Big data and smart digital environment 9783030120474, 9783030120481
286 41 10MB
English Pages 415 Year 2019
Contents......Page 6
Introduction......Page 10
1 Introduction......Page 11
2.2 Queuing Systems for the Parking Service......Page 12
3.2 Simulation System by ARENA Rockwell......Page 13
4 Result and Discuss......Page 14
References......Page 15
2 The Advantages of the Smart Water Management......Page 17
3.1 The Human Pillar......Page 18
3.3 The Economic Pillar......Page 19
4.1 The Components of the Smart Water Grid......Page 20
4.2 The Keys of the Smart Water Grid......Page 21
References......Page 23
1 Introduction......Page 25
2.1 Architecture of the System......Page 26
2.2 The Proposed Process of Adaptation......Page 27
3 Conclusion......Page 28
References......Page 29
1 Introduction......Page 30
3 Modified Torque Control Strategy......Page 31
3.2 Flux Control......Page 33
5 Results and Discussion......Page 34
References......Page 36
Abstract......Page 37
1 Introduction......Page 38
2 Versioning of Temporal Logical and Physical Characteristics......Page 39
3.2 Design Choices......Page 40
3.3 Operations Acting on the Whole Temporal Characteristics Document......Page 41
3.5 Operations Specific to the Temporal Logical Characteristics......Page 42
3.6 Operations Specific to the Temporal Physical Characteristics......Page 43
4 Application Example......Page 44
References......Page 48
1 Introduction......Page 50
2.1 ISysQ Model Background......Page 51
2.2 Aggregating Variables Questions into Variables Indicators......Page 53
3 Results......Page 54
4 Discussion......Page 56
Appendix: ISysQ Model Abbreviations......Page 57
References......Page 58
1 Introduction......Page 60
2.1 Line Segment Localization......Page 61
2.2 Clustering of Line Segments......Page 62
2.3 Segments Filtering......Page 63
3 Results......Page 64
3.2 Comparison with Existing Methods......Page 66
References......Page 68
1 Introduction......Page 70
2 Chipless RFID Tag Design......Page 71
3 Ultra Wide Band Antenna Design......Page 74
4 Conclusion......Page 75
References......Page 76
1 Introduction......Page 77
3 Review of Ye et al. Authentication Protocol......Page 79
4.2 Generating of Session Key......Page 81
4.4 Problem in Step 5......Page 82
References......Page 83
1.1 Bioinformatics......Page 85
1.3 DNA Methylation Datasets......Page 86
1.5 Formal Description of the Clustering Problematic......Page 87
2.2 Algorithmic Description......Page 88
2.4 Outline of the Proposed Method......Page 89
3.1.1 DNA-Methylation Datasets......Page 90
3.2.1 F-Measure......Page 91
3.3 Empirical Results and Comparison......Page 92
4 Conclusions......Page 94
References......Page 95
1 Introduction......Page 96
2.2 Variable Neighborhood Descent Method (VND)......Page 97
2.3 VNPSO Components of MSA Problem......Page 98
3 Simulation and Results......Page 99
References......Page 101
Abstract......Page 102
2.1 Adaptivity in E-learning Systems......Page 103
2.2 Recommender Systems......Page 104
2.3 Adaptivity and Personalization in MOOCs/SPOCs......Page 105
3.2 The Proposed Recommendation Approach......Page 106
4 Conclusion and Future Work......Page 107
References......Page 108
1 Introduction......Page 112
2.1 Intelligent Big Data Analysis......Page 115
2.2 Rapid Automatic Keyword Extraction (Rake)......Page 116
2.3 Latent Dirichlet Allocation (LDA)......Page 117
3 Hybrid Miner-Network Analyzer (HMNA)......Page 118
4.1 Collection Dataset......Page 120
4.3 Applying Pragmatic Method to Text Preprocessing......Page 121
4.4 Classify Document by Find Topic Distribution......Page 122
4.5 Building Multi-level Network......Page 124
4.6 Compute the Characteristic Measures......Page 125
5 Discussions and Conclusions......Page 126
References......Page 127
1 Introduction......Page 129
2.1 Measurement Results and Channel Models......Page 130
2.2 Challenges to Overcome......Page 132
3 5G NR Interface......Page 133
3.1 New Waveform and Optimized Modulation......Page 134
3.2 Beamforming......Page 135
References......Page 138
1 Introduction......Page 142
2.2 Battery Models......Page 144
3 Battery Management System (BMS) and the Mathematical Methods Used in SOC Estimating......Page 145
3.1 Direct Measurement......Page 146
3.3 Adaptive Systems......Page 147
4.1 State of Charge Modeling......Page 148
4.2 Power Consumption Modeling......Page 149
References......Page 152
Abstract......Page 154
1 Introduction......Page 155
2 Related Work......Page 156
4 Approach......Page 157
5.1.2 Pre-processing of the MTT Data......Page 158
5.2.2 Region Based Methods: Region Growing......Page 159
6 Evaluation and Discuss......Page 161
7 Conclusion......Page 162
References......Page 163
1 Introduction......Page 165
2 The Configuration Studied......Page 166
3.2 Hysteresis Control......Page 169
3.3 Direct Power Control Study of SAPF......Page 170
4 Simulation Results......Page 171
5 Conclusion......Page 177
References......Page 178
1 Introduction......Page 179
2 Previous Related Works......Page 180
3.1 Pseudo-random Key Stream Generator......Page 181
4.2 Key Space Analysis......Page 182
4.4 Entropy Correlation Coefficient Analysis......Page 183
5 Conclusion......Page 184
References......Page 185
1 Introduction......Page 186
2 OFDM Basis and OOB Emission......Page 187
3 Results and Discussion......Page 190
4 Conclusion......Page 194
References......Page 195
2 Experimental Results......Page 196
2.3 Comparison......Page 197
References......Page 199
1 Introduction......Page 200
2 Preliminaries......Page 201
3.1 TCP + AQM Dynamics......Page 202
3.2 Multi-class Model of TCP + AQM Dynamics......Page 203
4 Main Results......Page 206
5 Numerical Example......Page 208
References......Page 212
1 Introduction......Page 214
2 Related Works......Page 215
2.2 Divide and Conquer Approach......Page 216
2.3 Bitmap Approach......Page 217
2.5 Nearest Neighbor Approach......Page 218
3 Our Map Reduce Nearest Neighbor Approach (MR-NN)......Page 219
3.1 Comparison Between MR-NNS and NNS......Page 221
3.2 Example......Page 222
References......Page 223
1 Introduction......Page 225
2 Related Works......Page 226
3.1 Data Set and Attributes......Page 227
3.2 Classification Task......Page 228
3.3 Particle Swarm Optimization (PSO)......Page 229
4 Experiments and Results......Page 230
4.1 Effectiveness......Page 231
4.2 Accuracy Results......Page 232
References......Page 233
1 Introduction......Page 235
2 Related Categorical Clustering Approaches Using the k-Modes......Page 236
3.1 The Mode of a Categorical Cluster......Page 239
3.2 The Rough Set Theory......Page 240
3.3 The Rough Modes......Page 241
4 Experiments......Page 244
References......Page 245
1 Introduction......Page 247
2 Requirements......Page 248
3 Architecture Challenges......Page 249
3.1 Master and Slave Domain Challenges......Page 250
3.2 Network Domain Challenges......Page 251
References......Page 254
Abstract......Page 256
2 Literature View......Page 257
3.1 Intelligent Data Analysis (IDA)......Page 260
3.3 Prediction Techniques......Page 261
3.4 Churn Prediction......Page 263
4 Smart Customer Predictor (SCP)......Page 264
4.1 Main Stages of Design (SCP)......Page 267
5 Experiments and Results......Page 273
6 Conclusion and Future Work......Page 280
References......Page 281
1 Introduction......Page 283
2.2 The Pedagogical Object at the Heart of the Main Three Tensions......Page 284
2.4 How the Use of Pedagogical Objects in the Pedagogical Conception of an Adaptive cMOOC Can Be Optimized?......Page 286
3 Result......Page 287
5 Conclusion......Page 288
References......Page 289
1 Introduction......Page 290
2.1 Architecture and Operation of IoT......Page 291
2.2 IoT Platform......Page 292
3.3 Architecture of Cassandra......Page 293
4 Proposed Method......Page 294
5.1 Function of the Platform......Page 295
References......Page 296
1 Introduction......Page 298
2.2.1 Presentation......Page 299
2.3.2 Application Domain€......Page 300
3.1.2 Application Domain......Page 301
3.2.2 Application Domain......Page 302
3.3.4 Disadvantages......Page 303
References......Page 304
1 Introduction......Page 306
2.1 Definition of KM......Page 307
2.2 KM Process......Page 308
2.4 KM Methods and Tools......Page 310
2.6 KM Issues......Page 312
3.1 Current Situation Analysis......Page 313
4.1 Selection of the KM Strategy......Page 314
5 Conclusion......Page 315
References......Page 316
1 Introduction......Page 317
2 Inverter Topology......Page 318
3 The Control Strategies of the Inverter......Page 319
4 Simulation and Results......Page 320
References......Page 322
Abstract......Page 324
2.1 Dataset......Page 325
3 Results......Page 326
4 Conclusion and Discussion......Page 330
References......Page 331
1 Introduction......Page 332
2.1 Frequent Itemset Mining......Page 333
2.2 Prepost Algorithm......Page 334
3 The Proposed Algorithm......Page 336
4 Experiments......Page 339
References......Page 341
1 Introduction......Page 343
2.2 Word Cloud......Page 344
3.1 Grouped Data......Page 345
3.2 Measuring Parameters......Page 346
4.1 Word2vec......Page 347
4.2 Clustering......Page 348
5.2 Choose Number of Topics......Page 349
5.3 Topic Probabilities......Page 350
References......Page 351
2.1.1 Definition......Page 352
2.2 Scalogram......Page 353
3.2 Algorithm......Page 354
3.3.2 Heart Rate Based on RR Interval......Page 355
References......Page 356
1 Introduction......Page 357
2 Related Works......Page 358
3 Our Proposed Approach......Page 359
3.3 The Architecture of Our System......Page 360
3.3.4 Health Agent......Page 362
3.3.8 Geo-Location......Page 363
3.4 A Conceptual Model for Big Data and Social Media......Page 364
4 Experimentation and Results......Page 365
5 Conclusion and Future Works......Page 367
References......Page 368
1 Introduction......Page 369
2 Fundamental Concepts......Page 370
3.2 Risk Assessment Standard......Page 371
3.5 Trust Matrix Risk Assessment......Page 372
3.8 Risk Assessment as Service......Page 373
4 Synthesis and Discussion......Page 374
References......Page 376
1 Introduction......Page 379
2 Captologie......Page 380
3 Related Work......Page 381
4.1 Fogg’s Seven Principles of Persuasion......Page 382
4.2 Cialdini’s Six Principles of Persuasion......Page 383
5 Persuasive Technologies for Urban Mobility......Page 384
6 Comparison Study......Page 386
7 Conclusion et Perspectives......Page 387
Reference......Page 388
1 Introduction......Page 390
2 Fixed Monitoring Stations......Page 391
3 Crowdsourcing and Low-Cost Sensors......Page 392
4 Air Quality Remote Sensing......Page 393
5 Statistical and Machine Learning Based Models......Page 395
References......Page 396
2 Towards Conceptual Representations......Page 398
References......Page 399
1 Introduction......Page 401
1.1 Literature Review on the Marketing Information System......Page 403
2.1 The Need not to Negatize the Risk of Non-use of Computerized Applications Induced by the Marketing Information System......Page 406
3.1 Methodology of Quantitative Research......Page 407
4.1 Analysis of Quantitative Results......Page 409
5 Conclusion......Page 412
References......Page 413
Author Index......Page 414
Recommend Papers
![Smart Grids and Big Data Analytics for Smart Cities [1st ed.]
9783030521547, 9783030521554](https://ebin.pub/img/200x200/smart-grids-and-big-data-analytics-for-smart-cities-1st-ed-9783030521547-9783030521554.jpg)
- Author / Uploaded
- Farhaoui Y.
- Moussaid L (ed.)
File loading please wait...
Citation preview
Studies in Big Data 53
Yousef Farhaoui Laila Moussaid Editors
Big Data and Smart Digital Environment
Studies in Big Data Volume 53
Series editor Janusz Kacprzyk, Polish Academy of Sciences, Warsaw, Poland
The series “Studies in Big Data” (SBD) publishes new developments and advances in the various areas of Big Data- quickly and with a high quality. The intent is to cover the theory, research, development, and applications of Big Data, as embedded in the fields of engineering, computer science, physics, economics and life sciences. The books of the series refer to the analysis and understanding of large, complex, and/or distributed data sets generated from recent digital sources coming from sensors or other physical instruments as well as simulations, crowd sourcing, social networks or other internet transactions, such as emails or video click streams and other. The series contains monographs, lecture notes and edited volumes in Big Data spanning the areas of computational intelligence incl. neural networks, evolutionary computation, soft computing, fuzzy systems, as well as artificial intelligence, data mining, modern statistics and Operations research, as well as self-organizing systems. Of particular value to both the contributors and the readership are the short publication timeframe and the world-wide distribution, which enable both wide and rapid dissemination of research output. ** Indexing: The books of this series are submitted to ISI Web of Science, DBLP, Ulrichs, MathSciNet, Current Mathematical Publications, Mathematical Reviews, Zentralblatt Math: MetaPress and Springerlink.
More information about this series at http://www.springer.com/series/11970
Yousef Farhaoui Laila Moussaid •
Editors
Big Data and Smart Digital Environment
123
Editors Yousef Farhaoui Faculty of Sciences and Techniques, Department of Computer Science Moulay Ismail University Errachidia, Morocco
Laila Moussaid Department of Computer Science Hassan II University Casablanca, Morocco
ISSN 2197-6503 ISSN 2197-6511 (electronic) Studies in Big Data ISBN 978-3-030-12047-4 ISBN 978-3-030-12048-1 (eBook) https://doi.org/10.1007/978-3-030-12048-1 Library of Congress Control Number: 2018968014 © Springer Nature Switzerland AG 2019 This work is subject to copyright. All rights are reserved 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, express or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Contents
A Smart Parking for Invisible Disabilities . . . . . . . . . . . . . . . . . . . . . . . Aroua Amari, Laila Moussaid, and Saidia Tallal
1
Smart Water Management: Pillars and Technologies . . . . . . . . . . . . . . . Najat Abdeljebbar, Laila Moussaid, and Hicham Medromi
7
Integrating ICT in Education: An Adaptive Learning System Based on Users’ Context in Mobile Environments . . . . . . . . . . . . . . . . . . . . . . Soukaina Ennouamani, Laila Akharraz, and Zouhir Mahani Modified Strategy of Direct Torque Control Applied to Asynchronous Motor Based on PI Regulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Soukaina El Daoudi, Loubna Lazrak, Chirine Benzazah, and Mustapha Ait Lafkih
15
20
Managing Temporal and Versioning Aspects of JSON-Based Big Data via the sJSchema Framework . . . . . . . . . . . . . . . . . . . . . . . . . Safa Brahmia, Zouhaier Brahmia, Fabio Grandi, and Rafik Bouaziz
27
Specific Qualification for Information System Components from Managers and Technical Staff Perspective . . . . . . . . . . . . . . . . . . . Sarah Aouhassi and Mostafa Hanoune
40
A Clustering-Based Method for Detecting Text Area in Videos Recorded with the Aid of a Smartphone . . . . . . . . . . . . . . . . . . . . . . . . Hassan El Bahi and Abdelkarim Zatni
50
Chipless RFID Tag Using Multiple G-Shaped Resonators . . . . . . . . . . . Mohamed Amzi, Laila Moussaid, and Hicham Medromi Security Analysis of Ye et al. Authentication Protocol for Internet of Things . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mourade Azrour, Mohammed Ouanan, Yousef Farhaoui, and Azidine Guezzaz
60
67
v
vi
Contents
A Non-stochastic Method for Clustering of Big Genomic Data . . . . . . . Billel Kenidra and Mohamed Benmohammed An Enhanced Hybrid Model for Solving Multiple Sequence Alignment Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Lamiche Chaabane A Recommender System for Videos Suggestion in a SPOC: A Proposed Personalized Learning Method . . . . . . . . . . . . . . . . . . . . . . Naima Belarbi, Nadia Chafiq, Mohammed Talbi, Abdelwahed Namir, and Habib Benlahmar
75
86
92
Multi-level Network Construction Based on Intelligent Big Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 Samaher Al_Janabi, Mahdi Abed Salman, and Maha Mohammad What Will Millimeter Wave Communication (mmWave) Be? . . . . . . . . 119 Fatima Zahra Hassani-Alaoui and Jamal El Abbadi Application of the Battery Management System in a Multi-rotor Unmanned Aerial Vehicle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132 Atae Semmar, Fouad Moutaouakkil, and Hicham Medromi Brain Ischemic Stroke Segmentation from Brain MRI Between Clustering Methods and Region Based Methods . . . . . . . . . . . . . . . . . . 144 Fathia Aboudi, Cyrine Drissi, and Tarek Kraiem Comparison of the Control Strategies of an Active Filter of a Photovoltaic Generation System Connected to the Three-Phase Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155 Zoubir Chelli, Abdelaziz Lakehal, Tarek Khoualdia, and Yacine Djeghader A Multi-chaotic Fibonacci Algorithm for Digital Image Encryption . . . . 169 Lamiche Chaabane On the Out of Band Emission Reduction for Orthogonal Frequency Division Multiplexing Based Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 176 Naima Sofi, Fatima Debbat, and Fethi Tarek Bendimerad Monitoring of Resources Used by Java Mobile Applications . . . . . . . . . 186 Laila Fal, Laila Moussaid, and Hicham Medromi Anti-windup Compensation in TCP/IP Routers: A Multi-delay Feedback Systems Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190 Nabil El Fezazi, Ismail Er Rachid, El Houssaine Tissir, Fatima El Haoussi, Teresa Alvarez, and Fernando Augusto Bender A Parallel Nearest Neighbor Algorithm for Skyline Computation Using Map Reduce . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204 Brahim Bouderar, Larbi Alaoui, and Moulay Youssef Hadi
Contents
vii
K-Nearest Neighbour Model Optimized by Particle Swarm Optimization and Ant Colony Optimization for Heart Disease Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215 Youness Khourdifi and Mohamed Bahaj Rough Mode: A Generalized Centroid Proposal for Clustering Categorical Data Using the Rough Set Theory . . . . . . . . . . . . . . . . . . . . 225 Semeh Ben Salem, Sami Naouali, and Zied Chtourou Tactile Internet: New Challenges and Emerging Solutions . . . . . . . . . . . 237 Omaima Khalil and Anas Abou El Kalam Intelligent Big Data Analysis to Design Smart Predictor for Customer Churn in Telecommunication Industry . . . . . . . . . . . . . . . . . . . . . . . . . . 246 Samaher Al_Janabi and Fatma Razaq Providing Smart Content for Developing an Intelligent Adaptive cMOOC Platform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273 Soumaya El Emrani, Ali El Merzouqi, and Mohamed Khaldi Important Method of Exchange and Sharing of Massive Data Between Connected Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280 Kaoutar Makdad, Rafik Lasri, and Abdellatif El Abderrahmani Categories of Big Data Analytics Algorithms . . . . . . . . . . . . . . . . . . . . . 288 Salma Abarou, Abdellatif El Abderrahmani, and Khalid Satori Towards a Strategy of Knowledge Management Within the Agence Nationale de la Conservation Foncière du Cadastre et de la Cartographie (ANCFCC) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296 Ayoub Otmani and Taoufik Benkaraache A Comparative Study of Sinusoidal PWM and Space Vector PWM of an Induction Machine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307 Weam El Merrassi, Abdelouahed Abounada, and Mohamed Ramzi Predictive Analytics for Determining Patients’ Vitamin D Status . . . . . . 314 Souad Bechrouri, Abdelilah Monir, Hamid Mraoui, El-Houcine Sebbar, Ennouamane Saalaoui, and Mohamed Choukri Parallel Implementation of PrePost Algorithm Based on Spark for Big Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322 Yassir Rochd, Imad Hafidi, and Bajil Ouartassi Topics Classification of Arabic Text in Quran by Using Matlab . . . . . . 333 Abdelkrim El Mouatasim and Jaouad Oudaani An NMF Based Method for Detecting RR Interval . . . . . . . . . . . . . . . . 342 Said Ziani, Youssef El Hassouani, and Yousef Farhaoui
viii
Contents
Conceptual Framework for Analyzing Knowledge in Social Big Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347 Brahim Lejdel Security Risk Assessment of Multi-cloud System Adoption: Review and Open Research Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359 Saadia Drissi, Soukaina Elhasnaoui, Hajar Iguer, Siham Benhadou, and Hicham Medromi Exploring Persuasive Systems Using Comparative Study Between Actual Technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369 Alami Sara and Hanoune Mostafa Air Quality Monitoring Using Deterministic and Statistical Methods . . . 380 Noussair Lazrak, Jihad Zahir, and Hajar Mousannif Cognitive Lessons for Smart Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 388 Piet Kommers Marketing Information System: An Effective Tool for Participatory Management of Scarce Resources of Organizations. Case of SMEs in the City of Errachidia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391 Khalid Lali, Abdellatif Chakor, and Yousef Farhaoui Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
Introduction
Data is becoming an increasingly decisive resource in modern societies, economies, and governmental organizations. Data science inspires novel techniques and theories drawn from mathematics, statistics, information theory, computer science, and social science. This book reviews the state of the art of big data analysis and smart city. It includes issues which pertain to signal processing, probability models, machine learning, data mining, database, data engineering, pattern recognition, visualisation, predictive analytics, data warehousing, data compression, computer programming, smart city, etc. Papers in this book were the outcome of research conducted in this field of study. The latter makes use of applications and techniques related to data analysis in general and big data and smart city in particular. The book appeals to advanced undergraduate and graduate students, postdoctoral researchers, lecturers and industrial researchers, as well as anyone interested in big data analysis and smart city.
ix
A Smart Parking for Invisible Disabilities Aroua Amari(&), Laila Moussaid, and Saidia Tallal (EAS-LRI) Systems Architecture Team, Hassan II University, ENSEM, Casablanca, Morocco [email protected]
Abstract. In several works, the purpose of a smart parking is to find the most optimal location for a given situation, either in favor of the driver or in favor of the company offering this service. Thought in many papers, driver’s health is usually overlooked when it comes to invisible or temporary illnesses. However, in this article, we try to come up with a parking solution that takes into consideration normal people and invisible disabilities. Keywords: Smart parking Smart health
Queuing theory Arena Rockwell
1 Introduction Our motivating scenario deals with people with vulnerable health condition searching for a free parking place nearest to their destination. Our work builds upon a real life problem, i.e., the lack of consideration for invisible disabilities to find a parking place adequate to their situation. Indeed, the comfort of smart parking is proposed for two kinds of people normal people and handicapped people. What for people with invisible disabilities like chronic diseases or temporary illness? However, we try to propose a smart parking system which takes account all kinds of people. We propose a system that takes into consideration the health status of drivers when allocating available seats. Drivers with a vulnerable health status are given priority to access to the nearest parking spot from their destination. As for other drivers, normal health state, are managed according to the “first come, first served” principle. In this article, we simulate the propose parking system with arena Rockwell based on the queuing theory. In this article, we study the case of a parking lot of a private establishment where the majorities of the arrivals are already known by the system. This paper is organized as follows: Sect. 2 provide an overview of the most relevant research Works which served as an inspiration for this article. Sections 3 outline the proposed parking system with a simulation in Arena Rockwell. Section 4 provides the experimental results and Sect. 5 concludes the paper.
© Springer Nature Switzerland AG 2019 Y. Farhaoui and L. Moussaid (Eds.): ICBDSDE 2018, SBD 53, pp. 1–6, 2019. https://doi.org/10.1007/978-3-030-12048-1_1
2
A. Amari et al.
2 Overview 2.1
Mobile Phone Sensing for Pervasive Healthcare
The detection of health parameters such as activity and heart rate using the built-in or external sensors of the mobile phone has enabled the development of a wide range of applications for personalized health surveillance and management [1]. These applications are intended to strengthen the role of patients in the delivery of health care services, enabling them to cope with their health status while leaving the hospital and carrying out their daily activities. The value of frequent self-monitoring of health parameters has been demonstrated in several clinical guidelines [2, 3]. Patients (especially those with chronic conditions) may be frequently involved in the process of collecting personal health data in order to achieve better health outcomes [2]. These data may include vital signs measured by sensors - raw physiological data or aggregated information and monitoring of results after data processing and analysis. In addition, patients can generate information themselves by manually recording problems or symptoms encountered during their daily activities, their health behaviors, physiological measurements, laboratory examinations, etc. People’s health is interdependent [4], and the crowdsourcing of information can potentially lead to a form of “social wisdom” and participatory care that enables patients, particularly those with chronic conditions, to gain support and manage their health more effectively [5] and even their daily lives. Today, with clinical requirements, ubiquitous healthcare systems are required to be useful, effective and user-friendly to achieve widespread adoption and sustainability [6]. Systems that we can use to design a smart parking management system that helps people with chronic conditions get the best location to park their car in terms of distance. 2.2
Queuing Systems for the Parking Service
Since the arrivals of cars can be seen as random (stochastic) variables, it is possible to approximate them with the corresponding theoretical distributions. Flows of the arrivals of vehicles and the service time are simple random flows, for which theoretical distributions according to which the empirical data is assigned can be determined or approximated [7]. The key parking characteristics are occupancy, turnover, duration of stay per user, acceptability of parking charges, as well as potential identification of the modal split across alternative means of transport [8]. That is why parking systems can be analyzed as mass service systems. For that reason, it is possible to, by applying analytical methods, determine the characteristics of the input terminals and examine their permeable power for a different number of parking lot entrances.
A Smart Parking for Invisible Disabilities
3
3 Proposed Parking System 3.1
Flow of Proposed System
The system ranks the requests according to the vulnerability of the driver’s condition. Consequently, it prioritizes people who suffer from temporary or permanent physical fragility. For other drivers, they follow the “first come, first served” principle (Fig. 1).
Fig. 1. The flow of proposed parking system
3.2
Simulation System by ARENA Rockwell
In our case, we have two arrivals queuing processes: ordinary arrival and emergency arrival. In this simulation, we give the priority to the emergency arrival to take the nearest parking place. We talk about places in a garage or surface lot. Our parking has 5 parking spaces in ascending order according to the walking distance to the destination. According to the threshold of walking distance, we identify the first 3 places as being very favorable for drivers with invisible disabilities while the last two are very unfavorable. In this simulation, parking hosts two types of entities: normal drivers and drivers with invisible disabilities. When a normal driver arrives, the system assigned to the incoming entity the value 1 and when a driver with invisible disabilities arrives, the system assigned to the incoming entity the value 10. High values go first at each step. We consider that the drivers with invisible disabilities represent 8.2% of the drivers incoming. If there is no available space, the system hands the parking request in a loop so that it is treated first in the availability of seats if it has no driver with invisible handicap in the queue. In this simulation, the Poisson distribution defines the time between two successive arrivals of units to the service system. And the exponential distribution defines the duration or time of carrying out the service (Fig. 2).
4
A. Amari et al.
Fig. 2. Proposed model in arena Rockwell
4 Result and Discuss According to the simulation done on arena we notice that the people with reduced mobility can have more chance to occupy the places with a minimal distance with our priority system. The Fig. 3 below show the occupancy of places with priority with Fig. 4 show the occupancy of places without priority.
Fig. 3. Parking places occupancy with priority
A Smart Parking for Invisible Disabilities
5
Fig. 4. Parking places occupancy without priority
5 Conclusion This article presents a simulation model which aim to improve the quality of service of a smart parking by acting on the allocation system. Our perspective is to find a solution to such a big problem: the invisible disabilities in smart parking management. Designing a scenario simulated through simulation software “Arena Rockwell” using the queuing theory. In the future works, we will model this scenario by mathematic approach to analyze the performance of the proposed system.
References 1. West, J.H., Hall, P.C., Hanson, C.L., Barnes, M.D., Giraud-Carrier, C., Barrett, J.: There’s an app for that: content analysis of paid health and fitness apps. J. Med. Internet Res. 14(3), e72 (2012) 2. Riegel, B., et al.: State of the science: promoting self-care in persons with heart failure: a scientific statement from the American Heart Association. Circulation 120(12), 1141–1163 (2009) 3. Dickstein, K., et al.: ESC guidelines for the diagnosis and treatment of acute and chronic heart failure 2008. Eur. J. Heart Fail. 10(10), 933–989 (2008) 4. Smith, K.P., Christakis, N.A.: Social networks and health. Annu. Rev. Sociol. 34, 405–429 (2008) 5. Merolli, M., Gray, K., Martin-Sanchez, F.: Health outcomes and related effects of using social media in chronic disease management: a literature review and analysis of affordances. J. Biomed. Inform. 46(6), 957–969 (2013)
6
A. Amari et al.
6. Triantafyllidis, A.K., Koutkias, V.G., Chouvarda, I., Maglaveras, N.: Development and usability of a personalized sensor-based system for pervasive healthcare. In: Annual International Conference of IEEE Engineering in Medicine and Biology Society, pp. 6623– 6626 (2014) 7. Krpan, L., Maršanić, R., Milković, M.: A model of the dimensioning of the number of service places at parking lot entrances by using the queuing theory. Tehnički Vjesnik 24(1), 231–238 (2017) 8. Basarić, V., Mitrović, J., Papić, Z.: Passenger car usage for commuting to work as function of limited stay at car parks. Promet – Traffic Transp. 25(4), 323–330 (2013). https://doi.org/10. 7307/ptt.v25i4.322
Smart Water Management: Pillars and Technologies Najat Abdeljebbar(&), Laila Moussaid, and Hicham Medromi System Architecture Team, National Higher School of Electricity and Mechanics ENSEM, Hassan II University Casablanca, Casablanca, Morocco [email protected]
Abstract. The development of cities is related to the efficient management of water. The evolution of smart instruments, the progress made in Information and Communications Technology (Internet of things, Cloud Computing, Big Data analysis) improves water management efficiency and ensures the availability and preservation of this resource. In this article, we have first, proposed 4 major pillars to consider in smart water management, and second, we have explored the key points of the Smart Water Grid. Keywords: Smart water management Smart Water Grid Information and Communications Technology
1 Introduction The geographical position of Morocco makes it vulnerable to the effects of climate change. In fact, natural water resources are increasingly limited. Besides, the demand for water continues to increase because of the population explosion, the expansion of industrial activities and the development of irrigated agriculture. In addition, the natural resources in Morocco are among the lowest in the world. The potential of water resources is estimated at 22 billion m/year. That is the equivalent of 700 m3/inhabitant/year, a threshold that is considered critical and indicating the emergence of shortages and latent water crisis [1]. This alarming situation requires the use of information and communication technology (Internet of Things, Cloud Computing, Big Data analysis) in the management of water resources. In this document, we have proposed, in the first part, the 4 main pillars of smart water management: human, legal/political, economic and technological. In the second part, we focused on the smart water distribution system by identifying the key points of the Smart Water Grid.
2 The Advantages of the Smart Water Management Due to the growing demand for water and the challenges of climate change, water resources are gradually declining. Thus, it is important to find an effective solution for the conservation and preservation of these resources. All water re- sources must be considered as a precious one. A global, optimal, and efficient management of the water must be applied. Rainwater, groundwater, wastewater treatment, the water supply © Springer Nature Switzerland AG 2019 Y. Farhaoui and L. Moussaid (Eds.): ICBDSDE 2018, SBD 53, pp. 7–14, 2019. https://doi.org/10.1007/978-3-030-12048-1_2
8
N. Abdeljebbar et al.
chain, all these aspects must be taken into consideration through a global and effective vision. The smart water management is different of the traditional one by integrating technology into management systems. The development of smart instruments and meters, the progress made in Information and Communications Technology (ICT) (Internet of Things, Cloud Computing, Big Data analysis) improves management efficiency and optimizes process of water production and treatment [2].
3 Pillars of the Smart Water Management The smart water management contributes to the efficient preservation of natural resources. To make that management succeed with the maximum of benefits, it is necessary to identify all the aspects of the smart water management and engage all the stakeholders in its implementation. For this purpose, we have suggested four major pillars of the smart water management (Fig. 1).
Fig. 1. The 4 pillars of the smart water management
3.1
The Human Pillar
The smart water management projects have a direct impact on the lives of citizens. Indeed, smart water management cannot be entirely realized if the society is not aware of the major issues and the importance of preserving natural resources. The behavior of individuals and the rate of awareness are levers in the smart management of water. Water management technologies have been developed to influence the consumer behavior through the provision of real-time information and analysis of water
Smart Water Management: Pillars and Technologies
9
consumption. Interactive analytical tools for engagement have also been developed in this purpose [3]. In fact, home water management solutions are currently able to provide owners with total water consumption data using smart water meters and at intervals as short as 10 s. However, these tools are still rare compared to the electrical energy management solutions. Through the smart water management, the citizen goes from a simple consumer to an active and influential actor. Indeed, the consumer can monitor, control and optimize his consumption and contribute by himself to the preservation of this natural resource. Citizens must be fully sensitized and have an adequate level of education allowing them to assimilate and understand the importance of preserving resources, to change their behavior and to deal with new technologies and digital tools. This is a major challenge given that Morocco still has a high illiteracy rate. Indeed, despite the efforts made since independence, this rate is still estimated at 32% in 2014 [4]. 3.2
The Legal and Political Pillar
The development of smart water management requires an adequate legal and political framework. Political mobilization is needed to support the move towards smart water management and engage all stakeholders in its implementation. Morocco has made significant progress in the regulatory and institutional field, in this case Law 10-95, which consolidated the integrated, participatory and decentralized management of water resources [5]. This water law aims to put in place a national water policy based on a prospective vision that takes into account, on the one hand, the evolution of resources and, on the other hand, national water needs. It provides for legal provisions aimed at rationalizing the use of water, the generalization of access to water, interregional solidarity, the reduction of dis- parities between the city and the countryside within the framework of programs of which the objective is to ensure hydraulic safety throughout the Kingdom [6]. However, and given the importance of water conservation, the integration of the latest innovations in water management and the use of ICT, requires a strong will policy and the drafting of specific laws. Indeed, the smart water management must be used in a well-defined framework, in good harmony with the regulations and the restrictions. It must be framed by policy and consolidated by appropriate laws and regulations. 3.3
The Economic Pillar
The Population growth and pressure on water resources and infrastructure will rapidly exceed the current capacity of cities. Thus, the migration of traditional water management towards smart management must be included in an economic and financial perspective in order to prepare for the future by investing in today’s infrastructures. The economy is the driving force behind the development of smart water management. Large companies with large financial and human capital, small companies known for their flexibility and speed of execution, must all compete for profit maximization with an innovation spirit. However, the major challenge in the development of the smart water management market is the evaluation of profit. Indeed, the direct benefits should not include storage, supply, water treatment and flood control only, but also the benefits
10
N. Abdeljebbar et al.
of healthier and more livable cities. Quantifying indirect effects is even more difficult especially when it comes to the benefits related to the environment and citizens. The evaluation of the interest of smart water management is an important factor in its implementation. Innovations must be made in order to achieve a balance be- tween the benefits and the investment cost. In addition, it is imperative to prove the cost effectiveness of smart solutions over traditional methods to justify the use of smart water management. 3.4
The Technological Pillar
Smart water management is a combination of traditional management systems and the latest technologies. The collection, storage, analysis and transmission of information devices constitute the main lever of steering and action. In this context, several types of meters, sensors and software have been developed to improve the processes of storage, treatment, and distribution of water. The technologies developed are intended to facilitate the management of water. They allow advanced control from the source to the end use. Through the use of measuring instruments, sensors and smart devices, the quantity and quality of water can be monitored and controlled. Leaks can be detected and eliminated. The prediction of incidents and anomalies in the pipes is done in an easier and faster way. Besides, the use of Information and Communications Technology (Internet of Things, Cloud Computing, Big Data analysis) allows to collect, store and transmit a big mass of data, in real time. The information collected would be the basis of an analysis to detect weak points and thus determine the areas of improvement. Data analysis can also be used to anticipate anomalies, forecast demand, optimize storage and improve energy efficiency. Technology platforms, tools and interoperability standards provide the basis for solving real problems with innovative solutions [7].
4 The Smart Water Grid One of the main tools used in the smart management of water and precisely at the water distribution network, is the Smart Water Grid (SWG), the use of SWG as a means management of the water distribution network makes it possible to control, monitor and optimize the consumption of this resource. Indeed, thanks to the SWG, the quality, the pressure and the flow rate of the water can be monitored, the risks of bursting of the pipes can be prevented and the leaks can be detected. The use of SWG gives the user access to the various data and allows him to monitor and manage his consumption by himself. The SWG also provides better visibility and a high level of control, not only on consumption, but also on the state of the facilities, thus facilitating equipment maintenance, anomaly detection and incident prediction. In what follows, we will define the different components of the SWG and identify the three main keys of this system. 4.1
The Components of the Smart Water Grid
The SWG can be defined as an advanced system. It integrated ICT in the management of the distribution system [8]. The implementation of SWG technology allows for
Smart Water Management: Pillars and Technologies
11
optimized and efficient water management. The transmission and distribution of water is automated, monitored and controlled by instruments. The SWG consists of 5 interconnected layers [9] (Fig. 2).
Fig. 2. The components of the Smart Water Grid
4.2
The Keys of the Smart Water Grid
The commissioning of a SWG is imperative to improve the performance of the water distribution system and to ensure the preservation and the effective management of this resource. For this purpose, the SWG must meet the requirements for the water supply process. The four criteria of reliability, efficiency, security and availability constitute an indispensable asset for the implementation of the SWG. Using the latest technologies, the SWG makes it possible to collect a great deal of data, to provide real-time control, to react more quickly, and above all to anticipate any incident or anomaly. The SWG system is designed to respond to the major problems encountered in the management of the water distribution network. As a result, we have defined the three major keys on which the SWG is based (Fig. 3).
Fig. 3. The keys of the Smart Water Grid
Leak Management. One of the major problems of water distribution networks is the appearance of leaks. These leaks not only impact the environment but also generate significant losses and increase direct and indirect costs. The commissioning of SWG
12
N. Abdeljebbar et al.
allows to detect leaks more quickly, to prevent any incident or anomaly, to plan and schedule maintenance and replacement of equipment and pipes. Leakage or loss of water can be assessed as the difference between the volume of water entering the network and the permitted consumption. These leaks may be apparent leaks such as those due to count inaccuracies, or real leaks that caused by pipe bursting, or overflows [10]. Water leakage results from the deterioration of the water distribution system, this could be caused by environmental factors or physical factors related to the condition of the pipes or equipment, or due to operational factors such as pressure and flow velocity. As a result, leak detection is a major concern for improving network performance and ensuring the preservation of water resources. The equipment and methods used to detect leaks are multiple and vary according to the characteristics and properties of the network. • Apparent leaks: Apparent leaks are caused by measuring instrument anomalies or false alarms. To resolve that problem, various approaches have been taken to improve the process of identifying leaks. Indeed, by analyzing the acoustic leakage signals using the Fourier transform, an aggregative approach was developed based on the three models Naive Bayes (NB), Deep Learning (DL), Decision Tree (DT) and this to develop a single more precise answer. Through aggregation, the accuracy of leak detection has increased from 89% to 100% [11]. • Real leaks: To detect real leaks, several equipment and tools are used. The table below summarizes the main means with their advantages and disadvantages [3, 12]. Demand Management. Water demand management is an approach that focuses on controlling demand through the implementation of technological, regulatory or economic measures. The management of water demand is also essential for the evaluation of water prices and investments. Generally, total water demand differs across communities. It can be based on several criteria: the population, the cost of water, the geographical location, the climatic conditions, the development of local, commercial and industrial activity (Table 1). Table 1. Tools and methods for detecting leaks Tools/Methods
Advantages
Disadvantages
Acoustic sensors based on accelerometers Acoustic sensors based on hydrophone sensors High speed pressure sensors Virtual district metering areas (DMA)
Automatic identification of leak locations based on parameters such as pipe size, distance and pipe material Effective detection of leaks in large pipes
Precision for small diameters (less than 300 mm), Sensor range limited to 250 m
Wide detection range of 1.5 Km
High cost
A low resolution in a range of a few hundred meters Monitoring incoming and outgoing flows Water quality issues associated with of sections of the water system network “dead ends” Customer complaints due to the reduction of water pressure in the system Statistical analysis Reveal any discrepancies that may indicate Low precision Need to take into account of pressure and a new leak other parameters flow
Smart Water Management: Pillars and Technologies
13
Quality Management. Respect for water quality is a major issue in the water distribution network because this component directly impacts consumers. To control the quality of the water, several instruments are used. Indeed, a number of real-time water quality monitoring sensors currently exists on the market: • Conventional sensors that directly measure specific water quality parameters such as pH, turbidity and conductivity. • Sensors that detect any deviation from the generic properties of water (such as optics) to cover a broader spectrum of contaminants. • Biosensors that monitor the behavior of living organisms in the water to assess the toxicity of contaminants in water samples.
5 Conclusion Given the scarcity of resources and growing demand, it is now essential to use the future of technological advances. The four pillars: human, political/legal, economic and the key factors of smart water management. Indeed, focusing on the SWG, there are still a number of challenges that remain to go. Real-time sensors and meters readings generate massive amounts of data. It requires good organization and powerful analytical tools to extract useful information. In addition, the SWG implementation often has a lack of compatibility between the various components of the SWG (Sensors, wireless communication means, Data analysis tools….) Thus, the SWG remains a promising field of research.
References 1. Secretariat of State to the Minister of Equipment, Transport, Logistics and Water, in charge of Water. http://www.water.gov.ma/ressources-en-eau/presentation-generale/ 2. Byeon, S., Choi, G., Maeng, S., Gourbesville, P.: Sustainable water distribution strategy with smart water grid. Sustainability 7(4), 4240–4259 (2015) 3. Public Utilities Board Singapore: Managing the water distribution network with a smart water grid. Smart Water 1(1), 4 (2016) 4. Briefing Note from the Office of the High Commissioner for Plan on the Occasion of International Literacy Day, 8 September 2017. https://www.hcp.ma/Note-d-information-duHaut-Commissariat-auPlan-a-l-occasion-de-la-journee-internationale-de-l-alphabetisation-du 8a2009.html 5. State Secretariat to the Minister of Equipment, Transport, Logistics and Water, in charge of Water “Water Policy”. http://www.water.gov.ma/ressources-en-eau/politique-de-leau/ 6. Law No.10-95 on Water “Contributions of the Water Law”. http://www.clefverte.ma/ images/stories/espaceaide/reglementationloi/ReglementationEnvironnement/loi-10-95.pdf 7. Dickey, T.: Smart water solutions for smart cities. In: McClellan, S., Jimenez, J.A., Koutitas, G. (eds.) Smart Cities, pp. 197–207. Springer, Cham (2018) 8. Choi, G.W., Chong, K.Y., Kim, S.J., Ryu, T.S.: SWMI: new paradigm of water resources management for SDGs. Smart Water 1(1), 3 (2016)
14
N. Abdeljebbar et al.
9. Analysis of the water consumption of the Scientific Campus.pdf 10. The IWA Water Loss Task Force_Assessing NRW.pdf 11. El-Zahab, S., Asaad, A., Mohammed Abdelkader, E., Zayed, T.: Collective thinking approach for improving leak detection systems. Smart Water 2(1), 3 (2016) 12. Charalambous, B.: Leak detection and water loss management, p. 6 (2014)
Integrating ICT in Education: An Adaptive Learning System Based on Users’ Context in Mobile Environments Soukaina Ennouamani1(&), Laila Akharraz2, and Zouhir Mahani3 1
National School of Applied Sciences, Ibn Zohr University, Agadir, Morocco [email protected] 2 Faculty of Sciences, Ibn Zohr University, Agadir, Morocco [email protected] 3 School of Technology, Ibn Zohr University, Agadir, Morocco [email protected]
Abstract. The integration of Information and Communication Technologies in the educational area makes courses at universities more accessible anytime and anywhere. However, learners have different background knowledge, learning styles and they learn in different contexts and situations. For this purpose, we suggest in this paper a new model of adaptive mobile learning systems that considers a combination of learner’s characteristics and context in order to provide each learner with the most suitable learning content and format of presentation. Keywords: ICT Education Adaptive computing
Smart learning Mobile learning
1 Introduction Nowadays, the progress of mobile technologies is obviously required because of people mobility which has become a significant issue in many facets of our daily lives and in different domains. One of these domains is the educational sector, where mobile devices are used as a principal platform for teaching and learning [1]. Mobile learning is considered as a flexible learning approach that has the strength of making learning more personalized and adaptive. In fact, it has been recognized that individualized learning is much more effective than classroom learning [2] and that can change the traditional aspects of acquiring skills and knowledge. On the other hand, learners may have different preferences, learning styles and knowledge that should be considered for the design of adaptive learning assistance [3]. In addition to this, the massive quantity of available information makes a second challenge in adaptive learning domain. Therefore, we are supposed not only to provide learners with learning materials that they need, but also to select the most appropriate and relevant ones.
© Springer Nature Switzerland AG 2019 Y. Farhaoui and L. Moussaid (Eds.): ICBDSDE 2018, SBD 53, pp. 15–19, 2019. https://doi.org/10.1007/978-3-030-12048-1_3
16
S. Ennouamani et al.
“Learner adaptation is defined as matching content to the abilities and preferences of individual learners” [3]. In this perspective and according to [4], the adaptive systems could provide the most suitable contents according to learners’ characteristics and background knowledge. For the same reason, we aim in this research to integrate context parameters in order to enhance the adaptation of learning content and format, both together to satisfy learners’ requirements.
2 An Overview of the Proposed Framework Previous contributions [5–11] provide different m-learning systems that allow to adapt learning using different characteristics. Theses contributions are limited in term of integrating mobile characteristics to determine the learner’s context, and they are targeted to a specific field or domain of education (language teaching, museums, computer sciences, mathematics). For this reason, we suggest a new model that can be used for any subject and that integrates the missed functionalities of other works. It integrates knowledge level and its progress, learning styles, behavior and interactions with the system, learner’s preferences and satisfaction. Furthermore, environmental parameters are also considered such as location, noise detection and learner’s motions. 2.1
Architecture of the System
Three principal models of any adaptive mobile learning system are: learner model, domain model and adaptation model [12]. We have added other models to meet with the objective of our contribution, such as context model (Fig. 1). These models are described as follows: • Learner Model: This component performs the learner modeling using the characteristics provided by the learner. It gives a map of user’s learning styles, knowledge, behavior, and preferences. This model represents the core component of our proposal • Domain Model: It is a hierarchical representation of all learning resources that are stored in the learning materials database. This model defines the grouping of lessons, tests and their connections for each learning topic. • Adaptation Model: It is the main model of any adaptive system. It performs adaptation through collecting the learner model data with courses from the domain model in order to generate the suitable learning materials. • Context Model: It allows to collect the context values using mobile sensors in order to predict the learner’s current location (GPS sensor), physical activities (accelerometer sensor) and to detect if there is any noise around him/her (mobile microphone).
Integrating ICT in Education
17
Fig. 1. Mobile learning system architecture
2.2
The Proposed Process of Adaptation
Figure 2 represents three steps in the adaptation process which are: Step 1: Acquisition It deals with all parameters related to the learner and the device that he/she uses. These parameters can be detected and collected as follows: • Knowledge test: It helps to estimate the learner’s knowledge level by generating a pre-test once the course is selected. Pre-tests are created via questions stored in learning material database in accordance with the chosen courses. • Index Learning Styles (ILS): It is a multiple choice questionnaire of 44 questions created by Felder and Soloman [13] that aims to evaluate the learner preferences. It is a data collection tool based on the FSLSM [14] and classified according to 4 dimensions: Active/Reflective, Sensing/Intuitive, Visual/Verbal or Sequential/ Global. These dimensions respectively describe the variation of each learner’s information processing, information perception, information reception, and information understanding. • Context parameters (Proximity sensor, GPS sensor, microphone, accelerometer): The current learner’s location is detected by the GPS sensor to determine where he/she is using the application. For example, if the learner takes the course at home, then text or video format will be more suitable. The distance between learner and device is taken into account using the proximity sensor. It represents an important feature to decide whether text format should be generated when the device is near enough or not. The learner’s physical activities and motions are detected using the accelerometer sensor. This parameter has a significant impact on the learning format, say, text format is more suitable if the learner does not move
18
S. Ennouamani et al.
from a place to another during the learning activity. The noise detected by the mobile microphone is also required to indicate whether the learner is able to listen to an audio, to watch a video learning object or not. Step 2: Modeling The combination of the characteristics and data cited above allows to create an image about the learner’s context and characteristics in order to be used during the treatment level of the adaptation process. The generated model plays a major role in the following step, allowing to determine the adequate content and mode of presentation. Step 3: Treatment In this step, the adaptation engine receives the learner model as well as the available courses to perform the adaptation mechanism that consists of two stages: content adaptation and format adaptation. The first stage aims to select the appropriate learning contents based on learner’s knowledge and learning style. In addition to this, it also collects the related test questions in order to create test assessment. Afterwards, the selected contents need to be presented in a convenient format, according to the leaning styles, preferences and context parameters. The output of this treatment, is the adapted learning materials that will be received by the learner through the mobile application interface.
Fig. 2. The proposed framework adaptation process
3 Conclusion The choice of significant characteristics is an important task for increasing adaptive learning efficiency in a mobile environment. For this purpose, we have selected in this paper a set of parameters that meet with our objective which is the adaptation of learning contents and format. A learner diagnosis mechanism was applied to understand each learner’s needs and to take him/her as a specific case that requires a specific
Integrating ICT in Education
19
learning. Furthermore, our model has no limits concerning pre-determined fields or disciplines. It is an open framework that can be implemented for any kind of courses. Our future work will deal with the implementation of the proposed framework as well as its evaluation and test with a group of students. It will also help to have an available platform for further functionalities in accordance with adaptation aspects. Finally, this research area is continuing to grow and spread, and we are supposed to go by this current trends of educational applications.
References 1. Ali, A., Alrasheedi, M., Ouda, A., Capretz, L.F.: A study of the interface usability issues of mobile learning applications for smart phones from the users perspective. Int. J. Integrating Technol. Educ. 3, 1–16 (2014). https://doi.org/10.5121/ijite.2014.3401 2. Desmarais, M.C., Baker, R.S.J.D.: A review of recent advances in learner and skill modeling in intelligent learning environments. User Model. User-Adapt. Interact. 22, 9–38 (2012). https://doi.org/10.1007/s11257-011-9106-8 3. Huang, H.-C., Wang, T.-Y., Hsieh, F.-M.: Constructing an adaptive mobile learning system for the support of personalized learning and device adaptation. Procedia Soc. Behav. Sci. 64, 332–341 (2012) 4. Mampadi, F., Chen, S.Y., Ghinea, G., Chen, M.-P.: Design of adaptive hypermedia learning systems: a cognitive style approach. Comput. Educ. 56, 1003–1011 (2011). https://doi.org/ 10.1016/j.compedu.2010.11.018 5. Yang, G., Kinshuk, K., Graf, S.: A practical student model for a location-aware and contextsensitive personalized adaptive learning system. In: 2010 International Conference on Technology for Education, pp. 130–133 (2010) 6. Bouneffouf, D., Bouzeghoub, A., Gancarski, A.L.: Following the user’s interests in mobile context-aware recommender systems: the hybrid-e-greedy algorithm. In: 2012 26th International Conference on Advanced Information Networking and Applications Workshops, pp. 657–662 (2012) 7. Schiaffino, S., Garcia, P., Amandi, A.: eTeacher: providing personalized assistance to elearning students. Comput. Educ. 51, 1744–1754 (2008). https://doi.org/10.1016/j.compedu. 2008.05.008 8. Bachari, E.E., Abelwahed, E.H., Adnani, M.E.: E-learning personalization based on dynamic learners’ preference (2011) 9. Tortorella, R.A.W., Graf, S.: Considering learning styles and context-awareness for mobile adaptive learning. Educ. Inf. Technol. 22, 297–315 (2017). https://doi.org/10.1007/s10639015-9445-x 10. Dlab, M.H., Hoić-Božić, N., Botički, I.: A design-based approach to developing a mobile learning system. Int. J. Soc. Behav. Educ. Econ. Bus. Ind. Eng. 11, 2347 (2017) 11. Tseng, J.C.R., Chu, H.-C., Hwang, G.-J., Tsai, C.-C.: Development of an adaptive learning system with two sources of personalization information. Comput. Educ. 51, 776–786 (2008). https://doi.org/10.1016/j.compedu.2007.08.002 12. Ennouamani, S., Mahani, Z.: An overview of adaptive e-learning systems. In: 2017 Eighth International Conference on Intelligent Computing and Information Systems (ICICIS), pp. 342–347 (2017) 13. Felder, R.M., Soloman, B.A.: Index of learning styles questionnaire. https://www.webtools. ncsu.edu/learningstyles/ 14. Felder, R.M., Silverman, L.K.: Learning and teaching styles in engineering education. Eng. Educ. 78, 674–681 (1988)
Modified Strategy of Direct Torque Control Applied to Asynchronous Motor Based on PI Regulators Soukaina El Daoudi(&), Loubna Lazrak, Chirine Benzazah, and Mustapha Ait Lafkih Laboratory of Automatic, Energy Conversion and Microelectronics (LACEM), Faculty of Sciences and Technology, University of Sultan Moulay Slimane, Beni Mellal, Morocco [email protected]
Abstract. Direct Torque Control (DTC) is known to generate a fast and robust response in asynchronous engines. Yet, during steady state, observable flux, torque and current pulsations arise. This paper presents a study of modified direct torque control applied to a three-phase high performance asynchronous motor. The proposed strategy is based on PI regulators instead of the hysteresis comparators to control both the stator flux and torque whose parameters are determined from modulus and symmetrical optimum criterions. Since the performance of a feedback control system relies on the accuracy of the response signal, an estimator of stator flux and torque is presented. The control system benefits from the advantages of field oriented control (FOC) and conventional direct torque control (DTC) while avoiding some of the disadvantages of either of the two control methods. Performances of the asynchronous motor control are checked by simulations under MATLAB/SIMULINK software. Keywords: Modified DTC Conventional DTC FOC Asynchronous motor Modulus criterion Symmetrical optimum criterion PI regulator Torque/Flux estimator
1 Introduction The three-phase asynchronous motor is the most used electric motor worldwide for industrial applications. Simple in terms of design, robust and economical to use. It is the most promising drive solution at the level of low cost and high quality. Due to those advantages, numerous efforts by researchers have been made to develop the strategy of control for this type of electric motors. Among those strategies; there is the vector control strategies which consist of field oriented control (FOC), direct torque control (DTC), etc. In general, the vector control methods were based on the same idea that transforms the motor’s three phase variables into two phase variables to separate of the motor parameters as flux-component and torque-component [1]. In the same way, the FOC is a strategy that assigns high performance and quick dynamic response over a wide speed range to asynchronous motors by decoupling their variables; but it uses a rather high number of regulation loops which leads to a slow dynamics of the flux [2]. © Springer Nature Switzerland AG 2019 Y. Farhaoui and L. Moussaid (Eds.): ICBDSDE 2018, SBD 53, pp. 20–26, 2019. https://doi.org/10.1007/978-3-030-12048-1_4
Modified Strategy of Direct Torque Control
21
The direct torque control (DTC) method was proposed in 1980 by Takahashi [3], this method has become one of the high performance control strategies for motors to provide a very fast torque and flux control. It is based on comparing the reference values of the stator flux and torque to their estimated ones, the resultant errors are fed into a two level and three level hysteresis comparators to regulate the stator flux and the torque respectively, controlling the motor in the two directions of rotation. So, with this method, it is possible to directly control the inverter states by selecting the appropriate state through the “optimal switching table” in order to reduce the torque and flux errors within the prefixed band limits [2–4]. However, beside the advantage features such as fast dynamic response, there exist some problems associated with DTC, notably: lowspeed operation, high current and torque ripples, variable switching frequency and high noise level, as well as high sampling frequency needed for digital implementation of hysteresis comparators [5]. Most of above disadvantages can be eliminated by using a modified strategy. This strategy is constituted by linear PI regulators instead of the hysteresis comparators which calculate the required stator voltage vectors synthesized by pulse width modulator ‘PWM’ technique. The use of the PWM modulation ensures constant switching frequency which prevents the system from the high switching losses, ease of implementation and compatibility with recent digital microprocessors. The tuning methods used to define the regulators parameters are the symmetrical optimum and modulus criterions.
2 Modeling of the Asynchronous Motor in (d, q) Frame Respecting the classic simplifying assumptions, the model of the asynchronous motor described by Park’s equations in the (d, q) frame is presented below [6]: Vds ¼ Rs Ids þ
duds dt xs uqs dudr dt xg uqr
duqs dt þ xs uds duqr dt þ xg udr
; Vqs ¼ Rs Iqs þ
0 ¼ Rr Idr þ ; 0 ¼ Rr Iqr þ dX J dt ¼ Te Tr f X ; Te ¼ 34 P uds Iqs uqs Ids
ð1Þ
Where: Vds , Vqs are the stator voltages, Ids , Iqs , Idr , Iqr are the stator and rotor currents, udr , uqr are the rotor flux components in (d, q) reference frame, Rs , Rr stator and rotor resistances whereas xg is the slip frequency that is obtained from subtracting the rotor pulsation xr from the stator one xs. Ω is the Motor speed, while J, f and Tr are the Moment of inertia, Coefficient of friction and Load torque respectively. With P is the number of pole pairs.
3 Modified Torque Control Strategy To control the engine parameters; torque and stator flux, the proposed control is based on PI regulators which improves the dynamics of the system while eliminating the static error between the estimated magnitudes and the references. The block diagram of
22
S. El Daoudi et al.
the entire system is represented in Fig. 1. The regulation is done according to two loops; the first one is for controlling the torque and the second one is for the stator flux control.
Fig. 1. Block diagram of the modified direct torque control via two-level inverter applied to asynchronous motor.
The two closed loops of the stator flux and torque are based on five blocks each; the synoptic scheme is presented below in Fig. 2 [7]:
Fig. 2. Block diagram of the closed loop control taking into account the delays of the process.
It consists of a PI controller; processing and feedback delays, inverter and the asynchronous motor block within the open loop transfer function below: GðsÞ ¼ Kp
1 þ ss 1 1 1 ss Tp s þ 1 To s þ 1 Tf s þ 1
ð2Þ
Where: KP and s are respectively the proportional gain and the integration time. Tp is the processing/execution time of the algorithm, To is the inverter dead time and Tf is the time delay of the feedback filter. The block contained the asynchronous motor, design the equation of the parameter to settle, so for each parameter (stator flux or torque) there is a specified equation.
Modified Strategy of Direct Torque Control
3.1
23
Torque Control
Considering the equation of the electromagnetic torque, the control will be described using the following equation: 3 Te ¼ P uds Iqs uqs Ids 4
ð3Þ
Taking into account that the quadrature component of the stator flux is zero, this makes it possible to have a linear function between the torque and the quadrature stator current. Since the torque control proposed in this paper consists in extracting the stator voltage in quadrature, Eq. (3) can be rewritten using current expression as a function of the quadrature voltage (1): Te ¼
3P uds Vqs xs uds 4 Rs
ð4Þ
Equation (4) shows that the coupling between torque and flux is omitted, since the quantity is considered a disturbance and the direct component of the flux is a constant, hence the open loop transfer function then becomes: GTo ðsÞ ¼ Kp
1 þ ss 1 : :Kt ss Tu s þ 1
With: Kt ¼
3P u 4 Rs ds
ð5Þ
The technique used in the control of the torque is that of modulus criterion [8–10]. According to this criterion, the PI controller parameters are presented below where n is the damping factor of the system: 8 > < Kp ¼ 1 Kp ð1 þ Kt Þ2 > : ¼ 2 : s 4n Tu Kt
3.2
ð6Þ
Flux Control
Considering the equation of the direct stator voltage, the control will be described using the following equation: Vds ¼ Rs Ids þ
duds xs uqs dt
ð7Þ
By neglecting the voltage drop in the stator resistance and since the quadrature component of the stator flux is zero, the function between the stator flux and the direct stator voltage becomes linear. Since the sum of the small time constants which includes the static delay of the inverter (Dead Time) To, execution time of the algorithm Tp and
24
S. El Daoudi et al.
Tf the delay of the feedback filter is defined by: Tu ¼ Tf þ Tp þ To , the open loop transfer function will be written as: Guo ðsÞ ¼ Kp
1 þ ss 1 1 : : ss Tu s þ 1 s
ð8Þ
For the design of the stator flux controller parameters, the Symmetry optimal criterion is applied [8–10]. According to this criterion, the PI controller parameters are: 8 1 >
: :s ¼ 8ðTu þ To Þ2
ð9Þ
4 Torque and Flux Estimator From the point of view of control theory, the performance of a feedback control system relies on the accuracy of the feedback signal. For accurate and robust control characteristics, the most accurate signal must be selected as the feedback signal. In our case, the control used is based on an estimation of the torque and the stator flux, the latter can be evaluated more precisely than the rotor flux which has a certain limitation of performance due to the effects of “detuning”. The feedback parameters can be evaluated in the reference (a, b) and with the equations below [11]: Z ^ sa ¼ u
0
t
Z ^ sb ¼ ðVsa Rs Isa Þdt ; u
0
t
Vsb Rs Isb
^ sb u dt ; hs ¼ arctg ^ sa u
ð10Þ
Where: hs is the stator flux vector’s position angle. The motor electromagnetic torque estimator is based on the torque equation according to the estimated fluxes and the stator currents measured: 3 ^ as Ibs u ^ bs Ias : T^e ¼ P u 4
ð11Þ
5 Results and Discussion In order to prove the performance of the dynamics of the system control, simulations are made using MATLAB/SIMULINK software. The results of the simulations concern a three-phase squirrel cage asynchronous motor with a power of 1.5 kW and a switching frequency of 5 kHz. Initially it was powered up and had a vacuum start. Then a load is applied at t = 0.25 s.
Modified Strategy of Direct Torque Control
25
Fig. 3. Stator currents (a, b, c) Fig. 4. The Stator flux (Wb) Fig. 5. The electromagnetic torque with the load applied at 0.25 s
The starting current of the motor is controlled at 4 times its nominal value and a reference of the progressive stator flux is applied (Fig. 3). The steady state of the stator current is reached around 0.15 s, and then it stabilizes at its load value which is relatively large and very responsive but also reasonable since the engine is characterized by the presence of a significant air gap. The load is applied at t = 0.25 s, the current reaches rapidly its stationary value (5.2 A). Figure 4 shows that the flux’s magnitude is maintained at its reference, we can also note that it has a very low level of ripples. As can be clearly seen in Fig. 5, the electromagnetic torque quickly reaches its maximum value which is beneficial to achieve a quick start, then returns to zero since it is unladed. When the load is applied, the motor reaches its operating point quickly with a delay not exceeding 0.1 s. The figures above prove the effectiveness of the proposed approaches which give a good response with minimized ripples on the torque and stator flux (Table 1).
Table 1. Motor parameters used in the simulation Heading level Nominal speed/frequency Nominal current Stator resistance Stator inductance Rotor resistance Rotor inductance Mutual inductance Moment of inertia Friction factor Stator flux Sampling time
Font size and style 1.5 kW/50 Hz 6.4 A 4.85 Ω 0.274 H 3.805 Ω 0.274 H 0.29 H 0.031 kg.m2 0.00114 N.m.s/rd 0.9960 Wb 5 ls
26
S. El Daoudi et al.
6 Conclusion Selection of the best control scheme is crucial for motor applications on account of the motor requirements in industry. This paper has aimed to propose a modified strategy of the direct torque control where the hysteresis comparators were replaced by PI regulators to prevent the large unpleasant audible noises induced by the hysteresis comparators to maintain its constant bands. The regulators in use were tuned following the modulus and symmetrical optimum criterions. Those techniques improved significantly the system’s control by reducing the response time and eliminating the static error. Summarizing, the results show that the modified DTC scheme is convenient in both areas where the fast dynamic performance has primary importance and when high torque quality is demanded.
References 1. Korkmaz, F., Topaloğlu, İ., Çakir, M.F., Gürbüz, R.: Comparative performance evaluation of FOC and DTC controlled PMSM drives. In: 4th International Conference on Power Engineering, Energy and Electrical Drives, Istanbul, Turkey, pp. 705–708 (2013) 2. Liu, Y., Shi, L., Zhao, L., Li, Y.: The FOC and DTC scheme in a high power electrically excited synchronous motor based flywheel energy storage system. In: International Power Electronics and Application Conference and Exposition, Shanghai, China (2014) 3. Takahashi, I., Noguchi, T.: A new quick-response and high efficiency control strategy of an induction motor. IEEE Trans. IA 22(5), 820–827 (1986) 4. El Ouanjli, N., Derouich, A., El Ghzizal, A., Ali Chebabhi, A., Taoussi, M.: A comparative study between FOC and DTC controls of the Doubly Fed Induction Motor (DFIM). In: 3rd International Conference on Electrical and Information Technologies, Rabat, Morocco (2017) 5. Mohd, A.I., Nik Idris, N.R.: Torque ripple reduction and fast torque control in DTC of induction machine using overlapping triangular-based constant frequency torque controller. In: IEEE International Conference on Power and Energy (PECon) (2016) 6. Cherfia, N., Kerdoun, D.: Wind energy conversion systems based on a DFIG controlled by indirect vector using PWM and SVM. Int. J. Electr. Comput. Eng. (IJECE) 6(2), 549–559 (2015) 7. Bajracharya, C., Molinas, M., Are Suul, J., Undeland, T.M.: Understanding of tuning techniques of converter controllers for VSC-HVDC. In: Nordic Workshop on Power and Industrial Electronics, 9–11 June 2008 8. Kazmierkowski, M.P., Tunia, H.: Automatic Control of Converter Fed Drives. Elsevier, Amsterdam-London-New York-Tokyo (1994) 9. Papadopoulosa, K.G., Margarisb, N.I.: Extending the symmetrical optimum criterion to the design of PID type-p control loops. J. Process Control 22(1), 11–25 (2012) 10. Kazmierkowski, M.P., Krishnan, R., Blaabjerg, F.: Control in Power Electronics Selected Problems. Academic Press, Heidelberg (2002) 11. Ouboubker, L., Khafallah, M., Lamterkati, J., El Afia, A.: High torque control performance utilizing a three level inverters with a simple switching strategy of induction machine. In: International Conference on Electrical and Information Technologies (ICEIT) (2016)
Managing Temporal and Versioning Aspects of JSON-Based Big Data via the sJSchema Framework Safa Brahmia1, Zouhaier Brahmia1(&), Fabio Grandi2, and Rafik Bouaziz1 1
University of Sfax, Sfax, Tunisia [email protected], [email protected], [email protected] 2 University of Bologna, Bologna, Italy [email protected]
Abstract. Several modern applications (e.g., Internet of Things, online social networks), which exploit big data, require a complete history of all changes performed on these data and their schemas (or structures). However, although schema versioning has long been advocated to be the best solution for this issue, currently there are no available technical supports, provided by existing big data management systems (especially NoSQL DBMSs), for handling temporal evolution and versioning aspects of big data. In [14], for a disciplined and systematic approach to the temporal management of JSON-based big data in NoSQL databases, we have proposed the use of a framework, named sJSchema (temporal JSON Schema). It allows defining and validating temporal JSON documents that obey to a temporal JSON schema. A sJSchema schema is composed of a conventional (i.e., non-temporal) JSON schema annotated with a set of temporal logical and temporal physical characteristics. Moreover, since these two components could evolve over time to respond to new applications’ requirements, we have extended sJSchema, in [17], to support versioning of conventional JSON schemas. In this work, we complete the figure by extending our framework to also support versioning of temporal logical and physical characteristics. Indeed, we propose a technique for temporal characteristics versioning, and provide a complete set of low-level change operations for the maintenance of these characteristics; for each operation, we define its arguments and its operational semantics. Thus, with this extension, sJSchema will provide a full support of temporal versioning of JSON-based big data at both instance and schema levels. Keywords: Big data NoSQL JSON JSON schema sJSchema Conventional JSON schema Temporal JSON schema Temporal logical characteristic Temporal physical characteristic Schema change Schema versioning
© Springer Nature Switzerland AG 2019 Y. Farhaoui and L. Moussaid (Eds.): ICBDSDE 2018, SBD 53, pp. 27–39, 2019. https://doi.org/10.1007/978-3-030-12048-1_5
28
S. Brahmia et al.
1 Introduction Nowadays, big data [1, 2] are being stored and used in various applications like Internet of Things, healthcare applications, online social networks, big science projects, and smart cities. Moreover, both structures (or definitions) of big data and their instances are evolving over time and changing at a very high speed, to reflect changes in users’ requirements or in the reference world of the database. Moreover, several modern applications, which exploit big data, require a complete history of all big data versions and all changes performed on both these data and their schemas [3], in order to allow (i) recovering past big data versions, (ii) tracking changes over time, and (iii) executing temporal queries [4] on temporal big data. However, although the schema versioning technique (which consists in creating a new schema version, each time a schema change is applied, while preserving old schema versions with their corresponding data) [5] has long been advocated to be the best solution for this issue, currently there are no available technical supports, provided by state-of-the-art big data management systems (especially NoSQL database management systems [6–13]), for handling both temporal evolution and versioning aspects of big data. Therefore, the designers and developers of the aforementioned applications have to proceed in an ad hoc manner when they should deal with big data evolution while keeping track of all versions of big data and their schemas, or when they should allow time-slice queries to be evaluated, or when it is required e.g. to specify a schema for time-varying big data. In order to efficiently manage and query big data evolution over time, we think that we should have big data management systems with built-in temporal support. For that purpose, in our previous work [14], for a disciplined and systematic approach to the temporal management of JSON-based big data in NoSQL databases, we have proposed the use of a framework, named sJSchema (temporal JSON Schema). This latter is an infrastructure (i.e., a data model and suite of tools) that allows the NoSQL Database Administrator (NSDBA) to create and validate temporal JSON documents (which store time-varying big data) through the use of a temporal JSON schema (which defines the structure of these temporal big data and to which obey the temporal JSON documents). A sJSchema schema consists in a conventional (i.e., non-temporal) JSON schema [15] annotated with a set of temporal logical and temporal physical characteristics. (i) Logical characteristics identify whether a component (e.g., property, object) of the conventional schema varies over valid time and/or transaction time, whether its lifetime is described as a continuous state or a single event, whether the item itself may appear at certain times (and not at others), and whether its content changes. (ii) Physical characteristics specify the timestamp representation options chosen by the NSDBA, such as where the timestamps are placed and their kind (i.e., valid time or transaction time) and the kind of representation adopted. The location of timestamps is largely independent of which components vary over time. Timestamps can be located either on time-varying components (as specified by the logical characteristics) or somewhere above such components. Two JSON documents with the same logical characteristics will look very different if we change the location of their physical timestamps. Changing some aspect of only one timestamp can make a big difference in the representation.
Managing Temporal and Versioning Aspects of JSON-Based Big Data
29
It is worth mentioning that the temporal logical and physical characteristics are orthogonal and are independently maintained, while they are stored together in a single JSON document [16], named the temporal characteristics document and associated to the conventional JSON schema. Notice that a full description of the sJSchema framework (its architecture, its functioning, its tools …), which cannot find place in this paper due to space limitations, can be found in [14, Sect. 3]. In its initial definition [14], sJSchema was proposed as an infrastructure for managing JSON documents with time-varying instances that are valid to a static schema; only instance versioning is supported at that stage. Nevertheless, since each one of the three components of a sJSchema schema (i.e., conventional JSON schema, temporal logical characteristics, and temporal physical characteristics) could also evolve over time to respond to new applications’ requirements, we have augmented sJSchema, in a previous work [17], to support versioning of conventional JSON schemas. In this work, we complete the figure by extending our framework to also support versioning of temporal logical and temporal physical characteristics. Indeed, we propose a technique for temporal characteristics versioning, and provide a complete set of low-level change operations for the maintenance of these characteristics. For each one of the proposed operations, we define its arguments and its operational semantics. Thus, with this extension, sJSchema will fully support temporal versioning of JSON-based big data at both instance and schema levels, and consequently will provide a fully-fledged history of big data changes. The remainder of this paper is organized as follows. Section 2 presents our approach for versioning of temporal (logical and physical) characteristics. Section 3 introduces the schema change operations that we propose for the maintenance of temporal logical and physical characteristics. Section 4 illustrates our approach through an application example. The last section summarizes the paper and gives some remarks about our future work.
2 Versioning of Temporal Logical and Physical Characteristics In this section, we describe how sJSchema logical and physical characteristics are versioned in our approach. The first step of a schema versioning process is the creation of the first temporal JSON schema version: the NSDBA creates a conventional JSON Schema document (i.e., a classical JSON Schema file) annotated with some logical and physical characteristics in an independent document (which is stored as a JSON file). Consequently, the system generates the temporal JSON schema (also stored as a JSON file) that ties together the conventional schema and the temporal characteristics. In further steps of the versioning process, when necessary, the NSDBA can independently change the conventional schema, the temporal logical characteristics or the temporal physical characteristics. Changing the conventional schema leads to a new version of it. Similarly, changing temporal logical and/or physical characteristics leads to a new version of the whole
30
S. Brahmia et al.
temporal characteristics document. Therefore, the temporal JSON schema is automatically updated after each change to the conventional JSON schema or to the temporal characteristics document, in order to take into account the new version of the corresponding changed component. In this paper, we do not deal with changes to the conventional schema. Notice that change operations performed by the NSDBA are in general high-level, since they are usually conceived having in mind high-level real-world object properties. However, in this paper we have studied low-level change operations and not high-level ones, since we think that each high-level change operation can be expressed as a sequence of low-level change operations. Notice here that we have proposed all lowlevel operations that are necessary for performing any simple/complex change to the temporal characteristics document (based on the structure of this latter).
3 Operations for Changing Temporal Logical and Physical Characteristics In this section, we define low-level operations for changing temporal logical and physical characteristics in sJSchema. For each one of these operations, we provide a description of its arguments and its operational semantics. The definition of all these operations is based on (i) the schema (or the structure) of the temporal characteristics document (TCD) that we have constructed, as explained in the first subsection, titled “The Schema of Temporal Characteristics Documents”, and (ii) some common design choices that we have introduced in the second subsection, titled “Design Choices”. 3.1
The Schema of Temporal Characteristics Documents
In the architecture of our sJSchema framework [14], the schema for the temporal logical and physical characteristics is given by TCSchema (box 5) which is a JSON Schema [15] file that describes the structure of any temporal characteristics document. This schema has been only mentioned in our previous work [14] without being provided. In this paper, we define its JSON Schema code. Due to space limitations, the JSON Schema code of TCSchema can be found in the online appendix [18, Sect. A1]. The study of TCShema allows us determining the list of components of any TCD (e.g., logical, logicalItems, timeDimension, validTime, transactionTime, physical, stamps, stampKind, stampBounds). This list has permitted us proposing all possible change operations that could be executed on each component, by creating, modifying or dropping it. 3.2
Design Choices
The definition of the primitives will obey the following design choices: • All operations must have a valid temporal characteristics document (TCD) as input and must produce a valid TCD as output. • All operations need to work on a JSON file storing the TCD, whose name must be supplied as the first argument.
Managing Temporal and Versioning Aspects of JSON-Based Big Data
31
• For all operations, arguments which are used to identify the object on which the operation works are in the first place of the argument list. • Components which are just containers for other components (e.g., logical, physical) can be managed by the operations concerning the components, without specific operations acting on them (i.e., the container is created when the first subcomponent is created and is deleted when the last sub-component is deleted). • Operations adding objects with possibly optional properties have the values for all the properties as arguments; empty places in the argument list stand for unspecified optional properties. • We use Add…/Change… operations for all components (objects or properties) which have multiple occurrences (e.g., “logicalItem”, “stamp”); a single Set… operation is used for adding/changing components with occurrences K, then repeat the process from Step-1. • Step-7. Else if the number of objects in new-generation = K, then consider them as K clusters, and go to Step-9. • Step-8. Else, retrieve the previous content of new-generation from previous-newgeneration[], and keep merging every two clusters whose the distance between their associated centers is the closest. • Step-9. Exit. 2.3
Algorithm of Method
Given a set S of high-dimensional objects to be grouped into K clusters, and each object is getting weighted to W ← 0; let M[] be a set of objects whose the object N is the closest. Foremost, save S in new-generation[]. Algorithm 1: The Method Algorithm 1. 2. 3. 4. 5. 6. 7. 8.
elect[][]← Select(new-generation[]). elect[][]← Filter(elect[][]) {C1,C2,…,Cj}← GenerateClusters(elect[][]) previous-new-generation[]←new-generation[] new-generation[]← CalculateCenters({C1,C2,…,Cj}) if(Length(new-generation[]) > K) GotoStep(1) elseif(Length(new-generation[]) = K) GotoStep(9) else While(Length(previous-new-generation[]) > K) then MergeTheClosest (previous-new-generation[]) 9. Exit()
2.4
Outline of the Proposed Method
The purpose of step-1 is to detect representative objects in datasets, since the object being referenced as the closest to many other objects will be considered as a representative object. This quality could be quantified using the weight notion. Here the weight of object A is calculated according to the number of objects whose A is the closest. The greater the weight, the most representative the object will be. However, every object getting weighted to zero will be treated as an outlier.
80
B. Kenidra and M. Benmohammed
Step-2 eliminate all outlier objects and load each object whose the weight is different than zero in a matrix elect[][], each line of elect[][] is formed as: {W,N,M}, where M is a set of objects whose the object N is the closest, and W is the calculated weight of object N. In step-3, each line of elect[][] will form a cluster C, where C = {N} U {m} and W is the number of objects belonging to C. In step-4 and step-5 the centers associated to all generated clusters will be loaded into new-generation[], after saving its content in previous-new-generation[]. If the number of the centers is greater than the wanted clusters number then, it is necessary to repeat the process from the beginning of step-1, in order to reduce the number of clusters once again. And if the number of these centers equals to the wanted clusters number then, the associated clusters of these centers will be considered as final clusters. But if the number of these centers is less than the wanted clusters number, we must go back to the last saved content of previous-new-generation[], and we keep merging every two clusters whose the distance between their associated centers is the closest until reducing the number of clusters to K clusters, and that is what step-6, step-7, step-8 and step-9 are about.
3 Experiments and Results 3.1
Datasets Used in Experiments
To assess the performance of the proposed approach, an experimental study was conducted using 20 different publicly available gene expression dataset, having the proprieties shown in Table 1. 3.1.1 DNA-Methylation Datasets The first datasets comprise DNA methylation arrays downloaded from The Cancer Genome Atlas (TCGA). The data is composed of two groups, a group of patients (with breast cancer) and a group of 39 healthy individuals, each individual/patient is being associated to 17976 genes, thus, 17976 (215 + 39), i.e., 4 565 904 experimental measurements of gene expression levels have been taken. The clustering process is used to discriminate normal from tumor tissues. The major problem of all clustering algorithms for methylation analysis is the high dimensionality of input space. There are usually thousands to hundreds of thousands of genes treated as dimensions. Consequently, the high dimensionality would significantly increase the running time and the computational cost.
A Non-stochastic Method for Clustering of Big Genomic Data
81
Table 1. Characters gene expression datasets used in evaluation. Dataset name
Tissue
Methylation datasets alizadeh-v2 alizadeh-v3 armstrong-v1 bhattacharjee bredel chowdary
Breast, Cancer Blood Blood Blood Lung Brain Breast, Colon Bladder Lung Bone, Marrow Lung Multi-Tissue Colon Brain Brain Brain Blood Prostate Breast Bone Marrow
dyrskjot garber golub-v1 gordon khan laiho nutt-v3 pomeroy-v1 pomeroy-v2 shipp-v1 singh west yeoh-v1
Total samples 254
Classes number 2
Samples per class 215, 39
Total genes 17976
62 62 72 203 50 104
3 4 2 5 3 2
42, 9, 11 21, 21, 9, 11 24, 48 139,17,6,21,20 31, 14, 5 62,42
2093 2093 1081 1543 1739 182
40 66 72
3 4 2
9, 20, 11 17, 40, 4, 5 47, 25
1203 4553 1877
181 83 37 22 34 42 77 102 49 248
2 4 2 2 2 5 2 2 2 2
31, 150 29, 11, 18, 25 8, 29 7, 15 25, 9 10, 10, 10, 4, 8 58, 19 50, 52 25, 24 43, 205
1626 1069 2202 1152 857 1379 798 339 1198 2526
3.1.2 DNA-Microarray Datasets The second datasets comprise experimental measurements of gene expression levels in 19 different publicly available gene expression datasets, having the proprieties shown in Table 1 (from alizadeh-v2 to yeoh-v1). These datasets contain several genes and several samples, in this empirical study, samples are treated as data objects to be clustered, while genes are considered as features (dimensions). The significance of this clustering assists the diagnosis of the disease condition, and it discloses the effect of certain treatment on genes. 3.2
Evaluation Measures Used in Experiments
3.2.1 F-Measure It is a comparison between the obtained result and the expected result, if we have a reference partition P of the dataset (benchmark), which is probably derived from previously known domain knowledge, we can simply evaluate the cluster result C by comparing the similarity between P and C through some statistic such as F-measure.
82
B. Kenidra and M. Benmohammed
The F-measure values are within the interval [0,1] and larger values indicate higher clustering quality. F-measure equals 1, that means C is identical to P, and it is an optimal solution. F-measure combines the precision and recall concepts from information retrieval. We then calculate the recall and precision of that cluster for each class as [8]: Recallði; jÞ ¼
nij ni
Precisionði; jÞ ¼
nij nj
Were nij is the number of objects of class i that are in cluster j, nj is the number of objects in cluster j, and ni is the number of objects in class i. The F-measure of cluster j and class i is given by the following equation: F ði; jÞ ¼
2:Recallði; jÞ:Precisionði; jÞ Precisionði; jÞ þ Recallði; jÞ
3.2.2 Silhouette Index For a given cluster, Xj ðj ¼ 1; : :cÞ, the silhouette technique assigns to the ℎ sample of Xj a quality measure, ðiÞ ¼ ði ¼ 1; . . .mÞ, known as the silhouette width. This value is a confidence indicator on the membership of the ℎ sample in the cluster Xj and it is defined as: sði Þ ¼
ðbðiÞ aðiÞÞ Max faðiÞ; bðiÞg
Were a(i) is the average distance between the ℎ sample and all of samples included in Xj ; (i) is the minimum average distance between the ℎ and all of the samples clustered in Xk ðk ¼ 1; : :c; k 6¼ jÞ. Larger values indicate higher clustering quality [8]. 3.3
Empirical Results and Comparison
Since all the algorithms we compared with are stochastic, we performed multiple runs over all 20 benchmarks, and each value is the average of 50 runs. Table 2 shows the results of F-measure, obtained by each algorithm. F-measure is a quantitative comparison between obtained clusters and benchmarks clusters; Fmeasure = 1 means optimal solution. Consequently, the clustering of better quality is the one that maximizes F-measure (in bold). From this table, we can conclude that the proposed algorithm is the best in terms of F-measure. Table 3 displays the results for Silhouette index, obtained by each algorithm. This measure is higher as the classes are compact and far from each others. Consequently, the clustering of better quality is the one that maximizes this index (in bold). Once again, the proposed algorithm competes with K-Means algorithm in terms of Silhouette index.
A Non-stochastic Method for Clustering of Big Genomic Data Table 2. Results in terms of F-Measure Benchmark Methylation datasets alizadeh-v2 alizadeh-v3 armstrong-v1 bhattacharjee bredel chowdary dyrskjot garber golub-v1 gordon khan laiho nutt-v3 pomeroy-v1 pomeroy-v2 shipp-v1 singh west yeoh-v1
The proposed 0.9720 1 0.7565 0.6667 0.7906 0.8224 0.9808 0.6505 0.6304 0.6538 0.8562 0.7032 0.6791 0.6364 0.6560 0.6451 0.7341 0.6366 0.6524 0.9342
K-Means 0.6307 0.8766 0.6537 0.7153 0.5710 0.6723 0.6697 0.7554 0.5898 0.8186 0.6390 0.5777 0.7433 0.8909 0.6460 0.6235 0.6816 0.6286 0.8106 0.8117
DBSCAN 0.6791 0.8247 0.6902 0.7342 0.7003 0.6152 0.7568 0. 6709 0.5932 0.7982 0.6402 0.5643 0.7508 0.6584 0.6557 0.6620 0.7126 0.6451 0.7448 0.8204
Table 3. Results in terms of silhouette index Benchmark Methylation datasets alizadeh-v2 alizadeh-v3 armstrong-v1 bhattacharjee bredel chowdary dyrskjot garber golub-v1 gordon khan laiho
The proposed 0.2048 0.1815 0.1843 0.1753 0.1923 0.0813 0.1411 0.111 0.1016 0.1903 0.2314 0.1488 0.1028
K-Means 0.1004 0.1301 0.2110 0.1704 0.1810 0.2207 0.1784 0.1489 0.1886 0.1756 0.3132 0.1873 0.1582
DBSCAN 0.1195 0.1501 0.1864 0.1599 0.1906 0.1395 0.1568 0.1402 0.1966 0.1856 0.2307 0.1868 0.1709 (continued)
83
84
B. Kenidra and M. Benmohammed Table 3. (continued) Benchmark nutt-v3 pomeroy-v1 pomeroy-v2 shipp-v1 singh west yeoh-v1
3.4
The proposed 0.1524 0.2237 0.1757 0.7211 0.5135 0.2018 0.1012
K-Means 0.2250 0.2026 0.1941 0.2912 0.4626 0.1659 0.2716
DBSCAN 0.1483 0.1932 0.2056 0.3345 0.5037 0.1976 0.1009
Discussion and Evaluation
From these tables above, it amounts to conclude that even the proposed algorithm is ranked first in terms of F-measure, the obtained results in terms of Silhouette index show the competitiveness of the proposed approach compared to K-Means algorithm. The interpretation of this phenomenon amounts to conclude that even the resulting clusters of an algorithm are more similar than other algorithm’ ones to benchmark clusters, these resulting clusters don’t guaranty the best ratio of compactness and remoteness.
4 Conclusions Since biologists are intrinsically in need of efficient and effective computational methods to interpret the vast amounts of data that are constantly being gathered in genomic research, into knowledge, bioinformatics has arisen to fulfill this need. Bioinformatics is likely to be as a key to decipher encoded information in vast genomic structures. Going through high throughput datasets with high-dimension attributes in order to identify interesting biological features at a reasonable time, is a great challenge. The clustering of genomic datasets mainly aims at exploring the genetic relationships of deadly diseases and understanding the cell behavior. The conventional clustering techniques require large amounts of computational time when it comes to process high-throughput genomic datasets with high-dimension attributes as they are stochastic. The proposed algorithm could be used in big data (millions of objects), and it is more likely to provide very good results, on one hand, since the algorithm is fast as it is non-stochastic; just one run. On the other hand, it is simple to implement and easily parallelized.
A Non-stochastic Method for Clustering of Big Genomic Data
85
References 1. Miyoung, S., Jaeyoung, K.: Microarray Data Mining for Biological Pathway Analysis, p. 438 (2009). ISBN 978-3-902613-53-0 2. Chandrasekhar, T., Thangavel, K., Elayaraja, E.: Effective clustering algorithms for gene expression data. Int. J. Comput. Appl. 32(4), 25–29 (2011) 3. Macgregor, P.F., Squire, J.A.: Application of microarrays to the analysis of gene expression in cancer. Clin. Chem. 48, 1170–1177 (2002) 4. Marta, K., Manel, E.: DNA methylation and cancer. Adv. Genet. 70, 27–56 (2010) 5. Aggarwal, Ch.C.: An introduction to cluster analysis. Data Clustering: Algorithms and Applications, pp. 1–27 (2014) 6. Daxin, J., Chun, T., Aidong, Z.: Cluster analysis for gene expression data: a survey. IEEE Trans. Knowl. Data Eng. 16(11), 1370–1386 (2004) 7. Jain, A.K., Dubes, R.C.: Algorithms for Clustering Data. Prentice Hall, Englewood Cliffs (1988) 8. Rendon, E., Abundez, I., Arizmendi, A.: Internal versus external cluster validation indexes. Int. J. Comput. Commun. 5(1), 27–34 (2011)
An Enhanced Hybrid Model for Solving Multiple Sequence Alignment Problem Lamiche Chaabane(&) Computer Science Department, Mohamed Boudiaf University, M’sila, Algeria [email protected], [email protected]
Abstract. In this work, we aim to develop a novel hybrid system (VNPSO) for solving multiple sequence alignment (MSA) problem. The presented procedure is a hybridization of particle swam optimization (PSO) algorithm and variable neighborhood descent (VND) method. When the first metaheuristic is used to discover the search space, the VND procedure is exploited to improve the swarm leader (gbest) solution quality and to overcome the local optimum problem. Experimental studies on BaliBASE benchmark have shown the effectiveness of the proposed method and its ability to obtain good quality solutions comparing to those given by some literature published works. Keywords: Hybrid system Multiple sequence alignment Neighbourhood generation BaliBASE benchmark
PSO
VND
1 Introduction In recent years, multiple sequence alignment (MSA) problem is one of the most challenging tasks in bioinformatics [1]. MSA is generally the alignment of three or more biological sequences (protein or nucleic acid) of similar length. From the output, homology can be inferred and the evolutionary relationships between the sequences studied. Finding the optimal alignment of a set of sequences is known as a NPcomplete problem [2]. It classified as a combinatorial optimization problem [3], which is solved by using computer algorithms. Following to their advantages, metaheuristics have been largely used to solve the MSA problem. These approaches are based on the improvement of a starting solution through a series of iterations until the solution doesn’t become better any longer. They include genetic algorithm (GA) [4], simulated annealing algorithm (SA) [5], particle swarm optimization (PSO) [6], GA-ACO algorithm [7], Ant Colony Algorithm [8] and so on. In this research work, we present a hybrid approach based on PSO and VND metaheuristics to solve the MSA problem. The rest of the paper is structured as follows: Sect. 2 presents the proposed methodology. In Sect. 3, the simulation results are provided. Finally, the study is concluded in Sect. 4.
© Springer Nature Switzerland AG 2019 Y. Farhaoui and L. Moussaid (Eds.): ICBDSDE 2018, SBD 53, pp. 86–91, 2019. https://doi.org/10.1007/978-3-030-12048-1_11
An Enhanced Hybrid Model for Solving MSA Problem
87
2 Materials and Methods 2.1
Particle Swarm Optimization (PSO)
Particle swarm optimization (PSO) is a population-based stochastic optimization algorithm proposed for the first time by Kennedy and Eberhart [9]. The problem is tackled by considering a population (particles), where each particle is a potential solution to the problem. Initial positions and velocities of the particles are chosen randomly. In the commonly used standard PSO, each particle’s position is updated at each iteration step according to its own personal best position and the best solution of the swarm. The evolution of the swarm is governed by the following equations: V ðk þ 1Þ ¼ w:V ðkÞ þ c1 :rand1 : pbestðkÞ X ðkÞ þ c2 :rand2 : gbestðkÞ X ðkÞ
ð1Þ
X ðk þ 1Þ ¼ X ðkÞ þ V ðk þ 1Þ
ð2Þ
where: X is the position of the particle, V is the velocity of the particle, W is the inertia weight, pbest is the best position of the particle, gbest is the global best position of the swarm, rand1, rand2 are random values between 0 and 1, c1, c2 are positive constants which determine the impact of the personal best solution and the global best solution on the search process, respectively, k is the iteration number. Concerning the stopping condition, generally PSO algorithm terminates when the maximum number of iterations nbmaxiter is reached. In addition, all parameters of PSO algorithm are determined experimentally in order to have a good compromise between the convergence time of the algorithm and the final solution quality. 2.2
Variable Neighborhood Descent Method (VND)
A variable neighborhood descent (VND) procedure is a local search method. VND apply a local descent based on a number of different neighborhoods: when a local optimum is reached with the current neighborhood, the search resumes with a different neighborhood to escape from the current local optimum. This is repeated until the current solution cannot be improved anymore [10]. In this work, we propose to combine both PSO and VND algorithms in order to find an approximate solution for the MSA problem. At each iteration, our hybrid system apply PSO algorithm on the whole of the current population and improve its leader
88
L. Chaabane
particle quality by employing the VND procedure. The flowchart of our VNPSO approach is done in Fig. 1.
Fig. 1. VNPSO algorithm flowchart.
2.3
VNPSO Components of MSA Problem
Data Encoding: each particle in the swarm corresponds to a sequence alignment and it is encoded by the gap positions. These gaps are inserted in a random positions in each alignment. Scoring Function: in this work, we utilize the sum of pairs score (SPS) as a scoring function to evaluate each solution. SPS is given by the sum of the scores of the alignment of each pair of sequences (Eq. 3).
An Enhanced Hybrid Model for Solving MSA Problem
ScoreðAÞ ¼
k1 X k X
SðAi ; Aj Þ
89
ð3Þ
i¼1 j¼i þ 1
where the S(Ai, Aj) is the alignment score between two aligned sequences Ai and Aj. Particle Move: is the main mechanism for the evolution of the whole population. It is governed by using both equation 1 and 2 cited above. In order to obtain a validate particles, it is necessary to apply a supplementary operation to replace the negative values by choosing a random positive positions for each particle. Neighborhood Structure: In order to generate neighbors for each current solution when VND algorithm is applied, we choose two different neighborhoods: local shuffle operator and bloc gaps move operator. When the first one changes the position of one gap in a random selected sequence, the second one moves a set of consecutive gaps to the right or the left direction in the current sequence. The pseudo-code of VNPSO algorithm is:
3 Simulation and Results In order to verify the performance of our developed approach, we compare it with two of literature published works including RBT-GA algorithm [11] and GSAMSA method [12]. All SPS results using a set of benchmark instances coming from BaliBase database [13, 14] are summarized in Tables 1 and 2. Average results portrayed in Tables 1 and 2 demonstrated the superiority of our VNPSO approach compared with the all other cited methods in terms of SP score for the specified datasets.
90
L. Chaabane Table 1. SPS results on Reference 2 of Balibase 2. Name laboA lidy lcsy lr69 ltvxA lgtxA lubi lwit 2trx lsbp lhaveA luky 2hsdA 2pia 3grs kinase lajsA lcpt 1lvl lpamA lped 2myr 4enl Avg. Score
RBT-GA [11] GSAMSA [12] Our VNPSO 0.812 0.816 0.815 0.997 0.846 0.978 0.735 0.841 0.876 0.900 0.911 0.908 0.891 0.897 0.896 0.835 0.932 0.936 0.795 0.803 0.887 0.825 0.695 0.830 0.982 0.984 0.983 0.778 0.778 0.815 0.792 0.897 0.897 0.625 0.715 0.875 0.745 0.821 0.888 0.730 0.783 0.802 0.755 0.863 0.898 0.712 0.911 0.906 0.892 0.945 0.956 0.584 0.932 0.945 0.567 0.891 0.907 0.660 0.888 0.902 0.780 0.941 0.950 0.675 0.914 0.936 0.812 0.932 0.943 0.777 0.866 0.901
Table 2. SPS results on Reference 2 of Balibase 3. Name lidy lr69 Lubi Lwit Luky kinase lajsA lpamA Lped 2myr 4enl Avg. Score
RBT-GA [11] GSAMSA [12] Our VNPSO 0.546 0.611 0.558 0.374 0.777 0.812 0.310 0.439 0.465 0.780 0.756 0.811 0.350 0.512 0.602 0.697 0.869 0.806 0.180 0.441 0.546 0.525 0.795 0.798 0.425 0.814 0.870 0.330 0.467 0.513 0.680 0.885 0.910 0.472 0.669 0.699
An Enhanced Hybrid Model for Solving MSA Problem
91
4 Conclusion In this research study, we combined between PSO and VND techniques in order to tackle the MSA problem. The main advantage of this hybridization manner is the exploration ability of PSO and the local improvement of VND procedure. Obtained results showed the capability of our hybrid model compared to others of the literature. As a future work, the use of other population based algorithms or other efficient neighborhood structures is desired to enhance the performance of the proposed method. A comparison of the proposed method with some other state-of-the-art techniques such as Clustal W, SAGA or MULTALIGN is possible to verify its effectiveness.
References 1. Thompson, J.D., Thierry, J.E., Poch, O.: Rapid scanning and correction of multiple sequence alignments. Bioinformatics 19, 1155–1161 (2003) 2. Jiang, T., Wang, L.: On the complexity of multiple sequence alignment. J. Comput. BioI. 1, 337–378 (1994) 3. Papadimitriou, C.H., Steiglitz, K.: Combinatorial Optimization: Algorithms and Complexity. Dover Publications, New York (1998) 4. Horng, J.T., Wu, L.C., Lin, C.M., Yang, B.H.: A genetic algorithm for multiple sequence alignment. In: Proceedings of LNCS, pp. 407–420 (2005) 5. Hernández-Guía, M., Mulet, R., Rodríguez-Pérez, S.: A new simulated annealing algorithm for the multiple sequence alignment problem. The approach of polymers in a random media. Phys. Rev. E 72, 1–7 (2005) 6. Lei, C.W., Ruan, J.H.: A particle swarm optimization algorithm for finding DNA sequence motifs. In: Proceedings of IEEE, pp. 166–173 (2008) 7. Lee, Z.J., Su, S.F., Chuang, C.C., Liu, K.H.: Genetic algorithm with ant colony optimization (GA-ACO) for multiple sequence alignment. Appl. Soft Comput. 8, 55–78 (2008) 8. Chen, L., Zou, L., Chen, J.: An efficient ant colony algorithm for multiple sequences alignment. In: Proceedings of the 3rd International Conference on Natural Computation (ICNC 2007), pp. 208–212 (2007) 9. Kennedy, J., Eberhart, R.: Particle swarm optimization. In: Proceedings of IEEE International Conference on Neural Networks, Perth, vol. 4, pp. 1942–1948 (1995) 10. Mladenovic, N., Hansen, P.: Variable neighborhood search. Comput. Oper. Res. 34, 1097–1100 (1997) 11. Taheri, J., Zomaya, A.Y.: RBT-GA: a novel metaheuristic for solving the multiple sequence alignment problem. BMC Genom. 10, 1–11 (2009) 12. Yadav, R.K.: GSAMSA: gravitational search algorithm for multiple sequence alignment. Indian J. Nat. Sci. 6(33), 10523–10537 (2015) 13. Bahr, A., Thompson, J.D., Thierry, J.C., Poch, O.: BALIBASE (Benchmark Alignment dataBASE): enhancements for repeats, transmembrane sequences and circular permutation. Nucleic Acids Res. 29(1), 323–326 (2001) 14. Thompson, J.D., Koehl, P., Ripp, R., Poch, O.: BAliBASE 3.0: latest developments of the multiple sequence alignment benchmark. Proteins 61(1), 127–136 (2005)
A Recommender System for Videos Suggestion in a SPOC: A Proposed Personalized Learning Method Naima Belarbi1(&), Nadia Chafiq2, Mohammed Talbi3,4, Abdelwahed Namir1, and Habib Benlahmar1 1
Laboratory of Technological Information and Modelisation (LTIM), Faculty of Sciences Ben M’Sik, University Hassan II, Casablanca, Casablanca, Morocco [email protected], [email protected], [email protected] 2 Laboratory of Sciences, Information, Communication and Education Technology (LAPSTICE), Faculty of Sciences Ben M’Sik, Casablanca, Morocco [email protected] 3 Observatory of Research in Didactics and University Pedagogy (ORDIPU), University Hassan II, Casablanca, Casablanca, Morocco [email protected] 4 Laboratory of Analytical Chemistry and Physical Chemistry of Materials, Faculty of Sciences Ben M’Sik, Casablanca, Morocco
Abstract. Adaptivity, personalization and recommendation techniques are classic solutions recommended by many specialists for providing successful learning experiences by offering suitable adaptation that satisfy the learning preferences and meet heterogeneous characteristics of users. In the present paper, we propose a video recommender system across a Small Private Online Course (SPOC). We adopt a hybrid recommendation technique which consists on analyzing users’ video behavior while enrolling into a SPOC, estimating their interest in videos, finding learners with similar profile and finally recommending target user the same videos in which similar users are interested in. The proposed approach consist first on capturing and analyzing user’s video clickstream in order to construct a user profile with an implicit way to infer user’s interest in videos. Second, the unsupervised K-Means clustering algorithm is used to group users with similar video behavior into clusters. Finally, videos from similar profiles that could meet user’s interest can be recommended. Keywords: Adaptive learning Recommendation technique User profile Clustering Video clickstream analysis
SPOC
© Springer Nature Switzerland AG 2019 Y. Farhaoui and L. Moussaid (Eds.): ICBDSDE 2018, SBD 53, pp. 92–101, 2019. https://doi.org/10.1007/978-3-030-12048-1_12
A Recommender System for Videos Suggestion in a SPOC
93
1 Introduction The phenomenon of Massive open online courses (MOOCs) and Small Private Online Courses (SPOCS) are remarkably shaping online learning landscape especially in higher education [1]. However, heterogeneity of student profiles is a real challenge to the “one-size-fits-all” learning model provided by MOOCs or SPOCs [2, 3]. Personalization, adaptivity and recommendation techniques are known to offer great potential to overcome these challenges. These concepts are said to be able to increase learner satisfaction, to improve the learning process and to enable learners in finding relevant educational resources that best meet their personal preferences and needs [4–9]. In the last few years, these different approaches have invited growing interest in MOOCs [2, 10–17]. Therefore, technological advancements in a myriad of fields such as data mining, machine learning, techniques for managing big data, artificial intelligence have extended the use in the education field [18, 19]. In the present paper, we suggest a method to integrate personalization in a SPOC based on a recommendation technique. As Videos are an extremely important part of SPOCs, we suggest analyzing data obtained by observing users’ videos interactions at the click level while enrolling in the SPOC. Then, we adopt an implicit approach to estimate interests of users in videos by using the Bayesian method. Relevant Information concerning users’ videos interactions and their estimated interest in videos are stored in their profiles. We propose then to apply the K-means algorithm to cluster users with similar profiles in order to match a single user with preferences of similar users. Finally, for a given user, we propose to recommend videos which interest similar users and can provide a personalized experience learning that meet his preferences. The main contribution of the present work is the use of recommendation techniques across a SPOC in order to provide personalization of learning experiences to students. The structure of this paper is as follows: First, we present the various related works. Second, we describe our approach to recommend suitable videos to a learner which can offer a best experience by exploiting similar users’ profiles. Finally, we present our ongoing works.
2 Background 2.1
Adaptivity in E-learning Systems
Web systems generally and e-learning systems particularly suffer from the lack of their capacity to cater to the different needs and heterogeneous profiles of many users. To handle these challenges, adaptive systems has attracted increasing attention and become a special trend of research that aim to increase users’ engagement and motivation by providing personalized and individual experiences to users [20–22]. In fact, in [8], the authors reported that the goal of adaptive e-learning is delivering “the right content, to the right person, at the proper time, in the most appropriate way”. In an adaptive system, adaptation decision can be made accordingly to various individual characteristics of users [22] or with considering various sources of data relative either to the user, the usage or the environment [23].
94
N. Belarbi et al.
In the field of education, Adaptive learning systems are currently beginning an imperative alternative to the traditional “one size fits all” approach in the development of educational systems to address the heterogeneous needs and characteristics of the users [21]. In fact, Adaptive learning systems can improve content understanding [24], reduce navigation effort and improve the quality of learning [25], enhance learning [5] and increase efficiency, effectiveness and learner satisfaction [26]. Works on adaptive learning systems reveal various levels or factors to make the adaptation decision [7]. Adaptation can be made according to learner’s knowledge [27– 29], to learner’s learning level [16, 30], to learner’s interest [28, 29], to learner’s learning style [26, 29, 31–33], to learner’s preferences [6, 28], to learner’s degree of awareness and skills [6], to learner’s goals, [6, 29], to user’s background and experience [6], to learner’s historic data [6], to learner’s performance [32], to learner’s personality [30, 34], to learning behavioral type [29], to the student’s strength and weaknesses [35], to learner’s behavior in the learning environment [29], to learner’s errors/misconceptions [36, 37]. 2.2
Recommender Systems
Recommender systems (RSs) refer to any system which can provide the most appropriate recommendations or guidance to users to make choices that best meet their preferences. Over the last past years, the use of recommendations techniques has become more widespread in various areas such as Information retrieval, e-commerce, tourism websites, online libraries, etc.… [38–40]. RSs collect information and data concerning users’ preferences for items in order to make predictions and provide users with recommendations of items that a user may wish to utilize. The most used filtering methods used in RSs are (1) content based systems which find similarities by items’ properties; (2) collaborative-filtering systems where recommendation is delivered based on existing relations between users and items and (3) hybrid filtering which combines different recommendation techniques [39, 40]. In recent years, researches on adopting recommendations techniques in educational systems attracted increasing interest. In Fact, recommendation techniques can achieve personalization which is becoming an important feature in e-learning systems [41]. In addition, recommendation algorithms become more intelligent and can help to make smart decisions to achieve personalized learning experiences to learners [42]. In [43], the authors propose and validate by implementing a new recommendation mechanism in a web-based platform for lifelong learning. In [44], in order to improve course material navigation as well as to assist the online learning process, the authors suggest to recommend on-line learning activities or shortcuts in a course web site based on learners’ access history. An online recommendation platform (ALR) as a tracking tool for instructors to observe students’ learning activities in order to provide immediate assistance according to individual students’ learning strengths and weaknesses was proposed in [45]. The authors in [46] suggest a framework of a recommender system for e-learning to assist learners in finding and selecting the more relevant Learning Objects that meet their interest.
A Recommender System for Videos Suggestion in a SPOC
2.3
95
Adaptivity and Personalization in MOOCs/SPOCs
Open learning represent a new form of online learning but it has been found that the current model suffers from the lack of personalization. Adaptivity and personalized learning in MOOCs are recently a new trending of research within the area of adaptive systems. The number of works on adaptive MOOCs is increasing [14]. The most works considered personalization to meet static user profiles obtained by explicit approaches. Some works propose patterns or group profiles with the aim to learn more about learners and consequently to propose the pedagogical model that meet their interests. In [11], the authors propose an Adaptive learning strategy in a MOOC based on user’s learning style and assessment results. In [16], the personalization is provided accordingly to learners’ learning style in order to enhance their learning experience in MOOCS. The authors in [2] propose a framework for an adaptive MOOC which recommends appropriate learning resources that meet the user’s objectives and preferences and allows further personalization by adapting the contents during user’s course progression based on the users’ knowledge of the foregoing subjects. In [14], a framework was proposed for an adaptive MOOC (aMOOC) implemented in an adaptive platform so called iMOOC that includes a combination of six “adaptive pills” merged to fulfil diverse needs of adaptivity and to create personalized learning pathways: (P1) Self-assessment training; (P2) Adapted advance to the student’s learning speed; (P3) Adaptation of learning to different profiles/skills/interests; (P4) Contributing and sharing resources among a set of users with a common interest/profile; (P5) Adapted learning to the acquired knowledge; (P6) Monitoring student’s progress. In [17], the authors propose a model so called ahMOOC that merged the social advantages of cMOOCs, the organizational benefits of xMOOCs and the personalization of the learning to meet user’s gender, age, geographical location, educational background, profession. In [13], an adaptive MOOC model named PERSUA2MOOC was proposed. The model is based on a learner profile structured in five categories in: (1) the “ResourcesInteractions” section contains quantitative information on the use of MOOC resources by the learner; (2) The “moocInteractions” section proposes quantitative indicators concerning interactions with the MOOC platform; (3) The “behavior” section, which contains mainly qualitative indicators for obtaining information on the learner; (4) the “knowledge” section characterizes the knowledge and skills of the learner in the context of the MOOC; (5) “The learnerInformation” section contains information that can not a priori be derived from the traces of the learner, such as demographic data, the objectives of the learner. In [17], the authors propose an adaptive learning approach for a learning design in a cMOOC. Students with similar profiles are grouped and share the same resources and activities. Students can then choose the most appropriate resources that meet their own goals.
96
N. Belarbi et al.
3 Methodology 3.1
Study Context
The SPOC UNIVTICE is a course created and designed on Moodle. This course intends to accompany teachers at the University Hassan 2 – Casablanca Morocco, in order to acquire technical and pedagogical skills. We propose to implement the present recommendation technique across the SPOC UNIVTICE with the aim to provide personalized experiences to various profiles of teachers. 3.2
The Proposed Recommendation Approach
In the present work, we propose the use of a hybrid recommendation approach based on the method known as clique-based filtering method which consists on matching a single profile with profiles of similar users [23, 47]. As items target to recommend for a learner in the SPOC, we suggest videos as they are important components in a SPOC. We propose to observe and analyze users’ behaviors when interacting with videos to estimate implicitly their interests in videos as information characterizing users’ profile detailed in our previous work [48]. Previous research has already recognized the benefits of user-based analysis related to viewed videos. In fact, users’ video interactions such as Playing, Pausing, Replaying are useful to reveal the relationship that rely the complexity of videos and student video behavior in a MOOC [49], to identify interesting video segment [50], to reflect user’s experience difficulty [51], to tell more about learner video engagement in MOOCs [52–54], to reflect user’s interest in a video [55], to give important drop out indicators [54, 55] and to give insights about video production to MOOC instructors [51, 52, 54]. In our proposed approach, we retain the following events to track: Play (PL), Pause (PA), Replay (RP), Move Forward (FF), Move Back (RW), Download (DL) and Stop (ST). For each user enrolling within the SPOC, we propose the following steps: (1) With data captured from the tracking of a user’s video interactions, we construct first his Video Viewing Sequence (VVS) as a vector related to each viewed video in the SPOC. This vector is described as a sequence of events performed on a given video. Each event is characterized by its frequency in a first time, then, all frequencies are normalized as weight. ! ðVVSji ÞT ¼ ðPLij ; PAij ; RPij ; FFij ; RWij ; DLij ; STij Þ where i and j denote respectively subscripts for videos and users in the SPOC. (2) Then, we propose to construct the user Video Viewing History (VVH) as a matrix composed by all Videos Viewing Sequences related to a given learner for all videos in the SPOC.
A Recommender System for Videos Suggestion in a SPOC
V1 . . .. . . ! ½VVHj ¼ ðVVS1j ; 2 PL1j 6 PA j 6 1 6 RL j 6 1 6 ¼ 6 FF1j 6 6 RW1j 6 4 DL j 1 ST1j
V2 :: . . .. . . . . .. . . ! VVS2j ; ::; PL2j :: PA2j :: RL2j :: :: FF2j j :: RW2 j :: DL2 j :: ST2
Vi :: . . .. . . . . .. . . ! VVSij ; ::; PLij :: PAij :: RLij :: :: FFij j :: RWi j :: DLi j :: STi
97
Vn . . .. . . ! VVSnj Þ 3 PLnj PAnj 7 7 RLnj 7 7 FFnj 7 7 RWnj 7 7 DLnj 5 STnj
(3) For each learner, we suggest an implicit way to assess his interest in a given video. We retain four implicit interest indicators: (1) The Total Viewing Time for a user concerning a video, (2) The number of Move Back events performed by a user on a video, (3) The Download event performed by a user on a video and (4) Replay event performed by a user on a video. We propose the use of the Bayesian Method and a statistical analysis to assess the interest of a user in a given video. Hence, we obtain a vector containing estimated interests for each user in all videos. ! ðVIHj ÞT ¼ ðVIij Þ Where VIij denotes estimated Video Interest of the user j in video i. (4) Then, we propose to use one of the widespread unsupervised learning algorithms, the k-means algorithm [56] in order to segment users into clusters based on their video viewing sequences. (5) Finally, by finding learners with similar profiles, we can recommend target user the same videos in which similar users are interested in.
4 Conclusion and Future Work In the present work, we have described our approach to adopt a recommendation technique across a SPOC in order to provide personalization to users by offering them proposed videos to make their own choices based on choices made by similar users. In our approach, we proposed the use of data mining techniques in order to build a user profile that represents user’s video behavior at the click level when enrolling in the SPOC. This profile is used to make suggestions for selecting videos that can meet learners’ interests. We are currently implementing this recommender system across the SPOC UNIVTICE at the university Hassan II – Casablanca Morocco. Our future work aims also to evaluate the pertinence of recommendations provided for users by means of questionnaires.
98
N. Belarbi et al.
References 1. Kaplan, A.M., Haenlein, M.: Higher education and the digital revolution: about MOOCs, SPOCs, social media, and the Cookie Monster. Bus. Horiz. 59(4), 441–450 (2016). https:// doi.org/10.1016/j.bushor.2016.03.008 2. Onah, D.F.O., Sinclair, J.: Massive open online courses: an adaptive learning framework. In: The 9th International Technology, Education and Development Conference, Madrid, Spain, pp. 1258–1266. INTED2015 Proceedings, March 2015 3. Filius, R.M., De Kleijn, R.A.M., Uijl, S.G., Prins, F.J., Van Rijen, H.V.M., Grobbee, D.E.: Challenges concerning deep learning in SPOCs. Int. J. Technol. Enhanced Learn. 10(1/2), 111 (2018). https://doi.org/10.1504/ijtel.2018.10008600 4. Brusilovsky, P., Peylo, C.: Adaptive and Intelligent Web-based Educational Systems. Int. J. Artif. Intell. Educ. (IJAIED) 13, 159–172 (2003) 5. Alshammari, M., Anane, R., Hendley, R. J.: Adaptivity in E-learning systems. In: Eighth International Conference on Complex, Intelligent and Software Intensive Systems (2014). https://doi.org/10.1109/cisis.2014.12 6. Muruganandam, S., Srininvasan, N.: Personalised e-learning system using learner profile ontology and sequential pattern mining-based recommendation. Int. J. Bus. Intell. Data Mining 12(1), 78–93 (2017). https://doi.org/10.1504/ijbidm.2017.082704 7. Chrysafiadi, K., Virvou, M.: Advances in Personalized Web-Based Education. Intelligent Systems Reference Library (2015). https://doi.org/10.1007/978-3-319-12895-5 8. Shute, V., Towle, B.: Adaptive E-learning. Educ. Psychol. 38(2), 105–114 (2003). https:// doi.org/10.1207/s15326985ep3802_5 9. Henning, P., Heberle, F., Streicher, A., Zielinski, A., Swertz, C., Bock, J., Zander, S.: Personalized web learning: merging open educational resources into adaptive courses for higher education. In: 22nd International Conference User Modeling, Adaptation, and Personalization, vol. 1181, pp. 55–62. Aalborg, Denmark (2014) 10. Daniel, S.J., Vázquez Cano, E., Gisbert, M.: The future of MOOCs: adaptive learning or business model? RUSC. Universities and Knowl. Soc. J. 12(1), 64–73 (2015). https://doi. org/10.7238/rusc.v12i1.2475 11. Sonwalkar, N.: The first adaptive MOOC: a case study on pedagogy framework and scalable cloud architecture—Part I. MOOCs FORUM 1(P), pp. 22–29 (2013). https://doi.org/10. 1089/mooc.2013.0007 12. Lerís, D., Sein-Echaluce, M.L., Hernández, M., Bueno, C.: Validation of indicators for implementing an adaptive platform for MOOCs. Comput. Hum. Behav. 72, 783–795 (2017). https://doi.org/10.1016/j.chb.2016.07.054 13. Clerc, F., Lefevre, M., Guin, N., Marty, J. C.: Mise en place de la personnalisation dans le cadre des MOOCs. In: 7ème Conférence sur les Environnements Informatiques pour l’Apprentissage Humain EIAH, pp. 144–155, June 2015 14. Sein-Echaluce, M.L., Fidalgo-Blanco, Á., García-Peñalvo, F.J., Conde, M.Á.: iMOOC platform: adaptive MOOCs. In: Learning and Collaboration Technologies, pp. 380–390 (2016). https://doi.org/10.1007/978-3-319-39483-1_35 15. Yan-hong, L., Bo, Z., Jian-hou, G.: Make adaptive learning of the MOOC: The CML model. Paper presented at the 10th International Conference on Computer Science & Education (ICCSE) (2015). https://doi.org/10.1109/iccse.2015.7250398 16. Fasihuddin, H., Skinner, G., Athauda, R.: A framework to personalise open learning environments by adapting to learning styles. In: Proceedings of the 7th International Conference on Computer Supported Education (2015). https://doi.org/10.5220/000544 3502960305
A Recommender System for Videos Suggestion in a SPOC
99
17. García-Peñalvo, F.J., Fidalgo-Blanco, Á., Sein-Echaluce, M.L.: An adaptive hybrid MOOC model: disrupting the MOOC concept in higher education. Telematics Inform. 35(4), 1018– 1030 (2018). https://doi.org/10.1016/j.tele.2017.09.012 18. Dutt, A., Ismail, M.A., Herawan, T.: A systematic review on educational data mining. IEEE Access 5, 15991–16005 (2017). https://doi.org/10.1109/access.2017.2654247 19. Bakhshinategh, B., Zaiane, O.R., ElAtia, S., Ipperciel, D.: Educational data mining applications and tasks: a survey of the last 10 years. Educ. Inf. Technol. 23(1), 537–553 (2017). https://doi.org/10.1007/s10639-017-9616-z 20. Midgley, C.: Goals, Goal Structures, and Patterns of Adaptive Learning (2014). https://doi. org/10.4324/9781410602152 21. Brusilovsky, P.: Adaptive hypermedia for education and training. In: Adaptive Technologies for Training and Education, pp. 46–66 (2012). https://doi.org/10.1017/cbo9781139049580. 006 22. Brusilovsky, P.: Adaptive hypermedia. User Model. User-Adap. Inter. 11(1/2), 87–110 (2001) 23. Kobsa, A., Koenemann, J., Pohl, W.: Personalised hypermedia presentation techniques for improving online customer relationships. Knowl. Eng. Rev. 16(02), 111–155 (2001). https:// doi.org/10.1017/s0269888901000108 24. Boyle, C., Encarnacion, A.O.: Metadoc: an adaptive hypertext reading system. In: Adaptive Hypertext and Hypermedia, pp. 71–89 (1998). https://doi.org/10.1007/978-94-017-0617-9_3 25. Brusilovsky, P., Pesin, L.: Adaptive navigation support in educational hypermedia: an evaluation of the ISIS-Tutor. J. Comput. Inf. Technol. 6(1), 27–38 (1998) 26. Popescu, E., Badica, C., Moraret, L.: Accommodating learning styles in an adaptive educational system. Informatica 34(4), 451–462 (2010) 27. Baker, R.S.J.d., Goldstein, A.B., Heffernan, N.T.: Detecting the Moment of Learning. Lecture Notes in Computer Science, pp. 25–34 (2010). https://doi.org/10.1007/978-3-64213388-6_7 28. Halimi, K., Seridi-Bouchelaghem, H., Faron-Zucker, C.: An enhanced personal learning environment using social semantic web technologies. Interact. Learn. Environ. 22(2), 165– 187 (2013). https://doi.org/10.1080/10494820.2013.788032 29. Melesko, J., Kurilovas, E.: Personalised intelligent multi-agent learning system for engineering courses. In: IEEE 4th Workshop on Advances in Information, Electronic and Electrical Engineering (AIEEE) (2016). https://doi.org/10.1109/aieee.2016.7821821 30. Trikha, N., Godbole, A.: Adaptive e-learning system using hybrid approach. International Paper presented at the Conference on Inventive Computation Technologies (ICICT) (2016). https://doi.org/10.1109/inventive.2016.7824844 31. Al-Omari, M., Carter, J., Chiclana, F.: A hybrid approach for supporting adaptivity in elearning environments. Int. J. Inf. Learn. Technol. 33(5), 333–348 (2016). https://doi.org/10. 1108/ijilt-04-2016-0014 32. Schiaffino, S., Garcia, P., Amandi, A.: eTeacher: Providing personalized assistance to elearning students. Comput. Educ. 51(4), 1744–1754 (2008). https://doi.org/10.1016/j. compedu.2008.05.008 33. Dahbi, A., El kamoun, N., Berraissoul, A.: Conception d’un système hypermédia d’enseignement adaptatif centré sur les styles d’apprentissage: modèle et expérience. Revue Internationale Des Technologies En Pédagogie Universitaire 6(1), 55-71 (2009). https://doi. org/10.7202/039181ar 34. Kim, J., Lee, A., Ryu, H.: Personality and its effects on learning performance: design guidelines for an adaptive e-learning system based on a user model. Int. J. Ind. Ergon. 43(5), 450–461 (2013). https://doi.org/10.1016/j.ergon.2013.03.001
100
N. Belarbi et al.
35. Baschera, G.M., Gross, M.: Poisson-based inference for perturbation models in adaptive spelling training. Int. J. Artif. Intell. Educ. 20(4), 333–360 (2010) 36. Surjono, H.D., Maltby, J.R.: Adaptive educational hypermedia based on multiple student characteristics. In: Advances in Web-Based Learning, pp. 442–449 (2003). https://doi.org/ 10.1007/978-3-540-45200-3_41 37. Goel, G., Lallé, S., Luengo, V.: Fuzzy Logic Representation for Student Modelling. Lecture Notes in Computer Science 428–433 (2012). https://doi.org/10.1007/978-3-642-30950-2_55 38. Ricci, F., Rokach, L., Shapira, B.: Recommender Systems: Introduction and Challenges. Recommender Systems Handbook, pp. 1–34 (2015). https://doi.org/10.1007/978-1-48997637-6_1 39. Isinkaye, F.O., Folajimi, Y.O., Ojokoh, B.A.: Recommendation systems: principles, methods and evaluation. Egypt. Inf. J. 16(3), 261–273 (2015). https://doi.org/10.1016/j.eij. 2015.06.005 40. Bobadilla, J., Ortega, F., Hernando, A., Gutiérrez, A.: Recommender systems survey. Knowl. Based Syst. 46, 109–132 (2013). https://doi.org/10.1016/j.knosys.2013.03.012 41. Klašnja-Milićević, A., Ivanović, M., Nanopoulos, A.: Recommender systems in e-learning environments: a survey of the state-of-the-art and possible extensions. Artif. Intell. Rev. 44 (4), 571–604 (2015). https://doi.org/10.1007/s10462-015-9440-z 42. Drachsler, H., Verbert, K., Santos, O.C., Manouselis, N.: Panorama of recommender systems to support learning. In: Recommender Systems Handbook, pp. 421–451 (2015). https://doi. org/10.1007/978-1-4899-7637-6_12 43. Dascalu, M.-I., Bodea, C.-N., Mihailescu, M.N., Tanase, E.A., Ordoñez de Pablos, P.: Educational recommender systems and their application in lifelong learning. Behav. Inf. Technol. 35(4), 290–297 (2016). https://doi.org/10.1080/0144929x.2015.1128977 44. Zaiane, O. R.: Building a recommender agent for e-learning systems. International Paper presented at the Conference on Computers in Education (2002). https://doi.org/10.1109/cie. 2002.1185862 45. Tseng, H.-C., Chiang, C.-F., Su, J.-M., Hung, J.-L., Shelton, B.E.: Building an online adaptive learning and recommendation platform. Lecture Notes in Computer Science, pp. 428–432 (2017). https://doi.org/10.1007/978-3-319-52836-6_45 46. Fraihat, S., Shambour, Q.: A framework of semantic recommender system for e-Learning. J. Softw. 10(3), 317–330 (2015). https://doi.org/10.17706/jsw.10.3.317-330 47. Alspector, J., Kolcz, A., Karunanithi, N.: Feature-based and clique-based user models for movie selection: a comparative study. User Model. User-Adap. Inter. 7(4), 279–304 (1997). https://doi.org/10.1023/A:1008286413827 48. Belarbi, N., Chafiq, N., Talbi, M., Namir, A., Benlahmar, E.: User profiling in a SPOC: a method based on user video clickstream analysis. Int. J. Emerg. Technol. Learn. (2018, in press) 49. Van der Sluis, F., Ginn, J., Van der Zee, T.: Explaining student behavior at scale: the influence of video complexity on student dwelling. In: Proceedings of the Third ACM Conference on Learning @ Scale - L@S’16 (2016). https://doi.org/10.1145/2876034. 2876051 50. Chorianopoulos, K.: Collective intelligence within web video. Hum.-Centric Comput. Inf. Sci. 3(1), 10 (2013). https://doi.org/10.1186/2192-1962-3-10 51. Li, N., Kidzinski, L., Jermann, P., Dillenbourg, P.: How do in-video interactions reflect perceived video difficulty?. In: Proceedings of the European MOOCs Stakeholder Summit 2015, No. EPFL-CONF-207968, PAU Education (2015) 52. Guo, P. J., Kim, J., Rubin, R.: How video production affects student engagement. In: Proceedings of the First ACM Conference on Learning @ Scale Conference - L@S’14, pp. 41–50 (2014). https://doi.org/10.1145/2556325.2566239
A Recommender System for Videos Suggestion in a SPOC
101
53. Kizilcec, R.F., Piech, C., Schneider, E.: Deconstructing disengagement: analyzing learner subpopulations in massive open online courses. In: Proceedings of the Third International Conference on Learning Analytics and Knowledge – LAK 2013, pp. 170–179 (2013). https://doi.org/10.1145/2460296.2460330 54. Sinha, T., Jermann, P., Li, N., Dillenbourg, P.: Your click decides your fate: inferring information processing and attrition behavior from MOOC video clickstream interactions. In: Proceedings of the EMNLP 2014 Workshop on Analysis of Large Scale Social Interaction in MOOCs (2014). https://doi.org/10.3115/v1/w14-4102 55. Kim, J., Guo, P.J., Seaton, D.T., Mitros, P., Gajos, K.Z., Miller, R.C.: Understanding invideo dropouts and interaction peaks in online lecture videos. In: Proceedings of the First ACM Conference on Learning @ Scale Conference - L@S’14 (2014). https://doi.org/ 10.1145/2556325.2566237 56. MacQueen, J.: Some methods for classification and analysis of multivariate observations. In: Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability, vol. 1: Statistics, pp. 281–297. University of California Press, Berkeley, Calif. (1967). https://projecteuclid.org/euclid.bsmsp/1200512992
Multi-level Network Construction Based on Intelligent Big Data Analysis Samaher Al_Janabi(&) , Mahdi Abed Salman and Maha Mohammad
,
Department of Computer Science, Faculty of Science for Women (WSCI), University of Babylon, Babylon, Iraq {samaher,mahdi.salman}@uobabylon.edu.iq
Abstract. In this work, we present a hybrid miner-network analyzer (HMNA) system includes three main stages; the first stage called preparing and preprocessing stage that includes building initial network from citation file and Find Keywords through apply Rake and cleaning. The second stage involves building classification model including parameters detection and apply LDA for find topics, final stage, add the topics of document into initial network to construction multi-layer network, where each level represent community related of that topic. We can summarize the main points of HMNA system as: (i) It deals with real, complex, huge database of papers ‘citation’. (ii) The preprocessing stage involves retrieve keywords from corpus using Rake after add constructions on it and cleaning without using any feature selection method. (iii) It building digital corpus that combines with dictionary to clustering the clean text into multi groups based on LDA model. (iv) It reconstructed the initial network by add the labeling of topic results from above step to it. (v) It builds multi communities (multi-level network), each level in that network represent single communities, (vi) It computes the main characteristics measures for each level or communities to determine the similarity in it structure with cluster in citation network. Keywords: Intelligent data analysis Big data Text mining Complex network Citation network Characteristics network measures RAKE LDA
1 Introduction Recently, the logical community has realized that there are a few systems, both normal and artificial, which cannot be completely caught on by a reductionist approach (i.e., by analyzing their constituting components in a confined way). On the contrary, their simply visible properties appear to be characterized by the structures of intuitive between these components such frameworks are presently called complex Network. Cases have been found in many example logical fields (e.g. in social or transport systems, Web. Most surprising one of the cases a complex network is the brain; it is composed of more than 100 billion neurons, each one of them appears a very simple dynamic, the human capacity for thinking only emerges when these simple dynamics start to interact) [1]. © Springer Nature Switzerland AG 2019 Y. Farhaoui and L. Moussaid (Eds.): ICBDSDE 2018, SBD 53, pp. 102–118, 2019. https://doi.org/10.1007/978-3-030-12048-1_13
Multi-level Network Construction
103
Now, there are two approaches utilized to extricate data from complex networks as explained in Fig. 1; classical data mining methods, and complex network analysis. Born within physics, with considerable inputs of science and insights, the hypothesis of complex networks has demonstrated to be a capable device for the analysis of complex frameworks; it permits reducing them into basic structures of intelligence, which can effectively be examined by mean of scientific (logarithmic) tool, while ejecting all pointless subtle elements. Following this idea, a few imperatives results have been gotten, as, for instance, the discovery of critical genes in a living organism, or the definition of the leading procedures to halt the spreading of an irresistible malady. On the other hand, data mining mention to the prepare of discovering designs in expansive data sets, in arranging to naturally extract data and transform it into a justifiable structure. Born inside computer science, it involves strategies draw from applied mathematics and statistics. Each field can yield new thoughts and procedures that can clearly contribute to the change of the state of the ability of the other [2, 3]. In general, we can see the problem from two sides, first information discovery and data mining techniques may move forward the creation and analysis of complex systems by means of: (i) identification and choice of the most significant highlights in the initial data, (ii) standard methods for data pre-processing and (iii) examination of the significance of network-based result. On the other side, complex network analysis is basically expected better approach of representing and extracting information about the structure of frameworks characterized by interacting components, in this way, giving a new point of seeing to classical data mining task like classification and regression. Now, we are going to handle the issue of the integration of knowledge discovery and complex network analysis as Fig. 1.
Fig. 1. Interaction between complex systems and data mining
In Fig. 1(a) represent the nowadays approach to the study of complex networks, while (b) making intelligent among data mining and complex networks,
104
S. Al_Janabi et al.
There are several people worked on this problem as explained below: Oshawa et al. [4] “They present an algorithm for extracting keywords representing the asserted main point in a document, without relying on external devices such as natural language processing tools or a document corpus. Our algorithm Key Graph is based on the segmentation of a graph, representing the co-occurrence between terms in a document, into clusters. Each cluster corresponds to a concept on which author’s idea is based, and top ranked terms by a statistic based on each term’s relationship to these clusters are selected as keywords”. Our work similar this work by applying clustering but between words of documents rather than documents and representing it by based graph. The pragmatic method similar this work same goal is extracting keywords and representing it as graphs. Litvak and Last [5] “They introduced and compare between two novel approaches, supervised and unsupervised, for identifying the keywords to be used in extractive summarization of text documents. Both our approaches are based on the graph-based syntactic representation of text and web documents, which enhances the traditional vector-space model by considering some structural document features. In the supervised approach, we train classification algorithms on a summarized collection of documents with the purpose of inducing a keyword identification model. In the unsupervised approach, we run the HITS algorithm on document graphs under the assumption that the top-ranked nodes should represent the document keywords.” Zanin et al. [1] proposed novel hybrid between complex networks and data mining. The result show that procedures can viably utilized to creation of novel representations to the network, by pre-selecting the most critical components can be reduce the dimensionality of analyzed networks, and useful in analysis of distinctive network topologies. The pragmatic method similar this work by hybrid between the two approach data mining and complex networks but different techniques use. Abu-Errub [6] proposed a modern strategy to classification the Arabic content to compared the pre-defined records categories with document based on its substance utilizing TF.IDF strategy (Term Frequency times Inverse Document Frequency) degree, The record is classified to the fitting subordinate category utilizing Chi Square measure. Our work similar this work by performed the preprocessing but on English text rather than Arabic Tweet Extremity. Abilhoa and de Castro [7] “They propose a keyword extraction method for tweet collections that represents texts as graphs and applies centrality measures for finding the relevant vertices (keywords). To assess the performance of the proposed approach, three different sets of experiments are performed. The first experiment applies TKG to a text from the Time magazine and compares its performance with that of the literature. The second set of experiments takes tweets from three different TV shows, applies TKG and compares it with TFIDF and KEA, having human classifications as benchmarks. Finally, these three algorithms are applied to tweets sets of increasing size and their computational running time is measured and compared. Altogether, these experiments provide a general overview of how TKG can be used in practice, its performance when compared with other standard approaches, and how it scales to larger data instances”. The pragmatic method similar this work by hybrid between the two approach text mining and graph but different techniques uses.
Multi-level Network Construction
105
Jain and Mishra [8] “They presented modified Maximum Entropy based classifier. Most extreme Entropy classifiers give astonishing bargain of adaptability for parameter definitions and take after presumptions closer to genuine world situation. This classifier is at that point combined with a Naïve Bayes classifier. Naïve Bayes Classification is an exceptionally basic and quick method. The presumption show inverse to that of Maximum Entropy. The combination of classifiers is done through administrators that directly combine the result of two classifiers to foresee course of reports in query”. The pragmatic method similar this work by applying text classification but differ on it by the technique used. Brahimi et al. [9] shown method to determine the impact of applying stemming and n-gram procedures for Arabic writings (tweets) on opinion classification. And examining the impact of feature selection methods on the effective and performance of classifier. They utilized exactness and Review to assess their methods. The pragmatic method similar this work by applying text classification but differ on it by the technique used. Shah et al. [10] proposed the combination of classifiers where, Greatest Entropy classifiers give an extraordinary bargain of adaptability for parameter definitions and take after presumptions closer to genuine world situation. This classifier is at that point combined with a Naïve Bayes classifier. Naïve Bayes Classification is an exceptionally basic and quick strategy. The presumption show is inverse to that of most extreme Entropy. The combination of classifiers is done through administrators that directly combine the results about of two classifiers to anticipate course of records in inquiry. Combination of these two classifiers might result into superior accuracy. The pragmatic method similar this work by applying text classification but differ on it by the technique used. Rezaeian et al. [11] “They describe a three-step methodological framework for science foresight on the basis of published research papers, consisting of (i) life-cycle analysis, (ii) text mining and (iii) knowledge gap identification by means of automated clustering. The three steps are connected using the research methodology of the research papers, as identified by text mining. The potential of combining these three steps in one framework is illustrated by analyzing scientific literature on wind catchers; a natural ventilation concept which has received considerable attention from academia, but with quite low application in practice”. Our work similar this work by performed the preprocessing on paper and clustering but with different techniques used. The following table explained compare among the previous work.
2 Theoretical Concepts In this section, we will be showing the main concept used in design and implementation the suggest system. 2.1
Intelligent Big Data Analysis
Data is any object or single feature described that object or collection of features. In general, “data can be divided into the three types (Fig. 2) (i.e. small data, normal data, and big data) [16]. Although the term ‘Big Data’ has become popular, there is no general consensus about what it really means. But, we can definition from intelligent
106
S. Al_Janabi et al.
analysis as a way of extraction, transformation, and load (ETL) for huge datasets/databases. Big data challenges incorporate catching information, information stockpiling, information investigation, look, sharing, exchange, perception, questioning, refreshing, and data protection. There are three measurements of enormous information known as Volume, Variety, and Velocity, but this definition is developed to convert big data from 3 to 9 Vs for more detail see [12].” Intelligent analysis meaning extraction new, novel, and useful knowledge, patterns, relationships and models by nontraditional manner.
Fig. 2. Types of data [12]
2.2
Rapid Automatic Keyword Extraction (Rake)
It is a way to extract the keywords from corpus (set of documents) in a fast and useful way, it has three input parameters first the stop words list (or stop list), the phrase delimiters set, and word delimiters set. In the begin the documents is parsing its text into set of candidate keywords in order to extract keywords from it. First, it uses the words delimiters to split the text into sequence of words. Second, it used phrase delimiters and
Fig. 3. Word co-occurrence graph
Multi-level Network Construction
107
stop list position to split this array into sequence of contiguous words. Third, Rake assigned the same position to words within a sequence into text considered candidate keywords [13]. After determining the candidate keyword and constructing graph from the co-occurrences of words the score is calculated by compute the word frequency denoted as freq(w) and word degree denoted as deg(w) then ratio between them (Fig. 3).
Fig. 4. Calculate word scores from graph
The score of candidate keyword is calculate and define as sum of it member word scores. After scored the candidate keywords, the top T keywords are considered as keywords for that document [14] (Fig. 4). 2.3
Latent Dirichlet Allocation (LDA)
LDA is a generative probabilistic model that is more generally used in topic models for the collection of documents (Fig. 5).
Fig. 5. LDA model
108
2.4
S. Al_Janabi et al.
Complex Network
A network is consisting of components and edge connecting these components. The network is representing by a graph structure G (V, E) where V is vertices representing the components of the network, and E is a set of edges representing the links. The simplest networks do not have such features for example random graphs or lattices on other hand it occurs in graphs modeling for most real system. Due to the recent technological developments, such as increased computational power and storage space, has allowed the gathering of big amounts of data from these networks, therefore, the analysis of complex networks has become much easier than previously and with more detail [15]. 2.5
Characterizing Networks
In general, complex networks contain to topological properties which are can be explored in their experimental. Such as small world property indicate that distance is small between two nodes. Scale-free network it means that few nodes have high degree and the remainder node have low degree. Because complex networks consist of huge number of nodes it’s more difficult to represent as whole therefore, we need to schemes or methods rather than visualization to more understand the structure of it. Centrality is measure importance of network nodes and edges, which can be representing by the number of node’s neighbors or the number of connected edges to that node. Clusters or community structure of network is containing the nodes that are more similar to each other than the remainder of the network. In order to more understanding the structure and natural of network it must detecting such gather of nodes [15].
3 Hybrid Miner-Network Analyzer (HMNA) The citation network is one of the complex networks that growth continuous with the time. Therefore, it is very difficult of analysis and extraction useful knowledge from it quickly. This chapter considers with design and descript the main stages of a Hybrid Miner-Network Analyzer system by combines the advantages of both the text mining algorithms and complex networks to satisfy the objectives of this work. The initial citation network begins with only nodes represent the documents and links represent the citation on that documents. The goal is; how can handling corpus “set of documents”? corpus input into the preprocessing stage that includes two phases; first phase: “split the document to separated sentences and find the score for each one using Rake than, second phase cleaning the result of the previous phase”. The cleaning text grouping into multi clusters based on LDA model and find the topic for each document (i.e., “classified documents”). This step generated new knowledge about the contain of nodes “documents” by classified it based on topics, that satisfy through text mining techniques “Rake & LDA”. Next step, combination the label of document results from text mining analysis into initial network to reconstruction citation network as multi communities’ network. Where, each community or topic represent as level or sub network in that network.
Multi-level Network Construction
109
Analysis these “levels” or “sub networks” by each once appear one level and hidden all other levels “communities”. After that, compute the characteristic measures for it that knowledge represent the new type of knowledge extraction from “Complex network concept”. In general, we aim to show how knowledge obtained from text documents is connected to knowledge obtained from citation networks of their documents. The solvation methodology first classifying the content of documents then labeling each document by its class. Then compute the communities within citation network and see relation between detected communities and classes of documents. The overall solution is demonstrated in diagram in Fig. 6.
110
S. Al_Janabi et al.
Fig. 6. Architecture of HMNA System
4 Implementation Results of HMNA 4.1
Collection Dataset
We use in our work High Energy Physics (HEP) dataset. It is a publicly available dataset that are compiled by arXiv for the KDD Cup 2003 competition. It’s containing abstract of 29,555 articles in addition it includes citation network file. It is published between 1992–2003.
Multi-level Network Construction
111
The dataset involves the following information: (I) article’s abstract, (II) citation network for articles in form of (citing_articles_id, cited_articles_id), (III) Unique identifiers for each article and (IV) each article has SLAC dates in form (article_id, SLAC date). SLAC is referring to online publish dates for each article at the Stanford Linear Acceleration Center (SLAC) library. Each abstract has title of articles, articles id, date of publishing, author’s names, place of publishing, volume number, page number etc. 4.2
Building the Base Network
Citation networks consist of 27,771 vertices and 352,807 edges. The network is implemented by graph stream library. We keep only giant connected components (i.e. set of nodes that are belonging to the biggest connected component). The number of remaining nodes is 27400 and the number of edges is 352542 (Fig. 7). 4.3
Applying Pragmatic Method to Text Preprocessing
In this stage, we apply Rake on corpus than cleaning the results, and enter the clean text into LDA to determine the topics for each document. Using different number of topics (2, 6, 10) as explain Table 1. Table 1. Results of preprocessing stage Keywords Score virasoro highest weight conditions 5 23 higher degree potentials 16 genus expansion 6 8.5 main theorem 2 8.3 intersection numbers 2 8.3 2 expansion 5.3 kontsevich integrals 4.5 generating functions 4.5 moduli space 4 first part 4 3 proof 4 e equation 4.5
Keywords study algebraic aspects matrix airy equation schur function 2 kontsevich integral 2 theorem 3 painlev intersection theory singular behaviour kdv constraints kdv 4 virasoro
Score 9 8.5 8.3 8.3 4.5 4.5 4.5 4 4 4 3
Table 2. Results of LDA using number of topics 2 Cluster Group of words no. Cluster#1 0.028*“gauge” + 0.022*“theory” + 0.021*“quantum” + 0.020* + 0.011*“model” Cluster#2 0.025*“string” + 0.020*”theory” + 0.015*“field” + 0.010*“energy”
112
S. Al_Janabi et al. Table 3. Results of LDA using number of topics 6
Cluster no. Group of words Cluster#1 Cluster#2 Cluster#3 Cluster#4 Cluster#5 Cluster#6
0.052*“conformal” + 0.034*“boundary” + 0.033*“model” + 0.021*“integrable” 0.023*“theory” + 0.017*“function” + 0.016*“renormalization” + 0.015*“finite” 0.051*“string” + 0.048*“gauge” + 0.046*“theory” + 0.022*“supersymmetric” 0.044*“field” + 0.032*“boundary” + 0.024*“model” + 0.023*“scalar” ‘0.024*“black” + 0.019*“energy” + 0.018*“cosmological” + 0.013*“gravity”’ 0.032*“quantum” + 0.020*“theory” + 0.020*“field” + 0.016*“gauge”
Table 4. Results of LDA using number of topics 10 Cluster no. Group of words Cluster#1 Cluster#2 Cluster#3 Cluster#4 Cluster#5 Cluster#6 Cluster#7 Cluster#8 Cluster#9 Cluster#10
4.4
(1, ‘0.039*“quantum” + 0.030*“theory” + 0.030*“field” + 0.014*“classical”’) (2,‘0.019*“method” + 0.015*“function” + 0.014*“expansion” + 0.014*“perturbative”’) (3, ‘0.105*“string” + 0.050*“theory” + 0.022*“field” + 0.016*“tachyon”’) (4, ‘0.056*“boundary” + 0.032*“model” + 0.030*“theory” + 0.019*“limit”’) (5, ‘0.041*“effective” + 0.026*“potential” + 0.022*“energy” + 0.019*“action”’) (6, ‘0.026*“brane” + 0.016*“cosmological” + 0.013*“can” + 0.012*“hole”’) (7, ‘0.067*“gauge” + 0.023*“theory” + 0.017*“field” + 0.015*“model”’) (8, 0.020*“conformal” + 0.017*“charge” + 0.017*“topological” + 0.017*“field”’) (9, ‘0.031*“algebra” + 0.018*“space” + 0.017*“quantum” + 0.016*“group” + 0.013*“lie”’) (10, ‘0.039*“noncommutative” + 0.029*“gauge” + 0.028*“supergravity” + 0.024*“theory”’)
Classify Document by Find Topic Distribution
Compute the topic distribution for each document over the topic to classify documents on its topics. Then, we take topic documents as the maximum contribution value (Tables 2, 3 and 4).
Fig. 7. Topic_Distribution for document (9201001).
Then compute the Number of Documents in each Topic, in general, when divided corpus into two, six, and ten topics we get the results in Figs. 8, 9 and 10.
Multi-level Network Construction
Fig. 8. Number of documents in two topics
Fig. 9. Number of documents in six topics
Fig. 10. Number of documents in ten topics
113
114
4.5
S. Al_Janabi et al.
Building Multi-level Network
We obtained classification for each document on specific topic. We use this result to label node of the initial networks therefore, each node has its topic. The result is multicommunities networks each level represent specific topic within the dataset. Figure 11 represents second-communities in network after hidden other level.
Fig. 11. Second- level in multi level network
Figure 12 represents fifth-level in network after hidden all other levels.
Fig. 12. Level number five in multi level network
Figure 13 represents seventh-community in network after hidden all other.
Fig. 13. Seventh-level in multi-level network
Multi-level Network Construction
4.6
115
Compute the Characteristic Measures
We compute the characteristic measures for base network and for each community in the networks. First compute number of nodes in each topic and number of edges. Calculate the density of each community to see which community is highly dense and dominate in the network and Compute clustering coefficient (Table 5). Table 5. The properties measured for two-community networks Topic# Topic#0 Topic#1 Topic#2
N M 27400 352542 14983 95148 11295 149906
D 0.0093 8.47736 0.00235
CC 0.31590 0.31973 0.32856
Where, Topic #0 meaning the based network, while topic #1 and #2 results from LDA. N: number of node; M: number of edge; CC: clustering coefficient; D: density of graph (Tables 6 and 7); Table 6. The properties measured for six-community networks Topic# N M D CC Topic#0 27400 352542 0.0093 0.31590 Topic#1 9062 52591 0.00128 0.32722 Topic#2 766 1559 0.00532 0.34840 Topic#3 167 379 0.00273 0.02734 Topic#4 4799 15359 0.01146 0.32411 Topic#5 3406 30593 0.00527 0.31545 Topic#6 5531 47251 0.00308 0.33585
Table 7. The properties measured for ten-community networks Topic# N M D CC Topic#0 27400 352542 0.0093 0.31590 Topic#1 2859 8048 0.00196 0.28311 Topic#2 446 1140 0.01148 0.36531 Topic#3 1967 5282 0.00273 0.31110 Topic#4 923 4854 0.01146 0.32411 Topic#5 963 3272 0.00706 0.31543 Topic#6 1150 4897 0.00741 0.35111 Topic#7 6191 32627 0.00170 0.31927 Topic#8 595 1274 0.00720 0.29661 Topic#9 5181 61633 0.00459 0.33743 Topic#10 842 3081 0.00870 0.36605
116
S. Al_Janabi et al.
5 Discussions and Conclusions In this section, we will explain the main limitations and advantage of each tool, also specified the main parameters Influential in take decision of it. RAKE is “based on our observation that keywords frequently contain multiple words but rarely contain standard punctuation or stop words. Input of it are list of stop words (or stop-list), a set of phrase delimiters, and a set of word delimiters”. While output “is word co-occurrences “score candidate keywords”. Main advantage of this algorithm gets keywords only by compare frequency of keywords, it uses the simple and basic formula so the complex is low, this make system run faster than others when work with large document, extract more keyword and less bug. While, main limitations of it results because the keyword not only determine by frequency but also by its mean. In addition, this algorithm makes compound word get higher weight so they are nor accurate enough”. LDA is “A generative probabilistic model for collections of discrete data such as text corpora. LDA is a three-level hierarchical Bayesian model, in which each item of a collection is modeled as a finite mixture over an underlying set of latent topics. Each observed word originates from a topic that we do not directly observe. Each topic is, in turn, modeled as an infinite mixture over an underlying set of topic probabilities. LDA is the fitted model can be used to estimate the similarity between documents as well as between a set of specified keywords using an additional layer of latent variables which are referred to as topics. How is LDA related to text mining and other machine learning techniques? Topic models can be seen as classical text mining or natural language processing tools. Fitting topic models based on data structures from the text mining usually done by considering the problem of modeling text corpora and other collections of discrete data. One of the advantages of LDA over related latent variable models is that it provides well-defined inference procedures for previously unseen documents. While, Main limitations [i. Fixed K (the number of topics is fixed and must be known ahead of time), ii. Uncorrelated topics (Dirichlet topic distribution cannot capture correlations), iii. Non-hierarchical (in data-limited regimes hierarchical models allow sharing of data), iv. Static (no evolution of topics over time), v. Bag of words (assumes words are exchangeable, sentence structure is not modeled), vi. Unsupervised (sometimes weak supervision is desirable, e.g. in sentiment analysis)]”. Complex network is a graph (network) with non-trivial topological features, we can extraction useful pattern from that network. The input of that network (collection of nodes “documents” and links “citation of that documents” topic for each document, while the output relation shows the new types of knowledge need to prove and justification to indicated the authors of the suitable references for their works by take the results of HMNA to citation network. Main advantages of complex network can cover and give complete picture for huge\big database, it very difficult understand their details of that data but relationships between nodes and links can stablish our multi rules simplified understanding that data. Graph is considered as statistical model can extraction from it values of multi characteristic measures described the behaviours of that graph. While the main limitations of network are not considered the contain of node in extraction the relationships.
Multi-level Network Construction
117
In general, the main purposes of design HMN is attempting to answer on the following equations • Is the output of text mining useful in reconstructed the initial network? – Of course, yes because it provides the new information represent by the topics that used as label of nodes in network, this meaning text mining provide the network of the lack information “information about the contain of nodes that make it possible split the network into multi separated communities. • Is complex network can covert the text mining technique from black box to white box? – Yes, because the characteristic measures for each level give summary of the community that represent to that level without need return of the apply LDA in testing process. • Is classified text mining can provide useful information to avoid the problem of network “lack of information related of contain of nodes”? – Yes, as explain in equation number one, where the hybrid system gives me integration domain of problem and each technique used solve the drawback of another techniques to salsify the idea of integration system. • What is a new knowledge extraction from HMNA not appear into the initial citation network? – To topics of document not appear in traditional citation network, this topic makes the user can take answer for any queries based on the name of journal level, or the name of author level, or the date of publications. – Reduce number of preprocessing steps through using Rake and add condition on their results. – Prove RAKE give good results with the LDA Model Rather than FT\IDF that given bad result with LDA Model.
References 1. Zanin, M.: Complex networks and data mining: toward a new perspective for the understanding of Complex Systems (2014) 2. Ali, S.H.: A novel tool (FP-KC) for handle the three main dimensions reduction and association rule mining. In: IEEE, 2012 6th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), Sousse, pp. 951–961 (2012). https://doi.org/10.1109/SETIT.2012.6482042, http://ieeexplore.ieee. org/stamp/stamp.jsp?tp=&arnumber=6482042&isnumber=6481878 3. Ali, S.H.: Miner for OACCR: Case of medical data analysis in knowledge discovery. In: IEEE, 2012 6th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), Sousse, pp. 962–975 (2012). https://doi.org/ 10.1109/SETIT.2012.6482043, http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber= 6482043&isnumber=6481878 4. Ohsawa, Y., Benson, N.E., Yachida, M., Science, H.: KeyGraph: automatic indexing by co-occurrence graph based on building construction metaphor. In: International Forum on Research and Technology Research and Technology Advances in Digital Libraries, pp. 12–18 (1998)
118
S. Al_Janabi et al.
5. Litvak, M., Last, M.: Graph-based keyword extraction for single-document summarization. In: Proceedings of the workshop on Multi-source Multilingual Information Extraction and Summarization, pp. 17–24, August 2008 6. Abu-errub, A.: Arabic text classification algorithm using TFIDF and chi square measurements. Int. J. Comput. Appl. 93(6), 40–45 (2014) 7. Abilhoa, W.D., De Castro, L.N.: A keyword extraction method from twitter messages represented as graphs. Appl. Math. Comput. 240, 308–325 (2014) 8. Jain, A., Mishra, R.D.: Text categorization: by combining Naïve Bayes and modified maximum entropy classifiers. Int. J. Adv. Electron. Comput. Sci., 22–126, Sep 2016. Special Issue, ISSN: 2393-2835. http://www.iraj.in/journal/journal_file/journal_pdf/12-2951475473979122-126.pdf 9. Brahimi, B., Touahria, M., Tari, A.: Data and text mining techniques for classifying Arabic tweet polarity. J. Digit. Inf. Manag. 14(1), 15–25 (2016) 10. Shah, M., Shinde, S., Sawant, R.S., Wagh, P.P.: Analysis of text review using hybrid classifier, vol. 7, no. 4, pp. 10914–10916 (2017) 11. Rezaeian, M., Montazeri, H., Loonen, R.C.G.M.: Science foresight using life-cycle analysis, text mining and clustering: a case study on natural ventilation. Technol. Forecast. Soc. Change 118, 270–280 (2017) 12. Al_Janabi, S.: Smart System to create optimal higher education environment using IDA and IOTs. Int. J. Comput. Appl. (2018). https://doi.org/10.1080/1206212x.2018.1512460 13. Dutta, A.: A novel extension for automatic keyword extraction. Int. J. Adv. Res. Comput. Sci. Softw. Eng. 6(5), 160–163 (2016) 14. Rose, S., Engel, D., Cramer, N., Cowley, W.: Automatic keyword extraction from individual documents. In: Text Mining: Application and Theory, pp. 1–20. Wiley (2010). ISBN: 978-0470-74982-1. https://doi.org/10.1002/9780470689646.ch1 15. Erciyes, K.: Complex networks: an algorithmic perspective (2015) 16. Al-Janabi, S., Salman, M.A., Fanfakh, A.: Recommendation system to improve time management for people in education environments. J. Eng. Appl. Sci. 13, 10182–10193 (2018). https://doi.org/10.3923/jeasci.2018.10182.10193, http://medwelljournals.com/ abstract/?doi=jeasci.2018.10182.10193
What Will Millimeter Wave Communication (mmWave) Be? Fatima Zahra Hassani-Alaoui(&) and Jamal El Abbadi Smart Communications Research Team (ERSC), E3S Research Center, EMI, Mohammed V University Rabat, Rabat, Morocco [email protected], [email protected] Abstract. New research directions bring basic changes in the design of future fifth generation (5G) cellular networks. Millimeter wave communication (mmWave) has emerged as a key to enable the next generation. This perceivable survey focuses on identifying this new technology and its potential impact on 5G, by synthesizing high quality researches. This paper highlights the most recent and the newest substantive design choices, the features and the challenges to overcome in the mmWave area. We begin by defining the characteristics of mmWave communication. Using this framework, we evaluate the changes that will bring to the 5G new radio (5G NR) while using this frequency band. This article does not provide a final solution because researches are still in progress in this field, but it transmits the latest vision of the future architecture. Keywords: mmWave communication (mmWave) 5G cellular network
5G new radio (5G NR)
1 Introduction As more the existing technologies expand, consumers and businesses expect to see more opportunities in the future technology, this one has to be faster and have the ability to accomplish many services. 5G technology will operate the major fields of industries, namely: healthcare, education, transportation, smart homes and entertainment. Industrials and researchers started to clarify the 5G architecture after many experiences and resources. The commercial deployment of the 5G mobile network will be launched in 2019– 2020 according to The Third-Generation Partnership Project (3GPP). The first set of 5G standards, Release 15, was delivered in December 2017, and there still many Improvements to be accomplished to meet all the requirements needed. The 5G design includes a set of new and effective technologies that will be used in its networks [1–6]. Nowadays, smart phones and the other electronic devices are using a very specific frequencies on the radio frequency spectrum [3 kHz–6 GHz], but these frequencies are starting to get more crowded. This problem causes slower service and dropped connections. To solve this complication, it is necessary to transmit signals on new bands of frequencies which called millimeter waves [30 GHz–300 GHz]. On the one hand, this
© Springer Nature Switzerland AG 2019 Y. Farhaoui and L. Moussaid (Eds.): ICBDSDE 2018, SBD 53, pp. 119–131, 2019. https://doi.org/10.1007/978-3-030-12048-1_14
120
F. Z. Hassani-Alaoui and J. El Abbadi
large range of spectrum is needed to afford a combination of 5G requirements, namely: high data rates, high capacity, ultra-high reliability and omnipresent coverage. On the other hand, mmWave communications is the key to provide multi-gigabit used in new services [7]. In the initial phase, the principal worldwide spectrum options for 5G are at 3.3– 4.9 GHz and 24–28 GHz, 39 GHz for millimeter waves. More information about the licensed and unlicensed spectrum in the world are detailed in [8]. In this survey paper, we present the key technology of the future 5G networks. We have studied the newest and the latest progress in the mmWave communications area. The remainder of this review-paper is organized as follows. First, we introduce the signal propagation characteristics in mmWaves bands, from that point we describe the inconveniences in using those frequencies. Secondly, we highlight the 5G new radio (5G NR) which optimize the mmWaves based network, and a big attention is given to the Beamforming techniques due to the its important role. Finally, we conclude by the criterions that must be considered in the deployment of the future generation.
2 Characteristics of mmWave 2.1
Measurement Results and Channel Models
The creation of a new generation of mobile network, starts with understanding the channel characteristics of radio access technology (RAT), this step is important or even essential for the standardization, deployment and the design of 5G network [9]. The most influential and important project in the world in taking measurements and developing an advanced channel model is mmMAGIC project (mm-wave based Mobile radio Access network for fifth Generation Integrated Communication), It took 24 months and it was co-funded by the European Commission’s 5G PPP program, bringing together major infrastructure vendors (Samsung, Ericsson, Alcatel-Lucent, Huawei, Intel, Nokia), major European operators (Orange, Telefonica), leading research
Fig. 1. Overview of mmMAGIC measurements (black) and simulations (blue) [9].
What Will Millimeter Wave Communication (mmWave) Be?
121
Table 1. Overview of some researchers and academia measurement campaigns. Publication Detailed information on these measurement pursuits Scenario Objective The used Frequency bands in GHz [11]
15 and 28
[12]
60
[13]
60
[14]
28 and 38
[15]
28, 38 and 73
[16]
28 and 73
[17]
28, 38, 60 and 73
[18]
28
[19]
2.4, 28, 60 and 73
[20]
73
An outdoor microcellular environment with a valid Line of sight (LOS) link An urban test based on point-topoint predictions from different Base stations (BS) to User Equipment (UE) An urban environment for a road, surrounded by multistory buildings, single input single output (SISO) systems have been used The urban environments around the University of Texas at Austin and New York University
Dense urban environments (New York City and Austin), measurements taken for LOS and NLOS links Dense urban environments (New York City), measurements taken for free space LOS propagation and NLOS
Analysis of multipath propagation Obtain key statistical parameters for the mmWave channel
Investigate and characterize the mmWave communication link
Studying spread, path loss, building penetration and reflection characteristics for the design of future mmWave cellular systems Create mmWave large-scale path loss models for future 5G system simulation and design
Presenting a probabilistic path loss model which can be used to estimate signal coverage, interference and outage as a function of distance Base station to mobile, base station Analyzing and simulating the performance of mmWave to base station (backhaul), peer to peer, and vehicular (V2 V) scenarios networks that will rely on adaptive antennas and multiplein dense urban environments input and multiple-output (MIMO) Areas lined with trees similar to a Predicting path loss models vegetated residential environment through vegetated areas at 28 GHz Studying the coverage and A map-based model for the base stations (BSs) in a university campus interference effects in mmWave with a random distribution of users frequency bands Provide information about An urban Microcell open square in downtown Brooklyn, New York. coverage and interference for Measurements taken from Ten future mmWave small-cells that random receiver locations will exploit macro-diversity and CoMP
122
F. Z. Hassani-Alaoui and J. El Abbadi
institutes and universities (Fraunhofer HHI, CEA LETI, IMDEA Networks, Universities Aalto, Bristol, Chalmers and Dresden), measurement equipment vendors (Keysight Technologies, Rohde & Schwarz) and SME (Qamcom) [10]. The measurements and simulations obtained in this project were done in different environments cases, and in certain frequencies. Those scenarios are shown in Fig. 1. On the other hand, researchers and academia have led many measurement campaigns. Some of the latest and the most important projects are presented in the Table 1. 2.2
Challenges to Overcome
All the accomplished studies and measurements, (that were carried out in various frequencies, different scenarios and across different materials), demonstrate that mmWave communication has some challenges to overcome since it is the key to enable 5G mobile network. The shortcomings that mmWave faces are: high pathloss in the high frequencies, higher sensitivity to obstacles and blockage, and finally the decreased diffraction [21]. (a) The attenuation of rain and atmospheric gases or molecular The effects of rain, atmospheric gases or molecular are the principal factors that come into play in the propagation loss in the mmWave communications. Detailed simulations, which demonstrate the effects of those factors, are presented in [22–25]. Using small cells, which are considered an unavoidable technology in the deployment of the 5G network, can approximately understate the effects of rain and the atmospheric gases or molecular, in the propagation of mmWaves. This theory was established by many researchers in taking measurements in cellular under 1 km in distance, in 28 GHz, 38 GHz and 60 GHz frequencies. Taking into account that the size of cells, that will be installed in urban environments, are in the order of 200 m, we can consider that the new architecture of 5G will overcome this problem. (b) The sensitivity to obstacles and blockage Another important issue for using mmWave is the penetration loss experienced by radio transmissions traversing various objects (e.g., humans, buildings and furniture). In higher frequency bands, human blockage has a significant interest. The human body even the user’s own body can cause strong shadowing for the radio signal, in some cases it can block the communication link [26]. A person can induce more than 40 dB of loss when standing 50 cm from the transmitter or receiver antenna (which is shown in Fig. 2) [27].
Fig. 2. Body loss model for tablet use case (40 dB attenuation per ray at 73.5 GHz) [26].
What Will Millimeter Wave Communication (mmWave) Be?
123
A proposed cone blocking model to characterize the blocking effects by user’s body, is presented in [28]. In [29] Authors presented and studied the characteristics of mmWaves in the presence of the human body at 60 GHz. They introduced four models of different body parts to evaluate thermal effects of millimeter-wave radiation on the body. Simulation results show that about 34% to 42% of the incident power is reflected at the skin surface. On the other hand, mmWaves face penetration loss in crossing solid materials (e.g. glass, drywall, brick wall, wood, buildings …) [30], or vehicular [31]. Measurements in [32] clarify this serious issue. METIS project included a blockage model in its channel model, and the recent 3GPP (Release 14) channel model included “blockage” as an add-on feature [33]. (c) The decreased diffraction The phenomena of the weakening of waves around the corners of an obstacle, which is known as the diffraction, is decreasing while the frequency increases in outdoor scenarios, but it does not have an important impact above 10 GHz [34]. Interior buildings, diffraction loss has less impact due to reflection and transmission between walls [35]. For the indoor environment, a Knife Edge Diffraction (KED) model can be used to calculate diffraction loss, which is approximately 5 dB to 6 dB in standard deviation. For the outdoor building corner with rounded edges, diffraction loss can be predicted by a simple linear model with a fixed anchor point. 2.3
Deployment Environment of 5G mmWave
The environment where 5G will be deployed is conditioned by the criterion of mmWaves characteristics. Measurements and shortcomings of mmWaves conduct us to think about the cells characteristics of the 5G network. One of the important key aspect that characterizes the 5G architecture is deploying macro cells and small cells, and assuring the Cooperation between those two kinds. In this scenario, macro-cells provide wide area coverage while small base stations provide high data rates (approximately 1 Tb/s/km2 in up to 100 GHz band with 2 GHz carrier bandwidth [26]). The main benefits of the cooperation between heterogeneous cells are the energy efficiency and the low latency. Thus, it’s emergent to develop cooperation in multicast scenarios [36]. On the other hand, the 5G mmWave systems are targeted to be deployed in the following environments: Urban micro with cell radius less than 100 m; Suburban micro with cell radius around 200 m, and access points mounted at 6 to 8 m; and finally, indoor hotspots (offices, cubicles and shopping malls) which are three to five storeys high, and access points spaced at 2 to 3 m [21].
3 5G NR Interface On the one hand, using new frequency bands involve an emergent evolution towards 5G new radio (5G NR). This advancement can optimize the mmWaves based network. On the other hand, the 5G NR will enhance all the 5G requirements (long battery
124
F. Z. Hassani-Alaoui and J. El Abbadi
lifetimes, high data rate, low latency, new services), by using advanced techniques in modulation, waveforms and access technologies. 3GPP standards committee released the specifications of 5G NR in Release 15 (approved in December 2017). Release 16 is expected to be finalized in the middle of 2019. As any new technology, 5G NR will support limited use cases in the first phase. 3.1
New Waveform and Optimized Modulation
(a) Numerology 5G is designed to provide a wide variousness of services and utilities, by performance waveform parameters flexibly. The introduction of the flexibility in the waveform minimizes latencies, ameliorate reliability and QoS. This pliability was afforded by coexisting of numerologies, each numerology consists of a set of parameters specifying the frame structure of the waveform [37–39]. The flexible numerology in 5G NR is dissimilar from numerology in LTE. In spite of the advantages of this flexibility it brings new challenges in the manner waveforms are operated and built. Contemporaneous multi-numerology utilization was permitted by 5G NR. In the literature [40–43], Multi-numerology structures that were included in the 3GPP 5G NR standardization were enlarged. 5G NR takes up a flexible subcarrier spacing of: 2l 15 kHz (l = 0, 1, …, 4). In the mmWave, where the phase noise issue becomes serious, 5G NR supports 60 kHz and 120 kHz for data channels. 15 kHz and 30 kHz are suitable for lower frequencies below 6 GHz (Fig. 3).
Fig. 3. Structure of frames and slots in different frequencies [45].
What Will Millimeter Wave Communication (mmWave) Be?
125
(b) Frame structure In order to facilitate 5G NR and LTE coexistence, the structure of the frame will be similar to LTE: One frame (10 ms) contains ten subframes (1 ms for each one) and every slot own 14 OFDM symbols. The augmentation of the numerology maximizes slots in a subframe; as a result, rising the number of symbols sent in a determined time. In addition to that, while the frequency increases, the slot duration decreases. Slots can be DL, UL, or flexible. As a consequence, the network can dynamically scale UL and DL transmission. Thus, it is a solution to optimize traffic for different service types [44]. 5G NR allows transmission to start at any OFDM symbol and last as many symbols as required for the communication; in other word, scheduling works perfectly in a slot. This kind of slot transmission can aid in obtaining a very low latency for crucial data as well as reduce interference. A summarize of the waveform structure is presented in Fig. 3. (c) Optimized OFDM modulation Compared with other waveforms, orthogonal frequency division multiplexing (OFDM) technology is the excellent even the winner candidate which can be adopted, due to its advantages such as easy integration with MIMO, low complexity, low cost, and plain channel estimation [46]. In addition to that, OFDM is able to provide a more optimum parameter variety for every service group by permitting multiple parameter configurations, and hence better system efficiency. After long studies and evaluation of all the waveform proposals, 3GPP agreed to adopt OFDM with a cyclic-prefix (CP) for both DL and UL transmissions, in order to improve mobile broadband capacity [47]. On the other hand, 5G NR is able to use discrete Fourier transform (DFT) spread OFDM in the uplink to ameliorate coverage. In the mmWave communication, a single carrier (SC) based waveform is preferred to be used, so as to: increase power efficiency, allow efficient beamforming and decrease switching overhead [48]. A new transmission OFDM design considering the characteristics of mmWave communication has been studied in [49]. Case study of hybrid beamforming scheme for OFDM-based systems with large-scale antenna arrays is proposed in [50], which demonstrate the efficiency in using all those technologies in the mmWave communication. 3.2
Beamforming
(a) Massive MIMO Beamforming The implementation of Massive multiple input multiple output (MIMO) can be performed in various way. The implication of beamforming is a special technique used to combine multiple antenna elements to condense the power in a specific direction. The antenna system needs to dynamically steer the beam to the user devices in the cell area to insure the full coverage from a single radio and antenna (Fig. 4). The beamforming ameliorates the signal to noise plus interference ratio (SNIR); in other word, it can eliminate the interference, some designs for this objective are given in [51, 52]. A proposed beamforming technique which assuring atmospheric coverage is developed in [53].
126
F. Z. Hassani-Alaoui and J. El Abbadi
Fig. 4. Beamforming types
Advanced beamforming techniques can track the user. The paper [54] proposes a beam tracking robust mechanism for mobile mmWave communication systems which considered the system throughput and handoff probability. It is important to mention that the efficient adaptive beam tracking algorithms need to be employed for both BS and UE in order to be accurately aligned with acceptable latency. Fortunately, 5G NR can also leverage user equipment (UE) antenna variety to surmount the issue of hand-blocking [55]. The interesting work in [56] proposes a novel and highly configurable system design for 5G cellular UE, which is the distributed phased arrays based-MIMO (DPA-MIMO), this topology furnishes a solution to human body blockage. Depending on the hand position habits while using phones (or tablets) and his impact in the mmWave communications studied in [57] and summarized in the Fig. 5. An effective solution of placing BF modules in the phone devices is proposed. More precisely, it is mandatory to position BF modules (BFM) at the central part of the mobile device, top two corners and bottom two corners (Fig. 6). As a result, this tech can be used in device to device communication (D2D) [58]. (b) Initial access/Random access Initial Access. Initial access is the proceeding done by a user equipment to find a cell to camp on, to get the needful system information, and to request a linking through random access [59]. Different initial access protocols are studied in [60]. In order to choose the best initial access technique, a comparison analysis of several design is detailed in [61]. Owing to the propagation difficulties and obstacles in mmWave bands, 5G NR defined new initial access design for beamforming that will utilize beam scanning (or beam-sweeping); in this procedure, the base station can identify the strongest beam and establish the convenient connection. Therefore, Beamforming technology has a fundamental role in the access to the 5G NR, either for user plane or control plane, which can be performed at the transmitter side or receiver side. 5G NR inserted the synchronization signal block (SSB), which is role is to support the multi-beam process in high frequency scenarios. SSB is made up of a primary synchronization signal (PSS), a secondary synchronization signal (SSS), and a physical broadcast channel (PBCH), the lately one carries the basic system information. PSS, SSS and PBCH are transmitted together, so they all have the same periodicity and they share the same single antenna port. For the initial cell picking the UE can procure the physical cell identity. It is important to mention that the beams applied to an SSB are transparent to the UE, since it sees the equivalent SSs and PBCH after precoding and/or beamforming operations, that are up to the network implementation [62].
What Will Millimeter Wave Communication (mmWave) Be?
127
Fig. 5. Hand position holding mobile phone
Fig. 6. BF modules (BFM) positioning in the UE device.
Random Access. 5G NR will use Zadoff-Chu (ZC) sequences. The ZC sequence is used for generating NR random-access preambles, it is known to have an ideal autocorrelation property, including constant amplitude before and after Discrete Fourier Transform (DFT) operation, zero cyclic auto-correlation and low cross- correlation. This makes it appropriate for many applications such as the random-access preamble of the Physical Random-Access Channel (PRACH). The paper [63] exposes more details about ZC sequence design for random access, while the letter [64] proposes Improved ZC Sequence detection under unknown multipath. The PRACH transmits a random-access preamble from a device to the base station, so as to indicate a random-access attempt and to aid the base station to adjust the uplink timing of the UE (among other parameters). This new design has many advantages. The first benefit is the support of analog beam-sweeping during PRACH reception. The second one is allowing the base station receiver to use the same fast Fourier transform for random-access preamble detection and data. The third plus is the robustness, performed by the short preamble format, against time varying channels and frequency errors.
128
F. Z. Hassani-Alaoui and J. El Abbadi
4 Conclusion and Future Trends This paper aims to respond to the question: What will mmWave communications be? We have provided an overview of the characteristics of this communication, we have studied the evolution that will bring to the 5G networks. The idea gotten from this work describes the importance of optimizing this technology due to its value in the future generation. While academics and industrials are currently working on what will define 5G wireless networks, it is important, even essential, to think about Green Communications, which ensure energy efficient, low cost and especially “Environmentally Friendly” (in order to avoid the health effects caused by mmWaves). Limited researches were carried out in this area; therefore, it presents the objective of a future work.
References 1. Boccardi, F., Heath, R., Lozano, A., Marzetta, T., Popovski, P.: Five disruptive technology directions for 5G. IEEE Commun. Mag. 52(2), 74–80 (2014) 2. Chen, S., Zhao, J.: The requirements, challenges, and technologies for 5G of terrestrial mobile telecommunication. IEEE Commun. Mag. 52(5), 36–43 (2014) 3. Wang, C.-X., Haider, F., Gao, X., You, X.-H., Yang, Y., Yuan, D., Hepsaydir, E.: Cellular architecture and key technologies for 5G wireless communication networks. IEEE Commun. Mag. 52(2), 122–130 (2014) 4. Bangerter, B., Talwar, S., Arefi, R., Stewart, K.: Networks and devices for the 5G era. IEEE Commun. Mag. 52(2), 90–96 (2014) 5. Chin, W., Fan, Z., Haines, R.: Emerging technologies and research challenges for 5G wireless networks. IEEE Wirel. Commun. 21(2), 106–112 (2014) 6. Agiwal, M., Roy, A., Saxena, N.: Next generation 5G wireless networks: a comprehensive survey. IEEE Commun. Surv. Tutor. 18(3), 1617–1655 (2016) 7. Niu, Y., Li, Y., Jin, D., Su, L., Vasilakos, A.V.: A survey of millimeter wave communications (mmWave) for 5G: opportunities and challenges. Wirel. Netw. 21(8), 2657–2676 (2015) 8. Nokia: 5G new radio network: uses cases, spectrum, technologies and architecture (n.d.). https://onestore.nokia.com/asset/205407 9. Peter, M.: Measurement Results and Final mmMAGIC Channel Models (2017) 10. mmMAGIC – mm-Wave based Mobile Radio Access Network for 5G Integrated Communications (n.d.). https://5g-mmmagic.eu/. Accessed 7 Sept 2018 11. Sheikh, M.U., Lempiainen, J.: Analysis of multipath propagation for 5G system at higher frequencies in microcellular environment. In: 13th International Wireless Communications and Mobile Computing Conference (IWCMC), pp. 1660–1664 (2017) 12. Abdullah, N.F., Berraki, D., Ameen, A., Armour, S., Doufexi, A., Nix, A., Beach, M.: Channel Parameters and Throughput Predictions for mmWave and LTE-A Networks in Urban Environments. In: IEEE 81st Vehicular Technology Conference (VTC Spring), pp. 1– 5 (2015) 13. Kumari, M.S., Rao, S.A., Kumar, N.: Characterization of mmWave link for outdoor communications in 5G networks. In: International Conference on Advances in Computing, Communications and Informatics (ICACCI), pp. 44–49 (2015)
What Will Millimeter Wave Communication (mmWave) Be?
129
14. Rappaport, T.S., Sun, S., Mayzus, R., Zhao, H., Azar, Y., Wang, K., Gutierrez, F., Wong, G. N., Schulz, J.K., Samimi, M.: Millimeter wave mobile communications for 5G cellular: it will work! IEEE Access 1, 335–349 (2013) 15. Maccartney, G.R., Rappaport, T.S., Samimi, M.K., Sun, S.: Millimeter-wave omnidirectional path loss data for small cell 5G channel modeling. IEEE Access 3, 1573–1580 (2015) 16. Samimi, M.K., Rappaport, T.S., Maccartney, G.R.: Probabilistic omnidirectional path loss models for millimeter-wave outdoor communications. IEEE Wirel. Commun. Lett. 4(4), 357–360 (2015) 17. Rappaport, T.S., MacCartney, G., Samimi, M., Sun, S.: Wideband millimeter-wave propagation measurements and channel models for future wireless communication system design. IEEE Trans. Commun. 63(9), 3029–3056 (2015) 18. Ko, J., Noh, Y.-S., Kim, Y.-C., Hur, S., Yoon, S., Park, D., Whang, K., Park, D.J., Cho, D.H.: 28 GHz millimeter-wave measurements and models for signal attenuation in vegetated areas. In: 11th European Conference on Antennas and Propagation (EUCAP), pp. 1808– 1812 (2017) 19. Al-Dabbagh, R.K., Al-Raweshidy, H.S., Al-Aboody, N.A.: Performance comparison of exploiting different millimetre-wave bands in 5G cellular networks. In: International Conference on Performance Evaluation and Modeling in Wired and Wireless Networks (PEMWN), pp. 1–6 (2017) 20. MacCartney, G.R., Rappaport, T.S., Ghosh, A.: Base station diversity propagation measurements at 73 GHz millimeter-wave for 5G coordinated multipoint (CoMP) analysis. In: IEEE Globecom Workshops (GC Wkshps), pp. 1–7. IEEE (2017) 21. Nokia Networks White Paper: The 5G mmWave revolution (2016) 22. Yilmaz, T., Akan, O.B.: On the use of low terahertz band for 5G indoor mobile networks. Comput. Electr. Eng. 48, 164–173 (2015) 23. Drozdy, A., Kantor, P., Bito, J.: Effects of rain fading in 5G millimeter wavelength mesh networks. In: 10th European Conference on Antennas and Propagation (EuCAP), pp. 1–5 (2016) 24. Qingling, Z., Li, J.: Rain attenuation in millimeter wave ranges. In: 7th International Symposium on Antennas, Propagation & EM Theory, pp. 1–4 (2006) 25. Rappaport, T.S., Murdock, J.N., Gutierrez, F.: State of the art in 60-GHz integrated circuits and systems for wireless communications. Proc. IEEE 99(8), 1390–1436 (2011) 26. Nokia Networks White Paper: Ten key rules of 5G deployment (2015) 27. Rappaport, T.S., MacCartney, G.R., Sun, S., Yan, H., Deng, S.: Small-scale, local area, and transitional millimeter wave propagation for 5G communications. IEEE Trans. Antennas Propag. 65(12), 6474–6490 (2017) 28. Bai, T., Heath, R.W.: Analysis of self-body blocking effects in millimeter wave cellular networks. In: 48th Asilomar Conference on Signals, Systems and Computers, pp. 1921– 1925 (2014) 29. Wu, T., Rappaport, T.S., Collins, C.M.: The human body and millimeter-wave wireless communication systems: interactions and implications. In: IEEE International Conference on Communications (ICC), pp. 2423–2429. IEEE (2015) 30. Khan, F., Pi, Z., Rajagopal, S.: Millimeter-wave mobile broadband with large scale spatial processing for 5G mobile communication. In: 50th Annual Allerton Conference on Communication, Control, and Computing (Allerton), pp. 1517–1523 (2012) 31. Morimatsu, A., Kajiwara, A.: The attenuation of radio waves caused by body frame and glass of sedan vehicle in the communications of microwave/millimeter-wave band to be applied to 5G. In: IEEE Sensors Applications Symposium (SAS), pp. 1–4 (2018) 32. Anderson, C.R., Rappaport, T.S.: In-building wideband partition loss measurements at 2.5 and 60 GHz. IEEE Trans. Wirel. Commun. 3(3), 922–928 (2004)
130
F. Z. Hassani-Alaoui and J. El Abbadi
33. MacCartney, G.R., Rappaport, T.S.: A flexible millimeter-wave channel sounder with absolute timing. IEEE J. Sel. Areas Commun. 35(6), 1402–1418 (2017) 34. IEEE Webinar (2017). https://5g.ieee.org/images/files/pdf/5GmmWave_Webinar_IEEE_ Nokia_09_20_2017_final.pdf 35. Deng, S., MacCartney, G.R., Rappaport, T.S.: Indoor and outdoor 5G diffraction measurements and models at 10, 20, and 26 GHz. In: IEEE Global Communications Conference (GLOBECOM), pp. 1–7 (2016) 36. Araniti, G., Condoluci, M., Scopelliti, P., Molinaro, A., Iera, A.: Multicasting over emerging 5G networks: challenges and perspectives. IEEE Netw. 31(2), 80–89 (2017) 37. Yazar, A., Peköz, B., Arslan, H.: Flexible multi-numerology systems for 5G new radio (2018). https://pdfs.semanticscholar.org/21c1/20abb47e4a9c995f779b2e5eafceeb48ae77.pdf 38. Arslan, H., Ankaralı, Z.E., Peköz, B., Arslan, H.: Flexible radio access beyond 5G: A Future Projection on Waveform, Numerology & Frame Design Principles (2017) 39. TSGR: TS 138 211 - V15.2.0 – 5G; NR; Physical channels and modulation (3GPP TS 38.211 version 15.2.0 Release 15) (2018). https://portal.etsi.org/TB/ETSIDeliverableStatus. aspx 40. Zhang, L., Ijaz, A., Xiao, P., Quddus, A., Tafazolli, R.: Subband filtered multi-carrier systems for multi-service wireless communications. IEEE Trans. Wirel. Commun. 16(3), 1893–1907 (2017) 41. Zaidi, A.A., Baldemair, R., Tullberg, H., Bjorkegren, H., Sundstrom, L., Medbo, J., Da Silva, I.: Waveform and numerology to support 5g services and requirements. IEEE Commun. Mag. 54(11), 90–98 (2016) 42. Iwabuchi, M., Benjebbour, A., Kishiyama, Y., Wu, D., Tian, T., Gu, L., Cui, Y., Kashima, T.: 5G field experimental trial on frequency domain multiplexing of mixed numerology. In: IEEE 85th Vehicular Technology Conference (VTC Spring), pp. 1–5. IEEE (2017) 43. Guan, P., Wu, D., Tian, T., Zhou, J., Zhang, X., Gu, L., Benjebbour, A., Iwabuchi, M., Kishiyama, Y.: 5G field trials: OFDM-based waveforms and mixed numerologies. IEEE J. Sel. Areas Commun. 35(6), 1234–1243 (2017) 44. 5G Flexible Numerology – What is it?” Why … | Keysight Community (n.d.). https:// community.keysight.com/community/keysight-blogs/next-generation-wireless-communications/blog/2018/04/11/5g-flexible-numerology-what-is-it-why-should-you-care. Accessed 21 Sept 2018 45. Lin, X., Li, J., Baldemair, R., Cheng, T., Parkvall, S., Larsson, D., Koorapaty, K., Frenne, M., Falahati, S., Grövlen, A., Werner, K.: 5G new radio: unveiling the essentials of the next generation wireless access technology (2018) 46. Lien, S.-Y., Shieh, S.-L., Huang, Y., Su, B., Hsu, Y.-L., Wei, H.-Y.: 5G new radio: waveform, frame structure, multiple access, and initial access. IEEE Commun. Mag. 55(6), 64–71 (2017) 47. Zhang, C.J., Ma, J., Li, G.Y., Yu, W., Jindal, N., Kishiyama, Y., Parkvall, S.: New waveforms for 5G networks. IEEE Commun. Mag. 54(11), 64–65 (2016) 48. Ghosh, A., Thomas, T.A., Cudak, M.C., Ratasuk, R., Moorut, P., Vook, F.W., Nie, S.: Millimeter-wave enhanced local area systems: a high-data-rate approach for future wireless networks. IEEE J. Sel. Areas Commun. 32(6), 1152–1163 (2014) 49. Wang, W., He, S., Wu, Y., Wang, H., Huang, Y., Yang, L.: Performance evaluation and analysis of millimeter wave communication system. IEEE Syst. J. 1–12 (2018) 50. Sohrabi, F., Yu, W.: Hybrid analog and digital beamforming for mmWave OFDM largescale antenna arrays. IEEE J. Sel. Areas Commun. 35(7), 1432–1443 (2017) 51. Zhang, Z., Yu, H.: Beam interference suppression in multi-cell millimeter wave communications. Digital Communications and Networks (2018)
What Will Millimeter Wave Communication (mmWave) Be?
131
52. Berraki, D.E., Armour, S.M.D., Nix, A.R.: Benefits of the sparsity of mmWave outdoor spatial channels for beamforming and interference cancellation. Phys. Commun. 27, 170– 180 (2018) 53. Jo, O., Chang, S., Kweon, C., Oh, J., Cheun, K.: 60 GHz wireless communication for future Wi-Fi. ICT Express 1(1), 30–33 (2015) 54. Zhou, P., Fang, X., Long, Y.: Throughput and robustness guaranteed beam tracking for mmWave wireless networks. In: IEEE/CIC International Conference on Communications in China (ICCC), pp. 1–6 (2017) 55. Making 5G NR mmWave a commercial reality in your smartphone and beyond (2018). https://www.qualcomm.com/media/documents/files/making-5g-mmwave-a-commercial-reality-in-your-smartphone.pdf 56. Huo, Y., Dong, X., Xu, W.: 5G cellular user equipment: from theory to practical hardware design. IEEE Access 5, 13992–14010 (2017) 57. Vasilev, I., Plicanic, V., Lau, B.K.: Impact of antenna design on MIMO performance for compact terminals with adaptive impedance matching. IEEE Trans. Antennas Propag. 64(4), 1454–1465 (2016) 58. Informationstechnische Gesellschaft: 22nd International ITG Workshop on Smart Antennas, WSA 2018, Bochum, Germany (n.d.). https://ieeexplore.ieee.org/document/8385502/ 59. Dahlman, E., Parkvall, S.: NR - the new 5G radio-access technology. In: IEEE 87th Vehicular Technology Conference (VTC Spring), pp. 1–6. IEEE (2018) 60. Li, Y., Andrews, J.G., Baccelli, F., Novlan, T.D., Zhang, J.: On the initial access design in millimeter wave cellular networks. In: IEEE Globecom Workshops (GC Wkshps), pp. 1–6 (2016) 61. Giordani, M., Mezzavilla, M., Barati, C.N., Rangan, S., Zorzi, M.: Comparative analysis of initial access techniques in 5G mmWave cellular networks. In: Annual Conference on Information Science and Systems (CISS), pp. 268–273 (2016) 62. Liu, J., Au, K., Maaref, A., Luo, J., Baligh, H., Tong, H., Lorca, J.: Initial access, mobility, and user-centric multi-beam operation in 5G new radio. IEEE Commun. Mag. 56(3), 35–41 (2018) 63. Hyder, M., Mahata, K.: Zadoff–Chu sequence design for random access initial uplink synchronization in LTE-like systems. IEEE Trans. Wirel. Commun. 16(1), 503–511 (2017) 64. Tao, J., Yang, L.: Improved Zadoff-Chu sequence detection in the presence of unknown multipath and carrier frequency offset. IEEE Commun. Lett. 22(5), 922–925 (2018)
Application of the Battery Management System in a Multi-rotor Unmanned Aerial Vehicle Atae Semmar(&), Fouad Moutaouakkil, and Hicham Medromi (EAS-LRI) Systems Architecture Team, Hassan II University, ENSEM Casablanca, Casablanca, Morocco [email protected]
Abstract. The state of charge of a battery depends mainly on its capacity and the discharge current. This is an important parameter for the battery that protects it from deep discharges, overcharging and estimates its remaining energy. Thus, the more precise the estimation of the state of charge, the more energy that the battery can provide is better exploited. This can be part of a possible energy optimization of the UAV and the control of its performance. This article presents generalities about the battery and its management system mainly the mathematical methods for SOC estimation, as well as an application on Rotary-wing UAVs with a model of energy consumption. Keywords: Batteries State of charge Multi-rotor unmanned aerial vehicles Energy consumption Bayesian method
1 Introduction Unmanned aerial vehicle (UAV) is an aircraft without passengers or pilots that can fly autonomously or be remotely controlled from the ground. Rotary-wing UAVs are employed in military, civilian and scientific missions. Among its missions; there is surveillance, reconnaissance, mapping, cartography, border patrol, inspection, homeland security, search and rescue, fire detection, agricultural imaging and traffic monitoring. They can hover in place and take off and land vertically [1]. Drones can use several energy sources; Table 1 shows its sources of energy. Previous work has opted for a diversified electrical architecture that may contain more than one source of power to supply the UAV. Table 2 shows his work. According to Table 2, the battery is an essential source of energy for drones. In fact, the battery is one of the most attractive energy storage systems because of its high efficiency and low pollution [8]. There are several kinds of batteries currently being used in industry: lead-acid battery, Ni-MH battery, Ni-Cd battery, and Li-ion battery. The battery has the advantages of high working cell voltage, low pollution, low selfdischarge rate, and high-power density. UAVs are ideally suited for long endurance applications and the flight endurance is in direct relationship with the total weight of the craft. Therefore, Lithium Polymer (LiPo) batteries in electric UAVs are usually used as power source on account of their high density energy [9] and their low weight 100–265 Wh/Kg. © Springer Nature Switzerland AG 2019 Y. Farhaoui and L. Moussaid (Eds.): ICBDSDE 2018, SBD 53, pp. 132–143, 2019. https://doi.org/10.1007/978-3-030-12048-1_15
Application of the Battery Management System
133
Table 1. Energy sources of a drone Energy sources Batteries
Fuel cell
Supercapacitor
PV panels
Advantages
Drawbacks
– High energy density: 50–200 Wh/kg – Strong intensity of discharge – Low self-discharge and memory effect – Low internal resistance – High Energy Density: 150–1500 Wh/Kg – Non-polluting – Hydrogen is an inexhaustible source – Low maintenance
– Low density of power: 150 W/Kg – Risk explosion in short circuit or overload
– High Power Density: 1000–10 000 W/kg – Unparalleled durability: over a million cycles – Low internal resistance – Renewable energy, inexhaustible, non-polluting – Long life expectancy – Materials resistant to extreme weather conditions
– Highly flammable and explosive – Low density of power: 120 W/Kg – Limited lifetime: a few thousand hours – Slow response time – Low density energy: 4–6 Wh/Kg – Heavy
– Intermittent energy – Low efficiency
Table 2. Electrical architectures of a drone. Authors
Electrical architecture
Observations
Trenev and Mladenov [2] Albaker [3]
– Li-ion battery – Supercapacitor – PV panels – MPPT Module – Battery module – PV panels – Fuel cell – Rechargeable battery – Rechargeable battery – Li-po battery – Li-po battery
– Supercapacitors allow to restart the engine in flight in case of accidental stop – Depends on the type of MPPT algorithm
Hao and Khaligh [4]
Chéron et al. [5] Sanahuja [6] Sierra, Orchard, Goebel and Kulkarni [7]
– 500 W average power with a peak of up to 1500 and 800 W of take-off power – The size of the battery should be very large which will add extra weight – Flight time around 20 min – Flight time approximately 21.23 min
134
A. Semmar et al.
SOC estimation is a fundamental challenge for battery use. The SOC of a battery, which is used to describe its remaining capacity, is a very important parameter for a control strategy [10]. SOC cannot be measured directly and its values cannot be completely certain, therefore uncertainties regarding the estimation of the remaining charge in the battery are inevitable. Consequently, most flight plans are very conservative in nature and the flight time is shorter than what can provide the battery, hence the importance of the precision of SOC estimation. This paper is organized as follows. Section 2 describes the battery behavior and modeling. Section 3 presents a battery management system and the mathematical methods used in SOC estimation. Section 4 describes an application using a multi-rotor UAV 3DR IRIS. Section 5 ends with conclusions.
2 Battery Behavior and Models 2.1
Battery Behavior
Batteries are energy storage devices that facilitate the conversion, or transduction, of chemical energy into electrical energy, and vice versa [11]. When the battery delivers current, the voltage at its terminals is not equal to the no-load voltage. Static and dynamic phenomena intervene and are at the origin of these differences of tensions. These dynamic behaviors of the battery are closely related to the electrochemical phenomena inside it. Ohmic Effect. This refers to the internal resistance of the electrolyte related to the conductivity of the latter. This ohmic effect is modeled by a resistance “R1”. Load Transfer Effect. It appears at the electrode/electrolyte interfaces when the electrodes release and gain electrons. This effect is modeled by a resistance “R2”. Double Layer Effect. Electrons accumulate on one of the two electrode/electrolyte surfaces and holes on the other. Due to the ionic conduction of the electrolyte and the electric field effect created, the layer immediately adjacent to each electrode acquires an opposite charge. Then a momentary current appears and vanishes quickly. This polarizes the electrode. This behavior is modeled by a capacitor C. Diffusion Effect. During the passage of a current, the displacement of the species causes concentration heterogeneity between the electrolyte and the electrolyte/electrode surfaces. Under the effect of the chemical gradient, the species move in the opposite direction to this one to restore the homogeneity of concentration. It is modeled by a series of parallel RC circuits put in series. Other temporal effects also intervene in the behavior of the battery such as the magnetic and electrical effect; it is a very fast effect, and long-term aging. 2.2
Battery Models
There are three different types of modeling approaches to battery modeling, there are three main types of models to represent its behavior: electrochemical models, equivalent electric circuit models, and black box models (e.g. models using neural networks or fuzzy logic).
Application of the Battery Management System
135
Electrochemical Models. It describes the electrochemical phenomena of each component of the cell using the corresponding electrochemical laws. It is a complex model because of the detailed presentation of chemical mechanisms related to the generation of energy. It accurately estimates the behavior of the battery but it is difficult to simulate. The electrochemical model amounts to solving a complex system of 14 equations, most of which are partial differential equations, some parameters are difficult to know such as the geometry of the electrodes, the concentration of the electrolyte, the charge transfer coefficient and the diffusion coefficient. In addition, it is not suitable for developing control laws and for real-time applications. Black Box Models. This model uses learning models to estimate the behavior of the battery according to their experiences obtained during the learning phase with real data of the battery (current, voltage, temperature, SOC). It is used mainly in the estimation of SOC and SOH parameters. Equivalent Electric Circuit Models. It is an analogy between electrochemistry and electricity, each electrical element is correlated with an electrochemical phenomenon. The constituent elements of the different models vary with the state of charge, the temperature, and the current of the battery. Ideal Model. The battery is represented by a DC voltage source which represents the open circuit voltage (OCV). But it is a model far removed from the physical behavior of a real battery and the energy capacity is supposed to be infinite; the battery can provide energy for an infinite duration. Simple Model. This model serially puts an OCV voltage source and resistor into series. The dynamic phenomena of the battery are not taken into account. Thevenin Model. It is the first dynamic model, it takes into account the effects of charge transfer and double layer but the effects of transfer of matter are neglected. Randles Model. This model takes into account all dynamic phenomena. This is the closest to the actual behavior of the battery.
3 Battery Management System (BMS) and the Mathematical Methods Used in SOC Estimating The intensive use of Li-Po batteries in the electric vehicle industry has popularized the concept of Battery Management Systems (BMS). These systems are mainly aimed at using in a better way the energy stored in the batteries and provide real time diagnosis information for the benefit of craft operator. To accomplish these tasks, BMS must use information about the battery’s SOC and its Remaining Useful Life (RUL) [12]. The knowledge of these state variables is not only necessary to verify if the mission goal(s) can be accomplished but is also important to aid in online decision-making activities such as fault mitigation and mission re-planning [7].
136
A. Semmar et al.
The state of charge of a battery (SOC), is defined as the ratio between the residual capacity of the battery and its nominal capacity, and is expressed in percentage (%). It is 100% when the battery is fully charged and 0% when it is fully discharged. A SOC = 0% does not mean that the voltage across the battery is zero. SOCbatt ð%Þ ¼
Cr 100 Cn
ð1Þ
Where Cr, and Cn are respectively the residual and nominal capacities of the battery, expressed in Ah. The mathematical methods of estimating the state of charge could be divided into four categories; Direct measurement, Book keeping estimation, Adaptive systems and Hybrid methods. 3.1
Direct Measurement
It uses the physical properties of the battery. OCV Method. The OCV method based on the open circuit voltage of the batteries is proportional to the SOC when they are disconnected from the loads for duration greater than 2 h (used for lead-acid battery). There is approximately a linear relationship between the SOC of the lead-acid battery and its open circuit voltage (OCV) given by. VOC ðtÞ ¼ a1 SOCðtÞ þ a0
ð2Þ
Where SOC(t) is the SOC of the battery at t, a0 is the battery terminal voltage when SOC = 0%, and a1 is obtained from knowing the value of a0 and VOC(t) at SOC = 100%. Terminal Voltage Method. The terminal voltage method is based on the voltage drop of the terminals due to the internal impedances as the battery discharges, so that the electromotive force of the battery is proportional to the terminal voltage. Since the electromotive force of the battery is approximately linear to the SOC, the voltage across the battery is also approximately linear proportional to the SOC. Impedance Method. The impedance measurements allow knowing several parameters which size can depend on the SOC of the battery. Impedance Spectroscopy Method. The impedance spectroscopy method measures the impedances of the battery over a wide range of frequencies that are alternative to different charging and discharging currents. Model impedance values are found by the least squares corresponding to measured impedance values. The SOC can be indirectly inferred by measuring the current impedances of the battery and correlating them with the known impedances at different SOC levels [13, 14].
Application of the Battery Management System
3.2
137
Book-Keeping Estimation
It uses the data of the discharge current of the battery as input. This method allows including some internal battery effects such as self-discharge, loss of capacity, and discharging efficiency. Coulomb Counting Method. This method measures the discharge current of a battery and integrates the discharge current over time to estimate the SOC [15]. SOCðtÞ ¼ SOCðt 1Þ þ
IðtÞ Dt Qn
ð3Þ
Where SOC(t – 1) is the previous estimated SOC values, I(t) is the discharge current and Qn is the nominal capacity. Modified Coulomb Counting Method. This method uses the corrected current to improve the accuracy of the estimation. The corrected current is the function of the discharge current. There is a quadratic relationship between the corrected current and the discharge current of the battery. Ic ðtÞ ¼ k2 IðtÞ2 þ k1 IðtÞ þ k0
ð4Þ
Where k2, k1 and k0 are constant values obtained from the practice experimental data. In modified Coulomb counting method, SOC is calculated by the following equation: SOCðtÞ ¼ SOCðt 1Þ þ
Ic ðtÞ Dt Qn
ð5Þ
The experimental results show that the accuracy of the modified Coulomb counting method is superior to the conventional Coulomb counting method. 3.3
Adaptive Systems
They are self-designed systems that can be adjusted automatically in changing systems. Since batteries are affected by many chemical factors and have a non-linear SOC, adaptive systems offer a good solution for SOC estimation [16]. Back Propagation Neural Network. The prediction of the current SOC is done using the recent history of the voltage, current and ambient temperature of a battery [17]. Fuzzy Logic Method. This method provides a powerful way to model non-linear and complex systems. Radial Basis Function Neural Network. This method uses a useful estimation methodology for systems with incomplete information.
138
A. Semmar et al.
Fuzzy Neural Network (FNN). This method is used in many applications, including for the identification of unknown systems. In the nonlinear identification of the system, the FNN can efficiently adapt the nonlinear system by calculating the optimized coefficients of the learning mechanism [18]. Kalman Filter. The application of the Kalman filter method is shown to provide verifiable SOC estimates for the battery via real-time state estimation [19]. This filter is part of the Bayesian methods. Particle Filter. This method is an alternative to Kalman Filter and with enough samples; Particle Filter is more accurate than FK [7]. The objective is to estimate the posterior density of the SOC considering the observation variables. This filter is part of the Bayesian methods. 3.4
Hybrid Method
These methods benefit from the advantages of each SOC estimation method and allow for an overall optimized estimate of performance. Coulomb Counting and Electromotive Force Combination. In order to calculate the SOC and the remaining run-time (RRT) and to improve the SOC estimation system to cope with the effect of aging, a simple Qmax matching algorithm is introduced. In this algorithm, the stable conditions of the state of charge are exploited in order to adapt Qmax to the aging effect. Coulomb Counting and Kalman Filter Combination. Kalman Filter is used to converge the approximate initial value to its real value. Then the Coulomb Counting method is applied to estimate the SOC for the long working time [20]. Per-unit System and Extended Kalman Filter Combination. This method identifies the appropriate battery model parameters for the high precision SOC estimation of a degraded lithium-ion battery.
4 Application: Modeling SOC Estimation and UAV Consumption This part will summarize the work of Sierra, Orchard, Goebel, and Kulkarni [7] on the estimation of the state of charge of the battery and on the approximation of the energy consumption for a multi-rotor UAV. 4.1
State of Charge Modeling
The model of the state of charge estimation is based on model-based prognostics approaches that rely on the physics-based models that describe the behavior of systems and their components by state parameter estimation. The dynamics of the battery and the real-time availability of voltage and current measurements are assumed to be a discrete characterization by the model.
Application of the Battery Management System
139
Rint ðk þ 1Þ ¼ Rint ðkÞ þ w1 ðkÞ
ð6Þ
SOC ðk þ 1Þ ¼ SOC ðkÞ Pðk Þ:Dt:Ecrit ðk Þ1 þ w2 ðkÞ
ð7Þ
Ecrit ðk þ 1Þ ¼ Ecrit ðkÞ þ w3 ðk Þ
ð8Þ
VðkÞ ¼ Voc ðkÞ iðkÞ:Rint ðk Þ þ g ðkÞ
ð9Þ
Where P(k) is the power, V(k) is the battery voltage, Voc is the open circuit voltage, i(k) is the discharge current, Rint(k) is the internal resistance, Ecrit(k) is the expected total energy delivered by the battery, Process (w1, w2 and w3) and measurement η(k) noises are assumed Gaussian. The measurements of Voc ðkÞ and of i(k) are detailed in article [7]. The energy delivered by the battery is affected by the temperature and current discharge rate. The internal resistance of the battery varies according to the state of charge of the battery and is also affected by the temperature and current discharge rate. In order to estimate the value of its last two and in order to incorporate the current load dependence, temperature dependence and SOC dependence, the concept of artificial evolution is chosen. Bayesian methods provide good results in estimation and prognosis problems. These methods can adjust the SOC in real time for different load conditions [20] and they are able to assess the estimate’s confidence by a Probability Density Function (PDF) [21]. The effect of environmental factors such as temperature, battery degradation and aging are taken into account by using the concept of artificial evolution in conjunction with Bayesian methods. This association generates new parameter values at each time step by adding additional random disturbances to the sampled state vectors. Particulate filters that are part of Bayesian methods are used in estimating the state of charge of the battery since they are more accurate than Kalman filters. 4.2
Power Consumption Modeling
The tests were carried out with a Multi-rotor 3DR IRIS. The engines of the UAV are powered by Li-po 3S 5100 mAh batteries. The model of consumption is determined by an aerodynamic model based on momentum theory which includes the following maneuvers: climb, hover, horizontal flight and descent. This model takes into account the weight of the drone, disc actuator area, air density and therefore indirectly the temperature, but the humidity, the wind speed and the direction of the wind are not taken into account. The characteristics of the chosen drone are summarized in Table 3.
140
A. Semmar et al. Table 3. Multi-rotor parameters Parameter Number of rotors Propeller diameter Total disc actuator area Empty mass Maximum payload mass Empty weight Maximum payload weight Air density
Symbol n Dp At m0 mpmax W0 Wpmax q
Value 4 0.2413 0.1829 1.357 0.3 13.2986 2.94 1.15
Units m m2 kg kg N N kg/m3
The following assumptions have been taken into consideration; the total weight is equally distributed on n-rotors and At is the sum of n-disc actuator areas. W Vc Pc ¼ þ gc ðVc Þ 2 Ph ¼
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi! Vc2 W þ 2:qAt 4
W 3=2 pffiffiffiffiffiffiffiffiffiffiffi gh : 2:qAt
W ðVhor sinðav ðVhor ÞÞ þ thor Þ ghor ðVd Þ 0 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1 Vd2 W @Vd W A þ þ Pd ¼ gd ðVd Þ 2:qAt 2 4
Phor ¼
ð10Þ
ð11Þ ð12Þ
ð13Þ
Where Pc, Ph, Phor, Pd are the power required by the multi-rotor in climb, hovering, horizontal flight and descent respectively, q is air density, W is the total weight of the aircraft, At is the total disc actuator area, Vc is the vertical climb speed, Vd is the vertical descent speed, av is the rotor tilt, ηc is the efficiency factor in climb, ηd the efficiency factor in descent and thor is the induced velocity. The measurement of efficiency factors and the induced velocity are detailed in article [7]. The model of energy consumption of the drone is validated with different speeds taking as payload mp = 0 kg, mp = 0.1 kg and mp = 0.2 kg as shown in Figs. 1, 2 and 3.
Application of the Battery Management System
Fig. 1. Power required in climb
Fig. 2. Power required in descent
Fig. 3. Power required in horizontal flight
141
142
A. Semmar et al.
The three Figs. 1, 2 and 3 show that drone consumption is a non-linear function with respect to speed. They show that the higher the payload weight, the higher the consumption. Also, they show that the climb is the maneuver that consumes the most. The average maximum speed of a small size multi-rotor is about 15 m/s, and several applications just need speeds up 5 m/s [7].
5 Conclusion This work presents an overview of batteries and their system management. Also, it presents an application on a rotatory wing which has constraints associated with weight, size and cost. The battery has four main effects that describe its dynamic behavior and its electrochemical phenomena, namely; Ohmic effect, Load transfer effect, Double layer effect and Diffusion effect. In addition, the battery has three main models; namely electrochemical models, black box models and electrical models. Battery system management and SOC estimation are essential to know the state of health of the battery and then predict the battery’s Remaining Useful Life. A model has been presented to model SOC and another to model the drone’s consumption. Acknowledgments. The author would like to express his gratitude to his supervisors for supporting him with the work presented in this document.
References 1. Valavanis, K.P., Vachtsevanos, G.J. (eds.): Handbook of Unmanned Aerial Vehicles, 1st edn. Springer, Dordrecht (2015) 2. Trenev, V., Mladenov, M., Kanev, K., Petrov, E., Chavdarov, I.: UAV energy efficiency improvement by battery – supercapacitor (2010) 3. Albaker, B.M.: Preliminary architectonic design for a smart solar-powered UAV. In: IEEE Conference on Clean Energy and Technology (CEAT) (2013) 4. Hao, C., Khaligh, A.: Hybrid energy storage system for unmanned aerial vehicle (UAV). In: 36th Annual Conference on IEEE Industrial Electronics Society, IECON 2010 (2010) 5. Chéron, C., Dennis, A., Semerjyan, V., Chen, Y.: A multifunctional HIL testbed for multirotor VTOL UAV actuator. In: IEEE/ASME International Conference on Mechatronics and Embedded Systems and Applications (MESA) (2010) 6. Sanahuja, G.: Drone quadrirotor suivant une ligne par vision 7. Sierra, G., Orchard, M., Goebel, K., Kulkarni, C.: Battery health management for small-size rotary-wing electric unmanned aerial vehicles. An efficient approach for constrained computing platforms. Reliab. Eng. Syst. Saf. (2018). https://doi.org/10.1016/j.ress.2018.04. 030 8. Chang, W.Y.: State of charge estimation for LiFePO4 battery using artificial neural network. Int. Rev. Electr. Eng. IREE 7(5), 5800–5874 (2012) 9. Meyer, J., du Plessis, F., Clarke, W.: Design considerations for long endurance unmanned aerial vehicles. In: Aerial Vehicles. InTech, Chapters Published, University Johannesburg, South Africa, pp. 443–497 (2009) 10. He, H.W., Xiong, R., Guo, H.Q.: Online estimation of model parameters and state-of-charge of LiFePO4 batteries in electric vehicles. Appl. Energy 89(1), 413–420 (2012)
Application of the Battery Management System
143
11. Huggins, R.: Advanced Batteries: Materials Science Aspects, 1st edn. Springer (2008) 12. Pola, D., Navarrete, H., Orchard, M., Rabie, R., Cerda, M., Olivares, B., Silva, J., Espinoza, P., Perez, A.: Particle-filtering-based discharge time prognosis for lithium-ion batteries with a statistical characterization of use profiles. IEEE Trans. Reliab. 64(2), 710–720 (2015). https://doi.org/10.1109/TR.2014.2385069 13. Li, R., Wu, J.F., Wang, H.Y., Li, G.C.: Prediction of state of charge of lithium-ion rechargeable battery with electrochemical impedance spectroscopy theory. In: Proceedings of the 5th IEEE Conference on Industrial Electronics and Applications (ICIEA 2010), Taichung, pp. 684–688, June 2010 14. Bundy, K., Karlsson, M., Lindbergh, G., Lundqvist, A.: An electrochemical impedance spectroscopy method for prediction of the state of charge of a nickel-metal hydride battery at open circuit and during discharge. J. Power Sources 72(2), 118–125 (1998) 15. Ng, K.S., Moo, C.S., Chen, Y.P., Hsieh, Y.C.: Enhanced Coulomb counting method for estimating state-of-charge and state-of-health of lithium-ion batteries. Appl. Energy 86(9), 1506–1511 (2009) 16. Watrin, N., Blunier, B., Miraoui, A.: Review of adaptive systems for lithium batteries stateof-charge and state-of-health estimation. In: Proceedings of IEEE Transportation Electrification Conference and Expo, Dearborn, pp. 1–6, June 2012 17. Linda, O., William, E.J., Huff, M., et al.: Intelligent neural network implementation for SOCI development of Li/CFx batteries. In: Proceedings of the 2nd International Symposium on Resilient Control Systems (ISRCS 2009), Idaho Falls, pp. 57–62, August 2009 18. Li, I.H., Wang, W.Y., Su, S.F., Lee, Y.S.: A merged fuzzy neural network and its applications in battery state-of-charge estimation. IEEE Trans. Energy Convers. 22(3), 697– 708 (2007) 19. Xu, L., Wang, J.P., Chen, Q.S.: Kalman filtering state of charge estimation for battery management system based on a stochastic fuzzy neural network battery model. Energy Convers. Manag. 53(1), 33–39 (2012) 20. Chang, W.-Y.: The state of charge estimating methods for battery: a review. ISRN Appl. Math. (2013). https://doi.org/10.1155/2013/953792 21. Goebel, K., Saha, B., Saxena, A., Celaya, J., Christophersen, J.: Prognostics in battery health management. Instrum. Meas. Mag. IEEE 11(4), 33–40 (2008). https://doi.org/10.1109/MIM. 2008.4579269
Brain Ischemic Stroke Segmentation from Brain MRI Between Clustering Methods and Region Based Methods Fathia Aboudi1,2(&), Cyrine Drissi2,3, and Tarek Kraiem1,3 1 High Institute of Medical Technology in Tunisia (ISTMT), LRBTM Research Laboratory of Biophysics and Medical Technology, UTM, 9 Avenue Zouheir Essafi, 1006 Tunis, Tunisia [email protected], [email protected] 2 Department of Neurology, Mongi Ben Hmida National Institute of Neurology, Tunis El Manar University, Tunis, Tunisia [email protected] 3 Faculty of Medicine of Tunis, LRBTM Research Laboratory of Biophysics and Medical Technology, UTM, 15 Djebel Lakhdher Street La Rabta, 1007 Tunis, Tunisia
Abstract. Public health is one of the most concerns at the worldwide. Brain ischemic stroke is an acute cerebral vascular disease, which is likely to cause long-term disabilities and death. It occurs in most stroke patients. Brain Magnetic Resonance Imaging (MRI) is one of the essential non-invasive modalities that provide a contrast imaging to visualize and detections lesions. Brain ischemic stroke segmentation in MRI has attracted the attention of medical doctors and researches since variations in structural and contrast of medical data. Several proposals have been designed throughout the years comprising a different strategy of brain segmentation. In particular, in this paper we analyse a segmentation methods used for detection and localization brain ischemic stroke. That the goal has been presented to differentiate between the lesions with the normal region. The Spatial Fuzzy C Means (SFCM) and methods based regions are developed in order to obtain a robust, rapid, efficient, precious and precocious detection of acute stroke lesion from images data issues by MRI with diffusionweighted imaging (DWI) and perfusion-weighted imaging (PWI). The validation purpose was performed by comparing resulting segmentation to the manual contours traced by an expert. Results show that the SFCM appeared efficient in detection of acute with a accuracy value of 99.1% in PWI-MTT and of 47.44% in DWI and an timing average in order to one second. However, the accuracy rate of regions growing in order to 17.40% in DWI and to 71.30% in PWI. Keywords: Brain MRI Tissue segmentation Regions growing Brain stroke ischemic DWI
Spatial PWI
Fuzzy C means
© Springer Nature Switzerland AG 2019 Y. Farhaoui and L. Moussaid (Eds.): ICBDSDE 2018, SBD 53, pp. 144–154, 2019. https://doi.org/10.1007/978-3-030-12048-1_16
Brain Ischemic Stroke Segmentation from Brain MRI
145
1 Introduction Stroke brain disease is a sudden neurological deficit of vascular origin caused by an infarction or hemorrhage in the brain. Every year, 12 million people are affected by stroke brain in worldwide and the value is increasing [1]. It is a major cause of death and disability in both the more developed and the less developed world. It is placed into two categories an ischemia (cerebral infarction 80%) and a hemorrhage (20%) [2]. Ischemia is a physiological term indicating insufficient blood flow for normal cellular function. In ischemic stroke an obstruction of the cerebral blood supply causes tissue hypoxia, and advancing tissue death over the next hours, which deprives brain cells of oxygen and nutrients. Also the blood flow disturbances often originate from an embolism physically obstructing a cerebral artery or arteriole causing a critical reduction of blood flow to specific parts of the brain [3]. With a statistic values delivered by Pascale REISSER and confirmed by Dr Eric Revue, globally, every years, 12 million people are affected by stroke brain in worldwide, that means one 1 person every 5 s, in Europe 1 400 000. In France, 155 000 new people so one every four minutes and 62 000 will die [4]. In Tunisia, it is the first cause of physical disability in adults and the third cause of death. The average incidence around of 10,000 new people per year but the numbers are increasing, as it was declared by the Huffpost Tunisia [5]. The WHO is talking about a pandemic and is projecting an increase in the incidence of stroke: 23 million in 2030 (for memory 16 million in 2005) [4]. Brain Magnetic Resonance Imaging (MRI) has a high sensitivity for the acute ischemic lesions. It is often used to assess the presence of stroke lesion, her location, extent, age and other factors as this modality is highly sensitive for many of the critical tissue changes observed in acute stroke. Diffusion weighted (DW) and perfusion weighted (PW) MR imaging have been shown to detect acute stroke in early states. DWI measures self-diffusion of water molecules in a tissue. During ischemia, it is believed that cell breakdown impedes movement of water molecules, resulting in increased DWI signal. Lesions appear bright (high signal) on DWI. PWI allows for imaging of micro vascular hemodynamic. Therefore perfusion is normally measured in terms of flow, volume and mean transit time (MTT). Therefore accurate segmentation of area stroke lesions is important in understanding the evolution of diseases. Ischemic stroke segmentation is a challenging task for various reasons as her highly variable of size and location and clinical quality PWI_MTT images may have low resolution and have imaging artifacts. It has also many methods segmentation have been proposed in the past decade that her aim is separating the hyperintensed region and the infracted region from the brain image. In this work, we correlate a two of segmentation methods: one based clustering and the other based region methods applicate for different DWI maps (b0, ADC) and with PWI maps (MTT) with retrospective manual analysis done by expert in neuroimaging.
146
F. Aboudi et al.
2 Related Work Acute stroke brain segmentation is one of the major issues in biomedical image processing due to detect precociously the disease. Getting an overview of segmentation techniques for manual segmentation of images, we have found that several methods have been developed using clustering and region based methods. Ischemic stroke lesions in MR imaging are difficult to segment for various reasons, which include especially shape complexity and ambiguity. In addition, accurate segmentation of stroke lesions requires anatomical knowledge. In particular changes in MR images due to ischemic stroke follow the vascular territory of the occluded blood vessel, which is characteristic of cerebrovascular disease and helps in differentiating it from other disease entities. As a matter of fact, in the literature devoted to the segmentation of brain lesions, there has been few works done in the area of multimodal stroke segmentation [6]. Most of the work related concentrated on the detection of tumors [7, 8] and multiple sclerosis lesions [1]. Since 2012 numerous automatic tissue segmentation methods are published and already exist with varing performance. The segmentation performance depends on image acquisition factors, like MR field strength [9], and patient specific factors, such as brain abnormalities [10]. In the brain tumor segmentation (BRATS) challenges held in 2016, the dataset contains a number of subjects with gliomas and the task is to develop automatic algorithms to segment the whole tumor, the tumor core and the Gdenhanced tumor core based on multi-modal MR images. In the latest competition [11], over half of the methods were based on deep neural networks and they achieved top results. For instance, the hyper local features (original input image) are used prior to the final segmentation to improve the accuracy [12]. As a pixel-level segmentation problem, there are much more non-tumor pixels than the ones belong to part of the tumors, which means there is a significant label imbalance. To alleviate the imbalance, Lun and Hsu proposed a re-weighted loss function [13]. Randhawa and al also modified the crossentropy loss function so that the segmentations at tumor edges could be improved [14]. Instead of analyzing multi-modal MRIs in 2D, the Deep Medic approach performs segmentation of tumors in 3D while using extended residual connections [15]. In addition to deep learning algorithms, machine learning approaches based on the random forests [16] also demonstrate good performance using hand-crafted features. The segmentation of sub-acute ischemic stroke lesion is one of the tasks in Ischemic Stroke lesion segmentation (ISLES), which attracted many entries [2]. The challenge is to automatically segment sub-acute ischemic stroke lesions based on multi-modal MR images. Compared with the dataset in the BRATS, the dataset used in the ISLES is smaller. Similar to brain tumors, sub-acute ischemic stroke lesions are difficult to segment [16].
Brain Ischemic Stroke Segmentation from Brain MRI
147
3 Materials and Methods Patient Database: The MRI brain images are acquired from the service of neuroimaging in the national institute of neurology Mongi Ben Hmida in Tunisia. 5 patients (5 M, Age: between 42 and 81 years, body weight 53 and 103 kg) referred to our institute for suspected acute stroke and two patients are a reference. Every patient simulated by 71 sequences that every sequences have her protocol of acquisition. For this research the most significative sequence is DWI and PWI_MTT for reason of sensibility of detection the ischemic stroke brain disease in early stages. All MRI datasets were acquired on a 3 T of Siemens deposed. The MRI images are in DICOM format (Digital Imaging and Communications in Medicine). DWI Imaging Protocol Acquisition: Axial DWI trace images were acquired using a SE_EPI sequence, TE = 109 ms, TR = 5400 ms, FA = 90°, NEX = 2, 192 192 matrix, b = 0 s/m2 and a b = 1000 s/m2, FOV = 240 240 m2 and slice thickness = 3.5–7 mm. PWI Imaging Protocol Acquisition: After injection an intravenous bolus of paramagnetic contrast agent, Gadodiamide, the PWI trace images were acquired using a GE_EPI sequence, TE = 45 ms, TR = 2780 ms, FA = 90°, NEX = 2, 128 128 matrix, FOV = 240 240 m2 and slice thickness = 5–7 mm. Simulation: We used Matlab R2014a image processing toolbox for simulation of segmentation.
4 Approach Ischemic stroke segmentation is separating hyperintensed region from brain image. The approach was trained using images from MRI Brain. DWI images were processed directly following acquisition. However PWI_MTT passed by a specific pré-processing. Trends have shown, it being performed by experts, which is a critical task. Although assessment is performed precisely and with accuracy but this often gets difficult for them and will never be 100% accurate. The approach can be observed in Fig. 1. Figure 1 shows approach which is required to be followed in all kinds of segmentation methods. Medically, ischemic stroke region is of great importance as it provides concerned information like which region of brain and by how much percent it is affected and how much more clinically required features. Extracting useful information from multidimensional images is a vital task in medical images segmentation. Thus, accuracy of clinical information depends upon location region precisely, which will help for treatment by formulating right methodology as soon as it is diagnosed.
148
F. Aboudi et al.
Fig. 1. Brain ischemic stroke segmentation approach
5 Image Processing/Segmentation DW images were processed directly following acquisition. The acquired dynamic perfusion images were processed to create perfusion maps (rCBF, rCBV, MTT). MTT maps were chosen for analysis. The PWI- MTT images were processed as described in the flow chart in Fig. 2. 5.1
Program Implementation
5.1.1 Loading and Preparing Data for Processing Diffusion weighted imaging (DWI) images did not require any pre-processing prior to segmentation. However before performing the segmentation, the MTT data issues of perfusion weighted imaging (PWI) posted by pre-processing. 5.1.2 Pre-processing of the MTT Data In this work, we will firstly know the characteristics of images for segmentation. After, all negative values in the image were set to zero. The non-zero error codes produced in the MTT maps were assigned maximum MTT values on the assumption that poor perfusion (largeMTT, low or no flow) caused the error in those pixels (Fig. 2, steps 2–3). Although the lesion is visible at step 3, the MTT maps were still very noisy and
Brain Ischemic Stroke Segmentation from Brain MRI
149
Fig. 2. Steps of pre-processing of PWI-MTT images.
did not provide a clear lesion boundary. Therefore, the MTT maps were also smoothed and de- speckled slice by slice using in finally a 2Dmedian filter (8 8 kernel) (Fig. 2, step 4). Our objective was to build a tool n that would segment acute stroke brain lesions on DWI and PWI, images of MRI modalities, semi-automatically and with a minimal timing. The steps of pre-processing of the PWI-MTT images for segmentation of acute stroke are summarized as Fig. 2. 5.2
Segmentation Methods
5.2.1 Methods of Clustering: Spatial Fuzzy C-Means (SFCM) The Fuzzy C-Means (FCM) algorithm is a fuzzy clustering method based on the minimization of a quadratic criterion where clusters are represented by their respective centers. In the case of image segmentation, it assigns pixels to each category by using fuzzy membership functions. The FCM algorithm was proposed by Dunn (1973) and later improved by Bezdek [17]. Although FCM works well in many real-world applications, it can’t be applied to deal with spatial data directly, because there are two types of features in spatial clustering problems. During execution, it uses the notion of the probability of appearance. Figure 3 summarizes the principle of implementing the SFCM algorithm used in this work.
150
F. Aboudi et al.
Fig. 3. Principe of implementation of the SFCM algorithm.
5.2.2 Region Based Methods: Region Growing The region growing is a step that groups the pixels or sub-regions into larger regions based on a predefined criteria for growth. A “Region” forms pixels growing with the same intensity level which is used to calculate the area of white mater for the skull [18]. The first step in region growing is to select a set of seed points. Seed point selection is based on some user criterion the initial region begins as the exact location of these seeds. The regions are then grown from these seed points to adjacent points depending on a region membership criterion. Region growing methods can correctly separate the regions that have the same properties we define. And also they can provide the original images which have clear edges with good segmentation results [18]. The concept is simple. We only need a small numbers of seed point to represent the property we want, then grow the region. We can determine the seed point and the criteria we want to make and choose multiple criteria at the same time. It performs well with respect to noise. Also region growing is considered a semi-automated method for segmentation because it needed an operator for choose the initial region. The results of image simulation of all the methods have been presented in Table 1. Everyone have properly advantages and disadvantages.
Brain Ischemic Stroke Segmentation from Brain MRI
151
Table 1. Result of segmentation methods with DWI and PWI-MTT sequence
DWI
Sequence
PWI-MTT
Segmentation Methods Region Growing
Spatial Fuzzy CMeans(SFCM)
6 Evaluation and Discuss To demonstrate the efficiency of the proposed segmentation strategy, we compared our method with comparable methods from the published literature and with manual methods. The performance of the methods segmentation in platform design is further compared with the other classification methods by testing the entire techniques on the validation image dataset. A quantitative measure is used to provide the difference between the true regions Ma (constructed by experts in the National Institute of Neurology Mongi Ben Hmida Tunisia) and the classification results Ca [19]. The segmentation accuracy is estimated as: S T ½AðCaÞ AðMaÞ þ ½AðCaÞ AðMaÞ S Accuracy ¼ 100 1 ½AðCaÞ AðMaÞ
ð1Þ
152
F. Aboudi et al.
Where A (Ca) and A (Ma) are the areas of segmented regions Ma and Ca. The accuracy and the value timing of all the methods have been presented in the histogram (Fig. 4) below.
Fig. 4. Histogram of evaluation of segmentation methods.
The experimental results show good agreement between the results of the proposed segmentation methods and the expertise of neuroradiologists. In addition, experiments reveal that the SFCM approach applied to the PWI-MTT sequence is the most effective method for locating ischemic stroke injury with a precision of segmentation of 99.10% and 47.45% applied on the DWI sequence and an execution time of 0.09 s and 1.22 s respectively. By against the Growing Region accuracy rates on order of 71.30% with PWI-MTT and 17.40 with DWI respectively with an execution time of the order of 5.24 s and 4.68 s.
7 Conclusion MRI provides several advantages and challenges for systems neuroscience. It has a high sensitivity for the acute ischemic lesions. It is often used to assess the presence of stroke lesion. DW and PW sequences imaging have been shown to detect acute stroke in early states. If the detection is failed, so the disease passed a necrosis stage that it irreversible. For this, we presented two methods of segmentation of ischemic stroke brain disease bases on clustering and regions methods. We compare the two different methods in of accuracy criteria and a time of execution. Experiment shows that actually, Spatial Fuzzy C-Means is becoming very significant. This strategy achieves a quantization accuracy of over 99%. It is also the fastest for locating ischemic stroke injury from MRI images with a run time of 0.09 s. So this technique is the most reliable and may be recommended as the appropriate tool for segmentation of perfusion images.
Brain Ischemic Stroke Segmentation from Brain MRI
153
References 1. Guptal, N., Mittal, A.: Brain ischemic stroke segmentation: a survey. J. Multi Discip. Eng. Technol., 1 (2014) 2. Maier, O., Menzeh, B.H., Gablentz, J., Hani, L., Rangarajan, J.R., Reza, S.M.S., et al.: ISLES 2015 - a public evaluation benchmark for ischemic stroke lesion segmentation from multi spectral MRI. Med. Image Anal. 35, 250–269 (2016) 3. Bouts, M.J.R.J.: Prediction of tissue outcome after experimental stroke using MRI-based algorithms. Ph.D. thesis, Utrecht University, pp. 64–65 (2013). ISBN: 978-90-393-5955-6 4. Reisser, P., Revue, E.: Cours IFSI: Institut de formation de soin infirmiers: Journée européenne de l’AVC 2017. https://www.infirmiers.com/etudiants-en-ifsi/cours/coursneurologie-accident-vasculaire-cerebral.html 5. Ben Hmida, C.: la champagne de sensibilisation sur l’accident vasculaire cerebral, Huffpost Tunisie (2016) 6. Weinman, J.: Nonlinear diffusion scale-space and fast marching level sets for segmentation of MR imagery and volume estimation of stroke lesions. In: MICCAI 2003, LNCS 2879, pp. 496–504 (2003) 7. Prastawa, M., Bullitt, E., Ho, S., Gerig, G.: A brain tumor segmentation framework based on outlier detection. Med. Image Anal. 18(3), 217–231 (2004) 8. Capelle, S.: Evidential segmentation scheme of multi-echo MR images for the detection of brain tumors using neighbourhood information. Inf. Fusion 5, 203–216 (2004) 9. Heinen, R., Bouvy, W.H., Mendrik, A.M., Viergever, M.A., Biessels, G.J., De Bresser, J.: Robustness of automated methods for brain volume measurements across different MRI field strengths. PLoS ONE 10(11), e0165719 (2016) 10. Moeskops, P., De Bresser, J., Hugo, J.K., Adrienne, M.M., Geert, J.B., Josien, P.W.P., Ivana, I.: Evaluation of a deep learning approach for the segmentation of brain tissues and white matter hyperintensities of presumed vascular origin in MRI. NeuroImage Clin., 2 (2016) 11. Menze, B., Jakab, A.A., Bauer, S.S., Kalpathy-Cramer, J., Farahani, K.K., Kirby, J., et al.: The multimodal brain tumor image segmentation benchmark (BRATS). IEEE Trans. Med. Imaging, 34 (2015) 12. Chang, P.D.: Fully convolutional neural networks with hyperlocal features for brain tumor segmentation. In: Proceedings of MICCAI- BRATS 2016 Multimodal Brain Tumor Image Segmentation Benchmark, pp. 4–9 (2016) 13. Lun, T.K., Hsu, W.W.: Brain tumor segmentation using deep convolutional neural network. In: Proceedings of MICCAI-BRATS 2016 Multimodal Brain Tumor Image Segmentation Benchmark, pp. 26–29 (2016) 14. Randhawa, R.R., Modi, A.A., Jain, P.P., Warier, P.P.: Improving segment boundary classification for brain tumor segmentation and longitudinal disease progression. In: Proceedings of MICCAI-BRATS 2016 Multimodal Brain Tumor Image Segmentation Benchmark, pp. 53–56 (2016) 15. Kamnitsas, K., Ferrante, E., Parisot, S., Ledig, C., Nori, A., Criminisi, A., Rueckert, D., Glocker, B.: DeepMedic on brain tumor segmentation. In: Proceedings of MICCAI-BRATS Multimodal Brain Tumor Image Segmentation Benchmark, pp. 18–22 (2016) 16. Chena, L., Bentley, P., Rueckert, D.: Fully automatic acute ischemic lesion segmentation in DWI using convolutional neural networks. NeuroImage Clin., 3 (2017)
154
F. Aboudi et al.
17. Bezdek, J.C.: Pattern Recognition with Fuzzy Objective Function Algorithms. Kluwer Academic Publishers, Norwell (1981) 18. Omprakash, C., Devassy, N., Shushma, N., Raja, C., Mirajkar, P.: Stroke lesion detection in MRI images- a survey paper. Asian J. Appl. Sci. Technol., 2 (2017) 19. Mouelhi, A., Sayadi, M., Fnaiech, F., Mrad, K., Romdhane, K.B.: Automatic image segmentation of nuclear stained breast tissue sections using color active model and an improved watershed method. Biomed. Signal Process. Control, 10 (2013)
Comparison of the Control Strategies of an Active Filter of a Photovoltaic Generation System Connected to the Three-Phase Network Zoubir Chelli(&), Abdelaziz Lakehal, Tarek Khoualdia, and Yacine Djeghader University of Mohamed Chérif Messaadia, Souk-Ahras, Algeria [email protected]
Abstract. With the aim to improve the quality of the energy a photovoltaic generator is connected to the electrical network by associating the functionalities of shunt active power filter. In this work, a system consists of a field of solar panels, a three-phase voltage inverter connected to the grid and a non-linear load constituted by a diode rectifier bridge supplying a resistive load in series with an inductor is proposed. To compensate for harmonic currents and reactive power, as well as the injection of active solar energy into the network, Direct current and power controls are used. To find the maximum power point tracking (MPPT) the global method is applied. However, the compensation of the harmonic currents, the correction of the power factor and the injection of solar power towards the electrical network are guaranteed by the direct commands, the simulation of the system under Matlab/Simulink environment prove its robustness. According to the levels of the solar PV power injected and consumed by the nonlinear load, several regimes are approached. Keywords: Photovoltaic system MPPT
Direct power control Shunt active filter
1 Introduction This Solar energy captured with photovoltaic panels is a viable alternative to electricity generation, as it is a renewable source, both clean, unlimited and with a very low level of risk. Its potential is very important on the scale of the need for human activity; it is also very widely distributed throughout the globe, which gives it an interest shared by all. With the price of photovoltaic (PV) modules rising and the price of fossil fuels increasing, the exploitation of this resource with PV generation systems becomes viable and profitable [1, 2]. The rapid growth in the use of non-linear loads in power systems tends to degrade the quality of electrical energy supplied to consumers. Renewable energy has already attracted much interest of several researchers. Among them, Takagi and Fujita [3], to improve the quality of energy (PQ), introduced an application of the Active Power Filter (APF) in the photovoltaic (PV) renewable energy system, In order to prove it is powerful in detecting the reduction of total harmonic distortion (THD)
© Springer Nature Switzerland AG 2019 Y. Farhaoui and L. Moussaid (Eds.): ICBDSDE 2018, SBD 53, pp. 155–168, 2019. https://doi.org/10.1007/978-3-030-12048-1_17
156
Z. Chelli et al.
and the rapid regulation of voltage. They simulated their system with MATLAB/SIMULINK software package. As well, Khan, Ali, Shah, and Tariq [4] combined the cascaded multi-level inverter (CMI) with the quasi-Z-Cascade multilevel inverter (QZS-CMI) to minimize its complexity and switching which results in additional loss of voltage gain. Their approach is developed to improve the gain of voltage and the reduction of the total harmonic distortion (THD) according to IEEE Standard 519. Chen and Zheng [5], presented a control of a photovoltaic system connected to the network, to overcome the undesirable disadvantages of controlling the hysteresis current, and to obtain a constant switching frequency, they applied a modulated hysteresis control. A simulation studies under Matlab/Simulink is conducted to show the control performance of the grid connected photovoltaic system. However Suja and Raglend. [6] used a Neuro Fuzzy artificial inference system (ANFIS) controller, and unified power quality conditioner (UPQC) to improve the quality of power in a grid connected to the renewable energy system. Simulation results were used to analyze the effect of energy quality events at the Common Coupling Point in a grid connected renewable energy systems. In the same scopes, Xu et al. [7] presented a direct decay power control of the pulse width modulated constant current pulse width (PWM) rectifier. Their control method, called DPC-SVM, to generate the converter switching signal uses the SVM technique, and it has many advantages, including providing a line current very close to sine waveforms (TDH < 2%), and good DC bus voltage regulation is achieved using a PI controller. However, in order to overcome problems of harmonic pollution, active power filtering proves to be an adequate and efficient solution [8]. The purpose of this article is to examine the characteristics of an association between a photovoltaic generator, which aims to inject active power into an electrical network and a parallel active filter whose task is to eliminate disturbances present at this network [9].
2 The Configuration Studied The configuration studied consists of a solar PV generator connected to the DC bus of a three-phase voltage inverter, coupled in parallel to the network through an inductor [10, 11]. This electrical network supplies a non-linear receiver constituted by a rectifier PD3 having a load in series with an inductance [12, 13]. The diagram of “Fig. 1” illustrates this configuration. The analysis of the power fluxes is thus examined in various regimes imposed by the fluctuation of the level of irradiation during the diurnal period and the alternation with the nocturnal part where only the functions of the active filter are activated. It should be noted that with this principle the hardware investment is identical to a photovoltaic installation connected to the network but with the addition of the functionalities of an active filter in order to improve the quality of the energy on the network at the point of connection. It is therefore the control algorithm of the voltage inverter which is adapted in order to simultaneously ensure, at the level of the electrical network, the compensation of the harmonic pollution, reactive power, imbalances and the injection of power provided by PV panels [14, 15].
Comparison of the Control Strategies of an Active Filter
157
Fig. 1. Synoptic diagram of the configuration studied
The photovoltaic module (BP MSX-150) is chosen for modeling and simulation. It contains (72) multi crystalline silicon solar cells, and provides a maximum rated power of 150 W. The physical and electrical characteristics of this photovoltaic panel are given in the following “Table 1”: Table 1. Physical and electrical characteristics of the selected PV generator for modeling and simulation. Physical characteristics Number of cells in series (NS) Number of cells in parallel (NP) Electrical characteristics (STC) Maximum power (Pmax) Maximum point voltage (Vmpp) Current at maximum point (Impp) Open Circuit Voltage (Voc) Short circuit current (Isc)
BP MSX-150 NS = 72 NP = 1 Ga = 1000 w/m 2.25 °C.AM1.5 150 w 34.28 V 4.375 A 43.5 V 4.74 A
158
Z. Chelli et al.
The “Eq. 1” gives the mathematical model of a photovoltaic generator: V Voc:Ns þ 1Rs:Ns I ¼ Isc: 1 exp Vth
ð1Þ
Other expressions “Eqs. 2 and 3” have been given to express Isc and Voc respectively by: Isc ¼ C1: Ga 1 þ ðTc TcðSTC ÞÞ: 5:104
ð2Þ
Ga Voc ¼ VocðSTCÞ þ C3: Tc TcðSTCÞ þ Vth ln GaðSTC Þ
ð3Þ
To construct an equivalent model (by Simulink) of the PVG, the above expressions were used to subdivide the PV generator into blocks representing the various elements of its equivalent circuit model. The representative diagram of the mathematical current model of a photovoltaic module under matlab-simulink is given in “Fig. 2” and “Table 2” summarizes the simulation parameters of the shunt active power filter.
Fig. 2. Modeling PV generator under matlab-simulink
The values of the cell temperature T, the Ga irradiation, and the number of series photovoltaic cells Ns are accessible as external variables and can be changed during the simulation process. This makes it possible to observe and evaluate the reaction of the system to abrupt changes in operating conditions, such as variations in sunshine.
Comparison of the Control Strategies of an Active Filter
159
Table 2. The simulation parameters of the shunt active power filter System Power source
Nonlinear load
Shunt active power filter
Designations • The effective tension • Frequency • The internal resistance • The internal inductance • Bridge PD3 three-phase rectifier with R-L load • Inductance filtering at the input of the bridge PD3 • Storage capacity • Coupling inductance • Hysteresis band
Values Vs = 100 V F = 50 Hz Rs = 0.1 Ω Ls = 0.1 mH RL1 = 6.1 Ω RL2 = 10 Ω L = 20 mH Rc = 0.01 Ω Lc = 0.57 mH Cdc = 2200 µF Lf = 2 mH HB = 0.2 A
3 Shunt Active Power Filter Control Strategies 3.1
The Algorithm of the MPPT Method
The algorithm of the proposed MPPT method is shown in “Fig. 3”. With id (k) is the three-phase source current representation in the synchronous reference frame d − q, ΔiG (k) represents the variation in power caused by the change in illumination and can be defined as follows: DiG ¼ Te :e:ki ¼ Te :ki Vdcref ðk 1Þ Vdc ðk Þ
ð4Þ
The discretized writing of the current on the axis d lasts a sampling period Te in a situation of variation of illumination is written as follows: Did ðkÞ ¼ DiG ðkÞ þ Di# ðkÞ
ð5Þ
Δih (k) describes the variation of the current on the axis component generated by the perturbation increments of the MPPT algorithm (Incv). The first Incv is used when the output voltage of the panel is removed from the voltage of the MPP, and the second IncG in the presence of a variation of illumination. 3.2
Hysteresis Control
Hysteresis control, also known as all-or-nothing control, is a non-linear control that uses the error existing between the reference current Ifref and the current produced by the inverter If. The error is compared to a template called a hysteresis band. As soon as the error reaches the lower or higher band, a control command is sent in order to stay inside the band. The simplicity of implementation, as shown in “Fig. 3”, is the main advantage of this technique. Despite its simplicity of implementation, its robustness and its good dynamics, this order has some disadvantages namely:
160
Z. Chelli et al.
Fig. 3. Proposed MPPT algorithm for estimating the reference voltage
• The switching frequency is not fixed; it depends on the hysteresis band and of the current derivative. The command is applied separately on all three phases. The structure electro-technical system imposes at every moment that the sum of the three currents is zero. The result obtained on one current is not independent of the other two phases. Thus, the enslaved current cannot respect the limits imposed by the band of hysteresis [8]. 3.3
Direct Power Control Study of SAPF
The principle of direct control has been proposed at [16] and has been developed later in many applications. The objective is to eliminate the modulation block and internal loops by replacing them with a switchboard whose inputs are the errors between the reference values and the measurements. The first application developed was for the control of an electric machine and the control structure is known as Direct Torque Control (DTC). Subsequently, a similar power control technique (DPC) was proposed by [17] for a control application of the rectifiers connected to the network. With the DPC there is no current control loop or PWM modulation element, because the switching states of the inverter, for each sampling period, are selected from a switching table, based on the instantaneous error between reference values and those measured or estimated active and reactive powers, and the angular position of the source voltage vector. Generally, with this control strategy, the DC bus voltage is regulated for active power control and operation with a factor of unit power is obtained by imposing the reactive power at a zero value [18].
Comparison of the Control Strategies of an Active Filter
161
4 Simulation Results The photovoltaic compensation system consists of a PVG, a chopper in booster mode and an active shunt filter that connects to the network. The latter feeds a non-linear load. The proposed compensation system plays the role of a compensator reagent in the case of low illumination, and plays the role of a shunt active filter with a real power injection to the electricity network produced by the photovoltaic conversion chain in the case of strong illumination. The temperature and the illumination are fixed at standard conditions (STC) (Ga = 1000 w/m2 Ta = 25 °C) and the global system is simulated with two types of control of the inverter (active filter), the hysteresis control and the direct power control (DPC) so as to operate the system as a source of energy (injection of mains power) and an active shunt filter (harmonic compensation and reactive power). “Figures 4 and 5” shows the waveforms of the three-phase source current and the current consumed by the non-linear load, and the active and reactive powers of the three-phase source before the introduction of the photovoltaic compensation system.
Fig. 4. Current waveforms before photovoltaic compensation
Initially the system operates without SAPF, the load consumes an active power of 4 Kw, the source currents are identical to those of the nonlinear load (is = il = 27.84 A) characterized by a spectrum containing only harmonics of odd order (not multiples of three) and a THDi = 23.29% “Fig. 6”.
162
Z. Chelli et al.
Fig. 5. Active and reactive power characteristic
Fig. 6. Spectral analysis of the source current before commissioning of the SAPF
• Simulation of the system with the hysteresis control for the SAPF. “Figures 7 and 8” shows that the SAPF is put into operation, producing currents if which arrive, after a transient of t = 0.01 s, making the sinusoidal source currents and in phase with the corresponding voltages, the active power returns to its nominal value after a transient while the reactive energy continues to oscillate around zero. Therefore, the harmonic distortion rate of the source current is improved and is worth THDi = 2.51% “Fig. 9 (a)” and the harmonic distortion rate of the source voltage becomes THDv = 3.55% “Fig. 9 (b)”.
Comparison of the Control Strategies of an Active Filter
163
Fig. 7. Transient simulation results when closing the SAPF for a PD3- [RL1, L] nonlinear load
Fig. 8. Waveform of instantaneous powers after SAPF commissioning
The value of the DC bus voltage tends to its reference value Vdcref “Fig. 10”, obtained by the adaptation algorithm presented in Sect. 3.1 after a transient of Δt = 0.08 s. These results show the effectiveness of the proposed algorithm. “Figure 11” represent the voltage and the current of the PVG. From this simulation, we notice an injection of active power to the grid due to the PVG. This is characterized by a decrease of active power supplied by the three-phase network Ps = 3 Kw (current decrease of source is = 20 A), so one has a power of Pf = 1000 W produced by the PVG to meet the energy requirement of the nonlinear load “Fig. 12”.
164
Z. Chelli et al.
Fig. 9. Spectral analysis of signals after SAPF commissioning: (a) source current, (b) source voltage
Fig. 10. Waveform of the DC bus voltage and its reference voltage
Fig. 11. Characteristic of the PV Generator
Comparison of the Control Strategies of an Active Filter
165
Fig. 12. Waveform of the active power injected into the grid by the PV Generator
• Simulation of the system with the command DPC for the SAPF. After the commissioning of the SAPF we can notice that the source currents “Fig. 13”, after a transient of t = 0.01 s become sinusoidal with a THDi = 1.61% “Fig. 15 (a)”, and are in phase with source voltages with THDv = 2.77% “Fig. 15 (b)”. As regards the DC bus voltage, it tends towards its reference after a transient of “Fig. 16”. This control technique provides better energy quality compared to the previous technique “Fig. 14”.
Fig. 13. Transient simulation results at closure of SAPF for a PD3- [RL1-L] nonlinear load
Fig. 14. Waveform of instantaneous powers after SAPF commissioning (DPC)
166
Z. Chelli et al.
Fig. 15. Spectral analysis of the signals after SAPF commissioning: (a) source current, (b) source voltage (DPC)
Fig. 16. Waveform of the DC bus voltage and its reference voltage (DPC)
Fig. 17. Characteristic of the PV Generator (DPC)
Comparison of the Control Strategies of an Active Filter
167
Note that “Fig. 17” shows the PVG voltage and current waveforms at the MPP operating point, with power injection to the network “Fig. 18”.
Fig. 18. Waveform of the active power injected into the grid by the PV Generator (DPC)
5 Conclusion In this work, the association between a PV solar generator and an active filter is validated and allows simultaneous delivery of functionalities or services to the distribution network at the injection site without the addition of specific equipment. The results confirm the feasibility of the system and validate the various functionalities assigned to the voltage inverter, namely the compensation of harmonic pollution, reactive power and the transfer of the energy flow from the solar PV to the electricity grid. We exposed the results of the two control strategies. They are characterized by current control and direct power control. We have developed an adaptation algorithm, based on the incremental conductance technique, of the reference voltage of the continuous bus, which has proved its efficiency. The simulation results obtained in the various cases considered are satisfactory and confirm the theoretical study, and in particular the efficiency and robustness of the proposed system: a quasi-sinusoidal current absorption with a factor close to unity. In the case of the hysteresis control of SAPF, we observed a good signal quality in terms of harmonic distortion THDi = 2.51% of the currents and voltages of the three-phase source. Nevertheless, this current control technique induces at the spectrum level a wide frequency band due to the switching of the semiconductors, which is difficult to filter. While D.P.C control technique, in addition to its simplicity, a better control of instantaneous active and reactive power control is achieved. As well as a significant improvement in the current and voltage distortion rates relative to the hysteresis control THDi = 1.61%.
168
Z. Chelli et al.
References 1. Hassaine, L., Olias, E., Quintero, J., Haddadi, M.: Digital power factor control and reactive power regulation for grid-connected photovoltaic inverter. Renewable Energy 34, 315–321 (2009) 2. Chebabhi, A., Fellah, M., Kessal, A., Benkhoris, M.: A new balancing three level three dimensional space vector modulation strategy for three level neutral point clamped four leg inverter based shunt active power filter controlling by nonlinear back stepping controllers. ISA Trans. 63, 328–342 (2016) 3. Takagi, K., Fujita, H.: A three-phase grid-connected inverter equipped with a shunt instantaneous reactive power compensator. IEEJ Trans. Ind. Appl. 138, 530–537 (2018) 4. Khan, M., Ali, H., Shah, S., Tariq, D.: An efficient diode clamped multilevel inverter for reducing THD with selective harmonic elimination pulse width modulation (SHEPWM). Int. J. Sci. Eng. Technol. 4, 417–422 (2015) 5. Chen, Y., Zheng, Y.: Fuzzy control of Z-source PV grid-connected using constant frequency hysteresis comparison. Adv. Mater. Res. 852, 412–416 (2014) 6. Suja, K., Raglend, J.: Adaptive genetic algorithm/neuro-fuzzy logic controller based unified power quality conditioner controller for compensating power quality problems. Aust. J. Electr. Electron. Eng. 10, 351–361 (2013) 7. Xu, B., Yang, D., Jiao, J.: Coordinated control strategy based on DPC-SVM-DTC threelevel dual PWM converter. Int. J. Control Autom. 10, 327–338 (2017) 8. Chelli, Z., Toufouti, R., Omeiri, A., Saad, S.: Hysteresis control for shunt active power filter under unbalanced three-phase load conditions. J. Electr. Comput. Eng. 2015, 1–9 (2015) 9. Gururaj, G., Ramasamy, A., Prakash, D., Naganagouda, D.: Comprehensive approach of modeling and simulation of solar photovoltaic power plant. Int. J. Eng. Res. Appl. 07, 06–11 (2017) 10. Tsengenes, G., Adamidis, G.: Investigation of the behavior of a three phase grid-connected photovoltaic system to control active and reactive power. Electr. Power Syst. Res. 81, 177– 184 (2011) 11. Noroozian, R., Gharehpetian, G.B.: An investigation on combined operation of active power filter with photovoltaic arrays. Int. J. Electr. Power Energy Syst. 46, 392–399 (2013) 12. Esram, T., Chapman, P.L.: Comparison of photovoltaic array maximum power point tracking techniques. IEEE Trans. Energy Convers. 22, 439–449 (2007) 13. Zhang, F., Thanapalan, K., Procter, A., Carr, S., Maddy, J.: Adaptive hybrid maximum power point tracking method for a photovoltaic system. IEEE Trans. Energy Convers. 28, 353–360 (2013) 14. Liu, F., Duan, S., Liu, F., Liu, B., Kang, Y.: A variable step size INC MPPT method for PV systems. IEEE Trans. Ind. Electron. 55, 2622–2628 (2008) 15. Reisi, A.R., Moradi, M.H., Showkati, H.: Combined photovoltaic and unified power quality controller to improve power quality. Sol. Energy 88, 154–162 (2013) 16. Takahashi, I., Nunokawa, M.: Prediction control for a cycloconverter of a power distortion compensation system. IEEE Trans. Ind. Appl. 25, 348–355 (1989) 17. Sato, A., Noguchi, T.: Voltage-source PWM rectifier-inverter based on direct power control and its operation characteristics. IEEE Trans. Power Electron. 26, 1559–1567 (2011) 18. Adamidis, G., Tsengenes, G., Kelesidis, K.: Three phase grid connected photovoltaic system with active and reactive power control using ‘instantaneous reactive power theory. Renewable Energy Power Qual. J. 1, 1086–1091 (2010)
A Multi-chaotic Fibonacci Algorithm for Digital Image Encryption Lamiche Chaabane(&) Computer Science Department, Mohamed Boudiaf University, M’sila, Algeria [email protected], [email protected]
Abstract. Actually, images are widely used in various real applications, such as Internet communication, multimedia systems, medical imaging, telemedicine, monitoring, and military communication, so their security is becoming very important. Basically, encryption is one of the best alternative way to ensure information security. In this research work, we propose a hybrid encryption algorithm for the digital grayscale image. The developed approach is based on the combination of Fibonacci sequence with the chaotic logistic map to create the secret key in order to improve encryption quality. Numerical results using a set of image benchmarks have shown the capability of the proposed encryption model to produce a better image security compared to results given by some other recently literature works. Keywords: Internet communication Security Fibonacci sequence Chaotic logistic map
Encryption algorithm
1 Introduction Nowadays, the transmission of digital images over the Internet and wireless networks is in full effervescence development, due to the rapid developments in digital image processing and network communications fields. It needs serious protection of the communicated image information against illegal usage, especially for those requiring reliable, fast and robust secure systems to store and transmit, such as military image databases, confidential video conference, medical imaging system, online private photograph album, and so on [1]. Image encryption process try to convert original image to another image that is hard to understand by unauthorized people; to keep the image confidential between users, it is essential that nobody could get to know the content without a key for decryption [2]. Furthermore, a user can retrieve the initial image by applying a decryption method on the cipher image, which is usually a reverse execution of the encryption procedure [3]. Most traditional encryption techniques, such as Data Encryption Standard (DES), International Data Encryption Algorithm (IDEA), Advanced Encryption Standard (AES), Linear Feedback Shift Register (LFSR), etc. [4, 5] with high computational security consider plaintext as either block cipher or data stream and are not suitable for image/video encryption in real time for some reasons. Firstly, image data have high redundancy and strong correlation among image pixels. Secondly, they have usually a huge data volume. and strong correlation among image pixels. The implementation of © Springer Nature Switzerland AG 2019 Y. Farhaoui and L. Moussaid (Eds.): ICBDSDE 2018, SBD 53, pp. 169–175, 2019. https://doi.org/10.1007/978-3-030-12048-1_18
170
L. Chaabane
traditional algorithms for image encryption is even more complicated when they are realized by software. Therefore, these well known encryption algorithms are suitable for textual data not for multimedia data [6]. In this research work we aim to present a novel encryption method to enhance the image security. The developed approach consists in hiding the plain image bits with the bits of the pseudo-random key stream generated through the exclusive-OR operation (or XOR). The pseudo-random key stream is generated by performing XOR between two generators, the first one generator is chaotic logistic map and the second is modified Fibonacci sequence generator. The organization of the paper is as follow. In Sect. 2, some previous related works are briefly cited. We summarize principals concepts of chaotic logistic map and Fibonacci sequence in Sect. 3. In Sect. 3.1, we discuss the methodology of our proposed approach. The performances and analysis of the proposed image encryption scheme are studied in Sect. 4. Finally, conclusion remarks are drawn in Sect. 5.
2 Previous Related Works A brief review of some related works which are used in image encryption field is presented in this section. In Ref. [7], the authors proposed a new technique based on chaotic map encryption and iterated-random block transformation. Their algorithm uses three logistic map functions to get a highly confused, diffused and secured encrypted image. Experimental results showed that the developed approach has a large key space, and high-level security compared with similar chaotic encryption algorithms. Singh and Singh [8] developed a new technique based on the blowfish algorithm with two processes; the first is a key expansion, and the second is a data encryption. The offered Blowfish system includes two exclusive-or operations that are performed after 16 rounds and a swap operation. Experimental results demonstrated blowfish technique is fast and secure. Hua et al. [9] introduced an excellent new encryption technique based on twodimensional Logistic-Sine map (2D-LSM). Simulation results and security analysis demonstrate the potent of the proposed algorithm to protect images with a high security level. In Ref. [10], the authors developed an intelligent image encryption model based on three chaotic maps derived from some plan curves equations, which are described previously in [11, 12]. The main advantage of their work is the large space of the secret key, which allow the robustness of the image encryption scheme. An image encryption algorithm based on chaos and a balanced pixel algorithm to determine the times of image encryption is presented in [13]. In this work, the authors used a XOR operation to combine chaos function with an iterative equation to encrypt images. Experimental results showed the feasibility and effectiveness of the proposed method.
A Multi-chaotic Fibonacci Algorithm for Digital Image Encryption
171
3 Proposed Method The proposed image encryption algorithm called Fibo_Encrypt is decomposed in three steps: (1) pseudo-random key stream generator. (2) Encryption function. (3) Decryption function. The flowchart of the algorithm is presented in Fig. 1.
Fig. 1. Encryption function flowchart.
3.1
Pseudo-random Key Stream Generator
The pseudo-random key generator is realized by the pseudo-random number generator, which use the mathematical formulas as follows: Modified Fibonacci sequence generator: Xn ¼ ðXn1 þ Xn2 ÞMod M Initial parameters are: (L, K). Where: ðX0 ¼ LÞ; ðX1 ¼ KÞ: And M ¼ 255; ðL; KÞ 2 N:
172
3.2
L. Chaabane
Encryption Function
(1) Generate a pseudo random key stream as follows: A. Set the initial parameters (L, K) B. Generate a stream of pseudo-random number, K1 from modified Fibonacci sequence with same size of the plain image n n. C. Convert the generated stream K1 as bits. (2) Convert the plain image to a data stream as bits ðmi Þ. (3) Performing the operation XOR bit by bit between the data stream (of the plain image) and the pseudo-random key stream K1 to obtain a stream of encrypted data (encrypted image Ci) Ci ¼ mi K1:
3.3
Decryption Function
The decryption is the same like encryption function, except the initial parameters (L, K) must be the same where used in encryption function.
4 Experimental Results A good encryption procedure should be robust against all kinds of cryptanalytic, statistical and brute-force attacks. In this section, key space analysis, Histogram, information entropy and correlation coefficient analysis were carried out to clarify the good performance of the proposed model. 4.1
Image Database
The database image used in this paper is free available in university of WisconsinMadison [14]. Witch primarily support research in image processing, image analysis, and machine vision. 4.2
Key Space Analysis
For a secure image cipher, the key space should be large enough to make the brute force attack infeasible. The key used in our schema is pseudo-random numbers that was generated by the parameters (L, K, r, X), the size of the key is n n is the size of the original image. So each element in the pseudo-random numbers is encoded on 8 bits. Therefore the Key space is 28nn . For example an image of size 256 256 the key space is 28256256 .
A Multi-chaotic Fibonacci Algorithm for Digital Image Encryption
4.3
173
Histogram Analysis
The histogram of the ciphered image should be significantly different from the histogram of the plain image, and the histogram of the ciphered image should be as uniformed distribution as possible that will indicates more randomness [15]. The histogram analysis results for grayscale image lena are presented in Figs. 2 and 3.
Fig. 2. (a) Plain image; (b) Encrypted image.
Fig. 3. Histograms of images.
4.4
Entropy Correlation Coefficient Analysis
Entropy states the degree of uncertainty in a system. It can be done as follows: HðMÞ ¼
n X
pi x log2 pi
i¼1
For images with random pixels which are encoded by 8 bit, entropy should be equal to 8. However, entropy is usually smaller than 8, but a value closer to eight means that the possibility of predictability is less and the security level is higher [16]. Correlation is a statistical technique that can show whether and how strongly pairs of variables are related. The correlation coefficients are calculated by the following equation for two variables x and y of length N [17]: covðx; yÞ r ¼ pffiffiffiffiffiffiffiffiffiffipffiffiffiffiffiffiffiffiffiffi DðxÞ DðyÞ
174
L. Chaabane
where: covðx; yÞ ¼ EðxÞ ¼
N 1X ðxi EðxÞÞðyi EðyÞÞ N i¼1
N N X 1X xi ; DðxÞ ¼ ðxi EðxÞÞ2 N i¼1 i¼1
For images we randomly select pairs of two-adjacent pixels from plain images and ciphered images, and calculate the correlation coefficients, respectively by using the equations cited above. The smaller correlation means the better encryption effect. For the test case we used grayscale images, like Lena, Baboon, Cameraman and Peppers images. The obtained results are summarized in Table 1. Table 1. Entropy and correlation coefficient for encrypted image. Input images Cameraman Lena Peppers Baboon Average
Size 256 512 512 512
256 512 512 512
Entropy Correlation 7.9972 −0.0012 7.9992 0.0002 7.9992 −0.0001 7.9992 0.0022 7.9987 0.0009
The results show that after simulating 4 images, the average value of the entropy from the encrypted images is 7.9987. It is closer to the value 8. And the average value of correlation coefficients from the encrypted images is 0.0009, which is closer to 0, which means the effectiveness of our proposed encryption algorithm.
5 Conclusion In this paper, we have proposed a hybrid image encryption, which based on combination of chaotic logistic map and modified Fibonacci sequence. The experimental results showed that the proposed image encryption system has a very large key space, and high-level security. Thus the analysis proves the security, correctness, effectiveness. Also the results of the proposed algorithm are particularly suitable for Internet image encryption and transmission applications. In the future, we will revise our proposed key generator mechanism or to use another efficacy chaotic maps in order to increase the level security of the image. A comparison of the proposed method with some other encryption methods is possible to verify its effectiveness.
A Multi-chaotic Fibonacci Algorithm for Digital Image Encryption
175
References 1. Ruisong, Y., Haiying, Z.: An efficient chaos-based image encryption scheme using affine modular maps. Int. J. Comput. Netw. Inf. Secur. 7, 41–50 (2012) 2. Younes, M.A.B., Jantan, A.: Image encryption using block-based transformation algorithm. IAENG Int. J. Comput. Sci. 35(1) (2008). IJCS_35_1_03 3. Kumar, M., Aggarwal, A., Garg, A.: A review on various digital image encryption techniques and security criteria. Int. J. Comput. Appl. 96(13), 19–26 (2014). 0975 – 8887 4. Schneier, B.: Applied Cryptography: Protocols, Algorithms, and Source Code in C. Wiley, New York (1996) 5. Daemen, J., Rijmen, V.: The Design of Rijndael: AES - The Advanced Encryption Standard. Springer, New York (2002) 6. Faraoun, K.: Chaos-based key stream generator based on multiple maps combinations and its application to images encryption. Int. Arab J. Inf. Technol. 7(3), 231–240 (2010) 7. Tabash, F.K., Rafiq, M.Q., Izharrudin, M.: Image encryption algorithm based on chaotic map. Int. J. Comput. Appl. 64(13), 1–10 (2013) 8. Singh, P., Singh, K.: Image encryption and decryption using blowfish algorithm in MATLAB. Int. J. Sci. Eng. Res. 4(7), 150–154 (2013) 9. Hua, Z., Zhou, Y., Pun, C.-M., Philip Chen, C.L.: Image encryption using 2D Logistic-Sine chaotic map. In: 2014 IEEE International Conference on Systems, Man, and Cybernetics (SMC), 5–8 October 2014, San Diego, CA, USA, pp. 3229–3234 (2014) 10. Boriga, E., Dăscălescu, A.C., Diaconu, A.V.: A new fast image encryption scheme based on 2D chaotic maps. IAENG Int. J. Comput. Sci. 41(4), 249–258 (2014) 11. Boriga, R., Dăscălescu, A.C.: A novel pseudo-random bit generator based on some transcendental chaotic systems. Ann. Ovidius Univ. Econ. Sci. Ser. 11, 208–212 (2011) 12. Boriga, R., Dăscălescu, A.C., Priescu, I.: A new hyperchaotic map and its application in an image encryption scheme. Sig. Process. Image Commun. 29(8), 887–901 (2014) 13. Zhang, J., Zhang, Y.: An image encryption algorithm based on balanced pixel and chaotic map. Math. Prob. Eng. 2014, 1–7 (2014) 14. University of Wisconsin-Madison, Base de données d’images. https://homepages.cae.wisc. edu/*ece533/images/ 15. Rani, M., Kumar, S.: Analysis on different parameters of encryption algorithms for information security. Int. J. Adv. Res. Comput. Sci. Softw. Eng. 5(8), 104–108 (2015) 16. Schneier, B.: Applied Cryptography. Wiley, New York (2010) 17. Indrakanti, S.P., Avadhani, P.S.: Permutation based image encryption technique. Int. J. Comput. Appl. 28(8), 45–47 (2011)
On the Out of Band Emission Reduction for Orthogonal Frequency Division Multiplexing Based Systems Naima Sofi1(&), Fatima Debbat2, and Fethi Tarek Bendimerad1 1
2
LTT Laboratory, Department of Telecommunications, Tlemcen University, Tlemcen, Algeria [email protected] Department of Informatics, Mascara University, Mascara, Algeria
Abstract. Due to its ability to transmit a high rate data stream with maintaining robustness against multipath fading, high spectrum efficiency and efficient implementation, OFDM has been widely used in various communications systems such as DVB, WIFI and WIMAX. It is also considered as a candidate for the air interface of the current and future high speed mobile communications standards such as LTE, LTE advanced. It is also very appropriate for multi-band cognitive radio systems. However, OFDM suffers from some drawbacks. One of them is the Out of Band (OOB) leakage due to high spectral sidelobe which produces interference in neighboring bands and then degrade the system performance. Several schemes have been proposed to reduce the OOB emission such as filtering, windowing and precoding etc. in this paper, a hybrid approach based on two different precoding techniques is proposed to optimize this reduction. Simulation results verified the advantages of this scheme as compared with other well known techniques. Keywords: OFDM
OOB emission SVD Orthogonal projection
1 Introduction OFDM has attracted significant attention and shows powerful characteristics that make it suitable for different wireless standards. Thanks for its favorable properties such as the easy and efficient implementation through FFT/IFFT, its high spectral efficiency and multipath delay spread tolerance [1]. On the other hand, it is widely recognized that one of the major problems of OFDM, that would limit its performance with nest generation systems, is the large side lobe of its subcarriers. As the subcarriers constituting OFDM symbols can be shown in the time domain as a set of sinusoids multiplied by a rectangular window. In the frequency domain, those windowed sinusoids are a set of orthogonal Sinc-functions shifted in frequency. Thus, a high OOB due to the slow decay of OFDM spectrum occurs. This can be interfering with other communications systems that operate on neighboring bands. Several schemes have been proposed to suppress or reduce this OOB emission of OFDM. Cancellation carriers (CC) [2], constellation extension [3] and pre-coding [4–7] that can be classified as a frequency domain treatment, where the OOB emission © Springer Nature Switzerland AG 2019 Y. Farhaoui and L. Moussaid (Eds.): ICBDSDE 2018, SBD 53, pp. 176–185, 2019. https://doi.org/10.1007/978-3-030-12048-1_19
On the Out of Band Emission Reduction
177
reduction is performed before the Inverse Fast Fourier Transform (IFFT). The other treatment, where the OOB emission processes are carried out after the IFFT, is the time domain treatment. This treatment includes filtering [8] and windowing (also known as pulse shaping) etc. although any of those techniques can be efficient in OOB emission reduction, it suffers from some penalties. Such as cancellation carrier which requires an allocation of significant power on the cancellation carriers, resulting in a reduction of power efficiency. Filtering and windowing are inefficient due to the introduction of inband interference. And pre-coding that introduces an error floor in the error performance of the system by combining the independent data symbols. Another way for reducing OOB emission is to replace the OFDM system with lower OOB emission systems such as UFMC [9] or FBMC [10]. Although the performance of those techniques in the suppression of OOB emission, they have some disadvantages such as their high implementation complexity. Since each of the OOB reduction techniques suffer from some drawback, a good combination of two techniques might enhance the OOB reduction performance. In this paper, a hybrid approach to suppress the OOB emission in OFDM system based on combining two precoding approaches is proposed.
2 OFDM Basis and OOB Emission OFDM is a multi-carrier transmission by which a high data rate bit stream is caught and subdivided into N parallel bit streams that are transmitted simultaneously over a number of subcarriers. The ith OFDM transmitted symbol can be written as follows: si ðt Þ ¼
XN1 k¼0
Xi;k ej2pkt=T s I ðtÞ
ð1Þ
Where X i;k k¼0::N1 represent the N complex symbols that it carries. And the indicator function I ðt Þ ¼
1 0
for Tg t Ts elswhere
As it is shown from (1), each subcarrier is a rectangular windowed sinusoid in the time domain. Where the OFDM symbol si ðtÞ is the sum of these subcarriers. This becomes in the frequency domain, after the FFT: ( Si ðf Þ ¼ TFfsi ðtÞg ¼ TF ¼
XN1 k¼0
N 1 X k¼0
) X i;k e
2pkt Ts
I ðt Þ
n 2pkt o XN1 Xi;k :TF e Ts I ðtÞ ¼ Xi;k ak ðf Þ k¼0
ð2Þ
178
N. Sofi et al.
The spectrum of the OFDM signal s(t) can be written as: 2 o 1 XN1 1 n E jSi ðf Þj22 ¼ E k¼0 Xi;k ak ðf Þ ð3Þ T T 2 Where ak ð f Þ ¼ Tsinc pT f Tks is the representation of the kth subcarrier in the Pðf Þ ¼
frequency domain. Apparently, it is a Sinc function that decays as a factor of 1=f 2 . Where T ¼ Ts þ Tg . Thus, untreated OFDM causes significant OOB emission that we studied its suppression with some schemes. • Precoding approach The spectral precoding is a widely used technology in OFDM systems, to improve the performance of the transmission reliability. Also, it is a recent approach for the OOB emission reduction. There are many precoding methods that focus on the data modulating the subcarriers. Thus, the precoder pre-codes each data symbol Xi;k into X i;k , which in turn, modulates the corresponding subcarrier. This treatment can ensure a good OOB emission reduction. After precoding, the precoded OFDM symbol can be expressed as follow [4]: spi ðtÞ ¼
XN1 k¼0
j2pkt
Xi;k e Ts I ðtÞ
¼ PT ðtÞX i
ð4Þ
Where Xi , X i;k0 ; X i;k1 ; X i;k2 ; . . .; Xi;k1 is the result of precoding transmit data
T Xi;k , Xi;k0 ; Xi;k1 ; Xi;k2 ; . . .; Xi;k1 , and h j2pk0 t j2pk1 t i j2pkK1 t PðtÞ , e Ts ; e Ts ; . . .; e Ts I ðtÞ: Where: X i ¼ GX i
ð5Þ
G is the precoding matrix. The choice of this matrix depends on the perspective of the OOB emission, either in the time or the frequency domain. Some pre-coders view the OOB emission in the time domain as a consequence of the OFDM signal discontinuous nature (see Fig. 1), due to the rectangular windowed sinusoids constituting the OFDM signal. We focus on the time precoding techniques proposed in [11] and [12].
On the Out of Band Emission Reduction
179
Fig. 1. PSD of different precoding techniques
• Orthogonal projection pre-coding In this technique, the precoder observe the OOB emission in the frequency domain. The pre-coding matrix nulls completely the subcarriers in some well chosen frequencies ½f0 ; f1 ; . . .; fM1 , at each side of the frequency band. So that the Fourier transform of this symbol in these frequencies requires being zero: Si ðf M Þ ¼ 0;
f M ¼ f 0 ; f 1 ; . . .; f M1 :
ð6Þ
With (2), we can write: AX ¼ 0 Where A ¼ ðaðf0 Þ; aðf1 Þ; . . .; aðfM1 ÞÞT is an K * M matrix which represent the requirements in (4). As it is explained in [11], this pre-coding algorithm is generalized as a homogenously constrained least square problem, where its solution is a an orthogonal projection of the data symbol into the null space N ð AÞ of A, which is shown as N ð AÞ ¼ fx 2 CK jAx ¼ 0g. So then: X ¼ GX; Where G , I AH ðAAH Þ1 A
ð7Þ
With I is the identity matrix and H is the Hermitian operator. Since G is not a diagonal matrix, where it is an orthogonal projection matrix. Then, the process at the reception is no more invertible [13, 14]. Consequently, the precoding distortion can’t be compensated and an iterative receiver is necessary as presented in [14].
180
N. Sofi et al.
• SVD pre-coding In this technique, assuming that we want to minimize the transmission power of the frequencies corresponding to the OOB regions, we note these M frequencies y1 ; y2 ; . . .yM , by the design of a SVD pre-coding matrix Gs , such as: X i ¼ Gs X i
ð8Þ
Denoting the frequency sampling vector, Si ¼ ½Si ðy1 Þ; Si ðy2 Þ; . . .; Si ðyM ÞT : 0
1 P N ð yM Þ C .. .. A. . . P N ð y1 Þ P N ð yM Þ To minimize kSi k without caring about X i , and SVD decomposition of P must be accomplished. Where P ¼ URV H , U is a unitary matrix with of M * M dimension, R is a diagonal matrix of M * N and V is a unitary matrix with N * N where its columns are denoted v1 ; v2 ; . . .; vN . The pre-coding matrix Gs of dimension N * K is chosen to be s G ¼ ½vNK þ 1 vNK þ 2 . . . vN . With K/N is the coding rate and R = K − N is the redundancy. Therefore, for any R M, kSi k ¼ 0. P 1 ð y1 Þ B .. s So: Si ¼ PG Xi , P ¼ @ .
• Proposed approach (combining technique) Since each of the OOB radiation reduction techniques present some disadvantages and penalties, we try to find a good combination of two or more techniques to benefit from the advantage of each of them and to improve perfectly the system performance. As it is mentioned in [11] and [12], the main idea for the precoding is to force the power of chosen frequencies to be zero and to minimize the power in an optimized region. So, we must find a good combination to take advantages of both techniques and to obtain more reduction of OOB emission.
3 Results and Discussion In this section, simulation and analysis are run to show the performance of the hybrid scheme and to demonstrate the tradeoff between its requirements and the effect of each one on the other. The PSD is calculated to demonstrate the OOB emission reduction of this system. In our analysis we adopt an OFDM system conforming to 3GPP LTE 1 specifications. The system is configured with K = 500 subcarriers, Ts ¼ 15 ½ms, 7.5 MHz of bandwidth and Tg ¼ 4:7 ls. • In the first experiment, the PSD of different precoding techniques are presented to show their performance and to demonstrate the difference between them.
On the Out of Band Emission Reduction
181
As we can see from Fig. 1, the precoding techniques have a very important performance in OOB reduction as compared with original OFDM system. We can see also the performance of the chosen precoding techniques in this study (the orthogonal projection precoding and the SVD precoding). • In the second experiment, we study the performance of orthogonal projection precoding and the SVD precoding independently. Figures 2, 3 and 4 show the performance of SVD precoding scheme in terms of OOB reduction. Where it introduce an OOB emission reduction of 10 dB, and this reduction increase with the number of chosen frequencies. From the PAPR and BER simulations, there is no reduction of PAPR with this precoding technique as there is no
Fig. 2. PSD of SVD precoding
Fig. 3. CCDF of SVD precoding technique
182
N. Sofi et al.
Fig. 4. BER of SVD precoding technique
increase on it. The BER performance shows that this technique inhibits the OOB emission without additional error. So, without BER degradation as well as the OP precoding that improve the OOB emission reduction without any increasing or decreasing in PAPR. But contrary to the SVD precoding, the OP requires an iterative decoder at the receiver side, because of the error floor that can be introduced by the precoding processing (Figs. 5, 6 and 7). • Finally, we simulate our proposed schemes and comparing it with other existing combinations (Figs. 8 and 9).
Fig. 5. Orthogonal projection precoding technique
On the Out of Band Emission Reduction
183
Fig. 6. CCDF of orthogonal projection precoding technique
Fig. 7. BER of orthogonal projection precoding technique
We can see from the simulation above (Figs. 8 and 9) that proposed schemes outperforms other combinations existing in the literature with its capability in OOB suppression without any increasing of PAPR (with maximal PAPR = 13.55) and without any additional complexity or error floor. We can see that join PAPR and sidelobe suppression is not achieved with these combining schemes, as there is no PAPR reduction method employed like in [15]. It is for us a future work to find a scheme that combine precoding and PAPR reduction for joint PAPR and OOB suppression.
184
N. Sofi et al.
Fig. 8. PSD of proposed scheme and comparing with others
Fig. 9. CCDF of proposed scheme and comparing with others
4 Conclusion Even though OFDM is a promising and exciting technology in wireless communications, especially in Cognitive Radio, it has many important disadvantages that cannot be ignored, as the OOB emission. One of the awesome techniques that try to avoid this great disadvantage is the use of precoding with OFDM. However, each of the existing precoding techniques present some complexity depending on some parameters. Thus, in this paper we try to combine two techniques for a best reduction of OOB emission without more complexity and with the same PAPR and BER performance. So we take advantage of each reduction with an optimal manner. Therefore, the OOB is minimizing by using both techniques with minimum parameters. In the future works, we
On the Out of Band Emission Reduction
185
will try to optimize this scheme with AIS (artificial immunity system) optimization system, to benefit for more OOB reduction without increasing in complexity and without BER or spectral efficiency sacrifice.
References 1. Wu, Y., Zou, W.Y.: Orthogonal frequency division multiplexing: a multi-carrier modulation scheme. IEEE Trans. Consum. Electron. 41(3), 392–399 (1995). https://doi.org/10.1109/30. 468055 2. Brandes, S., Cosovic, I., Schnell, M.: Reduction of out-of-band radiation in OFDM systems by insertion of cancellation carriers. IEEE Commun. Lett. 10(6) (2006). https://doi.org/10. 1109/lcomm.2006.1638602 3. Pagadarai, S., Rajbanshi, R., Wyglinski, A.M., Minden, G.J.: Sidelobe suppression for OFDM-based cognitive radios using constellation expansion. In: Wireless Communications and Networking Conference, WCNC 2008. IEEE. https://doi.org/10.1109/wcnc.2008.162 4. van de Beek, J., Berggren, F.: N-continuous OFDM. IEEE Commun. Lett. 13(1) (2009). https://doi.org/10.1109/lcomm.2009.081446 5. Pan, L., Xiao, S., Qiu, Y., Li, B.: An adaptive precoder for out-of-band power reduction in OFDM based cognitive radio system. Int. J. Future Gener. Commun. Netw. 7(1) (2014). http://dx.doi.org/10.14257/ijfgcn.2014.7.1.14 6. Pan, L., Ye, J., Yuan, X.: Spectral precoding for out-of-band power reduction under condition number constraint in OFDM-based system. Wireless Pers. Commun. 95(2), 1677–1691 (2017) 7. You, Z., Fang, J., Lu, I.-T.: Combination of spectral and SVD precodings for out-of-band leakage suppression. In: 2013 IEEE Long Island Systems, Applications and Technology Conference (LISAT) (2013). https://doi.org/10.1109/lisat.2013.6578238 8. Abdoli, M.J., Jia, M., Ma, J.: Filtered OFDM: a new waveform for future wireless systems. In: IEEE 16th International Workshop on Signal Processing Advances in Wireless Communications (SPAWC), pp. 66–70 (2015). https://doi.org/10.1109/SPAWC.2015. 7227001 9. Vakilian, V., Wild, T., Schaich, F., ten Brink, S., Frigon, J.-F.: Universal-Filtered MultiCarrier Technique for wireless systems beyond LTE. In: IEEE Globecom Workshops (GC Wkshps) (2013). http://ieeexplore.ieee.org/document/6824990/ 10. Schellmann, M., et al.: FBMC-based air interface for 5G mobile: challenges and proposed solutions. In: 9th International Conference on Cognitive Radio Oriented Wireless Networks and Communications (CROWNCOM) (2014). http://eudl.eu/doi/10.4108/icst.crowncom. 2014.255708 11. van de Beek, J.: Sculpting the multicarrier spectrum: a novel projection precoder. IEEE Commun. Lett. 13(12), 881–883 (2009) 12. Jiao, B., Guo, Y.J., Ma, M., Huang, X.: Optimal orthogonal precoding for power leakage suppression in DFT-based systems. IEEE Trans. Commun. 59(03), 844–853 (2011) 13. Strang, G.: The fundamental theorem of linear algebra. Am. Math. Mon. 100(9), 848–855 (1993) 14. Mohamad, M., Nilsson, R., van de Beek, J.: An analysis of out-of-band emission and inband interference for precoded and classical OFDM systems. In: Proceedings of 21st European Wireless Conference (2015) 15. Alphan, Ş., Arslan, H., Tom, A.: Suppressing alignment: joint PAPR and out-of-band power leakage reduction for OFDM-based systems. IEEE Trans. Commun. 64(3), 1100–1109 (2016)
Monitoring of Resources Used by Java Mobile Applications Laila Fal(&), Laila Moussaid, and Hicham Medromi Systems Architecture, Hassan II University, ENSEM, Casablanca, Morocco [email protected], [email protected], [email protected]
Abstract. To follow up my study done in a first article [1] concerning the analysis of the volume of classes and threads allocated by Java applications, as well as the CPU used in embedded systems, I focused in this article on the behavior of the same applications [1] in a comparative study of VMZ and VMO virtual machines, on a Linux environment [3]. After monitoring the resources consumed, we noticed that the memory management change between the two VMZ [4] and VMO [5] machines and that it depends on several factors namely: the operating system, the VM in place and the application executed. Keywords: Virtual machine Embedded systems Garbage collector (GC) Eclipse Memory management Java micro edition, linux (Ubuntu)
1 Introduction This article presents a continuity of the study previously carried out, it is a question of following the resources used during the execution of a mobile java application. For this, we kept the same execution conditions, and the same applications for a meaningful comparison on two different operating systems. Subsequently, the experimental results will be collected and presented in the form of graphs and tables. A summary followed by perspectives will close the article.
2 Experimental Results The figures below were taken on a run time of 30 s.
© Springer Nature Switzerland AG 2019 Y. Farhaoui and L. Moussaid (Eds.): ICBDSDE 2018, SBD 53, pp. 186–189, 2019. https://doi.org/10.1007/978-3-030-12048-1_20
Monitoring of Resources Used by Java Mobile Applications
2.1
Results with VMO
See Table 1.
Table 1. Result with VMO. Application First Aid App 1.0 vPoker Zip Utility Skype (officiel) 1 KM Player 1.0 X Ray Scanner Best 1.0 Mobile GMaps 1.0 Calculatrice scientifique
2.2
Classes 3203 3152 3238 3244 3201 3160 3268 3155
Number of threads CPU (%) 40 1.8 26 1.4 38 0.7 37 0.4 37 2.7 41 3.2 45 2.2 44 1.2
Results with VMZ
See Table 2.
Table 2. Result with VMZ. Application First Aid App 1.0 vPoker Zip Utility Skype (officiel) 1 KM Player 1.0 X Ray Scanner Best 1.0 Mobile GMaps 1.0 Calculatrice scientifique
2.3
Classes 3203 3152 3238 3244 3201 3160 3268 3155
Number of threads CPU (%) 40 1.8 26 1.4 38 0.7 37 0.4 37 2.7 41 3.2 45 2.2 44 1.2
Comparison
Comparison of the two results (Figs. 1, 2 and 3):
187
188
L. Fal et al.
Fig. 1. Comparison of classes in VMO and VMZ
Fig. 2. Comparison of threads in VMO and VMZ
Fig. 3. Comparison CPU (%) in VMO and VMZ
Monitoring of Resources Used by Java Mobile Applications
189
3 Discussion Looking at the result comparison scheme on both VMs, we can see that for the most monitored applications, the classes and threads allocated by the VMO are greater than or equal to those allocated in the VMZ. The GC activity monitoring tools can help us choose and set the working environment, they specify the number of collections made to free memory space. There are many memory management GCs, each more suited to one application category: the GC serial is the most suitable for applications that do not tolerate pause time, however, the GC parallel should be used when many must be done and long breaks are acceptable.
4 Conclusion and Perspectives At the end of this study, we saw that memory consumption changes according to the work environment and that the need largely defines the operating system and the VM to use. The results of this study can be the basis of a work to be continued and improved for a much more in-depth study that can be the subject of a doctoral thesis. Thus, future prospects are the precise location of probable causes of memory waste in order to create an optimal memory management algorithm (GC).
References 1. Analysis of the allocation of classes, threads and CPU used in embedded systems for Java applications. https://doi.org/10.1016/j.procs.2018.07.181 2. Eclipse (install et configuration). https://stackoverflow.com/questions/41954192/how-toinstall-eclipse-neon-version-for-32-bit-system-in-ubuntu-14-04 3. https://tecadmin.net/install-oracle-java-8-ubuntu-via-ppa/ 4. https://www.azul.com/downloads/zulu/ 5. https://docs.oracle.com/javase/8/docs/technotes/guides/install/linux_jdk.html
Anti-windup Compensation in TCP/IP Routers: A Multi-delay Feedback Systems Approach Nabil El Fezazi1(&), Ismail Er Rachid1, El Houssaine Tissir1, Fatima El Haoussi1,2, Teresa Alvarez3, and Fernando Augusto Bender4 1 LESSI, Department of Physics, Faculty of Sciences Dhar El Mehraz, University Sidi Mohammed Ben Abdellah, BP 1796, Fes-Atlas, Morocco [email protected] 2 Department of Mathematic and Informatics, Polydisciplinary Faculty of Nador, IASCM, BP 300 Selouane, 62700 Nador, Morocco 3 Department of Systems Engineering and Automatic Control, University of Valladolid, 47005 Valladolid, Spain 4 Center of Exact Sciences and Technology, Universidade de Caxias do Sul, R. Francisco Getulio Vargas 1130, Caxias do Sul, RS 95070-560, Brazil
Abstract. The design of anti-windup compensators for TCP/IP congestion in the presence of multiple classes of traffic is addressed here: a methodology is presented that takes into account the constraints to achieve the desired queue size, guaranteeing the stability even in the presence of disturbances. Our proposed technique is based on using LMIs that include a tuning parameter, which makes possible to enlarge the domain of attraction to ensure the stability in the varying conditions inherent to TCP/IP traffic. This is so far the first control theory based approach for anti-windup in router that allows multi class Active Queue Management, considering delays, input saturation and variable link bandwidth (modeled as a time-variant disturbance). These characteristics are relevant for framing the network within Service Level Agreements, as data traffic is not homogeneous, and transmission delays are variable. Keywords: Anti-windup Congestion Domain of attraction Control theory
1 Introduction The increased demand for the Internet to transmit time-sensitive voice and video applications require new congestion control algorithms that take into account flows of different characteristics when regulating the demand of the network’s resources. Active Queue Management (AQM) - based algorithms are frequently used in routers to inform TCP’s senders about congestion, so senders fit their sending rate to the network characteristics. This is inherently a feedback control issue, so significant research is network characteristics. This is inherently a feedback control issue, so significant research is been devoted to the use of control theory to develop more efficient congestion control algorithms. A simple one is Drop Tail, which acts by dropping packets © Springer Nature Switzerland AG 2019 Y. Farhaoui and L. Moussaid (Eds.): ICBDSDE 2018, SBD 53, pp. 190–203, 2019. https://doi.org/10.1007/978-3-030-12048-1_21
Anti-windup Compensation in TCP/IP Routers
191
arriving at a router when the buffer is full. As emphasized in [1], this leads to flowsynchronization and performance degradation due to excessive time outs and restarts. As an improvement, RED was proposed [2–4]: it uses the EWMA of the queue length as a congestion indicator to drop/mark packets in order to achieve good balance between throughput and queuing delay. However, RED suffers from the lack of robustness with respect to tuning parameters which may easily lead to queue oscillation and instability. Different forms of controllers have been proposed in the literature based on Control Theory: for example, [5, 6] use the phase margin specification in the frequency domain to tune PI(D) controller gains. It must be pointed out that many of these Congestion Control methodology proposed in the literature based on Control Theory ignore the inherent time-varying delays, which is not adequate in practice: the packet drop/mark probability only has its effect on the senders after the Round Trip Time (RTT). It is then important to design controllers that do not ignore this delay, as there is a significant effect. Furthermore, the control variable (drop probability) is subject to nested saturation, which is not sufficiently considered by the previous works, degrading the performance [7–13]. In this sense, [7, 8, 11] proposed delay dependent anti-windup techniques to mitigate the saturation effect on stability of such delayed systems and to ensure closed loop stability of TCP/IP AQM routers. However, these works consider the traffic homogeneous, i.e. they do not differentiate one TCP flow from another, but there is a tendency in combining different classes of traffic in Internet links that have different requirements. For example, there are primarily secured data connections for allowing financial and e-banking transfers (where packets should not be dropped), combined with VoIP, that performs badly under variable delays, and standard e-mail or web page traffic. Using a unique, single delay AQM policy for the whole traffic does not address the Service Level Agreements (SLA) that telecom carriers and internet providers hold with high demanding customers nowadays. Thus, this is addressed in present work, considering multiple classes of traffic through the router, each class with its own packet discarting policy, queue size, bandwidth share, and constant RTT, to outpace the previous state of the art in this matter.
2 Preliminaries In this section, we introduce notations and lemmas (see [14, 15]) which are used in this paper. Thus, kðPÞ denotes the maximal eigenvalue n o of matrix P. The saturation function is satðycðkÞ ðtÞÞ ¼ signðycðkÞ ðtÞÞmin ycðkÞ ðtÞ; u0ðkÞ , Finally, k ¼ 1; . . .; m. Lemma 1. The following inequality holds for any matrices R [ 0; N1 ; N2 , and delay s: Z
t
tsðtÞ
n_ T ðsÞRn_ ðsÞds
nð t Þ nðt sðtÞÞ
T
N 2 1 N2
T ! nð t Þ N1 N1 1 N1 þ sðtÞ R nðt sðtÞÞ N2 N2 N2
192
N. El Fezazi et al.
Lemma 2. For a matrix GT , we define the following polyhedral set: n o S ¼ nðt sÞ