS

ZE

ZE

PB

PB

PB

PB

ZE

ZE

V

I Z

P

Table 2 Proposed rule base for FLC MPPT

N NB

PB

NB

NB

NM

PM

NM

NM NS

NS

PS

NS

Z

NS

Z

PS

PS

NS

PS

PS

PM

NM

PM

PM

PB

NB

PB

PB

3 Results and Discussion In this section, the simulation results based on dynamic behavior of PV system with two different MPPT techniques are presented for different operating conditions of the PV system (Fig. 7). On basis of MATLAB model shown in Fig. 3 for parameter extraction and simulation model for MPPT operation on five-parameter model, the following results are presented for comparative simulation study. Two different MPPT techniques at three different operating conditions are compared for analysis of power and voltage of the PV module for proper dynamic response.

366

P. Bahrani and N. Jain

Fig. 7 MATLAB block for simulation of MPPT on the five-parameter model of PV module

3.1 Comparative Simulation Results of DC-DC Boost Converter with MPPT Application on Test System The test system has been put under three different levels of irradiance, namely 1000, 500, and 250 w/m2 and the temperature for all three irradiances are 25 ° and 45 °C (Figs. 8, 9 and 10). The output power and voltage at different irradiances and temperatures are shown in Table 3. It is evident from the Table 3 that as the temperature increased the power output is decreased at the same level of irradiance as the voltage output is directly affected by the rise in ambient temperature.

4 Conclusions The five-parameter extraction model of PV cell and two MPPT methods, namely P&O and FLC are discussed in this paper and MATLAB/Simulink environment model has been developed as an illustrative example for performance analysis. In this study, the effect of temperature on each of the five parameters are considered for checking the performance of PV panel and comparison of SOC and dynamic conditions shows in simulation results that the FLC-based MPPT provides better output results as compared to P&O method in all operating conditions.

30 Performance Analysis of P&O and FLC Method …

367

Fig. 8 Comparative simulation result of boost converter at 1000 w/m2 a Power output 25 °C, b Power at 45 °C, c Voltage output at 25 °C and d Voltage at 45 °C

Fig. 9 Comparative simulation result of boost converter at 500 w/m2 a Power output 25 °C, b Power at 45 °C, c Voltage output at 25 °C and d Voltage at 45 °C

368

P. Bahrani and N. Jain

Fig. 10 Comparative simulation result of boost converter at 250 w/m2 a Power output 25 °C, b Power at 45 °C, c Voltage output at 25 °C and d Voltage at 45 °C

Table 3 Comparative results

Input Irradiance

1000

w/m2

500 w/m2 250

w/m2

FLC/P&O MPPT Power output (W) 25 ° C

45 ° C

894.5/842.1

777.7/732.2

812.2/767.6

700.6/662

610.6/608

551.4/540.8

References 1. Kanchev H, Di Lu, Colas F, Lazarov V, Francois B (2011) Energy management and operational planning of a microgrid with a PV-based active generator for smart grid applications. IEEE Trans Indus Electron 58:4583–4592 2. Eghtedarpour N, Farjah E (2012) Control strategy for distributed integration of photovoltaic and energy storage systems in dc micro-grids. Renew Energy 45:96–110 3. Ding T, Liang H, Wilsun Xu (2016) An analytical method for probabilistic modelling of the steady-state behaviour of secondary residential system. IEEE Trans Smart Grid, IEEE Early Access Articles 99:1–10 4. Dan Wu, Tang F, Dragicevic T, Vasquez JC, Guerrero JM (2014) Autonomous active power control for islanded ac microgrids with photovoltaic generation and energy storage system. IEEE Trans Energy Convers 29:882–892

30 Performance Analysis of P&O and FLC Method …

369

5. Seyedmahmoudian MM, Rahmani R, Mekhilef S, Oo AMT, Stojcevski A, Soon TeyKok, Ghandhari AS (2015) Simulation and hardware implementation of new maximum power point tracking technique for partially shaded PV system using hybrid DEPSO method. IEEE Trans Sustain Energy 6:850–862 6. Urtasun A, Sanchis P, Marroyo L (2015) Frequency-based energy management strategy for stand–alone systems with distributed battery storage. IEEE Trans Power Electron 30:1–15 7. Milanovi´c JV, Yamashita K, Villanueva SM, Djoki´c SŽ, Korunovi´c LM (2013) International industry practice on power system load modeling. IEEE Trans Power Syst 28:3038–3046 8. Li X, Hui D, Lai X (2013) Battery energy storage station (BESS)-based smoothing control of photovoltaic (PV) and wind power generation fluctuations. IEEE Trans Sustain Energy 4:464–473 9. Miao Z, Ling Xu, Disfani VR, Fan L (2014) An SOC-based battery management systems for microgrids. IEEE Trans Smart Grid 5:966–973 10. Mahmood H, Michaelson D, Jiang J (2014) A power management strategy for PV/battery hybrid systems in islanded microgrids. IEEE J Emerg Select Topics Power Electron 2:1–14 11. Golsorkhi MS, Lu DDC (2016) A decentralized control method for islanded microgrids under unbalanced conditions. IEEE Trans Power Deliv 31:1–9 12. Augustine S, Mishra MK, Lakshminarasamma N (2015) Adaptive droop control strategy for load sharing and circulating current minimization in low voltage standalone DC microgrid. IEEE Trans Sustain Energy 6:1–10 13. Wang Y, Tan KT, Peng XY, So PL (2016) Coordinated control of distributed energy storage systems for voltage regulation in distribution networks. IEEE Trans Power Deliv 31:1–10 14. Wang T, O’Neill D, Kamath H (2015) Dynamic control and optimization of distributed energy resources in a microgrid. IEEE Trans Smart Grid 6:1–11 15. Mahmood H, Michaelson D, Jin J (2015) Decentralized power management of a PV/battery hybrid unit in a droop controlled islanded microgrid. IEEE Trans Power Electron 30:1–15 16. Xiaonan L, Kai S, Guerrero JM, Vasquez JC, Lipei H (2015) Double quadrant state-ofcharge-based droop control method for distributed energy storage systems in autonomous DC microgrids. IEEE Trans Smart Grid 6:147–157 17. Vandoorn Tine L, Meersman B, Degroote L, Renders B, Vandevelde L (2011) A control strategy for islanded microgrids with DC-Link voltage control. IEEE Trans Power Deliv 26:703–713 18. Chen Y-K, Wu Y-C, Song C-C, Chen Y-S (2013) Design and implementation of energy management system with fuzzy control for DC microgrid systems. IEEE Trans Power Electron 28:1563–1570 19. Hiroaki K, Miura Y, Ise T (2010) Low-voltage bipolar-type DC micro-grid for super high quality distribution. IEEE Trans Power Electron 25:3066–3075

Author Index

A Ali, Mohammad Hanif, 133 Ameya, Dalvi, 1 Anand, Iyer, 1 Arora, Amit, 345

B Bahrani, Prakash, 357 Bala, Mousumi, 133 Bari, Abdul, 59 Bhadouria, Sujeet Singh, 123 Bhadu, Mahendra, 345 Bishnoi, S. K., 181 Biswas, Saurabh, 85

C Chakravorty, Jaydeep, 213 Chauhan, Vineeta, 213 Chinmay, Dani, 1 Chitara, Dhanraj, 331 Choudhary, Rahul, 193

Garg, Bhawana A., 301 Gawde, Rupali R., 301 Gupta, Manish, 159 Gupta, Keshav, 13 Gupta, Nandkishor, 223 Gupta, R. A., 223 Gupta, Shashikant, 123 Gupta, Sudha, 99 Gupta, Sunil Kumar, 267 Gupta, Tripurari Nath, 319

H Hasija, Yasha, 85 Hossain, Rifat, 25 Hossain, Sayed Muddashir, 25

I Iqbal, Mohammad Asif, 279

D Das, Rahulkumar, 37 Dhingra, Sunita, 149 Dhusia, Komal, 37 Dyanamina, Giribabu, 289

J Jadhav, Sharad P., 301 Jain, Naveen, 357 Jain, Sakshi, 171 Jain, Yash, 171 Jindal, Sonika, 71 Joshi, R. R., 245 Joshi, Yashwant, 253

G Gajrani, Jyoti, 171 Garg, Akhil Ranjan, 193

K Kapil, Mundada, 1 Khandelwal, Tarul, 171

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 R. C. Bansal et al. (eds.), Proceedings of International Conference on Computational Intelligence and Emerging Power System, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-16-4103-9

371

372 Khara, Mihir, 99 Khurana, Dhiraj, 149 Kulshrestha, Neha, 159 Kumar, Prashant, 235 Kumar, Vinod, 235, 245, 253 Kushwaha, Alok Kumar Singh, 71

M Mahela, Om Prakash, 193 Mangal, A., 223 Mann, Palvinder Singh, 13 Modani, Uma Shankar, 49, 159 Mohnish, Gujarathi, 1 Momen, Sifat, 25

N Naik, Devashri, 99 Niazi, K. R., 331

P Panda, Ashish Kumar, 289 Pareek, Kapila, 109 Patil, Niranjan, 37 Peeploda, Sanjay Kumar, 319 Prathmesh, Ketkar, 1 Prova, Ayesha Aziz, 133

R Raj, Shashi, 331 Ranjan, Vivek, 345 Rawat, Mahiraj Singh, 319

Author Index S Sachdeva, Monika, 71 Saggu, Guramritpal Singh, 13 Sameena, 59 Sanap, Varsha, 37 Sangtani, Virendra, 279 Shah, Dhruvik, 99 Shakil, Fahim Ahmed, 25 Sharanya, 59 Sharma, K. C., 223 Sharma, K. G., 223 Sharma, Krishan Gopal, 193 Sharma, Ravi Kant, 331 Singh, Pushpendra, 181 Singh, Rishi Kumar, 289 Singh, Surendra, 193 Singh, Vivek Kumar, 37 Swami, Rekha, 267 Swarnkar, Anil, 331

T Tazi, Satyanarayan, 159 Tharewal, Sumegh, 109

V Vardia, Monika, 253 Vijayvergia, Hemant Kumar, 49 Vyas, Megha, 253 Vyas, Shripati, 245, 253

W Waseem Mohammed, Sharfuddin, 59