Advances in Smart Vehicular Technology, Transportation, Communication and Applications: Proceedings of VTCA 2022 9819908477, 9789819908479

Citation preview

Smart Innovation, Systems and Technologies 347

Shaoquan Ni · Tsu-Yang Wu · Jingchun Geng · Shu-Chuan Chu · George A. Tsihrintzis   Editors

Advances in Smart Vehicular Technology, Transportation, Communication and Applications Proceedings of VTCA 2022

123

Smart Innovation, Systems and Technologies Volume 347

Series Editors Robert J. Howlett, KES International Research, Shoreham-by-Sea, UK Lakhmi C. Jain, KES International, Shoreham-by-Sea, UK

The Smart Innovation, Systems and Technologies book series encompasses the topics of knowledge, intelligence, innovation and sustainability. The aim of the series is to make available a platform for the publication of books on all aspects of single and multi-disciplinary research on these themes in order to make the latest results available in a readily-accessible form. Volumes on interdisciplinary research combining two or more of these areas is particularly sought. The series covers systems and paradigms that employ knowledge and intelligence in a broad sense. Its scope is systems having embedded knowledge and intelligence, which may be applied to the solution of world problems in industry, the environment and the community. It also focusses on the knowledge-transfer methodologies and innovation strategies employed to make this happen effectively. The combination of intelligent systems tools and a broad range of applications introduces a need for a synergy of disciplines from science, technology, business and the humanities. The series will include conference proceedings, edited collections, monographs, handbooks, reference books, and other relevant types of book in areas of science and technology where smart systems and technologies can offer innovative solutions. High quality content is an essential feature for all book proposals accepted for the series. It is expected that editors of all accepted volumes will ensure that contributions are subjected to an appropriate level of reviewing process and adhere to KES quality principles. Indexed by SCOPUS, EI Compendex, INSPEC, WTI Frankfurt eG, zbMATH, Japanese Science and Technology Agency (JST), SCImago, DBLP. All books published in the series are submitted for consideration in Web of Science.

Shaoquan Ni · Tsu-Yang Wu · Jingchun Geng · Shu-Chuan Chu · George A. Tsihrintzis Editors

Advances in Smart Vehicular Technology, Transportation, Communication and Applications Proceedings of VTCA 2022

Editors Shaoquan Ni School of Transportation and Logistics Southwest Jiaotong University Chengdu, Sichuan, China Jingchun Geng China Railway Economic and Planning Research Institute Beijing, China George A. Tsihrintzis Department of Informatics University of Piraeus Piraeus, Greece

Tsu-Yang Wu College of Computer Science and Engineering Shandong University of Science and Technology Qingdao, Shandong, China Shu-Chuan Chu College of Computer Science and Engineering Shandong University of Science and Technology Qingdao, Shandong, China

ISSN 2190-3018 ISSN 2190-3026 (electronic) Smart Innovation, Systems and Technologies ISBN 978-981-99-0847-9 ISBN 978-981-99-0848-6 (eBook) https://doi.org/10.1007/978-981-99-0848-6 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd. The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore

Preface

This volume composes the proceedings of the Fifth International Conference on Smart Vehicular Technology, Transportation, Communication and Applications (VTCA 2022), which is hosted by Shandong University of Science and Technology and is held in the virtual conference on December 24–26, 2022. VTCA 2022 is technically co-sponsored by Springer, Shandong University of Science and Technology, Southwest Jiaotong University (SWJTU), Fujian University of Technology (FJUT), Superconductivity and New Energy R&D Center (SWJTU), Fujian Provincial Key Lab of Big Data Mining and Applications, and National Demonstration Center for Experimental Electronic Information and Electrical Technology Education (FJUT). It aims to bring together researchers, engineers, and policymakers to discuss related techniques, to exchange research ideas, and to make friends. Forty-six regular papers were accepted in this proceeding. We would like to thank the authors for their tremendous contributions. We would also express our sincere appreciation to the reviewers and program committee members for making this conference successful. Finally, we would like to express our special thanks for the great help from all sponsors in organizing the conference. Chengdu, China Qingdao, China Beijing, China Qingdao, China Piraeus, Greece December 2022

Shaoquan Ni Tsu-Yang Wu Jingchun Geng Shu-Chuan Chu George A. Tsihrintzis

v

Contents

Part I 1

2

3

4

5

6

7

8

Smart Transportation Systems and Technologies

Research on Quality Management of Urban Rail Transit Vehicle Frame Overhaul Project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Zhong-De Zou, Ding Chen, and Bin Hu

3

Modeling and Analysis of Railway Passenger Flow Forecast During the Spring Festival . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Zhi-Cheng Zhang, Ding Chen, and Pei-Zhou Jiang

15

Seasonal and Period Division Method for Dynamic Passenger Flow of High-Speed Railway . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hui Han, Xinqian Zou, Yiyuan Gao, and Xiuyun Guo

27

A Study of High-Speed Railway Train Merger and Adjustment Based on Regional Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fangyu Shi, Zhi Wu, Haowen Tan, Min Yang, and Jinshan Pan

43

Research on Platform Door Setting of Suburban Railway of Mass Transit Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Zhang Wenxin, Yang Yilin, Song Qingguo, Zhu Chengli, and Shaoquan Ni Research on Equipment Operation and Maintenance Management Technology of Large Railway Passenger Station . . . . . Bozhou Wang, Lexi Li, Shaoquan Ni, and Dingjun Chen Research on Adaptability Evaluation Between Express and Local Train Operation Plan of Urban Rail Transit and Passenger Flow Demand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Tan Li Research on the Network Operation Mode of High-Speed Rail Express . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Yongcheng Wang, Yi Li, Yunhao Sun, and Tao Chen

57

67

79

93

vii

viii

9

Contents

Solving a Locomotive Routing Problem of Heavy Haul Railways . . . 111 Yongxin Li, Meng Wang, Zhen Liu, Chi Zhang, and Xueting Li

10 A Study of Optimization of Dynamic Freight Train Diagrams Based on Market-Orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133 Meng Wang, Fangyu Shi, Ziqi Dong, and Hongxia Lu 11 Research on Equipment Management System of Railway Passenger Station Based on High-Precision Positioning . . . . . . . . . . . 147 Lexi Li, Bozhou Wang, Zhen Liu, and Shaoquan Ni 12 Design of an Integrated System for the Train Working Diagram of Urban Rail Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 Fan Gao, Xu Chen, Xiaoxu Zeng, Li Bai, Xiaohe Song, and Xuze Ye 13 Research on Optimization of Operation Organization of Transship Trains in Railway Hub . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175 Zongying Song, Yi Li, Mengyuan Yue, Kun Liu, and Miaomiao Lv 14 Optimization Principle of Freight Train Operation Plan for Shenhua Railway . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189 Meng Wang, Qiuqi Liu, Wenhui He, and Xiuyun Guo Part II

Smart Vehicular Electronics, Networks, and Communications

15 Resource Recovery Vehicle Picking Up Resource Recovery Bin Robot Arm Structure Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203 Yu-Yang Yuan, Yi-Jui Chiu, Wen-Qi Yang, and Yung-Hui Shih 16 Reconfigurable Multibody Space Systems Based on Magnetic Flux Pinning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219 Lifeng Zhao, Qingyun Mao, Bo Zhang, Pei Wang, Jun Tao, Haige Qi, Jin Jiang, Yong Zhang, and Yong Zhao 17 Research on Supply Chain Financing Mode of New Energy Vehicle Industry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229 Cheng-Xiao Ju, Hui-Jun Xiao, and Mei-Feng Chen 18 Design of Intelligent Baby Walker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239 Yong Wu, Yi-Jui Chiu, Tian-Hang Deng, and Yung-Hui Shih 19 Research on the Method of Handling Missing ETC Transaction Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251 Songyang Wu, Fumin Zou, Feng Guo, Qiqin Cai, and Yongyu Luo 20 Highway Traffic Volume Prediction Based on GRU and Attention by ETC Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263 Shibin Huang, Fumin Zou, Feng Guo, and Qiang Ren

Contents

ix

21 Traffic Flow Prediction of Expressway Toll Station Exit Based on ETC Gantry Data and Attention Mechanism . . . . . . . . . . . . . . . . . . 277 Haolin Wang, Fumin Zou, and Feng Guo 22 Expressway Short-Term Traffic Flow Forecasting Considering Spatio-Temporal Features of ETC Gantry . . . . . . . . . . . . . . . . . . . . . . . 291 Gen Xu, Fumin Zou, Junshan Tian, Feng Guo, and Qiqin Cai Part III Artificial Intelligence—Innovation Technologies 23 Objectionable Image Content Classification Using CNN-Based Semi-supervised Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311 Shukla Mondal, Arup Kumar Pal, SK Hafizul Islam, and Debabrata Samanta 24 Software and Hardware Cooperative Implementation of the Rafflesia Optimization Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . 321 Zonglin Fu, Jeng-Shyang Pan, Yundong Guo, and Václav Snášel 25 A Hybrid Orthogonal Learning and QUATRE Algorithm Based on PPE Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333 Lulu Liang, Shu-Chuan Chu, Tien-Szu Pan, and Tsu-Yang Wu 26 Research on Gannet Optimization Algorithm and Its Application in Traveling Salesman Problem . . . . . . . . . . . . . . . . . . . . . . 343 Jeng-Shyang Pan, Fei-Fei Liu, Jie Wu, Tien-Szu Pan, and Shu-Chuan Chu 27 Artificial Hummingbird Algorithm with Parallel Compact Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353 Shu-Chuan Chu, Zhi-Yuan Shao, Chin-Shiuh Shieh, and Xiaoqing Zhang 28 Usability Testing Study of Meal Management APP for the Elderly Based on SHERPA and FMEA . . . . . . . . . . . . . . . . . . . 363 Li Hanji and Huang Jingjing 29 Directed Point Clouds Denoising Algorithm Based on Self-learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373 Yijie Fan, Linlin Tang, Yang Liu, and Shuhan Qi 30 NIST: Learning Neural Implicit Surfaces and Textures for Multi-view Reconstruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385 Xin Huang, Linlin Tang, Yang Liu, Shuhan Qi, Jiajia Zhang, and Qing Liao 31 Architecture Design of Equipment Warehouse Scheduling System Based on Software Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397 Xue Ting Zhang, Yan Peng Pan, Li Jie Yang, Chen Chen Xue, and Fu Quan Zhang

x

Contents

32 Multi-objective Firefly Algorithm for Hierarchical Mutation Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411 Zhi-bin Song, Ren-xian Zeng, Ping Kang, and Li Lv 33 DUWP: A Dynamic Unmanned Warehouse Partition Model for Balancing Commodity Allocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425 Ben Li, Lyuchao Liao, Wenqing Zhao, Hankun Xiao, and Youpeng He 34 Density Peaks Clustering Algorithm for Manifold Data Based on Geodesic Distance and Weighted Nearest Neighbor Similarity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439 Xin-Yue Hu, Jia-Zheng Hou, Run-Xiu Wu, and Jia Zhao 35 Optimizing the Layout of Nucleic Acid Test Sites for COVID-19 Based on Gannet Optimization Algorithm . . . . . . . . . 453 Ruo-Bin Wang, Rui-Bin Hu, Fang-Dong Geng, and Lin Xu 36 Two Factors that Influence Our Selection of Digital Avatars: Gender Performativity and Historical Culture . . . . . . . . . . . . . . . . . . . 463 Shutang Liu, Yongyu Li, Minggui Li, Yin Guan, Hang Jiang, and Lei Jiang Part IV Cybersecurity Threats and Innovative Solutions 37 Path Planning Method of UAV Cluster Against Forgery Attack Under Differential Boundary Constraint . . . . . . . . . . . . . . . . . 479 Jianchen Wang, Yanlong Li, Yabin Zhang, Jianjun Wu, Wei Sun, and Xuyang Zhou 38 To Analyze Security Requirements of Two AKA Protocols in WBAN and VANET . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 491 Haozhi Wu, Saru Kumari, and Tsu-Yang Wu 39 A Method of Expressway Congestion Identification Based on the Electronic Toll Collection Data . . . . . . . . . . . . . . . . . . . . . . . . . . . 501 Ziyang Lin, Fumin Zou, Feng Guo, Xiang Yu, Nan Li, and Chenxi Xia 40 Privileged Insider Attacks on Two Authentication Schemes . . . . . . . . 515 Yiru Hao, Saru Kumari, Kuruva Lakshmanna, and Chien-Ming Chen 41 Secure Communication in Digital Twin-enabled Smart Grid Platform with a Lightweight Authentication Scheme . . . . . . . . . . . . . . 525 Jiaxiang Ou, Mi Zhou, Houpeng Hu, Fan Zhang, Hangfeng Li, Fusheng Li, and Pengcheng Li

Contents

xi

42 A Secure Authentication Scheme for Smart Home Based on Trusted Execution Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 537 Houpeng Hu, Jiaxiang Ou, Bin Qian, Yi Luo, Yanhong Xiao, and Zerui Chen 43 Comments on “Two Authentication and Key Agreement Protocols in WSN Environments” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 551 Fangfang Kong, Saru Kumari, and Tsu-Yang Wu 44 Security Analysis of Two Authentication and Key Agreement Protocols Based on Wireless Sensor Networks . . . . . . . . . . . . . . . . . . . . 563 Liyang Wang, Saru Kumari, and Tsu-Yang Wu 45 Face Mask Detection Based on YSK Neural Network for Smart Campus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 575 Li Yu Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 585

About the Editors

Prof. Shaoquan Ni has served as the director of the Institute of Transportation Information Technology since 2000 and the executive deputy director of the Nationwide Railway Train Operation Diagram Compilation Research and Training Center since 2010. He is a member of the Academic Committee of Southwest Jiaotong University and one of the outstanding experts with prominent contributions in Sichuan Province. Professor Ni has demonstrated a long-term commitment to the study of computer-aided train operation diagram compiling and is now the leader in this field in China. He creatively proposed the “single-line and double-line in one” computer graphics unified algorithm ideas and established a set of technical methods based on the group coordination of the train operation diagram. And the railway train operation diagram compiling system presided over by Prof. Ni was recognized by the Ministry of Railways (China Railway Corporation) as a unified computer-aided train operation diagram compiling software, which was the technical means for compiling and managing the national railway train operation diagram in the past decade. And he has received the Zhan Tianyou Railway Science Achievement Award, Mao Yisheng Railway Science and Technology Award, and nearly 20 provincial or ministerial scientific and technological progress awards. Tsu-Yang Wu received the Ph.D. degree in the Department of Mathematics, National Changhua University of Education, Taiwan (R.O.C.). Currently, he is an associate professor at the College of Computer Science and Engineering, Shandong University of Science and Technology, China. In the past, he was an assistant professor at Harbin Institute of Technology, Shenzhen Campus, and an associate professor at Fujian University of Technology, China. He serves as executive editor in the Journal of Network Intelligence (Scopus, EI). Recently, Dr. Wu selected 2020 Single Year World’s Top 2% Scientists. His research interests include cryptography and network security.

xiii

xiv

About the Editors

Jingchun Geng is now working in China Railway Economic and Planning and Research Institute as a senior engineer, a national registered consulting engineer (investment), and a constructor (railroad), mainly engaged in research, consultation and review of railway transportation organization, logistics, and investment and financing and economic evaluation. In recent years, he has presided over more than 100 national major railway projects, more than 10 provincial and ministerial scientific research projects, and more than 20 bureau-level scientific research projects and won 5 provincial and ministerial science and technology awards, 1 excellent engineering design award, 7 national and provincial excellent consulting achievements, and nearly 20 bureau-level scientific and technological progress awards and excellent consulting achievements awards. He has published more than 30 academic papers in core journals such as Journal of Railway and Journal of Railway Engineering, has won 1 national invention patent and 3 software copyrights, and was awarded the first level candidate of “131” innovative talents training project in Tianjin in 2018, and the youth award of the 14th Zhan Tianyou Railway Science and Technology Award in 2019. Shu-Chuan Chu received a Ph.D. degree in 2004 from the School of Computer Science, Engineering and Mathematics, Flinders University of South Australia. She joined Flinders University in December 2009 after 9 years at the Cheng Shiu University, Taiwan. She has been a research fellow and associate professor in the College of Science and Engineering of Flinders University, Australia, since December 2009. Currently, she is a research fellow with a Ph.D. advisor in the College of Computer Science and Engineering of Shandong University of Science and Technology since September 2019. She also serves as an editorial board member for Engineering Applications of Artificial Intelligence (EAAI), Journal of Internet Technology (JIT), and Research Reports on Computer Science (RRCS). Her research interests are mainly in Evolutionary Computation, Intelligent Computing, Information Hiding, and Wireless Sensor Networks. George A. Tsihrintzis is a full professor in the Department of Informatics in the University of Piraeus, Greece, and served as its head from September 2016 through August 2020. He received the Diploma of Electrical Engineer from the National Technical University of Athens, Greece (with honors) and the M.Sc. and Ph.D. degrees in Electrical Engineering from Northeastern University, Boston, Massachusetts, USA. His current research interests include Pattern Recognition, Machine Learning, Decision Theory, and Statistical Signal Processing and their applications in Multimedia Interactive Services, User Modeling, Knowledge-based Software Systems, Human– Computer Interaction, and Information Retrieval. He has authored or co-authored over 350 research publications in these areas, which include 6 monographs and 35 edited volumes.

Part I

Smart Transportation Systems and Technologies

Chapter 1

Research on Quality Management of Urban Rail Transit Vehicle Frame Overhaul Project Zhong-De Zou, Ding Chen, and Bin Hu

Abstract Rail transit vehicle frame overhaul and maintenance is an important work to ensure the safe and reliable operation of vehicle equipment. To solve the problems of different frame overhaul modes and resource optimization of rail transit operation, this paper summarizes the frame overhaul and maintenance of major cities, and proposes a quality management method for urban rail transit vehicle frame overhaul project. The method is composed of a combination strategy, critical path method and PDCA cycle theory. The result shows that the method saves 4 day and 1 day maintenance times, respectively, by optimizing failure mode 1 and failure mode 2, which means it can improve maintenance efficiency, reduce maintenance cost and provide theoretical for the optimization of enterprise vehicle frame overhaul workshop.

1.1 Introduction The social Internet of vehicles provides better resources and services for the development of the Internet of vehicles and provides a better experience for users [1–5]. It is conducive to promoting the development of vehicle frame overhaul. Compared with other vehicle maintenance methods, the urban rail transit vehicle frame overhaul has the characteristics of large maintenance depth, high maintenance level, and uncontrollable cost [6]. Local rail transit operating companies adopt different frame overhaul modes, and mature rail transit operating companies are using with certain operating experience, stable and reliable maintenance mode [7]. Summarizing the Z.-D. Zou · D. Chen (B) School of Mechanical and Automotive Engineering, Xiamen University of Technology, No. 600, Ligong Rd, Xiamen 361024, Fujian Province, China e-mail: [email protected]; [email protected] Z.-D. Zou e-mail: [email protected] B. Hu Xiamen Rail Transit Group Co. Ltd., Xiamen 361010, Fujian Province, China © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_1

3

4

Z.-D. Zou et al.

frame overhaul modes and formulating scientific organization mode and maintenance regulation will help to coordinate the conflict between the maintenance quality, schedule, and cost of the frame overhaul, so as to improve operation efficiency and reduce operation cost. At present, vehicle frame overhaul organization is mainly divided into autonomous maintenance and outsourced maintenance two ways [8]. In today’s rail transit network operation, autonomous maintenance of the project management characteristics and laws is beginning to show incompatibilities with modern management methods [9]. Outsourced maintenance limits the operation unit’s mastery of the principle and performance of vehicle technology, and reduces the ability to control maintenance progress and quality. In the current rail transit equipment system integration, components of the whole life cycle factors complex background, the method will affect the daily maintenance of the fault judgment, are not conducive to the long-term development of operating enterprises [10]. In the frame overhaul and maintenance regulation, there is usually a strong correlation between the maintenance items, in order to reduce the maintenance cost and improve the maintenance efficiency, the formulation of the maintenance regulations needs to introduce the corresponding project quality management method for resource allocation and cost assessment [11]. Critical path method (CPM) is a representative project management method, which is widely used in engineering, scientific research, and enterprise management projects. It is also the main method used in the schedule and process refinement of urban rail transit vehicle project management [12]. In the quality management method, the cause–effect diagram is used to find out the main factors and secondary factors affecting the quality of the overhaul of the frame, and the causal relationship between the influencing factors is analyzed to obtain the key factors [13]. Analysis and quality control through PDCA cycle for factors affecting frame overhaul quality [6], ultimately achieve full use of vehicle life cycle efficiency, reduce operating and maintenance costs, and fine management purposes.

1.2 Vehicle Frame Overhaul Quality Management Method Based on the current situation of vehicle frame overhaul mode, from the perspective of project management, the essential problem of urban rail transit vehicle frame overhaul is transformed into the problem of project implementation path and its overall arrangement. The variables involved include the number, condition, and cost of implementation points. According to the enterprise’s own conditions, the reasonable design of the implementation path of the overhaul project is an effective means to realize the personalized customization of enterprise maintenance.

1 Research on Quality Management of Urban Rail Transit Vehicle Frame …

5

Pre-repair registration inspection

The system components overhaul

Vehicle each working condition debugging

Vehicle general assembly

Start overhaul Vehicle acceptance inspection Vehicle delivery and use

Fig. 1.1 Flow chart of vehicle frame overhaul project plan management

1.2.1 Maintenance Regulation for Vehicle Frame Overhaul Vehicle overhaul regulation is based on the preventive planned maintenance system, adopting the project quality management method that decomposes, sorts, and executes the maintenance items and dates according to the regulation. According to the difference in vehicle design demand, the frame overhaul repair program level regulation and cycle are generally divided into frame repair 5 years, 600,000 km or 4 years 500,000 km, overhaul 10 years, 1.2 million kilometers or 8 years 1 million kilometers to divide, such as Guangzhou and Nanjing metro companies. The content of the regulation is to determine the scope of maintenance, maintenance methods, maintenance standards, and other regulation as the main content, assist the process documents and instructions to implement the project plan management process, and repair the vehicle failure until the assembly delivery, as shown in Fig. 1.1.

1.2.2 Methods of Project Management The CPM realizes the control of project costs and expenses by assuming a definite value for the time of each process as shown in Fig. 1.2. Fig. 1.2 Critical path method path network diagram

20 LE EF 1 LS LF

40 ES EF 2 LS LF 10 ES EF 3 LS LF

20 ES EF 4 LS LF

15 ES EF 5 LS LF

6

Z.-D. Zou et al.

ES, EF, LS, LF, and TF in the diagram indicate the earliest start time, earliest end time, latest start time, latest end time, and total time, respectively. The numbers inside the network diagram are the project activity codes, the numbers above the network diagram are the project activity time. The ES and EF in each activity are derived by the forward extrapolation method; the LS and LF in each activity are derived by the backward extrapolation method, the TF is the difference between the latest end time and the earliest start time of the activity, the scientific computing software can be used to design the critical path algorithm that derives the critical path of the project activity.

1.2.3 Methods of Quality Management Safety, reliability, stability, technical state, and comfort level for the vehicle frame overhaul quality evaluation of five requirements. The quality evaluation indexes of urban rail transit vehicles are determined in two directions: qualitative and quantitative. Among them, qualitative quality evaluation is controlled by means of hierarchical analysis or grade classification, and quantitative quality evaluation can be determined based on theoretical calculation and data detection. Human factors, machine life, material selection, operating methods, and operating environment can cause different degrees of influence on the quality of frame overhaul. Taking the analysis of the cause of air conditioning debugging failure of Beijing Metro Line 4 as an example, the cause–effect diagram is used to draw the influencing factors of air conditioning failure and their correlation, as shown in Fig. 1.3, as a framework to find the source of the problem and complete the troubleshooting and quality improvement of air conditioning. PDCA cycle enables cycle optimization of deficiencies at existing maintenance levels and step-by-step output of optimization results, as shown in Fig. 1.4. In the

Sense of responsibility

Operation process

Lack of skills

Overhaul procedure Organizational Flow

Staff Experience

Overhaul Process

Operation error

Number of equipment Tools and equipment

Factors influencing the quality of rack overhaul

Unqualified quality

Laboratory temperature Unreasonable layout

Improper selection Untimely procurement

Too much dust

Equipment performance Quality of spare parts

Site confusion

Fig. 1.3 Vehicle frame overhaul failure analysis cause–effect diagram

1 Research on Quality Management of Urban Rail Transit Vehicle Frame …

7

Fig. 1.4 PDCA cycle schematic

rail transit industry, taking the cities that have opened metro lines first as experience reference, the maintenance regulation system that meets the requirements of their own enterprises through cyclic iteration is the key to generate advanced technical knowledge and reasonably promote scientific maintenance modes such as balanced repair. This content has been paid attention to by domestic metro operating companies such as Beijing, Shanghai, and Guangzhou. Combining the maintenance points for vehicles in the overhaul project of vehicle frame in Fig. 1.1 with the PDCA cycle, the quality management improvement of the overhaul project of rail transit vehicle frame can be divided into seven main steps, as shown in Table 1.1. The maintenance points are based on the principle of cyclic iteration to find deficiencies and improvements, and finally, determine the quality indicators. In this macro framework, introduction of cause–effect as a method of troubleshooting and tracing fault-associated factors, the optimal route between the execution of each point of the frame overhaul is determined by the critical path method, thus building a combined strategy for the project quality management of the frame overhaul, whose schematic diagram is shown in Fig. 1.5. This paper further develops the case study of the frame overhaul project with this combined strategy and verifies the feasibility of the method.

8

Z.-D. Zou et al.

Table 1.1 Steps of PDCA cycle in project quality management of vehicle frame overhaul PDCA

Step

P: Plan

1. Investigate the current situation and clarify the problem 2. Analysis of the causes of quality problems 3. Formation of control planning, measures

D: DO

4. Implementation of the plan and the implementation of measures

C: Check

5. Check

A: Act

6. Summary, modify, or develop new indicators 7. Carry the remaining problems into the next cycle

Fig. 1.5 The project quality management combination strategy for the frame overhaul diagram

Start

Quality index determination

Survey and develop related plans

Implementation plan Work breakdown Determine work hours

Critical path method

Optimal process content

PDCA

Shortcomings and improvements

Quality check

Start

1.3 Case Study on Project Quality Management of Frame Overhaul 1.3.1 Fault Determination and Quality Measures for Pants As an auxiliary electrical equipment of subway vehicles, battery is an important guarantee for the safe and stable operation of subway vehicles. This paper takes the battery maintenance project in the overhaul project of a subway company as a case study of project quality management and carries out the verification and result analysis of the quality management combination strategy of the overhaul project.

1 Research on Quality Management of Urban Rail Transit Vehicle Frame …

9

According to the combination strategy of overhaul project quality management, the quality index and standard of battery should be determined first. In order to ensure the safety of subway operation, the battery needs to meet the performance requirements of low price, low memory effect, fast charging ability, high discharge depth, long service life, good low-temperature performance, high overcharge and discharge ability, and strong reliability. The power scatter plot is drawn using the SOC (percentage of remaining capacity to fully charged capacity) detection values of the metro vehicle battery in the recent 1 year, as shown in Fig. 1.6. The three curves from left to right in Fig. 1.6 represent the monitored values of the battery SOC of the metro vehicle under normal operating conditions, fault mode 1 and fault mode 2 from January 1, 2019, to December 31, 2019, respectively. In the figure, the SOC value in failure mode 1 suddenly decreases, indicating that the battery is damaged by puncture during operation. The SOC value in failure mode 2 decreases faster, indicating that the quality of the plate is problematic. Targeted quality control measures are proposed based on the cause of the failure: Failure Mode 1 requires a strict external inspection of the battery to identify the cause, troubleshoot, and repair the failure of the enclosure and cover plate, and replace the battery if it cannot be recovered. Fault mode 2 needs to check the plate fault problem, measure the plate parameters to determine the fault location, and regularly measure the battery internal resistance; for the problem of capacity decline, it is necessary to test the electrolyte, replace the plate that does not meet the conditions in time, replace the plate that cannot restore the capacity, check whether the battery has a fault, check

Fig. 1.6 Scatter plot of SOC test values of a metro battery in the last year of a metro company

10

Z.-D. Zou et al.

Battery disassembly 1 Battery cleaning 1 External inspection 1

Battery charge and discharge test 1

Battery capacity test 2 Electrolyte level test 1 Electrolyte density test 1

Inspection and acceptance 2

Battery replacement 1 Electrolyte renewal 1 Pole plate renewal 1 Battery maintenance 2

Battery installation 1

Fig. 1.7 Process relationship diagram of battery overhaul project

whether the battery setting has a short circuit or wiring error, and if it exists, take it out and charge the monomer. If it cannot be restored, replace the monomer battery.

1.3.2 Determine the Content of the Process The cause of battery failure has now been determined, the battery work time, cost, and resources of the storage battery need to be determined according to the project quality management combination strategy, the minimum project decomposition of each maintenance project of the single battery and its process is represented by the relationship diagram, as shown in Fig. 1.7. The numbers in the grid in Fig. 1.7 represent the working days of the corresponding processes. The critical path of battery overhaul obtained by critical path method as follows: battery disassembly–battery cleaning–external inspection–battery charging and discharging test–electrolyte height test–electrolyte density test–inspection and acceptance–battery repair–battery installation, with a total working time of 13 days.

1.3.3 Quality Testing and PDCA Cycle Quality testing was performed to determine if the quality of work was acceptable based on the project quality management combination strategy for the frame overhaul. Failure mode 1 battery breakage problem has been solved, the breakage index has been improved to normal level, the SOC value after repair is more than 10% higher than before repair, the anti-pollution, maintenance, adaptability to ambient temperature and charge and heat conditions have been improved. In failure mode 2, the plate problem and capacity problem have been solved, the plate quality has been

1 Research on Quality Management of Urban Rail Transit Vehicle Frame …

11

improved to normal level, the SOC value after repair is more than 10% higher than that before repair, the anti-pollution, maintenance, adaptability to ambient temperature and charging situation have been improved. Based on the quality inspection results, the critical path method was used to optimize the process and time of the two failure modes and to develop targeted process maps, as shown in Figs. 1.8 and 1.9. Failure mode 1 battery is breakdown damage, main external inspection of the battery, battery capacity, pole plate quality inspection, and other irrelevant processes can be omitted. If the external inspection is not qualified and the quality requirement cannot be reached after repair, the battery will be replaced directly. The critical path algorithm finds out the critical path of failure mode 1 is battery disassembly–battery cleaning–external inspection–battery repair–battery replacement– battery installation, the total time is 9 days, compared with the time cost saving of 4 days before optimization. Failure mode 2 is mainly due to the failure of the pole plate resulting in insufficient battery capacity, the battery usually fails because the work of charge/discharge test and capacity test does not meet the requirements. Raising the priority of these two tasks can make these two tasks get enough attention to ensure the completion

Battery replacement

Battery cleaning Inspection and acceptance

Battery disassembly

Battery maintenance

External inspection

Battery installation

Fig. 1.8 Process relationship diagram of the battery project for failure mode 1

Battery disassembly

Battery charging and discharging test

Electrolyte height test Electrolyte density test Battery capacity test

Cleaning inspection Inspection and acceptance

Battery replacement Pole plate renewal Battery repair Electrolyte renewal

Fig. 1.9 Process relationship diagram of the battery project for failure mode 2

Battery Installation

12

Z.-D. Zou et al.

quality of both and finally improve the completion quality of the rack overhaul. The critical path algorithm finds out the critical path of failure mode 2 as follows: battery disassembly–battery charge/discharge test–electrolyte height test–electrolyte density test–cleaning, inspection–inspection and acceptance–battery repair–battery installation, with a total duration of 12 days, saving 1 day of time cost compared with before optimization.

1.4 Conclusion This paper discusses the project quality management methods that can be applied in the urban rail transit vehicle frame overhaul sector and proposes a reasonable vehicle frame overhaul project quality management combination strategy. Through the case analysis of the battery project in the subway frame overhaul, the feasibility of the quality combination strategy of the vehicle frame overhaul project was successfully verified. The results show that this project quality management combination strategy is suitable for enterprises with a certain scale of operation to build or expand the overhaul workshop, such as Guangzhou Metro and Shenzhen Metro, to share the construction cost in the future longer operation time. In this organization, further improvement of maintenance mode can maximize the quality of maintenance. Acknowledgements This project is sustained by the Natural Science Foundation of Fujian Province (No: 2021J011202), the Science and Technology Research Project of Xiamen University of Technology (No: YKJ19021R; No. XPDKQ20010), and the Graduate Science and Technology Innovation Projects (No: YKJCX2021028).

References 1. Li, Z., Miao, Q., Chaudhry, S.A., Chen, C.-M.: A provably secure and lightweight mutual authentication protocol in fog-enabled social Internet of vehicles. Int. J. Distr. Sensor Netw. 18(6) (2022) 2. Chen, C.-M., Xiang, B., Liu, Y., Wang, K.-H.: A secure authentication protocol for Internet of vehicles. IEEE Access 7, 12047–12057 (2019) 3. Wu, T.-Y., Lee, Z., Yang, L., Chen, C.-M.: A provably secure authentication and key exchange protocol in vehicular ad hoc networks. Secur. Commun. Netw. 2021, 9944460 (2021) 4. Wu, T.-Y., Lee, Z., Yang, L., Luo, J.-N., Tso, R.: Provably secure authentication key exchange scheme using fog nodes in vehicular ad-hoc networks. J. Supercomput. 77, 6992–7020 (2021) 5. Kumar, V., Kumar, R., Kumar, V., Kumari, A., Kumari, S.: RAVCC: robust authentication protocol for RFID based vehicular cloud computing. J. Netw. Intell. 7(3), 526–543 (2022) 6. Lopez, P., Centeno, G.: Integrated system to maximize efficiency in transit maintenance departments. Int. J. Product. Perform. Manage. 55(8), 638–654 (2006) 7. Hu, W., Zhang, S.: Determining Types of Vehicle Maintenance Customers Based on Normal Cloud Model. IEEE (2010) 8. Wu, S.: Assessing maintenance contracts when preventive maintenance is outsourced. J. Reliab. Eng. Syst. Saf. 98(1), 66–72 (2012)

1 Research on Quality Management of Urban Rail Transit Vehicle Frame …

13

9. Kathleen, E.M., Elliotn, N.W.: TPM: Planned and autonomous maintenance: Bridging the GAP between practice and research. Product. Oper. Manage. (2009) 10. Yamín, R.A., Harmelink, D.J.: Comparison of linear scheduling model (LSM) and critical path method (CPM). J. Constr. Eng. Manage. 127(5), 374–381 (2001) 11. Sime, M., Bailey, G., Hajj, E.Y., Chkaiban, R.: Road load based model for vehicle repair and maintenance cost estimation. Transp. Res. Rec. 2674(11), 490–497 (2020) 12. Komarov, V.A, Kurashkin, M.I., Sedoikin, S.V.: Research of quality indicators of motor vehicles for farmers during the warranty period (2021) 13. Koike, D., Yamakami, J., Miyashita, T., Kataoka, Y., Nishida, H., Yasuda, A.: Combining failure modes and effects analysis and cause–effect analysis: a novel method of risk analysis to reduce anaphylaxis due to contrast media. Int. J. Qual. Health Care (2022)

Chapter 2

Modeling and Analysis of Railway Passenger Flow Forecast During the Spring Festival Zhi-Cheng Zhang, Ding Chen, and Pei-Zhou Jiang

Abstract The sharp increase in railway passenger flow during the Spring Festival Travel Season has tested the organization and dispatching ability of the railway transportation system. In this paper, the advantages of least square support vector machine (LSSVM) in small sample data prediction are integrated, and the ARIMALSSVM hybrid model based on residual linear transfer superposition is proposed, which is verified by Xiamen Spring Festival railway passenger flow. The analysis results show that the average absolute errors of hybrid model are 0.565 × 104 and 0.979 × 104 person times, respectively, which are 22.50% and 12.43% higher than ARIMA model, and 28.30% and 18.35% higher than LSSVM model. This study plays a positive role in improving the railway passenger flow forecasting ability and adjusting the preparation time during the Spring Festival Travel Season.

2.1 Introduction Railway passenger flow during Spring Festival belongs to a typical manifestation of periodic traffic, because the Spring Festival transportation passenger flow surge is extremely easy to cause the crowds gathered in and out of the station and the resulting contingency [1]; therefore, a forecasting model that can effectively reflect the cyclical change trend of passenger flow is required to provide preparation time for the implementation and arrangement of the pre-Spring Festival work in railway stations. Short-term passenger flow data is also small sample data, and a large number of existing studies and practices have demonstrated the applicability of the least Z.-C. Zhang · D. Chen (B) School of Mechanical and Automotive Engineering, Xiamen University of Technology, No. 600, Ligong Rd, Xiamen 361024, Fujian Province, China e-mail: [email protected]; [email protected] Z.-C. Zhang e-mail: [email protected] P.-Z. Jiang Xiamen GNSS Development and Application C., Ltd., Xiamen 361102, Fujian Xiamen, China © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_2

15

16

Z.-C. Zhang et al.

squares support vector machine (LSSVM) model in predicting small sample data [2, 3]. Although the LSSVM model can achieve a certain accuracy in passenger flow prediction. But because the model lacks the sparsity of the solution. In order to obtain an optimal estimate of the support vector values, the model needs to satisfy the Gaussian distribution of the residual variables. Therefore, the requirements for the quality of training data are relatively high [4]. In the current research on passenger flow prediction, LSSVM model and the corresponding machine learning algorithm only ignore the linear trend term of passenger flow time series itself and the resulting error distribution characteristics [5], which limits the practical application effect of the model to a certain extent. In order to solve this problem, based on the railway passenger flow data during the Spring Festival travel rush in Xiamen, this paper establishes a passenger flow prediction model based on ARIMA and LSSVM, and comprehensively analyzes and discusses the applicability of the model in passenger flow prediction during the Spring Festival travel rush [6, 7].

2.2 LSSVM Model In order to solve the computational and real-time problems of support vector solving quadratic programming problems, researchers proposed the least square support vector machine (LSSVM) based on regularization theory and transformed the inequality constraints of support vector machine (SVM) into equality constraints. Thus, the LSSVM model can be expressed as [8] f (x) =

N 

ai k(x, xi ) + b

(2.1)

i=1

where xi ai k(x, xi ) b

input sample Lagrange multiplier kernel function estimation error

In Formula (2.1), the parameter selection of kernel function k(x, x i ) determines the computational characteristics of LSSVM model, which is usually obtained by means of trial and exhaustion search. Selecting k(x, x i ) parameters through particle swarm optimization (PSO) algorithm can avoid traversing all parameters and achieve the purpose of combining prediction accuracy with parameter global optimization [9]. The radial machine kernel function (RBF) is adopted and the kernel parameters σ and penalty factor γ in the kernel function are set as initialization parameters. The desired actual passenger flow k(x, x i ) is taken as input and output samples and the root mean square error eRMSE is taken as convergence condition to establish the optimization objective function of the PSO algorithm, as shown in Eq. (2.2).

2 Modeling and Analysis of Railway Passenger Flow Forecast During …

 eR M S E =

N 1 ( f (xi ) − yi )2 n i=1

17

1/2 (2.2)

When eRMSE in Formula (2.2) is less than the set error threshold, the PSO algorithm outputs optimized kernel function parameters and is used for LSSVM model construction and application analysis. Take the railway passenger volume data of Xiamen city during the 40-day Spring Festival transportation period from 2016 to 2018 as an example. Among them, the passenger volume data is the inbound and outbound passenger flow of Xiamen Station and Xiamen North Railway Station, and the data volume is the sample of 120 cumulative days in 3 years. The kernel function of the LSSVM model selects the radial basis kernel function (RBF) with relatively strong nonlinear characteristics [10]. The initialization parameter of the kernel function is the random number group, and the particle swarm size of the PSO algorithm is 20. The prediction effect of the LSSVM model is shown in Fig. 2.1. It can be seen from Fig. 2.1 that the predicted value of LSSVM model only represents the average change rule of data in each cycle, but there is a big difference between the predicted effect and the expected actual passenger flow in three discontinuous cycles. The prediction of mean absolute errors of Xiamen Station and Xiamen North Station are 0.788 × 104 and 1.199 × 104 person times, respectively. The reason for the large errors is related to the nonstationarity and autocorrelation of time series, among which, the nonstationarity includes the periodicity and trend factors of passenger flow change within 3 years. At the same time, the autocorrelation between passenger flow data is superimposed, which makes it difficult to construct independent variables in the process of regression calculation. Taking time series as the prediction object, there are still major influencing factors that have not been considered. Therefore, in order to solve this problem, it is necessary to analyze the nonstationarity and autocorrelation of time series, which also become the main reason for introducing ARIMA model.

(a) Xiamen North Railway Station Fig. 2.1 Prediction effect of LSSVM model

(b) Xiamen North Railway Station

18

Z.-C. Zhang et al.

2.3 Hybrid Model 2.3.1 ARIMA Model ARIMA model is composed of the difference term introduced by ARMA, which is used to transform nonstationary time series in ARMA into stationary time series by difference transformation. The difference transformation is as follows [11]: 

∇ yt = yt − yt−1 = (1 − B)yt ∇s yt = yt − yt−s

(2.3)

where ∇ ∇s yt B s

first-order differential operator periodic difference operator the time series of actual passenger flows lag factor number of cycles

According to Formula (2.3), a stationary ARMA process with autocorrelation order p and partial autocorrelation order q is formed after d difference of a group of time series [12], indicating that yt obeys ARIMA model and can be expressed as yt = μ +

p  j=1

γ j yt− j + et +

q 

θ j et− j

(2.4)

j=1

where μ γj θj et

constant term correlation coefficient of the autoregressive model correlation coefficient of the moving average model error term of the t time

In Formula (2.4), after d difference is required to eliminate the periodic trend of time series, the lag numbers of autocorrelation and partial autocorrelation coefficients are used to calculate q − 1 and p − 1, and order p and q are determined based on this. ARIMA model can also be expressed as ARIMA (p, d, q) and applied to practical prediction research. The same passenger flow time series is in FIG. 2.1 is analyzed based on ARIMA model, and the improvement of the model on the periodic change trend of time series is verified by stationarity test and residual analysis. The analysis results are shown in Figs. 2.2 and 2.3. Figure 2.2 shows the stationarity test and residual distribution of passenger flow time series of Xiamen Station. The autocorrelation coefficient and partial correlation

2 Modeling and Analysis of Railway Passenger Flow Forecast During …

(a) ACF

(c) Residual

19

(b) PACF

(d) Residual Distribution

Fig. 2.2 ADF and residual analysis of passenger flow in Xiamen Station

coefficient of passenger flow time series after d = 1 difference are calculated, as shown in Fig. 2.2a, b. The autocorrelation coefficient and partial correlation coefficient show a trailing phenomenon after the lag number is 3 and 1 order, respectively, and both fall between confidence intervals (−0.183, 0.183). Therefore, the constructed passenger flow time series of Xiamen station can be expressed as ARIMA (2, 1, 4). It can be seen from Fig. 2.2c, d that the normalized residual curve calculated by ARIMA model is distributed on both sides of the 0 bit of the mean with no obvious cyclical trend. Meanwhile, the statistical distribution of the normalized residual is gaussian, reflecting the noncorrelated random data whose residual is approximately white noise. Time series are considered to be aperiodic stationary and can be used in ARIMA model prediction. The stationarity analysis result of passenger flow time series of Xiamen North Railway Station is similar to that of Xiamen Station. In Fig. 2.3, after d = 1 difference, the autocorrelation and partial correlation coefficients show a trailing phenomenon after the lag number of order 1 and fall between confidence intervals. Therefore,

20

Z.-C. Zhang et al.

(a) ACF

(b) PACF

(c) Residual

(d) Residual Distribution

Fig. 2.3 ADF and residual analysis of passenger flow in Xiamen North Station

ARIMA model can be expressed as ARIMA (1, 1, 1), Residual analysis also shows the removal effect of periodic trends.

2.3.2 ARIMA-LSSVM Hybrid Model The ARIMA model is adopted to reflect the linear characteristics of passenger flow time series, and the prediction residual is taken as the input sample of LSSVM model in Formula (2.1), which is the main idea of constructing the ARIMA-LSSVM mixed prediction model. Y (t) represents the prediction result of passenger flow at time t, and the mixed model is shown in Eq. (2.5). Yt = yt + f (et )

(2.5)

The specific implementation steps of the ARIMA-LSSVM hybrid model are as follows:

2 Modeling and Analysis of Railway Passenger Flow Forecast During …

21

Step 1 Establish ARIMA model. The actual passenger flow data are used as time series yt for difference processing and stationarity test, and ARIMA (p, d, q) model is output. Step 2 ARIMA model predictive analysis. Time series yt (t = 1, 2 …, i/2) as a training set, the time series yt (t = i/2, i/2 + 1 …, i) Perform predictive analysis, and obtain the predictive residual series et of this time series. Step 3 Build the LSSVM model. Set the particle swarm size to 20, eRMSE = 1 × 10−4 . A set of kernel function parameters (σ, γ) are randomly generated as the initial parameters of the model. The normalized residual sequence et is used as the input parameter to train the LSSVM model, and the model parameters are obtained and saved. Step 4 ARIMA-LSSVM hybrid model. According to Formula (2.5), ARIMA (p, d, q) established in Step 2 and LSSVM model established in Step 3 are linearly added to establish the ARIMA-LSSVM mixed model, which can further calculate Y (t) at t = i + 1. Step 5 Cyclic optimization of ARIMA-LSSVM hybrid model. The expected value yt (t = N + 1, N + 2, …, M) as the training set and repeat Steps 2 to 4 to update ARIMA (p, d, q) and LSSVM kernel function parameters (σ, γ), so as to increase the possibility of extending the application of the mixed model from short time passenger flow to long period passenger flow.

2.4 Railway Passenger Flow Forecast During the Spring Festival in Xiamen In order to maintain the comparability of LSSVM model before and after mixing ARIMA model, Xiamen railway passenger flow data during the Spring Festival travel period from 2016 to 2018 in Fig. 2.1 is taken as the test data, with data sample i = 120. Test data are used for stationarity test and ARIMA model construction and the calculation results are shown in Figs. 2.2 and 2.3. In addition, 60 of the 120 data samples are used as the output of the ARIMA model to predict the residual and the training of the LSSVM model. A total of 40 data samples of the 2019 Spring Festival passenger flow are used as the predicted data of the ARIMA-LSSVM model. According to the calculation results in Figs. 2.2 and 2.3, the prediction results of 60 groups of data samples using ARIMA model are shown in Fig. 2.4. It can be found that compared with LSSVM model, ARIMA model has obvious advantages in reflecting periodic changes of data, and the prediction results still coincide with the actual change rule of expected value in the cross-cycle time sequence. The average absolute errors of passenger flow prediction of Xiamen Station and Xiamen North Station are 0.729 × 104 and 1.118 × 104 , respectively, which are 7.59% and 6.76% higher than those of LSSVM model in Fig. 2.1. Perform Step 2 and Step 3, and apply the residual sequence of ARIMA model calculation in Fig. 2.4 to LSSVM model training, and the calculation results are shown in Fig. 2.5. As shown in Fig. 2.5, the parameters kernel (σ, γ) and of LSSVM

22

Z.-C. Zhang et al.

(a) Xiamen Station prediction

(b) Xiamen North Station prediction

Fig. 2.4 Residual prediction results of ARIMA model

model kernel function of residual training in the two groups are (0.01, 19.9) and (0.01, 19.7), respectively. Neither the expected value nor the predicted value of the residual series has a significant trend of periodic change, indicating that the residual series is an irrelevant variable of the original time series. Compared with the calculation results in Fig. 2.1, the LSSVM model shows the advantage of small sample data prediction while maintaining adaptability with the residual series. Step 4–Step 5 in the ARIMA-LSSVM hybrid model are implemented, and the ARIMA-LSSVM hybrid model is constructed by combining the calculation results in Figs. 2.4 and 2.5. The railway passenger flow during the 2019 Spring Festival Travel rush is predicted and compared with the calculation results in Fig. 2.1. The comparison results are shown in Fig. 2.6.

(a) Xiamen Station prediction residual Fig. 2.5 Residual prediction results of LSSVM model

(b) Xiamen North Station prediction residual

2 Modeling and Analysis of Railway Passenger Flow Forecast During …

(a) Xiamen Station prediction

23

(b) Xiamen North Station prediction

Fig. 2.6 Prediction results of the ARIMA-LSSVM hybrid model

It can be seen from Fig. 2.6 that, compared with LSSVM model, the matching degree between the predicted value and the expected value of ARIMA-LSSVM hybrid model is effectively improved, while retaining periodic and nonlinear change law of passenger flow. The average absolute error of the hybrid model in predicting passenger flow of Xiamen Station and Xiamen North Station is 0.565 × 104, which is 28.30% and 18.35% higher than that of LSSVM model, 22.50% and 12.43% higher than that of ARIMA model. The validity and reliability of the hybrid model in railway passenger flow forecast for Spring Festival transportation are verified. The study needs to consider the model on the influence of traffic data volume and density change, such as number of days as the unit of time of traffic data refinement as to hours count, model calculation and calculation steps are matching problem, solve the problem of railway passenger traffic warning and transportation scheduling timely adjust has reference value. However, it is still challenging and worthy of further exploration.

2.5 Conclusion Aiming at the prediction of railway passenger flow during the Spring Festival travel rush, this paper carried out the modeling analysis and prediction research content of the ARIMA-LSSVM model, and drew the following conclusions: (1) LSSVM model has shortcomings in reflecting the periodicity of passenger flow during the Spring Festival. The average absolute error of passenger flow prediction for Xiamen Station and Xiamen North Station is 0.788 × 104 and 1.199 × 104, respectively. The reason for the large error is related to the nonstationarity and autocorrelation of time series of passenger flow during the Spring Festival.

24

Z.-C. Zhang et al.

(2) Compared with LSSVM model, ARIMA model has a great advantage in dealing with the periodicity of time series. After difference and stationarity tests, the prediction accuracy of ARIMA model improves by 7.59% and 6.76% compared with LSSVM model for passenger flow time series of Xiamen Station and Xiamen North Station. (3) Taking ARIMA prediction residual series as input parameters of LSSVM model can meet the purpose of aperiodicity and independence of input parameters, thus improving the applicability of ARIMA-LSSVM model in passenger flow time series. The average absolute errors of the prediction results of the mixed model are 0.565 × 104 and 0.979 × 104 , respectively, which are 22.50% and 12.43% higher than those of THE ARIMA model, and 28.30% and 18.35% higher than those of the LSSVM model. Acknowledgements This project is supported by the Natural Science Foundation of Fujian Province No. 2021J011202, the Education and Research Project of Young and Middle-Aged Teachers of Fujian Province No. JAT190661, the Science and Technology Project of High-Level Talents of Xiamen City No. YKJ19021R, and the Graduate Science and Technology Innovation Project of Xiamen University of Technology No. YKJCX2021028.

References 1. Wei, Y., Chen, M.C.: Forecasting the short-term metro passenger flow with empirical mode decomposition and neural networks. Transp. Res. Part C Emerg. Technol. 21(1), 148–162 (2012) 2. Zhang, F., Wu, T.-Y., Pan, J.-S., Ding, G., Li, Z.: Human motion recognition based on SVM in VR art media interaction environment. HCIS 9, 40 (2019) 3. Wu, Q., Zang, B.-Y., Zhang, Y., Qi, Z.-X.: Wavelet kernel twin support vector machine. J. Inf. Hiding Multimedia Signal Process. 12(2), 93–101 (2021) 4. Cortes, C., Mohri, M., Riley, M., et al.: Sample selection bias correction theory. In: Proceedings of the 19th International Conference on Algorithmic Learning Theory. Springer, Berlin, Heidelberg (2008) 5. Guo, S.: Prediction of short-term passenger flow on a bus stop based on LSSVM. J. Wuhan Univ. Technol. (Transportation Science & Engineering (2013) 6. Chen, C.-M., Chen, L., Gan, W., Qiu, L., Ding, W.: Discovering high utility-occupancy patterns from uncertain data. Inf. Sci. 546, 1208–1229 (2021) 7. Gan, W., Chen, L., Wan, S., Chen, J., Chen, C.-M.: Anomaly rule detection in sequence data. IEEE Trans. Knowl. Data Eng. (2021). https://doi.org/10.1109/TKDE.2021.3139086 8. Meng, P., Li, X., Jia, H., et al.: Short-time rail transit passenger flow prediction based on wavelet threshold denoising and support vector machine. Revista de la Facultad de Ingenieria 32(4), 292–302 (2017) 9. Yang, H.U.I., Yonggang, W.A.N.G., Hui, P.E.N.G., et al.: Subway passenger flow prediction based on optimized PSO-BP algorithm with coupled spatial-temporal characteristics. J. Traffic Transp. Eng. 21(04), 210–222 (2021) 10. Sun, Y., Leng, B., Guan, W.: A novel wavelet-SVM short-time passenger flow prediction in Beijing subway system. Neurocomputing 166, 109–121 (2015)

2 Modeling and Analysis of Railway Passenger Flow Forecast During …

25

11. Liu, S.Y., Liu, S., Tian, Y., et al.: Research on forecast of rail traffic flow based on ARIMA model. J. Phys: Conf. Ser. 1792(1), 012065 (2021) 12. Gu, D.-X., Chen, et al.: Forecasting holiday daily tourist flow based on seasonal support vector regression with adaptive genetic algorithm. Appl. Soft Comput. (2015)

Chapter 3

Seasonal and Period Division Method for Dynamic Passenger Flow of High-Speed Railway Hui Han, Xinqian Zou, Yiyuan Gao, and Xiuyun Guo

Abstract With the large-scale construction of China’s high-speed railway network, the demand for passenger flow is growing and varies in different seasons. Therefore, refinement operational requirements are increasingly prominent. However, the passenger flow itself has time-varying characteristics, and the division of passenger flow seasons and periods is divided by manual methods and experience subjectively. As a result, it is difficult to adjust it dynamically with the passenger flow changing dynamically. There are few studies on the seasons and periods of the division about high-speed railway passenger flow at home and abroad. In this study, the seasonal and period division methods of high-speed railway passenger flow were studied, the dynamic characteristics of passenger flow were analyzed, and the annual passenger flow fluctuation was considered. Then, the high-speed railway passenger flow was divided into seasons by using the firefly affinity propagation algorithm. Meanwhile, the high-speed railway passenger flow was divided into periods by employing the orderly clustering algorithm with the consideration of the daily passenger flow fluctuation.

H. Han China Railway Beijing Group Co., Ltd., Beijing 300010, China e-mail: [email protected] X. Zou · Y. Gao · X. Guo (B) School of Transportation and Logistics, Southwest Jiaotong University, Chengdu 610031, China e-mail: [email protected] X. Zou e-mail: [email protected] Y. Gao e-mail: [email protected] National and Local Joint Engineering Laboratory of Comprehensive Intelligent Transportation, Southwest JiaoTong University, Chengdu 610031, China National Engineering Laboratory of Integrated Transportation Big Data Application Technology, Beijing, China © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_3

27

28

H. Han et al.

3.1 Introduction With the large-scale construction of China’s high-speed railway network, the demand for passenger flow is increasing, its differences in different seasons are significant, and the refinement of operational requirements is prominent. In academic research, there are few studies on the seasons and periods division of high-speed railway passenger flow at home and abroad. It mainly divides the annual passenger flow seasonally at the macro level. For example, Wenxian [1] uses the nearest neighbor clustering algorithm to divide the annual passenger flow seasonally. There are few studies on the time division of high-speed railway full-day operation period at the micro level. This problem is essentially like the problem of time division for bus and subway operations. For example, in terms of bus operation time division, Xueyan et al. [2] used the ordered sample clustering method to divide bus operation time for different periods and proved the feasibility of the method by simulation. Peng [3] used a genetic algorithm to optimize the operation time division scheme by taking the time division point as the decision variable. In the subway operation time division, Xiaoxu et al. [4] used the orderly sample clustering method to divide the operation time. Dongyang et al. [5] constructed the characteristic variables of each period, used the K-means algorithm to cluster, and evaluated by clustering evaluation index to determine the optimal clustering number. The feasibility of the method was verified by examples. When solving engineering problems, deep learning, data mining, and intelligent computing are powerful tools. For example, Ying et al. [6] have introduced meta-heuristic algorithms and Ranfan et al. [7] have proposed an improved honey badger algorithm. There are also some studies to optimize existing algorithms. Ruishun et al. [8] proposed a non-inertial particle swarm optimization with elite mutation-Gaussian process regression (NIPSO-GPR) to optimize the hyper-parameters of GRP. Deep learning is also gradually applied to the field of transportation. Kumar et al. [9] proposed an LSTM Network for Transportation Mode Detection. Shunmiao et al. [10] proposed the application of long- and shortterm memory networks. From the above, there are few studies on the seasonal and time division methods of high-speed railway passenger flow at home and abroad, and the passenger flow itself has time-varying characteristics. It is formidable to adjust the division of passenger flow seasons and periods dynamically, with the passenger flow changing dynamically by manually relying on subjective methods such as experience to divide the passenger flow season and period. The method of high-speed railway passenger flow season and time division based on dynamic passenger flow characteristics remains to be studied.

3 Seasonal and Period Division Method for Dynamic Passenger Flow …

29

Fig. 3.1 Annual passenger flow distribution map of shanghai Hongqiao station

3.2 Analysis of Dynamic Characteristics of High-Speed Railway Passenger Flow The passenger flow demand of high-speed railway shows the characteristics of annual and daily dynamic change [1]. The passenger flow dynamics of the two points above are mainly analyzed in this section.

3.2.1 Dynamic Characteristics of Annual Passenger Flow Taking Shanghai Hongqiao Station of the Beijing–Shanghai high-speed railway as the research object, the daily arrival and departure passenger flow data from March 31, 2018, to March 20, 2019, are selected. The curve is displayed in Fig. 3.1. The horizontal axis is the time series, while the vertical axis is the passenger flow. From Fig. 3.1, the annual arrival and departure passenger flow of Shanghai Hongqiao Station generally presents seven peak periods, and the passenger traffic volume in other periods fluctuates within a certain range. Different peak periods correspond to different lengths of passenger flow periods. Among them, the duration of Qingming festival, Labor Day, Dragon Boat Festival, and New Year’s Day is short, while the duration of Summer Festival, National Day, and Spring Festival is longer. The annual passenger flow shows obvious seasonal differences such as peak, flat peak, and low peak.

3.2.2 Dynamic Characteristics of Daily Passenger Flow Taking the passenger volume of Shanghai Hongqiao Station and Beijing South Station of the Beijing–Shanghai high-speed railway as the research object, the data

30

H. Han et al.

of daily passenger volume at different times on Wednesday, Friday, and Saturday in peak season, flat peak season, and low peak season are selected, respectively. Draw the curve, the horizontal axis is the time series, and the vertical axis is the passenger flow, as shown in Figs. 3.2 and 3.3. It can be seen from Figs. 3.2 and 3.3 that the passenger traffic volume in different periods is significantly different, and there are obvious passenger flow peaks, showing a “two peak pattern” or “multiple peak pattern” overall. The passenger flow on different dates in the same period is also significantly different, and the fluctuation range is large. It is necessary to formulate the train diagram to adapt to different passenger flow demands.

Fig. 3.2 Time distribution map of passenger flow at Shanghai Hongqiao station

Fig. 3.3 Time distribution map of Beijing south railway station sending passenger flow

3 Seasonal and Period Division Method for Dynamic Passenger Flow …

31

Based on the above analysis, the annual passenger flow of high-speed railway shows obvious seasonal characteristics such as peak, flat peak, and low peak, and the passenger flow is quite different, but the passenger flow fluctuates within a certain range in the same season; the daily passenger flow of high-speed railway is significantly different in different periods, and there are different passenger flow distribution characteristics on different dates. Based on the analysis of the dynamic characteristics of high-speed railway passenger flow, divide the passenger flow of high-speed railway into seasons and periods from the annual passenger flow fluctuation and daily passenger flow fluctuation. In the seasonal and time division of high-speed railway passenger flow, the number of divisions cannot be too large or too small. If the number of divisions is too small, it is difficult to reflect the dynamic changes in passenger flow. If the number of partitions is too large, it makes the adjustment frequency of the train operation plan too high, increasing the complexity of the problem. At present, the manual method is usually used to subjectively divide the season and period of high-speed railway passenger flow, which lacks fine dynamic adjustment with the change in passenger flow and is difficult to meet the passenger flow demand. In academic research, there are few studies on the time division of high-speed railway passenger flow at home and abroad, mainly on the seasonal division of annual passenger flow at the macro level. Based on the existing research, this paper proposes to use the firefly affinity propagation algorithm (FO-AP algorithm) for the seasonal division of annual passenger flow fluctuations of high-speed railways. The orderly clustering algorithm is used to divide the period of daily passenger flow fluctuation of high-speed railway.

3.3 Season and Period Division of Dynamic Passenger Flow Based on the dynamic train diagram compilation process of high-speed railway, the annual passenger flow is divided into seasons to reflect the change law of passenger flow with off-season, peak season, holidays, and so on, which is the basis for compiling the dynamic train diagram adapted to the passenger flow demand of annual dynamic characteristics. The daily passenger flow is divided into periods to reflect the fluctuation of high-speed railway passenger flow during operation time. Based on this, make a match analysis of the passenger flow with the current basic operation diagram, and determine the adjustment period of the operation diagram, so that the daily operation diagram is adapted to the daily fluctuation of passenger flow demand. The seasonal division based on annual passenger flow fluctuation is based on the time range of 1 year, which is transformed into limited seasons with similar passenger flow fluctuation ranges. The period division based on daily passenger flow is based on 1 day as the time range, which is transformed into limited periods of similar passenger flow fluctuation ranges. This paper uses the firefly affinity propagation algorithm (FO-AP algorithm) for the seasonal division of annual passenger flow fluctuations of high-speed railways

32

H. Han et al.

and the orderly clustering algorithm for the period division of daily passenger flow fluctuations of high-speed railway.

3.3.1 Seasonal Division Method of Annual Passenger Flow Based on FO-AP Algorithm The affinity propagation clustering algorithm refers to the introduction of attraction and attribution to represent the exchange and transmission process of information between sample points without setting the cluster center point first, to achieve high similarity within the same class and low similarity between different classes. In an affinity propagation clustering algorithm, the regulatory parameters (bias parameters and damping) are usually selected manually, which is easy to limit the clustering effect. The firefly swarm intelligence optimization algorithm can adaptively search in the parameter space. Therefore, this paper combines the firefly swarm intelligence optimization algorithm with the affinity propagation clustering algorithm and proposes a firefly-based affinity propagation clustering algorithm (FO-AP algorithm). The algorithm can optimize the bias parameters and damping factors [11], thereby improving the clustering quality. Suppose that the annual operation time of a high-speed railway is divided into n unit time intervals. Count passenger volume per unit time interval for m high-speed rail stations, that is, m  variable indexes. The k index value ofthe i sample is xi,k . That is, the sample set is xi,k |i = 1, 2, · · · , m; k = 1, 2, · · · , n . The original sample is ⎧ x1,1 ⎪ ⎪ ⎨ x2,1 X= ⎪ ··· ⎪ ⎩ xm,1

x1,2 x2,1 ··· xm,2

··· ··· ··· ···

⎫ x1,n ⎪ ⎪ ⎬   x2,n = X 1 , X 2,··· , X n ··· ⎪ ⎪ ⎭ xm,n

(3.1)

Step 1: Read the data, including high-speed railway line data, passenger flow, and other related data, set the oscillation variable λ, algorithm maximum number of iterations T ; Step 2: Data normalization. Normalize x according to Formula (3.2) 

xi,k

  xi,k − min x1,k , x2,k , · · · , xn,k     = max x1,k , x2,k , · · · , xn,k − min x1,k , x2,k , · · · , xn,k

(3.2)

Step 3: Initialization. Set belonging degree x and attraction degree y to 0, construct similarity matrix according to Formulas (3.3) and (3.4), p is the value of similarity diagonal [12];

3 Seasonal and Period Division Method for Dynamic Passenger Flow …

s(i, k) = −di,k = −xi − xk  p=

 i= j

s(i, k) N × (N − 1)

33

(3.3) (3.4)

Step 4: Randomly select n groups p in the p space of preference parameters, and randomly select the oscillation  variable value λ, where the search range of the param    s(i, j) − max max{s(i, k), s( j, k)} , and the search eter p is P2m , max j

k

range of the damping factor [6] is [0.5, 1]. Step 5: update r (i, k) and a(i, k) according to formulas (3.5) and (3.6);     a(i, k ) + s(i, k ) r (i, k) ← s(i, k) − max  k =k

 ⎧    ⎪  ⎪ ⎨ min 0, r (k, k) + max 0, r (i , k) , i = k i ∈(i,k) / a(i, k) ←     ⎪ ⎪ max 0, r (i , k) , i = k ⎩

(3.5)

(3.6)

i  =k

Step 6: To avoid the local optimum of the algorithm, update r (i, k) and a(i, k) according to formulas (3.7) and (3.8): r (t+1) (i, k) ← (1 − λ)r (t+1) (i, k) + λr (t) (i, k)

(3.7)

a (t+1) (i, k) ← (1 − λ)a (t+1) (i, k) + λa (t) (i, k)

(3.8)

In the formula, t is the number of iterations. Step 7: The sum of r (i, k) and a(i, k) is calculated for all samples, and the sample center point of each sample is found according to Formula (3.9). The formula is as follows: E(k) = arg max(a(i, k) + r (i, k)) k

(3.9)

Step 8: iteration times t ← t + 1; Step 9: Determine whether the termination condition is reached, that is, if t > T , the algorithm terminates, outputs the seasonal division result, and otherwise returns Step 5. Step 10: Calculate the clustering results of p and λ of each group.

34

H. Han et al.

Step 11: Regard the parameter p as the relative brightness value of the firefly algorithm, that is, the objective function value of clustering; Use the damping factor λ as the attraction degree in the firefly algorithm, such as Formula (3.10). Obtain the next position of the firefly movement according to Formula (3.12). βik = β0 × e−γ

ri,k

rik

  D  = xi − xk  =  (xik − xk j )2

(3.10)

(3.11)

j=1

In the formula: β0 Maximum attraction degree; ri,k the Cartesian distance of firefly i and firefly k; γ the absorption coefficient of light.

xk (t + 1) = xk (t) + βik (xi (t) − xk (t)) + αεk 式中: xi (t) xk (t) εk α t

(3.12)

the location of the firefly i; the location of the firefly k; a random number vector; empirical value; iteration times.

Step 12: If the end condition is reached, then the end, otherwise turns Step 4. Step 13: Use the Silhouette index [7] to evaluate the clustering results according to Formulas (3.13) and (3.14). The Silhouette index range is [−1, 1], and the larger the value is, the better the clustering effect is. b(t) = min{d(t, Ci )} sil(t) =

[b(t) − a(t)] max{a(t), b(t)}

(3.13) (3.14)

n sample point is divided into k clusters Ci (i = 1, 2, . . . , k); Ci d(t, Ci ) in class C j , the distance from t to all samples in class Ci ; a(t) in class C, the distance between the sample point t and other samples. Step 14: Determine the seasonal division of the annual passenger flow of high-speed railway. Compare each day in each division season in pairs. If the date is adjacent, regard it as a passenger flow season; otherwise, regard it as another passenger flow season.

3 Seasonal and Period Division Method for Dynamic Passenger Flow …

35

3.3.2 Period Division Method of Daily Passenger Flow Based on Ordered Clustering Method The ordered sample clustering method is a method of searching for optimal segmentation based on keeping sample order unchanged. The basic idea of this method is to obtain the optimal segmentation with a high similarity between the same class and significant difference between different classes based on the calculation of loss function. The daily passenger flow period division method based on the ordered clustering method is as follows. Suppose that the daily operation time of a high-speed railway is divided into n unit time intervals. Count passenger volume needs per unit time interval for m highspeed rail stations, that is, m variable indexes. The j index value of the  i sample is  qi, j . That is, the sample set is qi, j |i = 1, 2, · · · , m; j = 1, 2, · · · , n . The original sample is ⎧ q1,1 ⎪ ⎪ ⎨ q2,1 Q= ⎪ ··· ⎪ ⎩ qm,1

q1,2 q2,1 ··· qm,2

⎫ · · · q1,n ⎪ ⎪ ⎬   · · · q2,n = Q 1 , Q 2,··· , Q n ··· ··· ⎪ ⎪ ⎭ · · · qm,n

(3.15)

Step 1: Normalization. Normalize according to Formula (3.16). 

qi, j =

  qi, j − min q1, j , q2, j , · · · , qn, j     max q1, j , q2, j , · · · , qn, j − min q1, j , q2, j , · · · , qn, j

(3.16)

Step 2:  Calculation of the diameter of the class. If a period G contains a sample Q = Q i , Q i+1 , · · · , Q j , the diameter of the class is defined as D(i, j). D(i, j) =

j 

(Q t − Q)2

(3.17)

t=i

Q is the mean vector of class G  1 Q= Qt j − i + 1 t=i j

(3.18)

Step 3: Classification loss function calculation. The loss function is L[b(n, c)] =

c  t

D(i t , i t+1 − 1)

(3.19)

36

H. Han et al.

b(n, c) n time intervals divided into c types of results. When n and c are known numbers, the smaller L[b(n, c)] is, the smaller D(i, j) is, that is, the more reasonable the classification is. Therefore, the minimal loss function is the optimal segmentation result b(n, c)∗ . Step 4: The recursive formula of classification function. The core part of Fisher’s optimal segmentation method is to search for the optimal segmentation results b(n, c)∗ by Formulas (3.20) and (3.21).   L b(n, 2)∗ = min{D(1, j − 1) + D( j, n)}, 2 ≤ j ≤ n       L b(n, c)∗ = min L b( j − 1, c − 1)∗ + D( j, n) , c ≤ j ≤ n

(3.20) (3.21)

Formula (3.20) represents the optimal segmentation results divided into two categories; formula (3.21) represents the optimal segmentation result divided into class c, which is the optimal segmentation result divided into class c − 1 based on the first j − 1 time intervals. Step 5: Determination of Optimal Period Division Scheme. The steps to determine the optimal number of segments are as follows: ➀ Determine the first period split point jC to satisfy       L b(n, c)∗ = min L b( jC − 1, c − 1)∗ + D( jC , n) c≤ jC ≤n

(3.22)

➁ Search for the second period split point to satisfy   L b( jC − 1, c − 1)∗ =

min

c−1≤ jC−1 ≤ jC −1

    L b( jC − 1, c − 2)∗ + D( jC−1 , jC − 1) (3.23)

➂ By analogy, repeat step ➁ to obtain the optimal number of segments. At the same time, according to the loss function change curve, the optimal number of segments can also be determined by its inflection point position. In summary, the passenger flow period division process based on the ordered clustering method is shown in Fig. 3.4.

3 Seasonal and Period Division Method for Dynamic Passenger Flow …

37

Fig. 3.4 Flowchart of passenger flow time division based on ordered clustering method

3.4 Case Analysis 3.4.1 Overview of Beijing–Shanghai High-Speed Railway The Beijing–Shanghai high-speed railway is the main line of ‘eight vertical and eight horizontals,’ starting from Beijing South Railway Station to Shanghai Hongqiao Station, with a total length of 1318 km. There are 24 stations and 6 lines in the whole line. It runs through the seven provinces and cities of Beijing, Tianjin, Hebei, Shandong, Anhui, Jiangsu, and Shanghai, connecting the two major economic zones of the Jing-Jin-Ji region and the Yangtze Triangle. It is a new benchmark for the construction of high-speed railways worldwide, as shown in Fig. 3.5. This paper considers the downward direction of the train for example analysis. Taking Beijing–Shanghai high-speed railway as an example, combined with the annual passenger flow and daily passenger flow distribution characteristics, use the firefly affinity propagation algorithm to divide the annual passenger flow season of the high-speed railway, and use the ordered clustering algorithm to divide the daily operation time of high-speed railway.

38

H. Han et al.

Fig. 3.5 Diagram of Beijing–Shanghai high-speed railway

3.4.2 Seasonal Division of Annual Passenger Flow Select the daily arrival and departure passenger flow data of Shanghai Hongqiao Station from March 31, 2018, to March 20, 2019, and divide the annual passenger flow period by the FO-AP algorithm. The results areshown in Table 3.1. It can be seen from Table 3.2 that the whole year is divided into 14 passenger flow periods, among which period c1 is the peak passenger flow period of Qingming Festival; period c2 is the spring peak passenger flow period; c3 is the peak passenger flow period of Labor Day; c4 is the low peak passenger flow period in June; c5 is the peak passenger flow period of the Dragon Boat Festival; c6 is the summer peak passenger flow period; c7 is located in the summer peak passenger flow period; c8 is the autumn peak passenger flow period; c9 is located in the peak passenger flow Table 3.1 The final clustering numbers 1

2

3

4

5

326,954

346,353

412,467

447,583

495,679

3 Seasonal and Period Division Method for Dynamic Passenger Flow … Table 3.2 Division table of annual passenger flow period

39

Period number Days crossed Period number Days crossed c1

1–5

c9

164–182

c2

6–30

c9

183–193

c3

31–35

c10

194–275

c4

36–76

c11

276–280

c5

77–81

c12

281–302

c6

82–92

c13

303–323

c7

93–147

c14

324–365

c8

148–163

period of National Day; c10 is located in the off-season peak passenger flow period after the National Day; c11 is the peak passenger flow period of the New Year’s Day; c12 is the flat peak passenger flow period between New Year’s Day and Spring Festival; c13 is the peak passenger flow period of Spring Festival travel; and c14 is the low peak passenger flow period after the end of the Spring Festival travel. By comparing with the actual passenger flow, the division results are in line with the fluctuation law of normal passenger flow, which verifies the effectiveness of the annual passenger flow period division based on the FO-AP algorithm, which can be used as a basis for the preparation of the basic operation diagram.

3.4.3 Period Division of Daily Passenger Flow Select the passenger volume data of each station of the Beijing–Shanghai high-speed railway, and the daily passenger flow of the Beijing–Shanghai high-speed railway is divided into periods based on the ordered clustering method. The results are shown in Table 3.3. Table 3.3 Division of daily passenger flow time of Beijing–Shanghai high-speed railway

Period number

Operation period

category

t1

6:00–8:00

Low peak period

t2

8:00–11:00

Peak period

t3

11:00–12:00

Flat peak period

t4

12:00–13:00

Low peak period

t5

13:00–14:00

Flat peak period

t6

14:00–18:00

Peak period

t7

18:00–20:00

Flat peak period

t8

20:00–24:00

Low peak period

40

H. Han et al.

Divide the daily passenger flow of the Beijing–Shanghai high-speed railway into eight periods, and the proportion of peak period, flat peak period, and low peak period are 2:3:3.

3.5 Conclusions Based on the dynamic characteristics of passenger flow and considering the annual fluctuation of passenger flow, this paper proposes a method of dividing the annual passenger flow season of a high-speed railway by the firefly affinity propagation algorithm. Considering the fluctuation of daily passenger flow, propose an orderly clustering algorithm to divide the daily operation time of high-speed railway into periods, which is the basis for the adjustment of the train diagram. Taking Beijing– Shanghai high-speed railway as an example, combined with the annual passenger flow and daily passenger flow distribution characteristics, use the firefly affinity propagation algorithm to divide the annual passenger flow season of the high-speed railway, and the ordered clustering algorithm is used to divide the daily operation time of high-speed railway. The results show that the paper can accurately divide the annual passenger flow into seasons, reflecting the change law of passenger flow with off-season, peak season, and holidays. It can accurately divide the daily passenger flow period to reflect the fluctuation of high-speed railway passenger flow during operation time, which is more accurate than the traditional division method. Funding This research was supported by the National Natural Science Foundation of China (Project No. 52072314; 52172321;52102391), Sichuan Science and Technology Program (Project NO. 2020YJ0268; 2020YJ0256; 2021YFQ0001; 2021YFH0175), Science and Technology Plan of China Railway Corporation (Project No.: 2019F002), China Shenhua Energy Co., Ltd. Science and Technology Program (Project No.: CJNY-20-02), China Railway Beijing Bureau Group Co., Ltd. Science and Technology Program (2021BY02, 2020AY02), and the National Key R&D Program of China (2017YFB1200702). Key science and technology projects in the transportation industry of the Ministry of Transport (2022-ZD7-132).

References 1. Wenxian, W.: Research on Evaluation and Adjustment of High-speed Railway Line Plan Based on Dynamic Demand. Southwest Jiaotong University (2017) 2. Xueyan, Z., Feng, H., Yugang, Z., Dongdong, Z.: Research and simulation on operation period division of urban bus system. Comput. Simul. 36(11), 126–129 (2019) 3. Peng, L.: Bus Passenger Flow Prediction and the Optimization of Bus Operation Period Division Based on Multi-source Data. South China University of Technology (2019) 4. Xiaoxu, Z., Lin, W., Xiandi, L., Ning, Z., Shengna, Z.: Application of ordinal clustering in the division of operation periods for urban ran transit. Urban Rapid Rail Transit 30(02), 108–112 (2017)

3 Seasonal and Period Division Method for Dynamic Passenger Flow …

41

5. Dongyang, C., Dewang, C., Shixiong, J., Ning, X.: Division of metro operation periods based on feature clustering of passenger flow. Comput. Syst. Appl. 30(03), 256–261 (2021) 6. Ying, S., Shu-Chuan, C., Pei, H., Junzo, W., Mingchao, S., Jeng-Shyang, P.: Overview of parallel computing for meta-heuristic algorithms. J. Netw. Intell. 7(3), 656–684 (2022) 7. Ran-Fan, C., Hao, L., Kuan, H., Trong, N., Jeng-Shyang, P.: An improved honey badger algorithm for electric vehicle charge orderly planning. J. Netw. Intell. 7(2), 332–346 (2022) 8. Lanlan, K., Ruey-Shun, C., Naixue, X., Yeh-Cheng, C., Yu-Xi, H., Chien-Ming, C.: Selecting hyper-parameters of Gaussian process regression based on non-inertial particle swarm optimization in internet of things. IEEE Access 7, 59504–59513 (2019) 9. Sachin, K., Agam, D., Aditya, K., Saru, K., Chien-Ming, C.: LSTM network for transportation mode detection. J. Internet Technol. 22(4), 891–902 (2021) 10. Shun-Miao, Z., Xin, S., Xin-Hua, J., Ming-long, C., Tsu, W.: A traffic prediction method of bicycle-sharing based on long and short term memory network. J. Netw. Intell. 4(2), 17–29 (2019) 11. Pianpian, M., Xingang, Z., Jingjing, L.: Research on affinity propagation clustering based on firefly algorithm. Netw. Secur. Technol. Appl. (12), 48–50 (2019) 12. Shiqi, M.: Research and Application of Distributed Semantic Neighbor Search Algorithm Based on Spark. Hangzhou Dianzi University (2019)

Chapter 4

A Study of High-Speed Railway Train Merger and Adjustment Based on Regional Network Fangyu Shi, Zhi Wu, Haowen Tan, Min Yang, and Jinshan Pan

Abstract Based on the “road network-region-cluster” three-level network, the optimization problem of train operation scheme under network condition is transformed into the problem of direct train operation and train combination between regions on cluster network. The design of train operation scheme based on the railway network realizes the passenger flow transportation within the regional network. However, the cross-network passenger flow on the regional network still needs to transfer at the overlapping nodes. In this paper, the train merging adjustment is carried out on the overlapping nodes identified by the regional network division, and the principle of train merging is studied. The train merging adjustment model is designed with the goal of the highest direct passenger flow, the largest train operation benefit and the lowest passenger travel cost, and the genetic algorithm is designed to solve the model. Finally, taking the high-speed rail network in Chengdu Bureau as an example, the feasibility and rationality of the model and algorithm proposed in this paper are verified. F. Shi · M. Yang · J. Pan (B) School of Transportation and Logistics, Southwest Jiaotong University, Chengdu 610031, Sichuan, China e-mail: [email protected] F. Shi e-mail: [email protected] M. Yang e-mail: [email protected] National and Local Joint Engineering Laboratory of Comprehensive Intelligent Transportation, Southwest Jiaotong University, Chengdu 610031, China National Engineering Laboratory of Integrated Transportation Big Data Application Technology, Chengdu 610031, Sichuan, China Z. Wu China Railway Chengdu Railway Group Co., Ltd., Chengdu 610000, Sichuan, China H. Tan School of Transportation Engineering, Chang’an University, Xi’an 710000, Shanxi, China © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_4

43

44

F. Shi et al.

4.1 Introduction As Chinese high-speed railway is intertwined, its network structure and passenger transport organization show high complexity and difficulty. With the increasing density of the network, the cross-frequency between each line increases, resulting in the complex operation of a large number of local and cross-line trains. At the same time, most of the lines have different speeds, different natures of the train running together, and the train operation interval time and station operation interval time are not uniform, causing the train operation organization of the road network extremely complex and cannot play a comprehensive network effect. In order to meet the diversified travel needs of passengers, it is of great theoretical and practical significance to study the optimization method of high-speed railway train operation plan under the condition of a large-scale road network, explore the method to effectively reduce the difficulty of compiling and solving the network high-speed railway train operation plan, and maximize the efficiency of high-speed railway network. The development of large-scale road network puts forward higher requirements for train operation plan. Passenger transfer, super-long passenger flow, and other phenomena gradually appear, and the complexity of solution increases. Therefore, on the one hand, some scholars study based on the idea of phased optimization. Scholl [1] proposed for the first time to construct a train operation plan alternative set and a “change & go” network. Based on the shortest path, an integer programming model was constructed and solved by branch and bound method, heuristic algorithm, and other algorithms. Based on the complexity of Chinese high-speed railway network, Huiling et al. [2] determined the step-by-step optimization idea, solved the train operation plan by constructing an alternative set, and then established a mixed integer programming model with the operation cost and travel time as the optimization objectives. Shouqing [3] constructed a cost–capacity network for passenger flow distribution and constructed a collaborative symmetric group crossover genetic algorithm for train operation plan to gradually approach the optimal solution. Huiping [4] studied the key issues of passenger transport organization under the condition of network formation and distributed passenger flow according to the analysis of passenger flow demand and line level. The planning model is constructed considering the optimization objectives of maximum train operation benefit, and the genetic algorithm is designed to solve the problem. Based on the high-speed railway network in Taiwan, Chang [5] constructed a multi-objective optimization model of train operation plan considering the objective constraints of operation cost and travel time under the condition of limited transport capacity and known OD passenger flow, and proposed a fuzzy mathematical programming method with high efficiency. Schobel et al. [6] constructed a model with passenger travel time as the goal. Hadas [7] analyzed the statistical distribution of passenger demand and travel time and considers increasing capacity utilization by reducing service frequency. With the development and research of high-speed railway network, the application of complex network theory to the research of high-speed railway transportation organization has become one of the trends. In order to study the high-speed railway

4 A Study of High-Speed Railway Train Merger and Adjustment Based …

45

transportation organization and passenger flow demand of passenger dedicated line, Yongtao [8] draws lessons from foreign research to construct the topological structure of the railway network, introduced the node importance system to analyze the stop plan problem, took the minimum operating cost as the optimization goal, and constructed the multi-objective optimization model of passenger dedicated line train operation plan from the perspective of transportation service network. Lanxia [9] innovatively used the network evolution of complex network theory to study the highspeed railway network, constructed the service network evolution model according to the railway department and passenger travel, and mapped it to the train operation plan. With the help of the concept of community structure of complex network, Yifei [10] divided the inter-city railway network, studied the passenger flow transportation method in the inter-city railway network, and put forward the method of line group division and the transportation method of different passenger flow transportation. With the goal of minimizing the cost of travel time and maximizing the benefit of operation, the optimization model of train operation plan was constructed considering the constraints of stations and sections. Ying et al. [11] studied the application of parallel computing in meta-heuristic algorithms. The combination of parallel computing and meta-heuristic algorithms can solve a variety of application problems, including path planning, large-scale optimization, and neural networks. There are still some deficiencies in the research on the train operation plan of networking high-speed railway, especially for the optimization of train operation plan with large network scale, diverse train operation routes, and complex passenger flow structure, which needs further research and exploration. After all, compared with a single line and a local network, it is necessary to consider other issues such as the rational distribution of passenger flow, the selection of operating routes, and the transfer of passengers under the network conditions, so as to meet the diversified needs of passengers and improve the efficiency of network transportation. Based on the complex network characteristics of high-speed railway, the optimization method of train operation plan considering subnet partition is studied to reduce the complexity of network solution.

4.2 Problem Description The train operation scheme design based on the coil network realizes the direct or transit transportation of passenger flow within the regional network. However, the cross-network passenger flow on the regional network also needs to transfer at the overlapping nodes, which artificially separates the long and large cross-line passenger flow, resulting in many problems such as excessive passenger transfer times and reduced satisfaction. Therefore, it is also necessary to consider the direct transportation of long-span cross-line passenger flow, which is not only convenient for passengers to travel as much as possible, but also saves the use of car bottom, reduces the operation cost, and improves the economic benefits of train operation.

46

F. Shi et al.

Based on this, this paper designs a train merging adjustment model on the overlapping nodes identified by the regional network division and studies the principle of train merging. On the one hand, it analyzes the impact of travel cost, transfer times, and direct passenger flow after train merging from the perspective of passenger flow. On the other hand, the rationality of train merging is analyzed from the perspective of trains, such as the influence of speed matching and running mileage restrictions, and the appropriate independent variables are determined to describe the principles and strategies of train merging. Taking the lowest passenger travel cost, the highest direct passenger flow and the highest train operation benefit as the optimization objectives, the train merging adjustment model is constructed considering the constraints of transfer times, limited mileage, and speed matching, then an adaptive genetic algorithm is designed to solve the problem. The principle of train merging: 1. Direction consistency principle. When merging trains, it is necessary to consider the direction of the train. If two trains are running in opposite directions, they cannot be merged. Therefore, direction consistency requires that both the frontsequence train and the rear-sequence train run in the direction away from the departure station.; 2. The principle of mileage continuity. The sum of the running paths of the frontsequence train and the rear-sequence train is the sum of the running paths of the whole trip. In order to ensure traffic safety, the running mileage is required to be lower than the limited mileage constraint, and it is not possible to merge without limit; 3. Passenger restrictive principle. The connection between the front and rearsequence trains must meet the passenger flow constraints. When the passenger flow is small and short, the economic and social benefits brought by the merger between the front and rear-sequence vehicles are not high, and there is no need for merger; 4. Speed matching principle. The merging of the front-sequence train and the rearsequence train needs to consider the matching of technical conditions such as the speed of the front and rear trains. Otherwise, the speed difference is too large. On the one hand, the technical conditions are not satisfied, and on the other hand, it affects the preparation of the train diagram.

4.3 Model Construction 4.3.1 A Subsection Sample H ∗ The set of nodes to be processed, that is, the community network divides and identifies overlapping nodes to form a set of nodes to be processed. λh tm

it refers to the arrival and departure line capacity of the node to be processed. refers to the previous train of node h, tm ∈ T Q X .

4 A Study of High-Speed Railway Train Merger and Adjustment Based …

tn mn qhc mn qhb lm ln stm stn Thh

Thz ξh λ L cycle x hmn

47

the subsequent train of node h, tn ∈ T H X . initial passenger flow at node h. connecting passenger flow after merging station h. running mileage of previous trains. running mileage of subsequent trains. refers to the speed level of the previous train. refers to the speed level of subsequent trains. the waiting time at the transfer station, which refers to the waiting time after passengers arrive at the transfer line. Generally, take the average waiting time, Thh = γh / f h , γe is a parameter, generally takes 0.5; f h is the departure frequency of the train number to be transferred on the platform vh . transfer travel time at transfer station. transfer time penalty factor. refers to the mileage limit ratio of multiple units, generally taken as 0.1 [12]. the limit value of the first level maintenance mileage of the EMU. this section focuses on the problem of train merging at overlapping nodes. Assuming that the decision variable is expressed as the merging of the frontsequence train lm and the rear-sequence train lm at station h, there is:  x hmn

=

1, trainsmerger , h ∈ H∗ 0, trains do not merge

4.3.2 Objective Function The principle of train merging is to recombine the previously broken long and longspan passenger flow, aiming to maximize the direct passenger flow: max F1 =



 h∈H ∗

 m∈T Q X

q mn x mn n∈T H X hb h

(4.1)

The train merging is based on the high-speed railway train operation scheme based on the line group. After the merging, the operation benefit and passenger travel cost of the train will change accordingly. In order to maintain consistency, the objective function still needs to consider the benefits of both transport departments and passengers. After the combination of trains, the use of the train bottom is reduced, the fixed cost is reduced, and the benefits and other costs are difficult to calculate and ignored, so the operating benefits are shown in Eq. (4.2). max F2 =

 r ∈R



 p,q∈Vr

t∈T

tr aticket d pq f pq +



 h∈H ∗

 m∈TkQ X

n∈TkH X

a f i x x hmn (4.2)

48

F. Shi et al.

Here is the coordination and optimization of the whole network. The passenger transfer cost should be added when the train is merged. The passenger travel cost is shown in Formula (4.3). minF3 =



 r ∈R

+



 t∈T



h∈H ∗

p,q∈Vr

tr f pq



m∈T Q X

   d pq  + f tr x tr tv p,q∈Vr r ∈R t∈T vn ∈V pq vn n g pq

n∈T H X

mn mn mn (qhc − qhb x h )(Thh + Thz )

(4.3)

4.3.3 Constraint Conditions 1. Maintenance mileage constraint The combined operation of trains can improve the service frequency of long-distance passenger flow, which is conducive to passenger travel. However, in order to ensure transportation safety, the operating mileage cannot exceed the maximum limit of the first-level maintenance mileage of EMUs. The constraints are shown in Formula (4.4). x hmn (lm + ln ) ≤ (1 + λ)L cycle , ∀h ∈ H ∗

(4.4)

2. Transfer times constraint In order to reduce the travel fatigue value of passengers, this paper limits the number of transfer times of passengers to no more than 2, as shown in Eq. (4.5). c( f irj ) ≤ 2, ∀vi , v j ∈ V ; r ∈ R

(4.5)

3. Speed constraint Because there are many mixed lines of high-speed railway in China, it is impossible to ensure the complete consistency of the design speed before and after, so the speed difference between the front train and the rear train is required not to exceed a certain threshold. stm − stn ≤ su , ∀tm ∈ T Q X ; tn ∈ T H X

(4.6)

4. Capacity constraint t If the front and rear trains are combined, the arrival–departure track occupation of the front train and the rear train shall be consistent. Therefore, the capacity of the arrival–departure track will affect the combined number of trains to avoid the problem of insufficient capacity of the arrival–departure track. It is necessary to limit the maximum combined number of trains.

4 A Study of High-Speed Railway Train Merger and Adjustment Based …



x mn m∈TkQ X ,n∈TkH X h

≤ λh , ∀h ∈ H ∗

49

(4.7)

4.4 Algorithm Design The train consolidation adjustment model is a multi-objective model with single decision variable. Because the model is relatively simple, it can be directly solved by genetic algorithm. According to the principle of biological evolution, the genetic algorithm regards the feasible solution as a biological individual, constructs chromosome genes through coding and randomly initializes to generate the first generation population, then selects excellent individuals according to the principle of survival of the fittest, changes and crosses the excellent individuals according to biological principles to generate new populations, and so on, and finally obtains the optimal solution [13]. In this paper, adaptive genetic algorithm is used to solve the problem, and binary coding is adopted. The specific algorithm flow is as follows: Step 1: Initialize the population; Step 2: Calculate the fitness value. The deviation standardization method is used to unify the dimension of the objective function and convert it into the minimized objective function to obtain the fitness function, as shown in Eqs. (4.8)–(4.11). F1∗ =

F1max − F1 F1max − F1min

(4.8)

F2∗ =

F2max − F2 F2max − F2min

(4.9)

F3∗ =

F3max − F3 F3max − F3min

(4.10)

min f = min(F3∗ − F1∗ − F2∗ )

(4.11)

Step 3: According to fitness selection, crossover, and variation, new species were generated. Step 4: Calculate the fitness value f of the new population. Step 5: judge whether the termination condition is met. If it is met, the cycle ends. If it cannot be met, Step 3 is returned.

50

F. Shi et al.

4.5 Case Analysis The research model and method in this paper involve a large number of passenger flow, line, and station data. Because the relevant data are difficult to be counted and processed, a small network example is analyzed with the high-speed railway line within the management of Chengdu Railway Bureau as the basic network data. On the basis of 2021 high-speed railway train operation in the railway operation diagram compilation system V4.0, Xu Zibo of the Bureau of Statistics on high-speed railway lines. There are 10 lines in total, some lines belong to several railway groups. Because only the lines within the jurisdiction of the Chengdu Railway Administration are considered, the boundary station is assumed to be the terminal station, and the distance beyond the boundary station is not considered. At the same time, because the connecting lines of Chengdu, Chongqing, and other hubs are complex, they are treated as a big node after being simplified by the hub station to facilitate modeling and solving, select 62 stations on 10 lines to build a physical network, as shown in Fig. 4.1. Access to relevant information and railway train diagram preparation system V4.0 statistics-related information, obtain the design speed and interval distance of 10 lines in Chengdu Bureau and then count the basic information of the station. Due to the difficulty of calculating the running and waiting time of the transfer, this paper assumes that the transfer time at each cross-line node station is a fixed value. Fig. 4.1 High-speed railway network layout

4 A Study of High-Speed Railway Train Merger and Adjustment Based …

51

Based on the passenger ticket data for the fourth quarter of 2021, the OD passenger flow data on the high-speed rail network within a certain day is counted. Then, the physical network is constructed according to the basic data of the station and the line interval. For the convenience of calculation and representation, the stations are numbered. Secondly, the train merging calculation of the cluster network is carried out. Firstly, the overlapping nodes H = {8,17,23,43,47} of the network are constructed, as shown in Fig. 4.2. The train merging model is solved at five overlapping nodes, and the relevant parameter settings are shown in Table 4.1. After the Python program operation, the objective function convergence diagram is obtained. It can be seen that when the iteration is about 150 generations, it gradually begins to converge and tends to be stable. At this time, the fitness objective function value is −0.412. Then the train merging results are output, and a total of 185 pairs

Fig. 4.2 Road network overlap nodes

Table 4.1 Genetic algorithm parameter setting table

Station

Values

Parameters

100

Population size

300

Maximum iterative algebra

0.8

Cross-probability

0.05

52 Table 4.2 Overlapping node merging (one-way)

F. Shi et al. Overlapping nodes

Combined number of trains

Yibin west node

59

Neijiang node

53

Suining node

42

Zunyi node

25

Guangyuan node

6

of trains within the merged network can be obtained at 5 overlapping nodes. The merging of each overlapping node is shown in Table 4.2. According to the results of the model decoding train merging model, we can get the train merging situation in the network, as shown in Fig. 4.3. According to the results, it can be seen that through the four nodes of Yibin West, Zunyi, Suining, and Neijiang, the passenger flow interruption caused by the regional division on the four highspeed railways of Chengdu–Guiyang, Chongqing–Guiyang, Chengdu–Chongqing, and Shanghai–Chengdu is connected (such as the green line), so that the passenger flow on the four lines is directly transported to meet the direct transportation. In addition, the five overlapping nodes also realize the cross-line connection between lines (such as red line representation), which satisfies the direct transportation of long and large cross-line passenger flow in the road network.

Fig. 4.3 Train merger schematic

4 A Study of High-Speed Railway Train Merger and Adjustment Based …

53

Table 4.3 Train operation by line (one-way) Line

Yu-Gui railway

Number of trains on this line

Number of cross-line trains

Total number of trains

Class A train

Class B trains

8

57

65

17

48

Chenggui passenger dedicated line

62

31

93

29

64

Chengdu–Chongqing high-speed railway

19

82

101

76

25

Guangzhou customer dedicated

18

47

65

54

11

Shanghai–Kunming high-speed railway

35

70

105

95

10

Hurong line

42

46

88

14

74

Xicheng passenger line 42

50

92

14

78

Yuwan passenger dedicated

14

13

27

7

20

Mianlu high-speed railway

0

11

11

10

1

Lanyu line

3

13

16

2

14

According to the combined situation, the road network can be a total of 437 lines in 1 day (one-way), each line of trains is shown in Table 4.3. According to the calculation of the internal operation plan model and the train merging model, a total of 437 high-speed trains (one-way) are operated in the highspeed railway network, including 227 trains of class A, 210 trains of class B, 184 trains of this line, and 289 trains of cross-line. Through the formulation of the train operation plan of the high-speed railway network in Chengdu Bureau, and in close connection with the passenger flow demand, the reasonable division of the region is realized, and the transfer and direct transportation of cross-line passengers are reasonably organized, which greatly reduces the difficulty of the preparation of the high-speed railway train operation plan under the network condition, and verifies the feasibility of the scheme and model algorithm proposed in this paper.

4.6 Conclusion An optimization method of high-speed railway train operation plan based on the three-level network division of road network–region–cluster is proposed. The optimization problem of train operation plan under the condition of network formation is transformed into the direct train operation based on the cluster network and the train merger adjustment problem on the regional network. On the road network, in order to improve the direct accessibility of long-span cross-line passenger flow, the

54

F. Shi et al.

principle of train combination is analyzed. According to the constraints of transfer times, train operation mileage, arrival and departure line capacity, speed, and so on, the train combination adjustment model is constructed with the lowest travel cost, the highest direct passenger flow, and the highest train operation benefit as the optimization objectives. Based on the genetic algorithm, the optimization of high-speed railway train operation plan under the condition of network formation is realized. Taking the small network composed of high-speed railways in Chengdu Railway Bureau as an example, the train operation plan is determined according to the optimization method studied in this paper. According to the train operation plan model within the group network and the train merging adjustment model on the regional network designed above, a total of 437 high-speed trains (one-way) are operated in the high-speed railway network, including 227 trains of class A, 210 trains of class B, 184 trains of this line, and 289 trains of cross-line, which verifies the feasibility and rationality of the optimization method of high-speed railway train operation plan based on three-level network division proposed in this paper. Acknowledgements The authors would like to thank Prof. Shaoquan Ni and Prof. Dingjun Chen for their constructive comments. This research was supported by the National Natural Science Foundation of China (Project No. 52072314; 52172321;52102391), Sichuan Science and Technology Program (Project NO. 2020YJ0268; 2020YJ0256; 2021YFQ0001; 2021YFH0175), Science and Technology Plan of China Railway Corporation (Project No.: 2019F002), China Shenhua Energy Co., Ltd. Science and Technology Program (Project No.: CJNY-20-02), China Railway Beijing Bureau Group Co., Ltd. Science and Technology Program (2021BY02, 2020AY02), the National Key R&D Program of China (2017YFB1200702), Chengdu Bureau Group Co., Ltd. Science and Technology Program (Project No.: CX2103),and Sichuan Science and Technology Program (2022020).

References 1. Scholl, S.: Customer-oriented line planning. Dissertation (2006) 2. Fu, H.L., Lei, N., Yang, H.: Research on the method for optimization of candiate-train-set based train operation plans for high-speed railways. J. China Railway Soc. 32(6), 1–5 (2010) 3. Dong, S.Q., Yan, H.F., Li, Q.R.: Research on genetic optimization of operation scheme for PDL trains based on network flow distribution. China Railway Sci. 33(4), 105–110 (2012) 4. Sun, H.P.: The Compilation of Passenger Train Operation Plan for High-speed Railway of Networked. Lanzhou Jiaotong University (2020) 5. Chang, Y.-H., Yeh, C.H., Shen, C.-C.: A multi-objective planning model for passenger train rail line: application to Taiwan’s high-speed research. Transp. Res. Part B (34), 91–106 (2000) 6. Schobel, A., Scholl, S.: Line Planning with Minimal Traveling Time. University of Gottingen, Germany (2005) 7. Hadas, Y., Shnaiderman, M.: Public-transit frequency setting using minimum-cost approach with stochastic demand and travel time. Transp. Res. Part B Methodol. 46(8), 1068–1084 (2012) 8. Niu, Y.T.: Research on Theory and Method of Passenger Dedicated Line Train Operation Plan Compilation under Network Condition. Beijing Jiaotong University (2012) 9. Zhang, L.X.: Study on Train Operation Plan of High-Speed Railway Network Based on Network Evolution. Beijing Jiaotong University (2017)

4 A Study of High-Speed Railway Train Merger and Adjustment Based …

55

10. Wang, Y.F.: Study on Train Operation Scheme of Intercity Railway Network. Southwest Jiaotong University (2020) 11. Sun, Y., Chu, S.-C., Hu, P., Watada, J., Si, M., Pan, J.-S.: Overview of parallel computing for meta-heuristic algorithms. J. Netw. Intell. 7(3), 656–684 (2022) 12. Xi, Z.W.: Study on Route Optimization of Passenger Train under Network Condition. Southwest Jiaotong University (2018) 13. Xu, Z.B.: Task Intelligent Scheduling Method for Cloud Data Center Based on Hybrid Particle Swarm Optimization. Beijing Jiaotong University (2020)

Chapter 5

Research on Platform Door Setting of Suburban Railway of Mass Transit Type Zhang Wenxin, Yang Yilin, Song Qingguo, Zhu Chengli, and Shaoquan Ni

Abstract The special function orientation and passenger flow characteristics of suburban railway put forward higher demands for its transportation organization, such as mass transit type, rapidness, diversification, and convenience. Under the condition of suburban railway of mass transit type, the characteristics of high density in and out of high-speed trains will affect the safety of waiting and landing at passenger platforms. Based on the current passenger organization and technical level, this paper discussed the necessity and technical feasibility of setting platform door in suburban railway. The results show that the installation of platform door can greatly improve the passenger experience, and the technical conditions for the development of platform door system in suburban railway have been met. It is suggested that the new suburban railway should consider the installation of platform gate or reserve the engineering conditions for the installation of platform gate.

Z. Wenxin · Y. Yilin · S. Ni (B) School of Transportation and Logistics, Southwest Jiaotong University, Chengdu 611756, China e-mail: [email protected] Z. Wenxin e-mail: [email protected] Y. Yilin e-mail: [email protected] S. Qingguo China Railway Chengdu Bureau Group Co. Ltd., Chengdu 610082, China Z. Chengli Yibin Railway Industry Investment Co. Ltd., Yibin 644000, China S. Ni National Engineering Laboratory of Integrated Transportation Big Data Application Technology, Chengdu 611756, China National and Local Joint Engineering Laboratory for Intelligent Integrated Transportation, Chengdu 611756, China © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_5

57

58

Z. Wenxin et al.

5.1 Introduction With the acceleration of urbanization and the improvement of people’s living standard in China, passengers have become more and more demanding on the convenience, public transportation, and comfort of transportation. The suburban railway (or municipal railway), as a special means of transport for fast passenger transport between inner city and sub-central city, important town and city group, greatly facilitates the daily commute to school, business, leisure, family visit, entertainment, and shopping travel, its fast operation speed can shorten the space and time distance between the core area of the city and the surrounding main areas, form a “one-hour traffic circle”, enhance the radiation role of the central city, and strengthen the division of labor and cooperation between various regions and towns. In recent years, China’s suburb has developed rapidly. At present, Beijing suburban railway, Jinshan Railway, Chengguan railway, Chengpu railway, Wenzhou suburban railway S1 line, and other suburban railways have been operated in China, and many suburban railways are also under construction. The characteristics of temporal and spatial passenger flow of suburban railway determine its high requirements for convenience and speed of travel process, so it is proposed to adopt “ mass transit type” to operate, that is, adopt high-density small marshaled trains, platform waiting, free seats, and other operating methods. Due to the high traffic density and high frequency of departure of suburban railway, a series of aerodynamic problems will be caused when the train enters the station at high speed, which threatens the safety of passengers and staff waiting at the platform. In view of this, it is necessary to refer to the setting of the platform door of urban rail transit and analyze the setting of the platform door under the mass transit type of suburban railway public transportation operation.

5.2 Requirements for Mass Transit Type of Suburban Railway Operation 5.2.1 Requirements for Suburban Railway Organization Suburban railway, also known as commuter railway and municipal railway, mainly serves between the central city and the suburbs, the major towns of the group type, and the central city and satellite cities. It is a passenger rail transit system within the metropolitan area characterized by fast, high density, small group, and mass transit type. The functional positioning of suburban railway is between the railway trunk system and the urban rail transit system, which can meet the commuting demand internally and effectively connect the railway trunk line externally. It can not only realize the rapid transportation between the central urban area and the peripheral towns but also realize passengers’ travel between the surrounding cities. Due to the special functions of municipal railway, compared with urban rail transit, it has the

5 Research on Platform Door Setting of Suburban Railway of Mass Transit …

59

following characteristics: longer transportation distance, larger station spacing, and higher transportation speed. These characteristics make suburban railway become an important factor in the construction of “one-hour traffic circle” of regional rail transit. In terms of passenger flow, suburban railway, as the main mode of transportation from satellite cities or outer suburbs to the urban center, has the characteristics of urban integration and commuting. At the same time, the passenger flow borne by the suburban railway has a certain relationship with the distance from the urban area. Different distances from the urban area, the purpose, and characteristics of the passenger flow are different. In the suburban areas, people who live in the suburbs and go to the central area to work and go to school are the main ones. This part of passenger flow has a wide range and a large number, with obvious peak period, and higher requirements for the convenience of travel. The average travel distance of the passenger flow in the outer suburbs is longer, and the requirements for the speed and comfort of the travel are higher. The special function orientation and passenger flow characteristics of suburban railway also put forward special requirements for transportation organization: (1) Mass transit type Due to the large proportion of commuter and general school passengers in the current suburban railway passenger flow, high requirements are put forward for the punctuality and convenience of trains, and they hope to arrive and leave on time to reduce the waiting time. Therefore, the operation and organization are required to simplify the entry process as much as possible and increase the frequency of trains. (2) Rapidness The basic goal of the suburban railway is to realize the 1-hour traffic circle between the regional central city and the important towns. Because the suburban railway in the municipal area has a long route, in order to attract more suburban residents to take the suburban railway, the travel speed should be increased as much as possible in the operation organization to meet the requirements of passengers’ speed. (3) Diversification Due to the diverse characteristics of passenger flow along the suburban railway, flexible operation organization should be adopted, such as the combination of fast and slow trains, large and small crossroads, etc., to meet the travel needs of different passenger flows as far as possible. (4) Convenience The suburban railway brings the passenger flow from the surrounding towns into the central city from different directions, and passengers need to evacuate through the public transport modes in the central city, which will cause great transfer pressure to the central city. Therefore, convenient transfer conditions between the municipal railway and other urban public transport modes are required.

60

Z. Wenxin et al.

5.2.2 Requirements for Mass Transit Type of Suburban Railway Operation To improve the commuter rail service commuter travel convenience, the National Development and Reform Commission in 2020 “Suggestions on accelerating the development of suburban (municipal) railways in metropolitan areas” pointed out that mass transit-type organizations should be carried out and some measures such as increasing the number of trains during rush hours should be taken to push the “station stop” and “big station stop” combination of flexible transport organization mode, provide diversified and convenient travel services [1]. There are two main characteristics of mass transit-type operation: (1) The interval between trains is less than 60 min, and passengers do not need to choose their travel time according to the train schedule. (2) No designated train number and seat, passengers can buy tickets at any time, and do not need to make a reservation in advance. Compared with the non-mass transit type, this operation mode cancels the link of waiting at the station hall and booking tickets, reduces the time of passengers buying tickets and waiting for buses, and improves the efficiency of daily travel. The suburban (municipal) railway mainly serves the circulation in the core areas such as the peripheral residents of the metropolitan area and the central business district and has higher requirements for timeliness and convenience. Compared with the traditional railway, mass transit type of operation is the prominent feature [2]. The requirements for mass transit type of suburban railway operation are as follows: (1) Ticket service Compared with the traditional railway booking in advance, the ticket purchase method of suburban railway with mass transit-type operation is closer to the subway and other urban rail transit, the ticket can be purchased at the station, no limit to the number of trains, and no limit to seats. And passengers can take credit card, mobile phone, face recognition and other more convenient and fast way to get into the station. (2) Train services Compared with the railway, the suburban railway provides higher density service, adjusts the departure frequency according to the peak hours of passenger flow, and maximally meets the needs of different passenger flows. At the same time, multiple doors are used to quickly get on and off the train, evacuate passengers in time, and alleviate the gathering degree of waiting crowd on the platform. (3) Platform waiting mode Traditional railway tends to adopt the waiting room waiting mode, cannot meet the demand of the peak passenger flow, high traffic density under the condition of the peak passenger flow, and train operation density. Now, the suburban railway gradually adopts the platform waiting mode, in order to reduce the passenger on and off time, improves not only transportation efficiency, but can also cause a large number of

5 Research on Platform Door Setting of Suburban Railway of Mass Transit …

61

passengers gathered waiting on the platform; so, it is necessary to optimize the design of the platform to achieve a multi-entrance, multi-direction platform layout.

5.3 Study on the Necessity and Feasibility of Platform Door Setting in Suburban Railway 5.3.1 The Necessity of Platform Door Setting in Suburban Railway Existing railway stations often adopt the waiting hall waiting mode, in which passengers need to arrive at the station at least half an hour in advance, wait in the waiting room after completing security and identity verification, and the station releases passengers to the platform a few minutes before the train arrives. The time for passengers to wait on the platform is short, stations only need to send a few staff to manage the platform order to basically ensure the safety of passengers during waiting time. At present, many suburban railways in operation are reconstructed from existing railways [3]. However, due to the high density and high efficiency required by the mass transit type of suburban railway operation, the waiting mode of waiting hall cannot meet the demand of suburban train passengers for efficiency. Therefore, the platform waiting mode becomes a more suitable waiting mode of suburban railways. In the platform waiting mode, passengers enter the platform area directly to wait for the train. The waiting time is long, and it is difficult to queue up for the train in an orderly manner for a long time, which increases the security risks of the platform waiting. Even if more staff are brought in to maintain order, it is difficult to completely eliminate safety risks. Platform door has become an inevitable choice to ensure the safety of waiting trains. The platform door is set at the edge of the station platform, which can safely isolate the waiting area from the rail area, minimize the risk of train operation to the waiting passengers on the platform, and prevent accidents. In addition to ensuring the safety of waiting trains, the platform door also has the following functions: (1) Reduce investment and save energy. Some suburban railways need to build new stations or landing stations, which requires a larger investment, a longer construction period, and a larger area. If the platform door is set up, the platform can be directly used to wait for trains after the canopy, ticket sales facilities, and guidance system are set up, and the investment and land resources of the station house construction can be reduced. In addition, using platform doors to ensure the safety of passengers waiting at the platform also reduces the pressure on staff. Underground station set up the platform closed door can be completely isolated from the train operation space station space, avoids large number of air conditioning cold air into the tunnel, to reduce the train brake when they radiate

62

Z. Wenxin et al.

Fig. 5.1 Streamline comparison of different waiting modes

heat into the waiting area, reduce the consumption of cold energy, achieve the goal of air-conditioning energy saving, reduce the air conditioning equipment capacity, investment construction area, and reduce the air conditioning room accordingly. (2) Simple ride procedures, save time for passengers. After the platform door is set, the waiting mode is changed from waiting at the waiting hall to waiting at the platform. The link between ticket booking and waiting at the station hall is canceled, which simplifies the process of passengers’ travel, saves the time of travel preparation, and improves the convenience of daily travel. The time required by different waiting modes is shown in Fig. 5.1. (3) Convenient transfer with suburban rail transit. The current subway system is equipped with platform doors centrally controlled by the signal system. The suburban railway with platform doors can unify the platform management with the subway to facilitate passengers’ transfer. (4) Reduce the impact of trains running on the platform. The safety gate system installed on the platform can reduce the influence of wind pressure, noise, and pulse power of the high-speed train on the waiting passengers. (5) Provide information. Safety doors can be embedded with visual devices such as LCD screens to display real-time information such as train arrival time and door number.

5.3.2 The Feasibility of Platform Door Setting in Suburban Railway Although platform doors have been widely used in suburban subway or light rail, they are rarely used in suburban railway. Therefore, in order to realize the safe application of platform doors system in suburban railway, the following technical problems caused by the difference between municipal train and subway must be solved. Platform doors for different types of train. In the urban rail transit system, a line usually operates only one kind of train, while the municipal railway has a variety of

5 Research on Platform Door Setting of Suburban Railway of Mass Transit …

63

trains, and the number and location of the doors of each type of train are different, which requires the platform door system to be compatible with a variety of trains. The Japanese scientific research team divided the platform door into translation type and lifting type. The prominent feature of the translational type is to make up for the accuracy error of the train door position and parking by retreating the platform edge to a certain distance, so as to realize the smooth ride and landing of passengers. The remarkable characteristic of lifting rope railing is that it adopts simple lifting rope or railing device for safety protection, which can meet the requirements of train stop when various models are mixed running. This protection method has short construction cycle and low construction cost. These two kinds of station doors are mainly to improve the compatibility of models by simplifying the structure. However, this way reduces the service level, cannot realize the effective partition, and block the high-speed wind pressure of waiting area, there is a great safety risk, cannot meet the development demand of Chinese municipal railway. In order to accommodate a variety of vehicle types, the platform doors of intercity railway in China are usually set at a certain distance back from the platform edge, but this method will expand the scale of the station, occupy underground suburban space, and increase the construction cost. In addition, a folding sliding door system compatible with multiple models has been developed, which realizes the compatibility to several fixed models, but it can only meet the combination of fixed models and has certain limitations. Therefore, combined with the Japanese lifting protective fence and the domestic platform door setting scheme, a lifting platform door design scheme adapted to multiple models is proposed [4]. The lifting platform door is composed of two largewidth doors in the vertical direction. When the doors are opened, they are overlapped above, as shown in Fig. 5.2. The lifting platform door system has a simple structure and strong compatibility with various trains, which can effectively meet the needs of interconnection between suburban railway and other rail transportation. Fig. 5.2 The lifting platform door

64

Z. Wenxin et al.

Vehicle–ground information transmission. In view of the characteristics of high traffic density and high speed of municipal railway, it is necessary to establish an orderly management order. Once passengers fall off the track and are squeezed by doors and platform doors, it will cause great confusion for passengers and bring a serious impact to transportation. If the platform door is not equipped with the corresponding vehicle–ground linkage system, the opening and closing of the train door and platform door must be manually confirmed and controlled by the passenger. Once the operator’s operation error, when the train does not stop at the prescribed position, the wrong opening of the door or platform door may cause the platform waiting passengers or passengers on the train to fall off the track or squeeze events. In addition, if the platform door fails and cannot be closed, if the linkage system is not set, the train doors may be closed and passengers are still crowded in the area between the platform door and the train. At this time, the train will start, and passengers may be crowded and fall off the track. The application experience of urban rail transit proves that the platform safety door screen door must be equipped with the corresponding vehicle-ground control system, in order to give full play to its role of ensuring safety and improving efficiency. The vehicle-ground joint control system will provide clear signals and open authorization to the crew after the “stationary” state have ensured to ensure that the doors open and close at the same time. The platform door without linkage system has certain security risks. Not only that, platform door linkage control can also reduce the number of platform personnel, to achieve the purpose of reducing staff efficiency. The platform doors and doors have been incorporated into the signal system ATO control in the newly built suburban rail transit projects. ATO system has vehicle– ground communication system, which can realize the vehicle-ground information exchange and meet the requirements for the linkage control of train doors and platform doors. According to the current technical status at home and abroad, there are two technical schemes that can realize vehicle–ground information transmission: dedicated 2.4 GHz spread spectrum data transmission station+dedicated secure communication protocol and GSM-R. Scheme 1: 2.4 GHz spread spectrum data transmission station + dedicated secure communication protocol [5, 6] 2.4 GHz ISM spread spectrum communication technology has been increasingly widely used at home and abroad, such as passenger information system (PIS), communication-based train control system (CBTC), and so on have adopted this technology. However, this technology can only support running speed below 120 km/h, which cannot meet the running speed requirements of suburban railways. In addition, a large part of suburban railway lines will be located on the ground, and the electromagnetic environment is complex, which is difficult to ensure the reliability of data communication. Therefore, it is not feasible for suburban railway to adopt WLAN-based vehicle–ground wireless communication .

5 Research on Platform Door Setting of Suburban Railway of Mass Transit …

65

Fig. 5.3 Vehicle–ground information transmission process

Scheme 2: GSM-R [7] GSM-R system is developed on the basis of GSM, which adds a special communication technology of railway special service on the basis of GSM technology. It has the advantages of mature and reliable technology, suitable for large-scale networking and high-speed railway. ATO realizes vehicle–ground communication based on the GSM-R network (Fig. 5.3). The onboard device or the driver triggers the command to open or close the train, which is sent to the CCS and the gating device through the GSM-R network. The CCS sends the command to the train control center (TCC) of each station. After receiving the platform door control command sent by the temporary speed restriction server (TSRS), the TCC drives the corresponding relay to control the opening/closing action of the corresponding platform door. TCC collects the gate state relay information provided by the platform gate control system and sends it to the TSRS and CTC station equipment, and TCC implements continuous and real-time monitoring of the station gate state relay.

5.4 Conclusions The mass transit type of suburban railway operation puts forward the demand of the platform waiting mode, and the establishment of the platform door is an effective means to prevent passengers from falling off the track when they are waiting on the platform, which plays a positive role in improving the safety of the passenger waiting, improving the conditions of the passenger waiting, improving the efficiency of the

66

Z. Wenxin et al.

bus, saving resources, and reducing the investment cost. Through the analysis of the existing platform door-related technology, the technical conditions for the development of platform door system in suburban railway have been met. It is suggested that the new suburban railway should consider the setting of platform door or reserve the engineering conditions for the installation of platform door system. Acknowledgements This research is supported by the Strategic Research and Consulting Project of Chinese Academy of Engineering (2021-DFZD-20) and Key science and technology projects in the transportation industry of the Ministry of Transport (2022-ZD7-132).

References 1. Suggestions on accelerating the development of urban (suburban) railways in metropolitan areas. China Metro (12), 54–56 (2020) 2. Liansong, G.: Research and countermeasures on high-quality urban region (suburban) railway development in metropolitan areas. Mod. Urban Transit 2022(06), 6–11 (2022) 3. Xiaoliang, S.: Research on the necessity of setting platform screen doors in high-speed railway and intercity railway stations. Constr. Des. Eng. 2019(01), 100–102 (2019) 4. Yinlong, Z.: Key technologies of smart platform door systems of urban railways. China Mech. Eng. 32(04), 475–480 (2021) 5. Zhihui, Z.: Discussion on the feasibility of train-ground integrated control of platform screen doors on intercity passenger dedicated lines. Railway Signal. Commun. 7(05), 1–4 (2010) 6. Xingrui, Lü., Yingying, Z.: Research on vehicle-ground wireless communication technology of CBTC system. J. Zhengzhou Railway Vocational Tech. College 33(03), 4–6 (2021) 7. Wei, Y.: Discussion on linkage between train doors and platform screen doors of CTCS2 +ATO system. Railway Signal. Commun. Eng. 19(07), 34–39 (2022)

Chapter 6

Research on Equipment Operation and Maintenance Management Technology of Large Railway Passenger Station Bozhou Wang, Lexi Li, Shaoquan Ni, and Dingjun Chen Abstract With the development of informationization and intelligence of railway passenger stations, problems such as inconvenient information interaction, missing operation and maintenance data, and difficulty in accurate positioning of equipment under the existing equipment operation and maintenance management mode have become the focus. In this paper, the operation and maintenance management process of equipment is divided into three stages: fault prediction and warning, fault diagnosis and processing, and fault rule summary. The implementation schemes of key technologies such as data warehouse, data mining, 5G fusion positioning, and electronic fence are given to realize functions such as condition assessment, fault prediction, fault diagnosis, precise positioning, fence warning, and auxiliary decision-making, which can meet the needs of managers and operations people. Research will help to improve the efficiency and safety of equipment operation and maintenance, and have a reference significance for integrated intelligent operation and maintenance technology and the construction of modern passenger stations.

B. Wang · L. Li · S. Ni (B) · D. Chen School of Transportation and Logistics, Southwest Jiaotong University, Chengdu Sichuan 610031, China e-mail: [email protected] B. Wang e-mail: [email protected] L. Li e-mail: [email protected] D. Chen e-mail: [email protected] S. Ni · D. Chen National and Local Joint Engineering Laboratory of Comprehensive Intelligent Transportation, Southwest Jiaotong University, Chengdu Sichuan 610031, China National Engineering Laboratory of Integrated Transportation Big Data Application Technology, Chengdu Sichuan 610031, China © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_6

67

68

B. Wang et al.

6.1 Introduction In order to meet the increasing travel demands of passengers, the scale of railway passenger stations is constantly expanding, the types of facilities and equipment are increasingly diverse, and the equipment operation and maintenance data have entered the era of massive quantification. The traditional equipment monitoring, warning, operation, and maintenance management mode is mainly based on manual inspection, which is costly and labor-intensive. Moreover, it is unable to detect and deal with equipment faults in time, resulting in certain security risks. With a large number of modern facilities and equipment put into use in passenger stations, the existing equipment operation and maintenance management mode is mainly faced with the following problems [1]: (1) the degree of equipment informatization is different, there is the phenomenon of information islands; (2) the equipment has not achieved full life cycle management, there is a phenomenon of missing data; (3) it is difficult to accurately determine the equipment area, there is a phenomenon of missing equipment detection. In order to solve the above problems, based on the whole process of equipment operation and maintenance management, through data warehouse, data mining, 5G fusion positioning, electronic fence and other technologies, assist managers, and operations people carry out related work, so as to improve the intelligent level of equipment operation and maintenance management of passenger stations.

6.2 Equipment Operation and Maintenance Management Stage Division The operation and maintenance management process of railway passenger station equipment can be divided into three stages: fault prediction and warning, fault diagnosis and processing, and fault rule summary [2], as shown in Fig. 6.1. Railway passenger station equipment operation and maintenance management is usually recorded in work orders or other forms, and a large number of rules are hidden in time-based work orders, such as the periodicity of maintenance time and the interrelationship of maintenance events [3]. Through data warehouse technology, the unified management of data in various heterogeneous data sources can be realized, and the data with poor quality can be eliminated and processed to lay the foundation for data analysis. Through data mining technology, hidden information can be found from a large amount of data, and then functions such as equipment failure prediction, condition assessment, regular statistics, and auxiliary decision-making can be realized [4]. In addition, 5G fusion positioning technology can accurately determine equipment failures and dangerous areas, and provide navigation services for operations people, further improving the efficiency and safety of equipment operation and maintenance [5].

6 Research on Equipment Operation and Maintenance Management …

69

Fig. 6.1 Passenger station equipment operation and maintenance management process

6.3 Key Technologies of Equipment Operation and Maintenance Management 6.3.1 Equipment Operation and Maintenance Data Warehouse Technology The equipment operation and maintenance data warehouse mainly integrates various data related to equipment operation and maintenance through ETL (extraction, transformation, and loading), and divides it into different subject fields. For equipment operation and maintenance management, subjects such as failure time, failure rate, and operations people can be selected. After subject division, a one-layer dimension star model can be used to organize data to improve large-scale operation efficiency. Figure 6.2 shows the star model of equipment operation and maintenance subject. After the foundation construction of the data warehouse is completed, OLAP and data mining technology can be used to analyze the equipment operation and maintenance data [6]. The comparison between the two technologies is shown in Table 6.1. Compared with data mining technology, OLAP is more intuitive and convenient, but it lacks the mining of hidden information and cannot achieve predictive operations. This paper focuses on the application of data mining technology in equipment operation and maintenance management.

70

B. Wang et al.

Fig. 6.2 Star model of equipment operation and maintenance subject

Table 6.1 Comparison of data analysis techniques Data analysis technology

Focus

Technical implementation

Analysis results

OLAP

Quickly analyze and process historical data, perform complex queries

Drill, slice, dice, rotate multidimensional cubes

Intuitive results such as charts

Data mining

Mining potential regular information, make predictions, or assist decision-making

Association rules, linear regression, neural networks, and other algorithms

Conclusion of association rules, knowledge base

6.3.2 Equipment Operation and Maintenance Data Mining Technology Equipment failure prediction can provide a basis for the formulation of equipment operation and maintenance plans and security plans. Through the correlation analysis, the main factors can be obtained from a large number of potential equipment failure factors, which lays the foundation for the equipment failure prediction analysis. For different types of equipment failure influencing factors, different correlation analysis methods can be used [7], as shown in Table 6.2. On the basis of correlation analysis, the quantitative relationship between equipment failure and various factors such as using time, environment, and storage conditions can be further obtained through regression analysis, and the equipment failure time and failure rate can be predicted. The multiple linear regression model is often used [8], and the form is as follows: y = β1 x1 + β2 x2 + · · · + ε

(6.1)

6 Research on Equipment Operation and Maintenance Management …

71

Table 6.2 Correlation analysis method of equipment failure influencing factors Equipment failure factors

Type of data

Correlation analysis method

Judgment criteria

Service time and load rate

Continuous

Person correlation analysis

Person correlation coefficient

Equipment model, temperature, and humidity

Classification

ANOVA

P value

Month, special weather, and large passenger flow

Unstructured

Statistical comparative analysis

Volatility

Among them, x is the independent variable, that is, the main failure factors of each equipment, y is the dependent variable, that is, the equipment failure time or failure rate, β and ε are the slope and intercept, respectively, obtained by the least squares method after substituting the data. By analyzing the existing equipment historical fault data set, the fault correlation law of each equipment can be obtained, which can assist in the diagnosis and processing of equipment faults. Apriori is a commonly used association rule mining algorithm, but it needs to scan the data set multiple times. When the amount of data is too large, the computational efficiency is low. Therefore, consider using the FPGrowth algorithm to mine frequent item sets, and then generate strong association rules [9]. If the equipment historical fault transaction data set is D, each piece of fault historical data is a transaction T , the fault type, fault cause, and other factors are items, and I = {I1 , I2 , · · · , In } is the set of all items. The association rule is X ⇒ Y , where X ⊂ I , Y ⊂ I and X ∩ Y = ∅. Support and confidence are important indicators to measure association rules [10], which are defined as follows: Suppor t(X ⇒ Y ) =

Count(XU Y ) |D|

Con f idence(X ⇒ Y ) =

Count(XU Y ) Count (X )

(6.2) (6.3)

where Count(X ) = |{Ti |X ⊆ Ti , Ti  D}|, |D| is the total number of transactions in the data set. The flow of the FP-Growth algorithm is as follows [11]: (1) Scan the equipment historical fault transaction data set and calculate the support of each item. Arrange all items in descending order of support to generate frequent one-item set and index tables. The frequent itemset is the set that satisfies the minimum support degree specified by people. (2) Construct FP-tree. Create the root node “null” of the tree, scan the historical fault transaction data set, add transactions according to the index table, and one transaction corresponds to one branch. Create an item header table, and each item points to the corresponding position in the FP-tree through the node chain.

72

B. Wang et al.

(3) Mining frequent item sets through FP-tree. The item header table is traversed according to the support degree from low to high, the conditional pattern base is constructed by the combination of each item prefix path, and the conditional FPtree satisfying the minimum support count is generated. Combining conditional FP-tree with suffix gets frequent K-item set. All proper subsets of frequent K-item set are arranged and combined, and the generated strong association rules meet the minimum confidence requirement specified by people. The generated strong association rules are stored in the fault knowledge base and retrieved when fault diagnosis is performed. In addition, when data mining of passenger station equipment operation and maintenance data involve confidential/sensitive information [12] or data generated under specific time periods such as holidays [13], or equipment added values, related utility, interest, and risk need to be considered [14, 15], suitable algorithms can be used in conjunction with data characteristics and application requirements.

6.4 5G Fusion Positioning Technology for Passenger Stations As an important transportation hub, the passenger station constitutes a complex interior space integrating multiple functions. Due to the weak GPS signal in indoor environment, commonly used navigation software cannot play a role. This paper uses 5G + WiFi-PDR fusion positioning technology to locate faulty equipment and operations people with high precision. WiFi positioning and PDR positioning are mainstream indoor positioning technologies, and their comparisons are shown in Table 6.3. Both positioning technologies rely on the built-in sensing devices of smartphones and have low comprehensive costs, which is suitable for implementation in large railway stations. The 5G network has the characteristics of “high bandwidth, low Table 6.3 Comparison of WiFi positioning and PDR positioning Indoor positioning technology

Advantages

Disadvantages

WiFi positioning

Faster data transmission and processing speed, lower cost

The positioning accuracy depends on the quality of the WiFi base station and access point (AP) equipment, the signal strength is affected by the environment, the amount of data collection is large

PDR positioning

Simple calculation, easy to implement using the built-in inertial measurement unit (IMU) of the phone

The positioning accuracy is affected by many factors, such as sensor accuracy and pedestrian step estimation

6 Research on Equipment Operation and Maintenance Management …

73

latency, and wide connection”, which can improve the efficiency of data transmission and processing and ensure the accuracy of real-time positioning. At the same time, the unscented Kalman filter algorithm (UKF) can be used to fuse the WiFi and PDR positioning results to eliminate the influence of environment factors [16]. The algorithm flow is as follows: (1) The initial position coordinates Ps = (xs , ys ) of equipment and operations people are measured through 5G + WiFi positioning, which is used as the initial value of PDR positioning and the starting value of UKF algorithm. 5G + WiFi positioning is carried out in two stages: • Offline training stage. The interior of the passenger station is divided into several spatial grids, and n wireless APs are arranged. The received signal strength of each AP point is measured at the grid intersection as a feature quantity, and a location fingerprint database is constructed. For any grid intersection, the location fingerprint database stores its coordinates Pi = (xi , yi ) and received signal strength vector V i = [vi,1 vi,2 · · · vi,n ], vi, j is the average signal strength received from the j-th AP device. • Online positioning stage. Measure the signal strength of a group of AP points V s = [vs,1 vs,2 · · · vs,n ] at the location to be tested, and calculate the Euclidean distance from each sampling point in the location fingerprint database:

Ds,i

  n   2 vs,k − vi,k =

(6.4)

k=1

Ds,i is the Euclidean distance between the position to be measured and the i-th sampling point. Use the K-nearest neighbor method (KNN) to determine the coordinates of the position to be measured, and set the coordinates of the sampling points corresponding to the first k smallest values as {P1 , P2 , · · · , Pk }, then the calculation formula of the coordinates of the position to be measured Ps is as follows: 1 Pi k i=1 k

Ps =

(6.5)

(2) Through the built-in inertial measurement unit (IMU) of the phone, it is judged whether the operations people or the staff who operate the equipment are walking. After the pedestrian is detected to complete one-step walking, calculate the pedestrian’s step length d and the orientation angle change θ , calculate the PDR positioning result, and perform 5G + WiFi positioning measurement synchronously. The initial position coordinate is Ps = (xs , ys ), then the PDR positioning calculation coordinate P1 = (x1 , y1 ) after walking one step, the formula is as follows:

74

B. Wang et al.



x1 = xs + dcosθ y1 = ys + dsinθ

(6.6)

(3) According to the 5G + WiFi positioning and PDR positioning results, the observation equation and the state equation are iteratively updated to obtain the UKF fusion positioning result. The state equation of the system: ⎤ ⎤ ⎡ ˜ xk−1 + dcosθ xk k−1 ⎢ ⎥ ˜ X k = ⎣ yk ⎦ = ⎣ yk−1 + dsinθ k−1 ⎦ + W k−1 ∼ θk θk−1 + θ ⎡

(6.7)

Among them, (xk , yk ) are the coordinates of the position of the pedestrian after the k step, θk is the orientation angle of the pedestrian after the k step, d˜ ∼

is the step length of the k-th pedestrian, and θ is the change amount of the k-th pedestrian’s heading angle. W k−1 is a three-dimensional system observation noise. The observation equation of the system: ⎤ ⎡ ⎤ xk xk ⎥ ⎢ y ⎥ ⎢ yk ⎥ ⎢ k ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ Z k = ⎢ dk ⎥ = ⎢ (xk − xk−1 )2 + (yk − yk−1 )2 ⎥ + V k ⎥ ⎢ ⎥ ⎢ ⎦ ⎣ θk ⎦ ⎣ θk − θk−1 θk θk ⎡

(6.8)

Among them, (xk , yk ) are the coordinates of the pedestrian’s position measured by the 5G WiFi positioning technology, dk is the step length of the k step of the pedestrian calculated by the accelerometer in the IMU, θk is the change amount of the pedestrian’s orientation angle at the k-th step, and θk is the pedestrian’s orientation angle after the k-th step. V k is a five-dimensional systematic observation noise. The precise location of the faulty equipment and operations people can be obtained through the above fusion positioning algorithm. On this basis, the shortest path algorithm can be used for path planning, and the navigation function can be realized through the GIS visualization platform. In addition, to ensure the security of relevant data storage and transmission during functional services, a client–server architecture scheme with bilinear pairings and elliptic curve cryptography techniques in 5G networks can be considered [17].

6.4.1 Electronic Fence Technology The electronic fence technology can ensure the safety of operation and maintenance by generating virtual fence areas. There are two main application scenarios in

6 Research on Equipment Operation and Maintenance Management …

75

Table 6.4 The judgment method of relative position Fence area shape

Judgment method

Round

Distance from location point to the center of circle

Convex polygon

Cumulative geometric relationship of location point and line equations

Concave Polygon

The sum of the angles formed by the line connecting the location point and the area vertices

the process of equipment operation and maintenance management [18]: (1) When equipment failure forms a dangerous area, the location of the faulty equipment can be determined first through 5G fusion positioning technology, and then the fault degree and the radius R of the dangerous area can be determined. Finally, a fence warning area with the faulty equipment as the center and radius R is formed. (2) When the operation people conduct related operations, operation scope is determined by operation people, and fence warning area is generated. The operation area is mostly irregular polygons. The key point of the electronic fence technology is to judge the relative position of the personnel and the fence area. The judgment method is shown in Table 6.4 [19]. Taking a convex polygon fence area as an example [20], asshownin Fig. 6.3. It is assumed that the position coordinates of the person P are x p , y p , the number of area edges is n, and the area vertices are P1 , P2 , · · · , Pn−1 , Pn , the corresponding position coordinates are (x1 , y1 ), (x2 , y2 ), · · · , (xn−1 , yn−1 ), (xn , yn ). The function corresponding to the i-th edge obtained by the two-point formula is f i (x, y) = (x i+1 − xi )y − (y i+1 − yi )x + xi yi+1 − yi xi+1

(6.9)

When i = n, let xi+1 = x1 , yi+1 = y1 , namely f n (x, y) = (x 1 − xn )y − (y 1 − yn )x + xn y1 − yn x1 To sum up Fig. 6.3 Convex polygon fence area

(6.10)

76

B. Wang et al.

 f i (x, y) =

(x i+1 − xi )y − (y i+1 − yi )x + xi yi+1 − yi xi+1 , i = 1, 2, · · · , n − 1 (x 1 − xn )y − (y 1 − yn )x + xn y1 − yn x1 , i = n (6.11)

Take any region vertex P j ( j = i, i + 1), if the point and the location point P satisfy:     f i x j , y j · f i x p , y p > 0(i = 1, 2, · · · , n), ( j = i, i + 1)

(6.12)

Then person P is located in the fenced area; otherwise, person P is located outside the fenced area.

6.5 Conclusion In order to improve the intelligent level of equipment operation and maintenance management of large railway passenger stations, this paper designs an integrated intelligent operation and maintenance process to solve the problems of inconvenient information interaction of various subsystems, lack of data, and difficulty in accurate positioning of equipment. At the same time, technical implementation schemes such as data warehouse, data mining, 5G fusion positioning, and electronic fence are given. The research will help to reduce the manual workload of the passenger station, reduce the production cost, and promote the construction of the modern passenger station. Acknowledgements This research was supported by the National Natural Science Foundation of China (Project No. 52172321) and the Science and Technology Plan of Sichuan Province (Project NO. 2020YJ0268) and Key science and technology projects in the transportation industry of the Ministry of Transport (2022-ZD7-132).

References 1. Liu, Z.: Design and Key Technology Research of Intelligent Operation and Maintenance Platform for Passenger Station Equipment. China Academy of Railway Sciences, Beijing (2021) 2. Wang, H., Shi, T., Jiang, H., et al.: A study on big data platform for monitoring the status of railway transportation equipment. Railway Transp. Econ. 40(2), 38–43 (2018) 3. Zhang, R.: Application of data mining in maintenance management of rail transit equipment. In: “Smart City and Rail Transit 2018”–Proceedings of the 5th National Annual Conference on Smart City and Rail Transit Technology Innovation, pp. 296–302. China Minzu University Press, Beijing (2018) 4. Wang, S., Geng, G., Zhou, M.: Research and application of data warehouse and data mining. Appl. Res. Comput. 9, 194–195+205 (2005) 5. Shi, F.: Research on safety supervision system of railway passenger station based on 5G highprecision fusion positioning. Railw. Transp. Econ. 44(6), 84–91 (2022)

6 Research on Equipment Operation and Maintenance Management …

77

6. Hou, X.: Data Analysis, Exhibition and Prediction Based on Data Warehouse, OLAP and Data Mining Technologies. Xidian University, Xi’an (2007) 7. Fan, L., Zhang, X., Su, W.: Research and application of substation equipment fault early warning based on big data mining technology. Power Syst. Big Data 22(01), 1–7 (2019) 8. Liu, H., Shen, G., Yang, Y., et al.: A research of manned equipment fault warning visualization based on the time error of running session. China Spec. Equip. Saf. 35(5), 1–5+24 (2019) 9. Zeng, X., Zhang, F., Shao, S., et al.: Fault feature analysis for CNC machine tools based on FP-growth algorithm. Mach. Tool Hydraul. 50(16), 174–180 (2022) 10. Liu, H., Guo, R., Jiang, H.: Research and improvement of Apriori algorithm for mining association rules. Comput. Appl. Softw. 26(01), 146–149 (2009) 11. Ji, H., Su, B., Lv, M.: Analysis of university mass emergency association rules based on FP-growth algorithm. China Saf. Sci. J. 22(12), 144–151 (2012) 12. Wu, T.-Y., Lin, J.C.-W., Zhang, Y., et al.: A grid-based swarm intelligence Algorithm for privacy-preserving data mining. Appl. Sci. 9(4), 774 (2019) 13. Chen, L., Gan, W., Lin, Q., et al.: OHUQI: Mining on-shelf high-utility quantitative itemsets. J. Supercomput. 78(6), 8321–8345 (2022) 14. Wu, T.-Y., Lin, J.C.-W., Yun, U., et al.: An efficient algorithm for fuzzy frequent itemset mining. J. Intell. Fuzzy Syst. 38(5), 5787–5797 (2020) 15. Chen, C.-M., Chen, L., Gan, W., et al.: Discovering high utility-occupancy patterns from uncertain data. Inf. Sci. 546, 1208–1229 (2021) 16. Chen, G., Zhang, Y., Wang, Y., et al.: Unscented Kalman filter algorithm for WiFi-PDR integrated indoor positioning. Acta Geodaetica et Cartographica Sinica 44(12), 1314–1321 (2015) 17. Yang, L., Chen, Y.-C., Wu, T.-Y.: Provably secure client-server key management scheme in 5G networks. Wirel. Commun. Mob. Comput. 2021, 4083199 (2021) 18. Ouyang, Z., Chen, Y., Song, Z.: Research and application of electronic fence technology based on high-precision Beidou combined positioning. Satell. Appl. 1, 32–33+36–39 (2019) 19. Yang, M., Pan, P.: Research on safety monitoring system for railway workers based on Beidou. China Saf. Sci. J. (S2 vo 29), 168–173 (2019) 20. Chen, Z.: Design and Realization of Operation Monitoring and Management System Based on the Beidou for Marshalling Stations. China Academy of Railway Sciences, Beijing (2018)

Chapter 7

Research on Adaptability Evaluation Between Express and Local Train Operation Plan of Urban Rail Transit and Passenger Flow Demand Tan Li Abstract The adaptability level of the operation plans and passenger flow demand has an influence on improving the passenger service efficiency and service level of urban rail transit express and local trains. In order to improve the matching degree between the operation plan and the passenger flow demand, the evaluation system is constructed from two aspects: passenger flow service quality and passenger flow service structure, and the AHP-set pair analysis method is used to quantitatively evaluate the adaptability of express and local train operation plan and passenger flow demand. Taking Tianjin Metro Line 9 as an example, this paper evaluates the adaptability of the passenger flow demand of four express and local train operation plans, and then obtains the evaluation results and proposes corresponding adjustment suggestions for the operation plan. The research shows that the plan P1 is most suitable for the passenger flow demand of Tianjin Metro during peak hours.

7.1 Introduction 7.1.1 Research Status With the advantages of high speed, economy, comfort and guiding urban development, urban rail transit has been valued by many large cities and has become an important travel way for urban residents. Express and local train is an optional mode of urban rail transit, which can well meet the different demands of passengers and improve the transportation serve quality, transportation efficiency and passenger service level of urban rail transit to a certain extent [1–3]. At present, express and local train mode has been widely used in some cities at home and abroad, overseas such as Reseau Express Regional A in Paris (RER A), Tsukuba Express and Toei Shinjuku Line in Japan, Long Island Railroad in New York and Seoul Metro Line 9 in T. Li (B) China Rail Way First Survey and Design Institute Group Ltd, Xian 710043, Shanxi, China e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_7

79

80

T. Li

South Korea [4], etc., and in China, such as Guangzhou Metro Line 14 [5], Shanghai Metro Line 16 [6] and Beijing Metro Line 6 [7], etc. The operation plan connects the transportation plan and the passenger flow demand, and the matching degree with the passenger flow has a direct impact on the quality of transportation services. At the theoretical level, the research on express and local train operation mode and the decision-making and evaluation of express and local train operation plan are in the initial stage, especially the research on the passenger flow adaptability of operation plan [8]. VUCHIC et al. analyzed the express and local operation, skip-stop operation and zonal operation of urban rail transit, and proposed the benefit evaluation method of different stop operation schemes [9]. In the research on the evaluation index system of the train operation plan, Zhang et al. established an evaluation index system from economic benefit and social benefit, and combined the data envelopment analysis method to design the input evaluation index and the output evaluation index of the train operation plan [10]. Huo et al. considered three aspects: cost control, economic benefit and social benefit to establish a passenger train comprehensive benefit evaluation model with multi-level index [11]. Wenxian Wang et al. established the evaluation index system from three aspects: volume adaptability, structural adaptability and quality adaptability [12]. Four factors are considered to establish an evaluation index system, which include average passenger income, load factor, the number of using train groups and the setting of train overtaking [13]. Zhao et al. used AHP to establish the comprehensive evaluation index system from four aspects: station cost, vehiclerelated cost, operation energy consumption and travel time [14]. And other scholars have used other methods of evaluation like neural network [15]. The existing research is mainly from the perspective of economic benefits, social benefits and comprehensive technology but the research rarely involves combining urban rail transit express and local train operation plan with passenger flow demand, and comparing the advantages and disadvantages of the operation plan and the adaptability of passenger flow demand.

7.1.2 Characteristic Analysis The adaptability between express and local train operation plan of urban rail transit and passenger flow demand is mainly reflected in two aspects: ➀ The transportation capacity generated by the operation plan can meet passenger transportation demand in all directions to the full extent, which provides passengers with comfortable, convenient, economic and time-efficient transportation services, namely, passenger flow service quality. ➁ The transportation capacity generated by the operation plan can be fully utilized, and reasonably matches the quantity demand and spatial distribution of the actual passenger flow transportation demand, namely, passenger flow service structure. Therefore, the higher the satisfaction degree in these two aspects , the more reasonable the passenger flow adaptability of operation plan of urban rail transit. Through the express and local operation plan of urban rail transit and passenger flow adaptability evaluation system, we use the AHP-Set Pair Analysis

7 Research on Adaptability Evaluation Between Express and Local Train …

81

method to evaluate the matching degree of operation plan and passenger flow demand, and take Tianjin Metro Line 9 as an example to verify the rationality of adaptability evaluation between the express and local train operation plan of the urban rail transit and passenger flow demand model.

7.2 The Evaluation Index Based on the characteristic analysis of the urban rail transit express and local train operation plan and passenger flow demand, the evaluation index system of urban rail transit express and local train operation plan and passenger flow adaptability is established from two aspects: passenger flow service quality and passenger flow service structure.

7.2.1 Passenger Flow Service Quality Index Passenger flow service quality index mainly includes passenger average travel lost time, passenger average transfer times, passenger non-transfer rate and train average full load factor. (1) Passenger average travel lost time p1 refers to the ratio of the lost time of all passengers who do not get off the train when all trains stop in all OD pairs to the total number of passengers, which is used to evaluate the matching degree of the stop plan and passenger flow and it is described as follows, p1 =

N  U 

tnm qnm

(7.1)

n=1 m=1

where tnm is the stop time of train n at the m-th station, qnm is the number of passengers who don’t get off the train when train n stops at the m-th station, N is the number of trains in the operation plan, U is the number of stations that trains pass through, qe is the number of passengers in passenger flow OD pair e. (2) Passenger average transfer times p2 is the ratio of the sum of the passenger total transfer times in all OD pairs to the total number of passengers, evaluating the matching degree of express and local train operation plan and passenger flow demand, and it is given as follows, p2 =

E  K  e=1 s=1

qes · s

(7.2)

82

T. Li

where qes is the number of passengers who transfer s times to reach the destination in passenger flow OD pair e, E is the total number of passenger flow OD, and K is transfer times of transferring passengers. (3) Passenger non-transfer rate p3 is the ratio of the number of passengers who don’t transfer in all OD pairs to the total number of passengers at the current express and local train operation plan, evaluating the matching degree of express and local train operation plan and passenger convenience, which is given as follows,  p3 = 1 −

 E K 

 qes

e=1 s=1

E 

 × 100%

qe

(7.3)

e=1

(4) Train average full load factor p4 is the ratio of the sum of full load factor of all trains to the number of all trains, in which the sum of the full load factor of all trains is the ratio of the sum of the product of the full load rate of the train operation section and the section length to the total length of the train operation section. It evaluates the matching degree of passenger flow transport capacity generated by the express and local train operation plan and real passenger flow demand, and also reflects the comfort level of passengers. And it is given as follows, p4 =

 N  R   n=1

  αrn · L r /L n

N

× 100%

(7.4)

r =1

where αrn is the full load rate of the train n in the r-th section, L r is the length of the r-th section, L n is the operation distance of the n-th train.

7.2.2 Passenger Flow Service Structure Index Passenger flow service structure index mainly includes passenger flow demand satisfaction rate, passenger flow structure matching rate, passenger flow average transport rate, passenger seats utilization of train, station service frequency and node coverage of station. (1) Passenger flow demand satisfaction rate p5 is the ratio of actual passenger flow to total passenger flow demand at the current express and local train operation plan, evaluating the matching degree of express and local train operation plan and passenger flow demand [12], and it is given as follows, p5 =

 E  e=1

 qe

E  e=1

 qe × 100%

(7.5)

7 Research on Adaptability Evaluation Between Express and Local Train …

83

where qe is passenger flow transported by all trains in the scheme in the passenger flow OD pair e. (2) Passenger flow structure matching rate p6 is the matching or deviation degree of the maximum passenger transport capacity and actual passenger flow demand at the current express and local train operation plan [12], which is given as follows, p6 =

E 

 F(βe ) E × 100%

(7.6)

e=1

where βe is the deviation degree of the maximum passenger transport capacity provided by all trains and actual passenger flow demand in the passenger flow OD pair e, F(βe ) is the matching degree of the express and local train operation plan and the actual passenger flow demand in the passenger flow OD pair e. (3) Passenger flow average transport rate p7 is the weighted ratio of the passenger flow transported by trains in each section to the fixed number of passenger of the train [12], which is given as follows,  R  N 1   n p7 = q /Cn · An ×100% N n=1 r =1 r

(7.7)

where An is the number of sections that train n runs in. (4) Train transport capacity utilization p8 is the ratio of passenger turnover quantity to passenger place kilometers, that is the number of man-kilometers completed per train per kilometer. It can evaluate utilization of the transport capacity of the operation plan, which is given as follows, p8 =

 R 

qrn

· Lr

 N 

e=1

 Cn · L n

× 100%

(7.8)

n=1

where qrn is the passenger flow that train n transports in the section r, L r is the length of the section r, Cn is the fixed number of passenger of train n, L n is the running distance of train n. (5) Station service frequency p9 is the average of the ratio of the number of stopped train pairs at each station excluding the origin and destination to the number of running trains at the current express and local train operation plan, which is the frequency of train stops at intermediate stations, and it is given as follows,  p9 =

 N U0  

m=0

  xnm

N

 U0 × 100%

(7.9)

n=1

where xnm is 0–1 variables, and the value is 1 when train n stops at the station m, otherwise the value is 0, U0 is the total number of line stations.

84

T. Li

(6) Node coverage of station p10 refers to the ratio of times of stops of all trains to the total number of station nodes, that is the ratio of the number of nodes that trains stop at to the total number of nodes at the current express and local train operation plan[12], and it is given as follows, p10 =

 N U 0 

 xnm

 N · U0 × 100%

(7.10)

n=1 m=1

7.3 Evaluation Model The AHP-set pair analysis method is used to quantitatively evaluate the passenger flow adaptability of the express and local trains [16–19]. The AHP-set pair analysis method can comprehensively investigate the certainty and uncertainty of a system by assuming that the essential attribute of all things is uncertainty. Then we can use mathematical theories and models to deal with the subsystems, and combine the construction of the same-degree decision-making matrix to comprehensively evaluate the selected plans. Therefore, we can select the most matching plan combination among the operation plans.

7.3.1 The Same-Degree Decision-Making Matrix (1) We suppose a line has i operation plans to be selected from P1 , P2 , ..., Pi , and each of the operation plans includes j evaluation indicators p1 , p2 , p3 , ..., p j . So the decision-making matrix W of various schemes can be obtained. ⎡

p1,1 · · · ⎢ .. . . W =⎣ . . pi,1 · · ·

⎤ p1, j .. ⎥ . ⎦

(7.11)

pi, j

(2) The optimal value is selected from each indicator of i alternative operation plans and P0,h is used to represent the h-th optimal index of the ideal operation plan and form an ideal operation plan P0 .

 P0 = p0,1 , p0,2 , ..., p0, j

(7.12)

  where the calculation formula is p0,h = max pg,h when the index is a benefit index and the value should be selected. The calculation formula is   maximum p0,h = min pg,h when the index is a cost index and the minimum value should

7 Research on Adaptability Evaluation Between Express and Local Train …

85

be selected. Passenger travel lost time and passenger average transfer times are cost indexes, and others are benefit indexes among the above indexes. (3) The same-degree bg,h is gotten by the optimal index of the ideal operation plan and the evaluation index of each operation plan. Therefore, the unweighted same-degree matrix B of the ideal scheme evaluation index based on set pair analysis and the evaluated scheme index can be obtained. ⎡

b1,1 · · · ⎢ .. . . B=⎣ . . bi,1 · · ·

⎤ b1, j .. ⎥ . ⎦

(7.13)

bi, j

where when p0,h is the benefit index, bg,h = pg,h / p0,h .When p0,h is the cost benefit index, bg,h = p0,h / pg,h .

7.3.2 Determination of Evaluation Index Weights with AHP (1) We construct a hierarchy structure model and compare and analyze each index in the model by the method of 1–10 and the reciprocal scale. And the judgment matrix Q is obtained. ⎤ a1,1 · · · a1, j ⎥ ⎢ Q = ⎣ ... . . . ... ⎦ ai,1 · · · ai, j ⎡

(7.14)

where ah,g = 1/ag,h . (2) We solve the eigenvector R and largest eigenvalue λmax of the judgment matrix R. The eigenvector is the weight value obtained after sorting the importance j of each index between one layer and the previous layer η = ag / g=1 ag , in 

√j which ag = ag,1 ag,2 ag,3 · · · ag, j . And the eigenvector R = η1 , η2 , ..., η j is j obtained in turn. The largest eigenvalue is λmax = g=1 (Q R)g /jηg . (3) We verify the judgment matrix and calculate the consistency index C I = (λmax − j) . The average random consistency index value is obtained by looking ( j−1) up the table, which is shown in Table 7.1. Then calculate the random consistency ratio C R = C I /R I . If the random consistency ratio is less than 0.1, we can consider that the obtained sorting result satisfies the above conditions, so a reasonably allocated weight coefficient can be obtained. Table 7.1 The average random consistency index Order

1

2

3

4

5

6

7

8

9

10

11

RI

0.00

0.00

0.58

0.90

1.12

1.24

1.32

1.41

1.45

1.49

1.51

86

T. Li

If the conditions cannot be met, the values in the judgment matrix Q need to be readjusted.

7.3.3 Comprehensive Evaluation Model We solve the weighted same-degree matrix H of the evaluation scheme Pi and the ideal scheme P0 , and the calculation formula is given as follows, ⎡

b1,1 · · · ⎢ .. . . T H = BR = ⎣ . . bi,1 · · ·

⎤ b1, j .. ⎥ η , η , ..., η T = ( f , f , ..., f ) j 1 2 i . ⎦ 1 2

(7.15)

bi, j

j where f g = h=1 ηg bg,h , g = 1, 2, 3, ..., i. f g is the sum of the g-th weighted same degree of the evaluated plan and the ideal plan. The order of weights of the evaluated schemes can be obtained by the value f g in the weighted same-degree matrix H . The larger the value of f g , the better the corresponding scheme .

7.4 Case Analysis This paper quantitatively evaluates the urban rail transit express and local train operation plan and passenger flow adaptability taking Tianjin Metro Line 9 as an example.

7.4.1 Operation Plan Analysis According to the passenger flow data of Tianjin Metro Line 9 during peak hours, as shown in Fig. 7.1, there are 4 operation plans for express and local trains to choose from during peak hours, as shown in Fig. 7.2. It can be seen from Fig. 7.2 that the operation plan P1 and P2 adopt a combination of long-short route and express and local train, and the operating section of the short route is from Dong Li Kai Fa Qu Station to Tian Jin Station. In the plan P1 , there are 10 local trains per hour on the long route, 5 local trains per hour on the short route, and 5 express trains per hour on the long route which designs that the express trains don’t stop at the four stations including Tai Hu Road Station, Gang Guan Company Station, Xiao Dong Zhuang Station and Zhang Gui Zhuang Station. The number of trains per hour in the plan P2 is the same as the plan P1 , but the express trains on the long route don’t stop at the five stations: Tai Hu Road Station, Gang Guan Company Station, Xiao Dong Zhuang

7 Research on Adaptability Evaluation Between Express and Local Train …

87

Station, Zhang Gui Zhuang Station and Yi Hao Qiao Station. The operation plan P3 and P4 only use the express and local train mode. In the plan P3 , there are 13 local trains per hour on the long route, and 7 express trains per hour on the long route which designs that the express trains don’t stop at the four stations including Tai Hu Road Station, Gang Guan Company Station, Xiao Dong Zhuang Station and Zhang Gui Zhuang Station. Compared with the plan P3 , the number of trains per hour in the plan P4 is the same, but the express trains don’t stop at the five stations: Tai Hu Road Station, Gang Guan Company Station, Xiao Dong Zhuang Station, Zhang Gui Zhuang Station and Yi Hao Qiao Station.

7.4.2 Evaluation Index Analysis (1)

(2)

(3)

(4)

(5)

(6)

Passenger average travel lost time p1 . According to the formula (7.1), passenger average travel lost time of the four operation plans P1 to P4 is respectively 47.5856 s, 48.3135 s, 48.0244 s, and 48.5886 s. Passenger average travel lost time of the plan P1 is the shortest, and passenger average travel lost time of the four operation plans is all within 50 s. Passenger average transfer times p2 . According to the formula (7.2), passenger average transfer times of plan P1 and P3 is less than 400, which is respectively 307 and 343. Passenger average transfer times of plan P2 is P4 are more than 400, which is respectively 404 and 423. And passenger average transfer times of plan P1 is the least. Passenger non-transfer rate p3 . According to the formula (7.3), passenger nontransfer rate of four transferring ways is all more than 95%, in which plan P1 is the highest. Passenger non-transfer rate of the other plans is respectively 98.81%, 98.99%, and 98.76%. Train average full load factor p4 . According to the formula (7.4), the train average full load factor of four operation plans is all over 50%, which is 56.77%, 56.94%, 52.30%, and 52.47%. The train average full load factor of the plan P1 and P2 on the short route is up to 77.76% and 78.06%, which is related to passenger flow demand during morning peak. Passenger flow demand satisfaction rate p5 . We can get from the formula (7.5) that the passenger flow demand satisfaction rate of the plan P1 is the highest, reaching 95.36%, and secondly the plan P2 is 94.15%. The passenger flow demand satisfaction rate of the plan P3 and P4 is respectively 85.59% and 86.46%. Therefore, it shows that four operation plans can well meet the passenger flow demand. Passenger flow structure matching rate p6 . The critical value of the deviation degree can be taken as 0.3 after expert consultation. According to the formula (7.6), we can obtain that among the 420 passenger flow ODs in Tianjin Metro Line 9, the passenger flow OD numbers in which the passenger flow demand and the train transportation capacity of the four operation plans are mutually compatible are 301, 293, 263, 259 respectively. This shows under the passenger

88

T. Li

a. Passenger volume in section during peak hour.

b. Passenger flow of boarding and Alighting during peak hour. Fig. 7.1 Passenger flow data of Tianjin Metro Line 9 during peak hour

7 Research on Adaptability Evaluation Between Express and Local Train …

89

a. Operation plan of express and local train and long-short route.

b. Express and local train operation plan. Fig. 7.2 Operation plan during peak hour of Tianjin Metro Line 9

(7)

(8)

(9)

flow demand of Tianjin Metro Line 9 during peak hours, more than half of the passenger flow OD of the four operation plans are in a state of mutual match, and the adaptability of passenger transport capacity and passenger flow demand which the plan P1 provides is the highest. Passenger flow average transport rate p7 . According to the formula (7.7), passenger flow average transport rate of four operation plans is all more than 50%, and is 55.80%, 55.97%, 52.30%, and 52.47% respectively. Train transport capacity utilization p8 . According to the formula (7.8), train transport capacity utilization of four operation plans is all over 65%, and is respectively 70.24%, 70.33%, 65.08%, 65.17%. The transportation capacity and passenger flow can be balanced and matched, and the plan P2 has the highest transport capacity utilization. Station service frequency p9 . We can get from the formula (7.9) that station service frequency of plan P3 is the highest and is 92.63%. The plan P3 has the highest matching degree between the number of train pairs and the stopping

90

T. Li

Table 7.2 The adaptability evaluation index of the operation plan and passenger flow demand Plan

p1

p2

p3 (%)

p4 (%)

p5 (%)

p6

p7 (%)

p8 (%)

p9 (%)

p10 (%)

P1

47.5

307

99.1

56.7

95.3

301

55.8

70.2

80.6

82.4

P2

48.3

404

98.8

56.9

94.1

293

55.9

70.3

74.6

77.0

P3

48.0

343

98.9

52.3

85.5

263

52.3

65.0

92.6

93.3

P4

48.5

423

98.7

52.4

86.4

259

52.4

65.1

90.7

91.6

demand of each station, and the station service frequency of the plan P2 is the lowest, which is 74.63%. And it is 80.60% and 90.79% for the plans P1 and P4 . (10) Node coverage of station p10 . According to the formula (7.10), node coverage of station of the four plans is 82.43%, 77.03%, 93.33%, and 91.67% respectively. The plan P3 has the highest node coverage of station and the highest satisfaction degree of stopping demand of each station. In summary, the adaptability index of the express and local train operation plan of Tianjin Metro Line 9 and passenger flow demand are shown in Table 7.2.

7.4.3 Evaluation on the Adaptability of Express and Local Trains Operation Plans and Passenger Flow Demand We evaluate express and local trains operation plans of Tianjin Metro Line 9 with the AHP-set pair analysis method. (1) Construct the same-degree decision-making matrix. The decision-making matrix W is constructed from Table 7.2, which is shown as follows, ⎡

47.59 307 ⎢ ⎢ 48.31 404 W =⎢ ⎣ 48.02 343 48.59 423

0.9910 0.9881 0.9899 0.9876

0.5677 0.5694 0.5230 0.5247

0.9536 0.9415 0.8559 0.8646

301 293 263 259

0.5580 0.5597 0.5230 0.5247

0.7024 0.7033 0.6508 0.6517

0.8060 0.7463 0.9263 0.9079

0.8243 0.7703 0.9333 0.9167

⎤ ⎥ ⎥ ⎥ ⎦

According to the benefit index and cost index of the decision-making matrix, the ideal vector P0 is obtained as,   P0 = 47.59 307 0.9910 0.5694 0.9536 301 0.5597 0.7033 0.9263 0.9333 The same-degree matrix B without weight is obtained by calculation as follows

7 Research on Adaptability Evaluation Between Express and Local Train … ⎡

1 ⎢ ⎢ 0.9849 B=⎢ ⎣ 0.9909 0.9794

1 0.7599 0.8950 0.7258

1 0.9971 0.9989 0.9966

0.9971 1 0.9186 0.9215

1 0.9874 0.8976 0.9067

1 0.9737 0.8759 0.8603

0.9971 1 0.9345 0.9374

0.9987 1 0.9253 0.9265

0.8701 0.8056 1 0.9801

91 ⎤ 0.8832 ⎥ 0.8253 ⎥ ⎥ ⎦ 1 0.9821

(2) Determine the evaluation index weights. According to the questionnaire statistics of experts related to urban rail transit passenger flow transportation, we use AHP to calculate the eigenvector R of the evaluation index:   R = 0.2497 0.1674 0.1674 0.0612 0.1011 0.0841 0.0656 0.0606 0.0215 0.0215

Eigenvalue λmax = 10.26., Consistency indicator C R = 0.0195 < 0.1. So it meets consistency requirements. (3) Evaluate comprehensively. According to the formula (7.15), the weighted samedegree matrix H of the evaluation plan Pi and the ideal plan P0 is obtained as   H = B R T = 0.9793 0.9292 0.9155 0.8831 It can be seen from the same-degree matrix that the adaptability order of the operation plan and the passenger flow demand is P1 > P2 > P3 > P4 , and applying the plan is P1 the most economical and reasonable, which is no adjustment required. In addition, reducing the number of local trains on the long route and adding trains on the short route can be considered for the plan P3 and P4 , which can improve the passenger flow demand satisfaction rate and passenger flow structure matching rate, and reduce the waste of transportation capacity.

7.5 Conclusion The reasonable evaluation of the adaptability of the express and local train operation plan of urban rail transit and passenger flow demand is a test for the train operation plan science, and it is also the premise of the reasonable adjustment for the train operation plan. This paper’s analysis from passenger flow service quality and passenger flow service structure by constructing the evaluation model of adaptability of the operation plan and passenger flow demand, and the evaluation results can be used to optimize and adjust the operation plan. Furthermore, we can extend the evaluation model, and evaluate the adaptability of passenger flow demand of the same operation plan in different time periods. Through the evaluation of different operation plans of Tianjin Metro Line 9, the conclusions of this study is conducive to further improving the passenger flow service level of the urban rail transit, and improving the passenger satisfaction and service efficiency.

92

T. Li

Acknowledgements This research was supported by the Science and Technology Research Project of China Rail Way First Survey and Design Institute Group Ltd (Project No. 2021KY55YB-09).

References 1. Gao, D., Hu, C., Zong, J.: Study on express and local train operation program of regional rapid rail transit. Railw. Transp. Econ. 36(2), 73–78 (2014) 2. Rashedi, Mahmoud, M., Hasnine, S., et al.: On the factors affecting the choice of regional transit for commuting in Greater Toronto and Hamilton area: application of an advanced RP-SP choice model. Transp. Res. Part A Policy Pract. 105, 1–13 (2017) 3. Zhang, H.: esearch on the Express and Local Operation of Urban Rail Transit. Beijing Jiaotong University, Beijing (2015) 4. Zhou, Q.: Analysis of express and local train operation of urban rail transit in the world. Urban Rapid Rail Transit 26(2), 18–22 (2013) 5. Sun, Y., Shi, H.: Research and practice on express and local train operation of city line. Urban Rapid Rail Transit 26(2), 14−17 (2013) 6. Cheng, X.: On the express and local train operation program on Shanghai rail transit line16. Urban Mass Transit 17(5), 68–72 (2014) 7. Dong, S.: Research on Optimization Algorithm of Express and Local Line for Urban Rail Transit. Beijing Jiaotong University, Beijing (2015) 8. Wang, L.: Study on the Operation Organization of Express and Local Train for Urban Rail Transit. Beijing Jiaotong University, Beijing (2013) 9. Vuchic, V.: Urban Transit: Operations, Planning and Economics, pp. 20–25. John Wiley & Sons, New Jersey (2006) 10. Zhang, Y., Zhang, H.: DEA method for analyzing and evaluating railway passenger train plan. J. Lanzhou Jiaotong Univ. 27(6), 83–86 (2008) 11. Huo, L., Tian, Z., Bao, L.: General benefits evaluation method for passenger trains. Railw. Transp. Econ. 31(10), 15–18 (2009) 12. Wang, W., Cheng, L., Chen, T., Ni, S.: A Research on adaptability evaluation of train operation plan and passenger flow. Railw. Transp. Econ. 41(7), 65–71 (2019) 13. Sun, Y., Ran, X., Yang, F., Chen, S.: Design and evaluation on operational plan of express and local trains for urban rail transit. J. Railw. Sci. Eng. 15(1), 233–239 (2018) 14. Yi, Z., Tan, X., Chen, F., Cao, S.: On comprehensive evaluation system for mixed express and slow train operation of urban rail transit. Railw. Standard Des. 61(9), 65–67+75 (2017) 15. Zhang, F., Wu, T.-Y., Wang, Y., Xiong, R., Ding, G., Mei, P., Liu, L.: Application of quantum genetic optimization of LVQ neural network in smart city traffic network prediction. IEEE Access 8, 104555–104564 (2020) 16. Xu, L.: The study on programing passenger train operation scheme by using level analysis method. Railw. Econ. Res. 3, 46–48 (2001) 17. Liu, Y., Li, Q., Lv, X., Zhou, S., Wang, Z.: Research on weight of railway passenger transfer scheme selection based on analytic hierarchy process. Railw. Comput. Appl. 28(11), 9–12 (2019) 18. Wang, H., Du, Y., Wang, W.: Comparison on high-speed railway train operating schemes based on set pair analysis. Railw. Transp. Econ. 37(9), 13–17 (2015) 19. Zhao, K.: Set Pair Analysis and Its Preliminary Application. Hangzhou: Zhejiang Science & Technology Press (2000)

Chapter 8

Research on the Network Operation Mode of High-Speed Rail Express Yongcheng Wang, Yi Li, Yunhao Sun, and Tao Chen

Abstract Taking high-speed rail express as the research object, the research is carried out on the network operation mode of high-speed rail express. Taking the minimum cost of transportation enterprises as the optimization goal, the decision model of the network operation mode of high-speed rail express is constructed by taking the remaining capacity limit of the middle section of the high-speed rail express line network, the carrying capacity and operating cost restrictions corresponding to the high-speed rail express operating mode, and the overall transportation task limit as constraints. Taking the train operation map in the third quarter of 2021 as a reference, the decision-making model for the construction of the operation mode is verified, and the ARIMA model is used to calculate the OD of each city node in the prospect year, and the carrying capacity and operation cost of different operation modes are analyzed and calculated. The results show that: my country should adopt a high-speed rail express network operation mode combining direct transit. Transit transportation completes the collection and distribution of small batches of goods, improves the full load rate of high-speed rail freight trains, expands the scope of services, and completes long-distance transportation of large batches of goods through direct transportation, improving economies of scale.

Y. Wang Department of Transportation, China Railway Beijing Group Co., Ltd, Beijing 100860, China Y. Li · Y. Sun · T. Chen (B) School of Transportation and Logistics, Southwest Jiaotong University, Chengdu 610031, China e-mail: [email protected] Y. Li e-mail: [email protected] National and Local Joint Engineering Laboratory of Comprehensive Intelligent Transportation, Southwest Jiaotong University, Chengdu 610031, China National Engineering Laboratory of Integrated Transportation Big Data Application Technology, Chengdu, China © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_8

93

94

Y. Wang et al.

8.1 Introduction With the development of e-commerce, the volume of express delivery business in China has maintained a strong growth, but the share of railway transportation in the domestic express mail market is less than 5%. At present, China’s high-speed rail express has the problems of imperfect construction of supporting facilities, unclear layout of transfer nodes, and uncertain operation mode. Therefore, to study the influencing factors of China’s high-speed rail express network operation mode, build a decision model for high-speed rail express network operation mode, and put forward the network operation mode of China’s high-speed rail express transportation, which will improve the coordination ability of China’s high-speed rail express network and expand high-speed rail express transportation. The market share of the product is of great significance. Scholars at home and abroad have carried out research on the operation mode of high-speed rail express around market potential, transportation methods and operational safety. Liang [1, 2] studied from two aspects of market potential and transportation methods, and believed that the use of high-speed rail confirmation cars and passenger train piggybacks are the two main forms of high-speed rail express transportation in the short term, and these two methods have low marginal cost and profit. Considering the operational safety of high-speed railways, Noh et al. [3] proposed a multi-section load planning algorithm for high-speed freight trains in order to ensure the internal load balance of the carriages. Qin [4] modeled the combination of transportation organizations with the goal of maximizing transportation benefits and minimizing the generalized cost of cargo owners, and constructed a method to solve the combination of high-speed rail express freight transportation forms; Zhang [5], Wang [6] Based on different forms of freight transportation, the decision-making problem of high-speed rail express transportation is studied, and the decision-making problems such as operation route, choice of travel mode and freight volume are solved. Li [7] proposed the concept of high-speed rail express system, analyzed and constructed the system framework and determined the function of highspeed rail express system. Guo [8] made a theoretical analysis and summary of the cost of the four modes of high-speed rail express, and proposed that cooperation with large logistics enterprises should be strengthened, the express loading and unloading equipment should be improved, and the small express market should be expanded. To sum up, in recent years, most of the research on the operation mode of highspeed rail express is a simple analysis of the four modes of high-speed rail confirmation car and passenger train piggyback, adding freight carriage, and running highspeed rail freight train. When constructing the decision model of high-speed rail express operation mode, the optimization goal is based on the expenses of the cargo consignor and the carrier, without considering the scale and overall benefits brought by high-speed rail express, and cannot give full play to the combined advantages of each operating mode. On the basis of the existing research, this paper constructs a decision-making model of the travel mode from the factors of high-speed rail express network capacity, market OD factors, carrying capacity and cost factors of the travel

8 Research on the Network Operation Mode of High-Speed Rail Express

95

mode. As a basic research unit, a reasonable combination of the four operating modes is allowed.

8.2 Problem Analysis The high-speed rail express network operation mode is aimed at rationally optimizing the high-speed rail express operation mode, so as to realize the rational use of the high-speed rail express network capacity, complete the cargo transportation task without affecting the passenger operation, and reach the railway department. The overall operating expenses are minimal. There is a certain amount of goods that need to be transported between each city node. The idea of the model construction in this paper is to take the transportation mode adopted by the goods between two adjacent nodes in the line network as the basic research unit. According to the transportation distance and the characteristics of the four travel modes, a suitable network travel mode between the two nodes is selected, and the index values such as the transportation cost, the number of trains, and the carrying capacity under the travel mode are determined. When making a decision on the network operation mode of high-speed rail express, in addition to completing the transportation task and fulfilling the OD requirements between nodes, constraints such as the network capacity corresponding to the travel path and the capacity limit of each operation mode should also be considered.

8.3 Model Building 8.3.1 Parameter Definition The high-speed rail express network in China is simplified into a directed graph G = (V, E), in which the set V = {v|1 ≤ v ≤ 31} is the node city set of E = {e(i, j)|i, j ∈ V } 31 provinces, municipalities, autonomous regions and municipalities in China, the transportation interval between de(i, j) adjacent nodes i and node j of the high-speed rail express network, and the adjacent nodes i and distance between nodes j. Let the set be the set M = {m|m = 1, 2, 3, 4} of high-speed rail express network operation modes, and the value of m ranges from 1 to 4. The corresponding operation modes from small to large are: high-speed rail confirmed car piggyback mode, passenger sports car group piggyback mode, passenger-cargo mixed mode, high-speed rail freight special train model. Let the OD matrix of high-speed rail express in the prospect year be O D = {od(i, j)|(i, j) ∈ E}: od(i, j)e(a,b) the OD volume Q(i, j)m e(a,b) in the interval e(a, b)

96

Y. Wang et al.

from node i to node j, and let it represent the freight volume of the goods i sent from node to node j in the interval e(a, b) using the mth travel mode. Let Cθm denote the corresponding running cost under the mth running mode, θ [1, 2, 3], which correspond to fixed cost, loading and unloading cost, and variable cost, respectively.   k |i, j ∈ V, k = 1, . . . , n of walking paths Let the set be the set R K = rod(i, j) k k from dod(i, j) node i to node j, which means i is the length of the d(i, j)e(a,b) kth K path from node i to node j; the  length of the middle section of the e(a, b)Z =  k z od(i, j) |i, j ∈ V, k = 1, . . . , n kth path from node to node j; the length of the kth path from node i to node j A collection of nodes along the path. Let n m e(i, j) denote e(i, j) the number of trains that can travel corresponding to the mth travel mode in the interval; denote the number of trains n(i, j)m e(a,b) that use the i mth travel mode in the interval forthe goods e(a, b) sent from the node to the node j; the set is S = se(i, j) |e(i, j) ∈ E the average travel speed of the adjacent node i and node j interval Set; αm is the maximum carrying capacity of the mth travel mode. decision variable is set to x(i, j)m e(a,b) ∈ [0, 1] indicate i and whether the goods e(a, b) sent by the node to node j are transported by the mth travel mode in the interval; if the mth travel x(i, j)m e(a,b) = 1 mode is used for transportation in the interval e(a, b), otherwise it is 0. k Set the variable yod(i, j) ∈ [0, 1] to indicate i whether the goods sent from the node to the node j choose the k -th travel path, if the goods i sent from the node to the k k node j choose the k -th travel path yod(i, j) = 1, otherwise yod(i, j) = 0.

8.3.2 Objective Function The overall transportation cost of the railway sector can be divided into three categories: fixed cost, loading and unloading cost, and variable cost. This paper builds a single-objective optimization model based on the minimum overall transportation cost. The objective function of the model is as follows: minC =

4  O D od(i,  j)

m C1m n(i, j)m e(a,b) x(i, j)e(a,b)

m=1 od(i, j) e(a,b)

+

4  O D od(i, n  j) 

4  OD 

C2m Q m od(i, j)

m=1 od(i, j) k m C3m d(i, j)ke(a,b) x(i, j)m e(a,b) yod(i, j) n(i, j)e(a,b)

(8.1)

m=1 od(i, j) e(a,b) k=1 m m m where: C1m n(i, j)m e(a,b) x(i, j)e(a,b) is the fixed cost of transportation C 2 Q od(i, j) in the adjacent interval (·) for the i, j mth travel mode, is the loading and unloading m k cost C3m d(i, j)ke(a,b) x(i, j)m e(a,b) yod(i, j) n(i, j)e(a,b) of the mth travel mode, and is the variable cost of the mth travel mode in the adjacent interval ((i, j)).

8 Research on the Network Operation Mode of High-Speed Rail Express

97

8.3.3 Constraints 1. Remaining capacity restrictions in the middle section of the high-speed rail express network under each operating mode. The operation of high-speed rail freight trains is limited by the remaining capacity of the high-speed rail express network. The idea of constructing the model in this paper is to take the operation mode adopted by the goods between two adjacent nodes in the network as the basic research unit. Therefore, the capacity constraint of the minimum interval should be considered. The capacity range of the line network cannot be exceeded; the high-speed rail confirms that there is only one train in the same section every day, so the high-speed rail confirms that the maximum number of trains in the piggyback mode is 1; the number of trains using the piggyback mode should be less than the actual number of passenger cars. Translating the above constraints into mathematical language can be expressed as: m k m k n(i, j)m e(a,b) yod(i, j) ≤ n e(a,b) yod(i, j) x(i, j)e(a,b)

(8.2)

∀m ∈ M, e(a, b) ⊆ od(i, j), k ∈ R K , od(i, j) ⊆ OD OD 

m n(i, j)m e(a,b) ≤ n e(a,b)

(8.3)

od(i, j)

∀m ∈ M, e(a, b) ⊆ od(i, j), od(i, j) ⊆ OD k In the formula: n(i, j)m e(a,b) yod(i, j) is the number of trains i that use the mth m k travel mode n m e(a,b) yod(i, j) x(i, j)e(a,b) in the interval for the goods e(a, b) sent from the node i to the node j, is the number of trains that can travel corresponding to m k the mth travel mode in the interval, yod(i, j) and is the e(i, j) decision x(i, j)e(a,b) variable. 2. The weight of the goods transported by the trains operating in each operating mode should be within the carrying capacity range m Q(i, j)m e(a,b) ≤ n(i, j)e(a,b) αm

(8.4)

∀m ∈ M, e(a, b) ⊆ od(i, j) where: is n(i, j)m e(a,b) αm the overall maximum carrying capacity of the mth travel mode in the interval.e(i, j) 3. The transportation task must be completed The sum of the goods transported under each travel mode in the interval should match the OD

98

Y. Wang et al. 4 

k Q(i, j)m e(a,b) ≥ yod(i, j) od(i, j)e(a,b)

(8.5)

m=1

∀m ∈ M, e(a, b) ⊆ od(i, j) 4. Other constraints

˄

˅

˄

x(i, j)m e(a,b) ∈ [0, 1]

(8.6)

k yod(i, j) ∈ [0, 1]

(8.7)

˅

(8.8)

8.4 Solving Algorithm China’s high-speed rail express network is simplified as a directed graph G = (V, E), in which the set V = {v|1 ≤ v ≤ n} is the node city set, which is v ∗ used to mark the points on the temporary shortest path, which is E = {e(i, j)|i, j ∈ V } the transportation interval between de(i, j) adjacent nodes i and node j of the high-speed rail express network, and is the adjacent node. distance from i node j. The set U ⊆ V stores the known shortest path nodes, the set W ⊆ V stores the nodes other than the 1 known shortest path nodes, U ∪ W = V ; dod(i, j) (v) represents i the length of the 1 shortest path Z od(i, j) from the starting node to the node between i the node v and the i node j, which is the node set of the shortest path from the node to the node j, Using Dijkstra’s method to determine the shortest path between high-speed rail express network nodes The specific steps are as follows: 1 1 1 Step 1: Set the initial value, dod(i, j) (v) = ∞, U = {i}, Z od(i, j) = {i}, dod(i, j) (i) = 0. Step 2: Starting from the starting node i, check the nodes  ∗  in the set1 of nodes to 1 1 1 be selected, W dod(i, e(i,v) , ask dod(i, j) vi = min{dod(i, j) (v), v ∈ j) (v)= dod(i, j) (i) + d   ∗ 1 ∗ W }, then at this time Z od(i, j) = (i, vi ) , U = i, vi . Step 3: From the vi∗ beginning, continue to check the nodes in the  ∗ set of 1 1 ∗ 1 ∗ v + d , seek d nodes to be selected, W dod(i, = d (v) e(v ,v) j) od(i, j) i od(i, j) vi+1 = i 1 ∗ 1 min{dod(i, j) (v), v ∈ W}, and vi+1 write them into the set Z od(i, j) and the set U , at  ∗   ∗ ∗    1 ∗ i, vi , vi , vi+1 , U = i, vi∗ , v i+1 . this time Z od(i, j) = Step 4: When the node set to be selected W is empty, the shortest path solution is terminated, otherwise go to step 3. The steps for determining the next shortest path and the path length are as follows: 1 Step 1: Determine the shortest path between the Z od(i, j) high-speed rail express network node i and node j by using the Dijkstra method;

8 Research on the Network Operation Mode of High-Speed Rail Express

99

1 Step 2: Remove Z od(i, j) the shortest edges in turn, and then use the Dijkstra method 2 2 to determine the shortest path Z od(i, j) and its length at this time dod(i, j) ; Step 3: Repeat the above steps, remove one shortest edge each time, and obtain k the Z od(i, j) corresponding path which is the kth shortest path. using Dijkstra’s method to obtain the shortest path and sub-shortest path set k between the high-speed rail express network Z od(i, j) node i and node j, where the minimum cost model is solved. The specific steps are as follows: Step 1: Calculate the remaining capacity of China’s high-speed rail express network, the OD amount of each city node in the prospect year, the carrying capacity of different operating modes, and the operating cost; Step 2: Select city nodes, and refer to the current high-speed rail express network, build a network topology model, and establish a set of decision-making schemes J , and select the high-speed rail freight train operation mode for the OD value between nodes that satisfy the constraints of the high-speed rail freight train. Priority is given according to the size of the OD value. Arrange a walking path. Step 3: Input the known parameters and constraints, and solve the decision-making scheme of the opening mode between a certain ji OD; Step 4: Judging whether the number of trains running on the route in the scheme ji satisfies the constraint of the remaining capacity of the line network, if so, go to k step Z od(i, j) 6; otherwise, go to step 5; k+1 Step 5: ji Replace Z od(i, j) the walking path in the plan with the walking path k Z od(i, j) , and go to step 3; k Step 6: Record the walking path Z od(i, j) and ji ,and end the cycle.

8.5 Case Analysis This part firstly calculates the capacity and cost of China’s high-speed rail express network, the OD market in the prospective year, and the carrying capacity and cost of each operation mode, and then verifies the operation mode decision-making model by taking China’s high-speed rail express network as an example.

8.5.1 Calculation of Network Capacity of High-Speed Rail Express In this paper, the remaining capacity of the line is used to conduct an overall evaluation of the backbone network of high-speed railways in China. Taking the train operation map in the third quarter of 2021 as an example, the difference between the number of pairs of high-speed trains actually running in the network and the passing capacity of the parallel operation map is calculated. The formula is as follows [9]:

100

Y. Wang et al.

Table 8.1 Remaining capacity of China’s high-speed railway network Line name

Segment Start

Terminal

Number Segment Average Remaining of trains length speed capability (columns) (km) (km/h) (column)

Beijing-Guangzhou Beijing west Shijiazhuang 135 Expressway Shijiazhuang Zhengzhou 118 East

294

230.7

62

397

230.7

57

Zhengzhou East

Wuhan

129

514

230.7

52

South Beijing

Tianjin South

125

133

232.5

69

Tianjin South

Jinan West

137

312

232.5

62

Jinan West

Xuzhou East 148

295

232.5

57

353

232.5

52

Beijing-Shanghai Expressway

Xuzhou East Nanjing South

155

Nr = N − Na N=

1440 − Tw 60S − I VI

(8.9) (8.10)

In the formula: Nr is the remaining capacity of the line; Na is the logarithm of the actual running high N - Tw speed trains; The average speed V (km/h) of the high S-speed train in the section; it is the length of the high-speed operating section (km). The “train operation diagram system 4.0” is used to collect data on the train operation diagram in the third quarter of 2021 in China, and the residual capacity of each line section in China’s high-speed railway network can be calculated by using formula (8.9) and formula (8.10), and some results are shown in Table 8.1.

8.5.2 Annual OD Volume of High-Speed Rail Express In order to avoid the impact of the epidemic, this paper uses the data of 2019 for calculation. Consult the statistical yearbook to obtain the GDP and SR values of each city node in China in 2019 [10]. The sharing rate is set to account for 70% of the total volume of off—site express delivery, and the average weight of each express delivery is 1 kg. The OD matrix of high-speed rail express products between nodes in each city in the prospect year is shown in Table 8.2.

8 Research on the Network Operation Mode of High-Speed Rail Express

101

Table 8.2 OD matrix of high-speed rail express in national urban nodes in the prospect year (part) (unit: ton) City node Beijing Shanghai Tianjin Chongqing

Beijing

Shanghai

Tianjin

Chongqing

Harbin

Changchun

Shenyang

0.00

84.24

55.82

52.44

50.53

54.12

112.79

129.08

0.00

75.54

83.55

50.04

50.32

92.46

44.57

39.36

0.00

22.44

21.08

22.83

48.99

8.96

9.32

4.80

0.00

2.39

3.02

5.16

Harbin

38.88

25.13

20.31

10.75

0.00

64.59

64.64

Changchun

27.70

16.81

14.63

9.04

42.96

0.00

48.30

108.81

58.21

59.19

29.11

81.05

91.04

0.00

15.66

6.82

6.84

5.98

4.22

4.16

7.51

317.26

126.70

170.01

90.02

56.43

57.97

112.71

Shenyang Hohhot Shijiazhuang Taiyuan Jinan

29.97

16.23

15.11

14.14

7.27

7.30

13.61

312.64

230.15

216.15

114.68

77.01

80.92

164.14

8.5.3 Calculation of the Carrying Capacity and the Cost of Running in Different Modes of Operation 1. Confirm the carrying capacity of the vehicle and the calculation of the running cost (1) Carrying capacity The high-speed rail confirmation car is the same as the daily operating highspeed rail express train, and is mainly responsible for the safety inspection of the high-speed rail line behind the sunroof. When confirming that the goods are transported on the vehicle, the goods are concentrated in the express cabinet, the area between the seats, the large luggage storage area at the connection of the carriages, etc. The reference [5, 11] can be obtained. The average carrying capacity of the confirmed vehicle running mode of the high-speed rail is 27.6 tons. (2) Opening cost High-speed rail confirmed trains consist of fixed costs and loading and unloading costs. References [5, 12] can be obtained that the fixed cost of confirming the operation of high-speed rail includes station freight service fees and service personnel wages, and the value in this paper is 5263.55 yuan; the loading and unloading cost includes loading and unloading costs and on-site operation fees, and the value in this paper is 154.33 yuan./Ton. Therefore, the total cost of confirming the driving of high-speed rail can be expressed as: C1 = 5263.55 + 154.33Q 1

(8.11)

102

Y. Wang et al.

In the formula: the total cost of C1 confirming the driving of the highspeed rail, Yuan; Q 1 Confirm the tonnage of goods loaded in the piggyback mode for high-speed rail, tons; 2. Calculation of piggybacking capacity and running cost of high-speed rail passenger trains (1) Carrying capacity Similar to the high-speed rail confirmation train mode, when the highspeed rail passenger train is piggybacking, the goods are concentrated in the express locker, the area between the seats, the large luggage storage area at the connection of the carriages, etc., but consider the passengers on the train. With the use of the area, the actual remaining area for placing goods is greatly reduced. References [5, 11] found that the average carrying capacity of high-speed rail passenger trains is 2.43 tons. (2) Opening cost Similar to the high-speed rail confirmation car, the goods are piggybacked, and the variable cost is not included. Taking the fixed cost of the highspeed rail passenger train piggybacking mode as 152.27 yuan, the loading and unloading cost is 35.5 yuan/ton. The total cost of the high-speed rail passenger train piggybacking can be expressed as: C2 = 152.27 + 35.5Q 2

(8.12)

In the formula: C2 is the total cost of piggybacking high-speed rail passenger trains, yuan; Q 2 Tonnage of cargo loaded in piggyback mode for HSR passenger trains, tons; 3. Calculation of combined passenger and cargo carrying capacity and cost of operation (1) Carrying capacity The passenger-cargo mixed operation mode generally adopts the form of reserved carriages. In this paper, an eight-group passenger train is used as an example. Seven carriages are used for normal passenger transportation, and the remaining one is used for high-speed rail express business. Reference [5] can obtain the average carrying capacity of this mode is 12.67 tons. (2) Opening cost Different from the opening cost under the first two piggybacking modes, a certain amount of variable cost will be generated when the passenger and cargo mixing operation is carried out. In reference [5], the fixed cost of starting a train with passenger and cargo is 2 346.34 yuan; the variable cost is 18.61 yuan per vehicle kilometer; the loading and unloading cost is 58.87 yuan per ton, so:

8 Research on the Network Operation Mode of High-Speed Rail Express

C3 = 2346.34 + 58.87Q 3 + 18.61D3

103

(8.13)

In the formula: C3 the total cost of opening the line for the mixed arrangement of passenger and cargo, Yuan; Q 3 Tonnage of cargo loaded for passenger and cargo mixed operation mode, tons; D3 The travel distance for the passenger and cargo mixed travel mode, kilometers. 4. Calculation of carrying capacity and running cost of high-speed rail freight trains (1) Carrying capacity CRRC’s newly developed high-speed rail freight train with four movements, four trailers and eight formations have a rated load of 110t, a cargo volume of not less than 800 m3 , a cargo space utilization rate ≥ of 85%, and can adapt to ambient temperature −25 °C to 40 °C. (2) Opening cost References [5, 12] shows that the fixed cost of high-speed rail freight trains is 18770.68 yuan; the loading and unloading cost is 54.2 yuan/ton; the variable cost is 148.9 yuan/train kilometer. Therefore, the total cost of high-speed rail freight trains can be expressed as: C4 = 18770.68 + 54.2Q 4 + 148.9D4

(8.14)

In the formula: C4 the total costs of the high-speed rail freight train, Yuan; Q 4 Tonnage of cargo loaded for high-speed rail freight train mode, tons; D4 It is the travel distance of the high-speed rail freight train.

8.5.4 Determination of the Network Operation Mode of High-Speed Rail Express The high-speed rail express network formed by the combination of 19 city nodes such as Beijing, Tianjin, Taiyuan, Shijiazhuang, etc. to study the network operation mode of high-speed rail express. The layout of the node cities and the simplified network are shown in Fig. 8.1. Taking the train operation map in the third quarter of 2021 as a reference, in the high-speed rail express network constructed in this paper, the number of trains that can be piggybacked by using high-speed rail confirmation vehicles in the downstream direction are: DJ 111 Beijing South-Tianjin, DJ 1021 Beijing WestShijiazhuang, DJ 1 Beijing South-Jinan West, DJ 1055 Beijing West-Wuhan, DJ 491 Shijiazhuang-Jinan West, DJ 33 Zhengzhou East-Xi’an North, DJ 1133 Xi’an North-Chengdu East, DJ 135 Zhengzhou East-Hefei South, DJ 1061 Nanjing SouthHefei South, DJ 740/37 Hefei South-Hangzhou East, DJ 713 Nanjing-Shanghai, DJ 27 Wuhan-Changsha South, DJ 29 Changsha South-Guangzhou South, DJ 10003 Guiyang North-Guangzhou South, DJ 11/4 Shanghai Hongqiao- South of Fuzhou.

104

Y. Wang et al.

Fig. 8.1 High-speed rail express network and simplified schematic diagram

The following will use the constructed minimum cost model to compare the three decision-making schemes for the operation of high -speed rail express, and determine the network operation mode of China’s high-speed rail express from the feasibility and economy of the scheme. 1. Confirm the driving mode of the car and passenger sports car group piggybacking (1) Confirm the feasibility of the driving mode scheme piggybacked by car and passenger sports car groups Substitute data such as OD prediction value and line network capacity into the optimized model to solve. It is found that the amount of OD in coastal cities is relatively large. For example, the predicted value of OD sent from Hangzhou to Fuzhou every day is 1 067 tons, while the Hangzhou-Shenzhen line via Hangzhou-Shenzhen line has a large amount of OD. There are 70 passenger trains scheduled for the section to Fuzhou, but there are no confirmed trains available. In the constructed decision-making model of travel mode, the residual capacity constraints in the high-speed rail express network section limit the number of trains using the piggyback mode should be less than the actual number of passenger cars. Therefore, the HangzhouFuzhou section of the Hangzhou-Shenzhen Line can carry a maximum of 170.1 tons of goods through piggyback transportation, which is far less than the predicted value of 1067 tons. Therefore, only the driving mode of piggybacking by confirmed vehicles and passenger sports vehicles cannot meet the express market demand in the prospect year. The mode is suitable for the initial stage of the development of high-speed rail express with small overall cargo volume.

8 Research on the Network Operation Mode of High-Speed Rail Express

105

(2) The economy of the driving mode scheme piggybacked by car and passenger sports car groups confirmed that the piggybacking mode of trains and passenger trains can make use of the existing high-speed rail equipment and line network capacity resources. 2.

Direct high-speed rail express network operation mode (1) The feasibility of the network operation mode of direct high-speed rail express Substitute the line network capacity, the OD value of the prospective year, the carrying capacity of the four operating modes of the high-speed rail express, and the operating cost into the optimized model constructed to solve the problem, and it can be confirmed that the trains and passenger EMUs are piggybacked, the passenger and cargo are mixed, and the highspeed rail freight trains are directly operated. Table 8.3 shows part of the results of the decision-making plan for the networked operation mode of high-speed rail express. The scheme adopts the transportation modes of confirmed trains and passenger EMUs, mixed passenger and freight trains, and direct high-speed rail freight trains. It can meet the constraints of line network capacity and OD constraints, and can meet the market needs of future large-scale cargo transportation. It is feasible. (2) Economical plan of direct high-speed rail express network operation mode

Table 8.3 Development bank mode decision scheme (Part) (Unit: Column) Zhengzhou Open model

Changsha

2

3

4

1

2

3

Beijing

2

5

1

1

5

Tianjin

2

6

4

Taiyuan

2

3

Shijiazhuang

4

5

Jinan

2

5

Xi’an

2

7

3

2

3

5

3

3

3

5

3

4

2

1

1

2

2

2

1

1

1

2

4

5

3

5

4

1

3

0

1

3

7

2

4

4

1

1

5

6

3

0

5

2

3

Zhengzhou

1

0

Nanjing

0

3

2

5

4

1

2

2

1

3

0

6

2

4

1

2

Chengdu

3

6

1

2

5

1

0

0

1

1

0

2

3

7

1

Hefei

1

1

1

0

0

1

Hangzhou

1

3

8

3

3

7

Shanghai

2

4

1

5

5

1

Chongqing

4

1

0

1

4

7

Changsha

4

1

Nanchang

2

Wuhan

1

Wuhan

1

4

1

4

1

106

Y. Wang et al.

Table 8.3 correspond to the number of trains operating in different operating modes. In this operating mode decision-making plan, a total of 169 high-speed rail freight trains will be operated, with 18,590 tons of freight transported accounting for more than 70% of the overall OD volume, and 505 carriages. 426 high-speed rail passenger trains are responsible for piggybacking, and the total cost of the trip is 33,477,731 yuan. Therefore, the network operation mode scheme of direct high-speed rail express can meet the needs of the transportation market under the constraints of online network capacity. This mode has relatively simple operation organization and short overall transportation time, which is feasible, but cannot give full play to China’s high-speed rail express hub-and-spoke network advantages, higher transportation costs, and general program economy. 3. High-speed rail express network operation mode combined with direct transit (1) The feasibility of the high-speed rail express network operation mode scheme combined with direct transfer The network operation mode of high-speed rail express combined with direct transit is: on the basis of the network operation mode of direct high-speed rail express, under the condition of satisfying the constraints of line network capacity, carrying capacity and OD constraints, the OD flow with the same travel path will be the integration of distribution and distribution at the transfer node, and the completion of long-distance transportation through high-speed rail freight trains, thus forming economies of scale, reducing the overall transportation cost, and making it feasible to operate. (2) The economy of the high-speed rail express network operation mode scheme combined with direct transfer When constructing the decision-making model of the opening mode, the impact of transit transportation on the decision-making model was not considered. Taking the OD flow from Beijing to Zhengzhou and the OD flow from Tianjin to Zhengzhou as examples, the model calculated the result that the high-speed rail freight from Beijing to Zhengzhou opened. 1 special train, and the rest of the OD volume is piggybacked by 5 passenger-cargo mixed trains and 2 passenger trains. The goods sent from Tianjin to Zhengzhou are piggybacked by 6 passenger-cargo mixed trains and 2 passenger trains. (1) Calculate the minimum mileage that can be used for transfer and distribution Since the confirmed vehicle piggybacking is limited by the route and quantity of the confirmed vehicles, only the piggybacking of passenger and sports trains and the transit transportation between passenger and freight mixed trains and high-speed rail freight trains are considered in the optimization of transfer distribution. Calculated according to the average price of 2.14 yuan/ton kilometer of goods, in the case of high-speed rail freight trains, mixed passenger and freight trains, and passenger-cargo trains being fully loaded at the same time, the highspeed rail freight trains will start to make profits when the running mileage

8 Research on the Network Operation Mode of High-Speed Rail Express

107

exceeds 285 km, and the operation will start at this time. The profit of highspeed rail freight trains is greater than the profit of passenger and freight trains. When the operating mileage exceeds 302 km, the profit of operating high-speed rail freight trains is greater than the profit of passenger and sports trains. The city nodes will carry out transfer and distribution operations. (2) Optimization plan after transfer and distribution The operating mileage from Beijing to Zhengzhou is 617 km, far exceeding the profit mileage of 302 km. Therefore, the goods transported by the 6 passengercargo mixed trains from Tianjin to Zhengzhou in the original plan can be transferred to Beijing for transit, distribution, and then high-speed rail freight. special trains, thereby improving the overall efficiency of the railway sector and giving full play to the advantages of China’s high-speed rail express network. To sum up, the networked operation mode of high-speed rail express combined with direct transfer is feasible to implement. In the same travel path of two or more OD flows, suitable transfer nodes can be selected for distribution operations, resulting in economies of scale and reducing the overall cost of transportation. The above three schemes are compared as shown in Table 8.4. By comparing the three schemes of China’s high-speed rail express network operation mode, it can be concluded that China’s high-speed rail express should adopt a combination of direct transit and operation mode, and the four modes are optimized and combined with each other. The main channels of high -speed railways such as Guangzhou Expressway, Shanghai-Chengdu Expressway, and Xulan Expressway are used for trunk line transportation. The other three modes are used as supplements. They are responsible for the transportation of small and medium-sized goods between regions. The full load rate expands the scope of services, direct transportation Table 8.4 Comparison of three open mode schemes Open mode scheme

Advantage

Shortcoming

Confirm the driving mode of the car and passenger sports car group piggybacking

By confirmed cars and passenger sports trains, with low input cost and no occupation of line capacity

The carrying capacity is small and cannot meet the needs of the future development of the high-speed rail express market

Direct drive mode

Direct driving, simple driving organization and short overall transportation time

Unable to give full play to the advantages of China’s high-speed rail express hub-and-spoke network, the transportation cost is high

Combination of direct and transit mode

The transfer and distribution can be used to improve the utilization rate of vehicle bottom and the full load rate of high-speed rail freight trains, thereby bringing economies of scale and reducing transportation costs

The layout of the transit nodes needs to be planned, and the traffic organization is more complicated

108

Y. Wang et al.

to complete long-distance transportation of large quantities of goods, and improves economies of scale.

8.6 Conclusions This paper analyzes the decision-making factors of the high-speed rail express network operation mode from the three aspects of the high-speed rail express network capacity, the high-speed rail express market OD, the carrying capacity of the operation mode, and the operation cost. The remaining capacity limit, carrying capacity limit and overall transportation task limit of the high-speed rail express network corresponding to the mode are used as constraints to build a high-speed rail express operation mode decision-making model. Then, the carrying capacity and the cost of different travel modes are analyzed and calculated, and the decision-making model of the travel mode is verified, and the networked mode of high-speed rail express transportation combined with direct transit in China is proposed. In this paper, when constructing the decision-making model of high-speed rail express network operation mode, in order to facilitate the model solution, it is assumed that there is no transfer operation, and only when the network operation mode combination scheme is obtained, it analyzes the effect of the transfer operation on the high-speed rail express network through an example. The influence of the decision-making of the CDB model has not been reflected in the decision-making model. How to incorporate the transit factor into the decision-making model of the HSR Express network-based CDB model needs further research. Funding This research was supported by the National Natural Science Foundation of China (Project No. 52072314; 52172321;52102391), Sichuan Science and Technology Program (Project No. 2020YJ0268;2020YJ0256;2021YFQ0001;2021YFH0175), Science and Technology Plan of China Railway Corporation (Project No.: 2019F002), China Shenhua Energy Co., Ltd. Science and Technology Program (Project No.: CJNY-20–02), China Railway Beijing Bureau Group Co., Ltd. Science and Technology Program (2021BY02, 2020AY02),the National Key R&D Program of China (2017YFB1200702), Key science and technology projects in the transportation industry of the Ministry of Transport(2022-ZD7-132).

References 1. Xiao, H.L., Ke, H.T.: Market potential and approaches of parcels and mail by high-speed rail in China. Case Stud. Transp. Policy 7(3), 583–597 (2019) 2. Xiao, H.L., Ke, H.T., Anthony, W., et al.: Parcels and mail by high-speed rail–A comparative analysis of Germany, France and China. J. Rail Transp. Plann. Manag. 6(2), 77–88 (2016) 3. Dong-jin, N., Byung-In, K., Hyunbo, C., et al.: A multi-leg load-planning algorithm for a high-speed freight train. Int. J. Ind. Eng. 23(3), 183–194 (2016) 4. Qin, M.: Research on Combined Optimization of High-Speed Rail Express Demand Forecast and Organizational Model. Southwest Jiaotong University, Chengdu (2017)

8 Research on the Network Operation Mode of High-Speed Rail Express

109

5. Zhang, C.: Research on High-Speed Rail Freight Transportation Decision-Making Based on Different Transportation Organization Modes. Southwest Jiaotong University, Chengdu (2018) 6. Wang, X.: Research on the Organization Mode of Express Freight Transportation by High-speed Railway. Lanzhou Jiaotong University, Lanzhou (2020) 7. Li, S.: Research on Express Freight Transportation System of High-speed Railway. Southwest Jiaotong University, Chengdu (2017) 8. Guo, X.: Evaluation of My Country’s High-Speed Railway to Develop Express Business Model. Beijing Jiaotong University, Beijing (2015) 9. Chen, T., Tian, F., Ren, R., et al.: A review of research on high-speed railway network capacity. J. Transport. Eng. Inform. 19(3), 51–58 (2021) 10. Li, H., Zhang, C.: Railway Stations and Hubs. China Railway Press, Beijing (2018) 11. Zhang, H.: Discussion on the Feasibility and Operation Mode of High-Speed Railway to Develop Fast Freight Transportation. Lanzhou Jiaotong University, Lanzhou (2015) 12. Hou, J., Zheng, G., Cao, J.: Research on high-speed railway passenger and freight cost conversion based on activity-based costing method—Taking the PQ high-speed railway passenger line as an example. Logist. Eng. Manag. 3(29), 27–31 (2017)

Chapter 9

Solving a Locomotive Routing Problem of Heavy Haul Railways Yongxin Li, Meng Wang, Zhen Liu, Chi Zhang, and Xueting Li

Abstract Locomotive routing problem of heavy haul railways is more difficult than that of regular railways because it has some unique characteristics including different numbers of locomotives utilized for different train type, mixed using of various locomotive types, and unpaired locomotive turn-around. This paper extends the existing model of LRP. Locomotive attached variable is introduced to ensure locomotive turn-around in pair. Tractive force supply and demand balance constrain is used to transform the complex effects of multi-locomotive traction and various locomotive types. Locomotive servicing and maintenance requirements is transformed into constrains utilizing conditional discrimination. Then we develop a MIP model which has a huge scale. It is difficult to solve it utilizing the exact algorithm. We design a two-stage heuristic algorithm. In the first stage, the initial solution is produced by the heuristic algorithm based on dynamic table of locomotives staying at stations. In the Y. Li · M. Wang · Z. Liu · X. Li (B) School of Transportation and Logistics, Southwest Jiaotong University, Chengdu 610031, China e-mail: [email protected] Y. Li e-mail: [email protected] M. Wang e-mail: [email protected] Z. Liu e-mail: [email protected] X. Li National and Local Joint Engineering Laboratory of Comprehensive Intelligent Transportation, Southwest JiaoTong University, Chengdu 610031, China National Engineering Laboratory of Integrated Transportation Big Data Application Technology, Chengdu 610031, China M. Wang China Energy Railway Equipment Company, Beijing 100011, China C. Zhang China Railway Shanghai Design Institute Group Corporation Limited, Shanghai 200070, China e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_9

111

112

Y. Li et al.

second stage, Tabu Search is used to optimize the initial solutions deeply with the search space limit method and the feasible solution transformation method which are used to handle the model constraints. We conduct many computational experiments and comparative experiments to prove the rationality of the model and the effectiveness of the algorithm. The results of this paper can provide an effective reference for the study on the complex locomotive routing problem.

9.1 Introduction 9.1.1 Background In order to satisfy various tractive force demands of different types of heavy haul train, heavy haul railways are always equipped with locomotives of various types, and most of heavy haul trains need multi-locomotive traction. It makes the locomotive routing problem (LRP) of heavy haul railways a “multi-multi” matching problem. Compared with the “one-to-one” problem of ordinary railways, LRP of heavy haul railways is more complicated. In the actual scheduling work, locomotive scheduling also has low plan fulfillment rates. Therefore, the results of the study on the locomotive scheduling problem (LSP) of heavy haul railways are difficult to apply in practice. It is more practical to study the LRP of heavy haul train.

9.1.2 Literature Review LSP and its extension LRP have always been the focus of railway transportation research field. Florian et al. [1] propose the general LSP first, which is how to assign the locomotives to each train considering several locomotive scheduling constraints to ensure the lowest cost. Scholars have proposed multiple methods to solve this problem. Forbes et al. [2] design a two-stage exact algorithm. In the first stage, relax the locomotive type constraint, so that the model can be solved by the exact algorithm. In the second stage, Branch and Bound Method is used to allocate locomotive types and obtain the complete optimal solution. Cordeau et al. [3] focus on the locomotive and car assignment problem, build a MIP model based on a time–space network, and use Benders Decomposition Approach to solve this model. Ahuja et al. [4] build a MIP model of a big size in order to solve the locomotive scheduling problem faced by CSX Transportation, and design an a algorithm using problem decomposition, Integer Programming, and very large-scale Neighborhood Search to solve this model in stages. Zhang et al. [5] build a three-dimensional assignment mathematical model to express LSP with multi-type locomotives and design an algorithm for formulating the locomotive working diagram under the condition of a fixed number of locomotives and a heuristic algorithm for locomotive routing connection exchanging to

9 Solving a Locomotive Routing Problem of Heavy Haul Railways

113

achieve overall optimization of locomotive turnaround time and locomotive type assignment. Wang et al. [6] design a genetic—ant colony algorithm with genetic strategy to improve the traditional algorithms which are easy to fall into the local optimal solution. Based on the LSP, LRP additionally considers the influences of locomotive servicing, maintenance, and recharging, etc. Its research results are easier to apply in practice. Vaidyanathan et al. [7] generalize this kind of problem as LRP, and develop a modeling framework and a multi-stage solution approach. Wang et al. [8] transform the LRP into a route choice problem in the time–space network, and build a route choice optimization model, which is solved by the Genetic Algorithm. Frisch et al. [9] propose two algorithms to solve the MIP model for the LSP with locomotive maintenance constraints. The Overlapping Rolling Horizon Approach can obtain high-quality solutions, but the calculation time is long; the Two-Stage Matheuristic can obtain solutions in a short time, and can meet the demands of various situations. Miranda et al. [10] transform the LRP into a multi-commodity network flow problem with side constraints, build an integer linear programming model, and use current the state-of-the-art mixed integer programming solvers to solve this model. The solution speed and quality of their method are verified by computational experiments. The LRP research of heavy haul railways needs to consider many factors including various locomotive types, multi-locomotive traction, and unpaired locomotive turnaround. Compared with regular railways, this problem has a larger scale and has become more difficult to solve. At present, there are few studies for this problem. Su et al. [11] build a MIP model for the LSP faced by Da-qin Railway, and transform the problem into a paired, single-locomotive traction problem by adding virtual train, then use the Hungarian algorithm to solve the problem. In summary, the existing research has achieved fruitful results in solving the LRP of regular railways. There are several methods that can be applied in practice already. However, the number of researches for heavy haul railways is still few. There are only a few results for the LSP of heavy haul railways.

9.1.3 Contributions This paper focuses on the LRP of heavy-haul railways and considers the constraints of locomotive maintenance and servicing. An effective method is proposed to obtain a high-quality locomotive routing plan in a short time. Specifically, the main contributions of this paper are as follows: • We deeply summarize the key difficulties in solving the LRP of heavy haul railways. The difference between the operating rules of heavy haul railways and regular railways is analyzed. The requirements of locomotive servicing and maintenance which is to ensure safety and stability are summarized in detail. • We build a MIP model for the LRP of heavy haul railways. The application of conditional discrimination is to convert locomotive servicing and maintenance

114

Y. Li et al.

requirements into model constrains. The application of tractive force supply and demand balance constraint is to transform the influences of the multi-locomotive traction and various locomotive types. The application of locomotive attached 0–1 variable is to realize locomotive turn-around in pair. • We develop a two-stage heuristic algorithm to solve the model. In the first stage, the initial solution is produced by the heuristic algorithm based on dynamic table of locomotives staying at stations. In the second stage, Tabu Search is used to optimize the initial solutions deeply with search space limit method and feasible solution transformation method which are used to handle the model constraints. This algorithm provides new ideas for solving LRP of heavy haul railways. The rest of this paper is organized as follows. Section 9.2 describes the analysis of the LRP of heavy haul railways in detail and summarizes the solving ideas. Section 9.3 prevents the building process of the MIP model. Section 9.4 prevents the design ideas and algorithm flow of the two-stage heuristic algorithm. Section 9.5 verifies the rationality of the model and the effectiveness of the algorithm through computational experiments.

9.2 Problem Analysis The LSP of heavy haul train has the following features. • Multi-locomotive traction. Most of heavy haul trains are towed by a locomotive group which consists of multiple locomotives. Trains with different tractive tonnages require different number of locomotives. In the actual locomotive dispatching, it is necessary to disassemble and reconnect the locomotive groups in order to ensure the efficient locomotive operation. • Various locomotive types. Heavy haul railways are always equipped with various locomotive types, and the tractive force of each type is different. It is necessary to use a variety of locomotives to ensure that all demands of tractive force can be met without wasting. • Unpaired locomotive turn-around. Heavy haul railways generally exist in the heavy cars direction and empty cars direction. The demand of the number of locomotive in these two directions is different. The heavy cars direction need more locomotives obviously. That causes the unpaired locomotive turn-around. In order to ensure locomotive turn-around in pair, locomotives are always attached on the opposite train to return. There are several maintenance classifications including minor repair, light repair, intermediate repair, etc. each classification has a period. Every time the accumulated running kilometers of a locomotive reaches the end of one maintenance classification period, this locomotive should be maintained at this classification. The demand of locomotive servicing is similar to maintenance. The period of servicing is called “Continued mileage for servicing”.

9 Solving a Locomotive Routing Problem of Heavy Haul Railways

115

Based on the above analysis, the research questions of this paper can be summarized as follows: Base on given train schedule, the locomotives with various types are assigned to provide tractive force for each heavy haul train and be attached on the opposite train to return in an economical and reasonable manner. At the same time, according to the real-time status of the locomotives, arrange locomotives for servicing and maintenance.

9.3 Model Building 9.3.1 Model Assumptions 1. The research only considers a single locomotive tractive district. Locomotives operation in each tractive district of heavy haul railways is independent. The locomotive routing plan of each tractive district can be prepared separately. 2. The locomotive running mode of the tractive district is cycle locomotive running mode. Study on LRP with cycle locomotive running mode is one of the most complex problems. It is easy for Study on LRP with other mode to obtain solutions by modifying this study. 3. The heavy haul trains don’t stop at the station in the tractive district.

9.3.2 Mathematical Symbol Description Model related parameters I is the set of trains in the train schedule, which consists of two sets. I1 is the set of real trains. I2 is the set of virtual trains. In order to represent the locomotives leaving the depot and put back into operation, the virtual trains are used. Their arrival time is the time of the locomotives leaving the depot. Their terminal station is the depot. Their departure station and time can select arbitrarily. i and j are elements in I . di is the direction of i. When its value is 1, it means that the direction of i is down; when the value is −1, it means that the direction of i is up. K is the set of locomotives, k is a element in K . M is the set of locomotives types, m is a element in M, m k is the type of k. S is the set of stations. There are three elements in S according to the features of cycle locomotive running mode, including s1 , s2 , s3 . s1 and s3 is the station with the turn-around depot. s2 is the station with the locomotive depot. s is a element in S. P is the set of maintenance classifications. p is an element in P. q expresses locomotive servicing. Tisj is the connection time of i and j at s which is the difference between arrival time of i and departure time of j at s. Tik is the running time of k on i. Ti1 , Ti2 are s the running time of i in the s1 − s2 section and s2 − s3 section. Twor k is the minimum

116

Y. Li et al.

operating time of the locomotive at s. Tback is the minimum turn-around time. Tr e f is the minimum servicing time. tisa , tisb are the arrival and departure time of i at s. lqk is the accumulated running kilometers of k after the last servicing. lqmkmax is the servicing period of m k . l kp is the accumulated running kilometers of k after the last k maintenance of p. l mk p max is the period of the maintenance of p for m k . li is the running kilometers of k on i. l1 , l2 are the length of the s1 − s2 section and s2 − s3 section. l is the warning value of locomotive servicing and maintenance, and its value is equal to l2 . N f is the maximum number of locomotives that can be attached to a train. Fmk is the tractive force that the locomotive whose type is m k can provide. Fi is the tractive force demand of i. Zero–one variable xsik j is locomotive connection variable that means whether k tows j after towing i at s. f sik j is locomotive attached variable that means whether k is attached on j after towing i at s. u qk means whether k is going to servicing. u kp means whether k is going to maintenance of p.

9.3.3 Objective Function During the operation of the locomotive, the state of locomotives can be divided into the operating state and non-operating state. The operating state is the state in which the locomotives tow the train. The non-operating state is the state in which the locomotives are attached on the train or stay at the station. In order to improve the efficiency of locomotive operation, the time when the locomotives are in nonoperating state should be reduced as much as possible, so the objective function is to minimize the non-operation time of the locomotive which is expressed by the summation of the total stop time and the total attached time as follows. min Z =

  (xsik j + f sik j ) · Tisj + f sik j · T jk

(9.1)

s∈S k∈K i∈I j∈I j

Tisj , Tk can be calculated by the following formula. Tisj = t bjs − tisa

(9.2)

9 Solving a Locomotive Routing Problem of Heavy Haul Railways

⎧ 1 k ⎪ ⎨ T j , l j = l1 j Tk = T j2 , l kj = l2 ⎪ ⎩ T 1 + T 2, l k = l + l 1 2 j j j

117

(9.3)

9.3.4 Constraints 1. Tractive force supply and demand balance constraint The locomotive routing should satisfy the tractive force demands of all trains in the train schedule. It can be expressed by the following formula. ∀ j ∈ I1



(xsik j · Fmk ) = F j

(9.4)

s∈S k∈K i∈I

2. Cycle locomotive running mode constraint The timing of turn-around and servicing must meet the requirements of the cycle locomotive running mode. Locomotives should turn around immediately at the stations with the turn-around depot. In other words, locomotives should choose an opposite train to tow or be attached after the last task is done at the stations with the turn-around depot. It can be expressed by the following formula. ∀s = s1 , s3 , k ∈ K , i ∈ I1



(xsik j + f sik j ) = 1, di · d j = −1, l kj = l1 + l2

j∈I1

(9.5) The timing of servicing is determined by the accumulated running kilometers of the locomotives which is calculated by the following formula. ∀xsik j = 1 lqk = lqk + l kj

(9.6)

When the accumulated running kilometers reach the servicing warning value, the locomotive should be serviced.  (xsik j + f sik j ) = 1, u qk = 0, lqk = 0 (9.7) ∀u qk = 1, s = s2 , i ∈ I1 j∈I1

The servicing at the locomotive depot can be regarded as a locomotive connection. After it is complete, the value of lqk is cleared. The following formula can express it. ∀u qk = 1, s = s2 , i ∈ I1

 j∈I1

(xsik j + f sik j ) = 1, u qk = 0, lqk = 0

(9.8)

118

Y. Li et al.

The direction of i and j in formula (9.8) should be constrained by the following formula. l ,d = 1 l ,d = 1 ∀u qk = 1, s = s2 , xsik j + f sik j = 1 lik = 1 i , l kj = 1 j l2 , di = −1 l2 , d j = −1 (9.9) s 3. Operating time constraint ∀xsik j = 1 Tisj ≥ Twor k The operating time of the connection of k at s should be greater than or equal to s Twor k. s ∀xsik j = 1 Tisj ≥ Twor k

(9.10)

s The specific value of Twor k depends on the station, as shown in the following formula. ⎧ ⎨ Tback , s = s1 , s3 & i ∈ I1 s = (9.11) Twor T , s = s2 & i ∈ I 1 k ⎩ ref 0 , i ∈ I2

4. Number of attached locomotives constraint The number of locomotives attached to each train must be less than or equal to Nf. ∀ j ∈ I1



f sik j ≤ N f

(9.12)

s∈S i∈I k∈k

5. Locomotive maintenance constraint When the accumulated running kilometers reach the maintenance warning value, the locomotive should be maintained. The accumulated running kilometers of a locomotive for different maintenance classifications should be counted separately, as shown in the following formula. ∀xsik j = 1, p ∈ P l kp = l kp + l kj

(9.13)

When the accumulated running kilometers for p reach the warning value, the locomotive should return to the locomotive depot and do the maintenance of p. While the maintenance is complete, the corresponding accumulated running kilometers should be cleared. l 1 , s = s1 k k − l, +∞) l = , u = 1, l kp = 0 (9.14) ∀l kp ∈ (l mk p max j l 2 , s = s3 p After the maintenance is complete, the locomotive should leave the depot and perform new tasks. This process can be represented by xsik j and f sik j as a

9 Solving a Locomotive Routing Problem of Heavy Haul Railways

119

locomotive connection of a virtual train and a real train at the station with the locomotive depot. ∀i ∈ I2 , s = s2



(xsik j

+

f sik j )

=

1, u kp

k∈K j∈I1

=

0, l kj

=

l1 , d j = −1 (9.15) l2 , d j = 1

9.4 A Two-Stage Heuristic Algorithm 9.4.1 Applicability Analysis of Tabu Search Algorithm The problem studied by this paper is a NP-hard problem. It is difficult to solve it using exact algorithms. The calculation time is also too long that it cannot meet the timeliness requirement of the locomotive operation. From the point of view of practical application, it is a more reasonable idea to design the meta-heuristic algorithm based on the characteristics of the problem to quickly obtain high quality solutions, such as GA, PSO, etc. [12, 13]. The model in this paper is a model with complex constraints. Tabu Search algorithm has great advantages in solving this model. For Tabu Search, initial solutions is the key to the performance of the algorithm. Tabu Search requires that initial solutions not only have high quality but also have randomness. We design a heuristic algorithm to get high quality and random initial solutions.

9.4.2 Algorithm Design Tabu Search algorithm elements include encoding, fitness function, initial solution, neighborhood, tabu list, selection strategy, aspiration criterion, and stopping rule. The specific design is as follows. 1. Encoding and decoding The solution should be able to fully represent a locomotive routing plan. Therefore, the encoded content of the solution should contain: • All trains towed by each locomotive. • All trains attaching to each locomotive. • The departure and arrival station of towing or attached for each locomotive. To encode the solution, all trains and locomotives are numbered. Trains are numbered in order of departure time. Among them, down trains are numbered by odd d numbers, like 1, 3, 5, …, i, …, i max and up trains are numbered by even numbers, like u d u is the total number of down trains and up trains. 2, 4, 6, …, i, …, i max . i max and i max Locomotives are numbered like 1, 2, 3, …, k, …, kmax . kmax is the total number of

120

Y. Li et al.

locomotives. The code of k towing can be realized by arranging the numbers of the trains towed by k in order of time, like the following formula. codek = (i 1 , i 2 , i 3 , · · · )

(9.16)

The embryonic form of solution code can be obtained by combining the codes of all locomotives, as shown in the following formula. code = (codek ), k = 1, 2, · · · , kmax

(9.17)

Locomotive attached can be regarded as a special form locomotive towing. Then the numbers of the trains to which the locomotive is attached on can be directly added to formula (9.17) in order of time. Although the encoding form of the locomotive attached is same as of locomotive towing, it can also be decoded indirectly as follows. K i is the set of the locomotives on i. Any group of locomotives that just meet the f tractive force demands of i can be selected from K i as the elements of K ix . K i is the f set of the locomotives which are attached on i. K i is the universal set, and K i can x be expressed by the complementary set of K i . f

K i =  K i K ix

(9.18)

The running length of locomotive towing and attached is divided into full running and half running. Full running refers to the locomotive running between two turnaround depots. Half running refers to the locomotive running between turn-around depot and locomotive depot which is in the middle of the two turn-around depots. Code “0” is used to encode this message. It is stipulated that the two codes after "0" is half running, and the tractive force demands of these two codes are double in order to avoid misjudgment. 2. Fitness function The objective function of the model can be directly used as the fitness function. The total stop time and total attached time of each locomotive can be obtained by decoding, and the fitness can be obtained by accumulating one by one locomotive as follows.  f (Tkx + Tk ) (9.19) Tcode = k∈K

3. Initial solution generation algorithm The initial solution of the Tabu Search algorithm must be feasible and random, so we design an initial solution generation algorithm based on the dynamic table of locomotives at stations.

9 Solving a Locomotive Routing Problem of Heavy Haul Railways

121

Table 9.1 Dynamic table Station

Available locomotives

s1

k1

k2

k3

…

Unavailable locomotives k10

k11

k12

…

s2

k4

k5

k6

…

k13

k14

k15

…

s3

k7

k8

k9

…

k16

k17

k18

…

The dynamic table of locomotives in stations (hereinafter referred to as “dynamic table”) is a table that records the message of locomotives in stations in real time (see Table 9.1). The record content of the dynamic table is the available locomotives and unavailable locomotives staying at each station at a certain time. Unavailable locomotives are the locomotives which are turning around or being serviced or maintained. After the operation is completed, they become available locomotives. Based on this table, the initial solution generation algorithm is designed as follows. Step 1. Input the data required by the algorithm, including the initial state of the locomotives at 18:00, and train schedule and fill in the dynamic table at 18:00. Step 2. Determine the algorithm advance moment and initialize the code. The algorithm is advanced through events, that is, when any train departs or arrives at the station, the algorithm is advanced. Therefore, all time points of all trains are arranged by time to obtain the advancing moment of the algorithm, and the first time among them is taken as the current time of the algorithm. The number of code rows is determined by the locomotive number, and the code of each line is blank. Step 3. Renew accumulative running kilometers of locomotives and dynamic table. According to the running situation of the locomotives not at stations from the previous moment to the current moment, the accumulative running kilometers is renewed. Turn the unavailable locomotives which complete the operation into available locomotives. Step 4. Determine whether the event at the current moment occurs at the turnaround depot or the locomotive depot. If it is at the turn-around depot, turn to Step 5. Otherwise, turn to Step 8. Step 5. Determine whether the event at the current moment is the departure or arrival of the train. If it is departure, turn to Step 6. Otherwise, turn to Step 7. Step 6. Assign locomotives to the departure train. Randomly select a group of locomotives that meet the tractive force demand from the available locomotives at the departure station to tow the departure train. If there are not enough available locomotives, look for it at other stations and assign the found locomotives to the departure station by attaching. Renew the code finally. Step 7. Renew the dynamic table according to the situation of locomotives entering and leaving the station. Add the entering locomotives to the corresponding position in the dynamic table and delete the leaving locomotives from the dynamic table. Step 8. Determine whether the locomotives need to be serviced or maintained. If they need, it is indicated that the train needs to change locomotives at the locomotive depot, and turn to Step 6. Otherwise, turn to Step 9.

122

Y. Li et al.

Step 9. Determine whether the current moment is the final moment. If not, make the next moment the current moment, and turn to Step 3. Otherwise, stop the algorithm. Its algorithm flow is shown by Fig. 9.1. 4 Neighborhood and candidate solution selection The neighborhood solutions are generated by exchanging the locomotives of two adjacent trains in the same direction (see Fig. 9.2). Due to the complex constraints of the model, we use the search space limit method and feasible solution transformation method to deal with the neighborhood moving. The specific processing is as follows. • In order to meet the constraints of the train traction demand, only one locomotive is selected for exchange. Start Data input Determine advancement time and initialize code Renew mileage and dynamic table

Turn-around deport

N

Y N

Departrue train Y

Y Next moment

Find locomotive Y Renew code

N

Servicing/maintenance

Find from other station Assign attached

Renew dynamic table N

Final moment Y Output code End

Fig. 9.1 Algorithm flow of 3. Initial solution generation algorithm

N

9 Solving a Locomotive Routing Problem of Heavy Haul Railways k3 k2 k1

Exchange k1 , k2

123 k3 k2 k1

Fig. 9.2 Neighborhood solution generation diagram

• In order to meet the locomotive running mode constraint and maintenance constraints, it is stipulated that the locomotives entering and leaving the depot shall not be exchanged; • In order to meet the operating time constraint, after the exchange, it is judged whether the locomotive meets the operating time constraint, if not, the neighborhood solution is discarded. Neighborhood generation in the algorithm requires a large amount of computation. In order to improve the search speed, the candidate solution selects the first non-tabu improved solution found, or the solution that satisfies the aspiration criterion. 5. Tabu list Take the exchange of the two locomotives during as tabu objects. Tabu

√ the iteration  size is related to problem size. Its value is i max − 1 . i max is the total number of trains. 6. Aspiration criterion Aspiration criterion is based on fitness. When the fitness of a candidate solution is better than that of the historical optimal solution, its tabu state is ignored and it is taken as the current solution. 7. Search strategy In order to prevent the algorithm from falling into a local optimal solution, the algorithm sets a decentralized search strategy. If no new optimal solution is found after 50 iterations, the new initial solution is generated and the tabu list is cleared. The algorithm restarts the optimization at a new starting point. 8. Stopping rule The stopping criterion is set to a maximum number of iterations which is 500. The algorithm flow of the two-stage heuristic algorithm is shown by Fig. 9.3.

9.5 Case Analysis The model and algorithm in this paper are verified by case analysis. The case is a tractive district with the cycle locomotive running mode (see Fig. 9.4). Station A and C are the stations with the turn-around depot. Station B is the station with the

124

Y. Li et al.

Start

Fig. 9.3 The algorithm flow of the two-stage heuristic algorithm

Initialize data Generate initial solution decentralized search

Y

Clear tabu list

N

Generate neighborhood, select candidate solution Meet Aspiration criterion

Y Candidate solution

as current solution

N

Determining the properties of the candidate solution taboo Determine current solution and Renew state N

Stop Y

Output results End

locomotive depot. The length of the A-B section is 406 km. The length of the B-C section is 183 km. The tractive district is equipped with 32 SS4 locomotives which can supply tractive force equivalent to 5000 t and 31 HXD1 locomotives which can supply tractive force equivalent to 10,000 t. The minimum turn-around time is 40 min. The minimum servicing time is 225 min. Continued mileage for servicing is 2200 km. The period of light repair and minor repair are 70,000 km and 200,000 km. The initial state of the locomotive is shown in Table 9.2. There are 28 trains in schedule. The train schedule is shown by Table 9.3. There are only heavy trains in down direction, including eight 10,000-ton trains and twenty 20,000-ton trains. There are only empty trains which have half tractive force demands of heavy trains with same formation in the up direction. Fig. 9.4 Tractive district diagram

A

B

C

9 Solving a Locomotive Routing Problem of Heavy Haul Railways

125

Table 9.2 The initial state of the locomotive Locomotive number

Initial position

501

W20029

502

W20029

503 504

Running kilometers after servising/km

Running kilometers after light repair/km

Running kilometers after minor repair/km

774

17,312

105,116

774

60,309

66,046

W20029

577

8655

92,762

W20029

577

50,360

109,483

505

Station C

137

58,756

122,210

506

Station C

137

63,431

96,066

507

W10011

714

22,029

113,758

508

W10011

714

24,863

78,805

509

W20033

815

26,583

103,549

510

W20033

815

56,417

129,441

511

W20033

2093

19,697

128,187

512

W20033

2093

45,792

96,439

513

W10012

1566

46,650

65,114

514

W10012

1566

41,224

107,827

515

Minor repair

0

0

0

516

Minor repair

0

0

0

517

Station A

1598

38,763

71,903

518

Station A

1598

52,063

152,273

519

W10012 attached

128

24,245

129,302

520

W10012 attached

128

32,525

72,648

521

W20032

248

62,785

84,612

522

W20032

248

20,757

154,258

523

W20034

742

33,982

91,581

524

W20034

742

51,682

140,719

525

W20034 attached

1547

53,320

92,337

526

W20034 attached

1547

34,191

134,689

527

W20036

992

10,425

156,753

528

W20036

992

40,936

106,411

529

W20040

330

58,487

134,332

530

W20040

330

17,067

67,785

531

W20040

868

51,612

144,602

532

W20040

868

26,303

95,594

1001

W20025

186

26,848

82,028

1002

W20025

186

5984

156,568

1003

W20027

396

26,231

152,446 (continued)

126

Y. Li et al.

Table 9.2 (continued) Locomotive number

Initial position

1004

W20027

1005

Station C

1006 1007

Running kilometers after servising/km

Running kilometers after light repair/km

Running kilometers after minor repair/km

396

40,912

60,275

199

7605

115,067

W20031

870

21,370

152,957

W20031

870

4209

159,610

1008

W10013

911

38,962

168,386

1009

W20035

580

48,053

92,908

1010

W20035

1578

9597

137,499

1011

W20037

1578

44,630

65,429

1012

W20037

239

18,882

199,395

1013

W10015

1002

47,870

98,262

1014

W20039

869

24,566

62,201

1015

W20039

130

9524

66,551

1016

W20028

513

41,869

69,284

1017

W20028

976

37,099

89,726

1018

W20030

552

40,580

151,207

1019

W20030

639

57,255

98,262

1020

W10014

51

42,143

128,147

1021

W10014

964

11,985

143,796

1022

W10016

930

69,388

80,400

1023

Light repair

0

117,809

1024

W20038

894

51,905

76,777

1025

W20038 attached

653

24,503

107,408

1026

W20002

972

39,141

122,837

1027

W20002 attached

332

24,350

124,368

1028

W20004

366

5271

139,547

1029

W20004 attached

995

4085

129,944

1030

Light repair

0

15,569

79,454

1031

Light repair

0

32,962

132,416

0

The algorithm is realized using Matlab 2016a. The optimal solution is obtained by the algorithm with the above data input (see Table 9.4). The calculation time is 343 s, which can meet the timeliness requirement of locomotive operation. The total non-operation time of the optimal solution is 22142 min. The total stop time is 12620 min. The total attached time is 9522 min. The performance of the algorithm is proved by comparative experiment with the Genetic Algorithm. The GA is designed as follows.

Departure

18:39

22:06

0:55

2:46

5:10

12:44

14:02

16:27

18:27

19:30

20:09

21:27

22:21

23:36

1:19

2:34

3:28

3:55

5:25

W10001

W10003

W10005

W10007

W10009

W10011

W10013

W10015

W20001

W20003

W20005

W20007

W20009

W20011

W20013

W20015

W20017

W20019

W20021

A

Train number

Down direction

Table 9.3 Train schedule

11:59

10:29

10:02

9:08

7:53

6:10

4:55

4:01

2:43

2:04

1:01

23:01

20:36

19:18

11:44

9:20

7:29

4:40

1:13

Arrival

B

12:24

10:54

10:27

9:33

8:18

6:35

5:20

4:26

3:08

2:29

1:26

23:21

20:56

19:38

12:04

9:40

7:49

5:00

1:33

Departure

15:08

13:38

13:11

12:17

11:02

9:19

8:04

7:10

5:52

5:13

4:10

2:05

23:40

22:22

14:48

12:24

10:33

7:44

4:17

Arrival

C

W20022

W20020

W20018

W20016

W20014

W20012

W20010

W20008

W20006

W20004

W20002

W10016

W10014

W10012

W10010

W10008

W10006

W10004

W10002

Train number

Up direction

5:58

4:48

3:14

2:24

1:03

23:54

22:48

22:06

20:50

19:49

18:01

14:35

11:05

9:47

8:09

5:40

0:21

23:38

20:35

Departure

C

8:33

7:23

5:49

4:59

3:38

2:29

1:23

0:41

23:25

22:24

20:36

17:10

13:40

12:22

10:44

8:15

2:56

2:13

23:10

Arrival

B

8:58

7:48

6:14

5:24

4:03

2:54

1:48

1:06

23:50

22:49

21:01

17:30

14:00

12:42

11:04

8:35

3:16

2:33

23:30

Departure

(continued)

15:12

14:02

12:28

11:38

10:17

9:08

8:02

7:20

6:04

5:03

3:15

23:44

20:14

18:56

17:18

14:49

9:30

8:47

5:44

Arrival

A

9 Solving a Locomotive Routing Problem of Heavy Haul Railways 127

Departure

6:19

7:34

8:52

11:53

12:59

14:30

15:12

16:06

17:21

W20023

W20025

W20027

W20029

W20031

W20033

W20035

W20037

W20039

A

Train number

Down direction

Table 9.3 (continued)

23:55

22:40

21:46

21:04

19:33

18:27

15:26

14:08

12:53

Arrival

B

0:20

23:05

22:11

21:29

19:58

18:52

15:51

14:33

13:18

Departure

3:04

1:49

0:55

0:13

22:42

21:36

18:35

17:17

16:02

Arrival

C

W20040

W20038

W20036

W20034

W20032

W20030

W20028

W20026

W20024

Train number

Up direction

17:16

16:09

14:19

13:09

12:00

10:08

9:00

7:45

6:36

Departure

C

19:51

18:44

16:54

15:44

14:35

12:43

11:35

10:20

9:11

Arrival

B

20:16

19:09

17:19

16:09

15:00

13:08

12:00

10:45

9:36

Departure

2:30

1:23

23:33

22:23

21:14

19:22

18:14

16:59

15:50

Arrival

A

128 Y. Li et al.

9 Solving a Locomotive Routing Problem of Heavy Haul Railways

129

Table 9.4 The optimal solution Locomotive number

Locomotive route

501

W10004→W20023→Turn-around at station C

502

W10004 attached→W20023→Turn-around at station C

503

W20010→W20025

504

W20010→W20025

505

W20010 attached→W20025

506

W20010 attached→W20025

507

W20012→W20029

508

W20012→W20029

509

W20014→W10011

510

W20014→W10011

511

W20014 attached→Servicing at station B→W20035

512

W20014 attached→Servicing at station B→W20035

513

W20003→Servicing at station B→W20020→W20033

514

W20003→Servicing at station B→W20020→W20033

515

W20020 attached→W20033

516

W20020 attached→W20033

517

W10001→Servicing at station B→W20007→W20028

518

W10001→Servicing at station B→W20007→W20028

519

W20003→W20026

520

W20003→W20026

521

W10003→W10012

522

W10003→W10012 attached

523

W20009→W20030

524

W20009→W20030

525

W20009→Servicing at station B→W20017→W20038

526

W20009→Servicing at station B→W20017→W20038

527

W10005→W10014→Servicing at station B

528

W10005→W10014 attached→Servicing at station B

529

W20013→W20036

530

W20013→W20036

531

W20013→W20036 attached

532

W20013→W20036 attached

1001

W10002→W10009→W20040

1002

W20006→W20019→W20002

1003

W20008→W20021→Turn-around at station C

1004

W20008 attached→W20021→Turn-around at station C (continued)

130

Y. Li et al.

Table 9.4 (continued) Locomotive number

Locomotive route

1005

W20006→W20019→W20002

1006

W20012 attached→W20027

1007

W20012 attached→W20027

1008

W10006→W20029

1009

W20016→W20031

1010

Servicing at station B→W20001→W20024→W10015

1011

Servicing at station B→W20003→W20026

1012

W20018→Minor repair at station B

1013

W20020→Servicing at station B→W10001→W20022→W20037

1014

W10008→W20035

1015

W10008 attached→Servicing at station B→W20015→W10016

1016

W20001→W20022→W20037

1017

W20001→Servicing at station B→W20009→W20030

1018

W20005→W10010→Turn-around at station A

1019

W20005→W10010 attached→Turn-around at station A

1020

W20007→W20028

1021

W20007→Servicing at station B

1022

W10007→Light repair at station B

1023

W10007→W20032

1024

W20011→W20034

1025

W20011→W20034 attached

1026

W20015→Servicing at station B→W10014

1027

W20015→W10016 attached

1028

W20017→W20038 attached

1029

W20017→Servicing at station B

1030

W20018→W10013

1031

W20035→W20016 attached→W20031

• The encoding, fitness function, and stopping rule are the same as those of the Tabu Search algorithm. • Population: The number of individuals is 20, and the initial solution generation algorithm of the Tabu Search algorithm is used to generate 20 initial solutions as the initial population. • Genetic operator: The type of crossover is two-cutting crossover. For two selected chromosomes, two adjacent coding with the same parity are selected randomly to swap. The probability of crossover is 0.9. Mutation is same as neighborhood generation of the Tabu Search algorithm which is the exchange of two adjacent co-coding positions in chromosomes. The probability of mutation is 0.05.

9 Solving a Locomotive Routing Problem of Heavy Haul Railways

131

-22000 -22200 -22400

fitness

-22600 -22800 -23000 -23200 -23400 -23600

TS GA

-23800 0

50

100

150

200

250

300

350

400

450

500

iteration ordinal number

Fig. 9.5 Iterative convergence comparison diagram

Genetic operators use the same search space limit method and feasible solution transformation method as the tabu search algorithm to handle the constraints. After multiple comparison experiments and taking the average value, the comparison of iterative convergence of the two algorithms is shown in Fig. 9.5. It can be seen from the comparative experiments that the algorithm of this paper has a faster convergence speed and higher solution quality. It shows that the algorithm has certain advantages in the optimization performance.

9.6 Conclusion This paper comprehensively considers the influence of the features of heavy haul railway locomotives including multi-locomotive traction, various locomotive types, and unpaired locomotive turn-around, and locomotive servicing and maintenance on locomotive operation. The MIP model whose object is to minimize the non-operation time is built to quantify the problem. Due to the characteristics of a large solution space and complex constraints of the model, the two-stage heuristic algorithm is designed to solve the model successfully. Case analysis shows that the method designed in this paper can solve the LRP of heavy haul railways in a short time, and has a certain reference value for the research of intelligent locomotive dispatching on heavy haul railways. This paper does not consider the situation of overlapping of major and minor locomotive routing which is very common in reality, and it might be the focus of our future research.

132

Y. Li et al.

Acknowledgements This research was supported by the National Natural Science Foundation of China (Project No. 52072314; 52172321; 52102391), China Shenhua Energy Co., Ltd. Science and Technology Program (Project No. CJNY-20-02), Sichuan Science and Technology Program (Project NO. 2020YJ0268; 2020YJ0256; 2022YFH0016; 2021YFQ0001; 2021YFH0175, 2022YFQ0101), Sichuan Science and Technology Program (Project No. 2022020), Key science and technology projects in the transportation industry of the Ministry of Transport (2022-ZD7-13).

References 1. Florian, M., Bushell, G., Ferland, J., Guerin, G., Nastasky, L.: The engine scheduling problem in a railway network. Infor 14, 121–138 (1976) 2. Forbes, M.A., Holt, J.N., Watts, A.M.: Exact solution of locomotive scheduling problems. J. Oper. Res. Soc. 42(10), 825–831 (1991) 3. Cordeau, J.F., Soumis, F., Desrosiers, J.: A benders decomposition approach for the locomotive and car assignment problem. Transp. Sci. 34(2), 133–149 (2000) 4. Ahuja, R.K., Liu, J., Orlin, J.B., Sharma, D., Shughart, L.: Solving real-life locomotive scheduling problems. Transp. Sci. 35(4), 503–517 (2005) 5. Zhang, J., Ni, S.Q., Lv, M.M., Wu, H.G.: Model and algorithms for formulation of multi-type locomotives working diagram based on optimization of locomotive operation costs scheduling. J. China Railw. Soc. 36(10), 1–6 (2014) 6. Wang, W.X., Chen, D.J., Chen, B.Y.: Locomotive working problem based on integrating of GA—ACO. Comput. Simul. 32(3), 183–185 (2015) 7. Vaidyanathan, B., Ahuja, R.K., Orlin, J.B.: The locomotive routing problem. Transp. Sci. 42(4), 492–507 (2008) 8. Wang, L., Ma, J.J., Lin, B.L., Chen, L., Wen, X.H.: Method for optimization of freight locomotive scheduling and routing problem. J. China Railw. Soc. 36(11), 7–15 (2014) 9. Frisch, S., Hungerländer, P., Jellen, A., Primas, B., Steininger, S., Weinberger, D.: Solving a real-world locomotive scheduling problem with maintenance constraints. Transp. Res. Part B: Methodol. 150, 386–409 (2021) 10. Miranda, P.L., Cordeau, J.F., Frejinger, E.: A time–space formulation for the locomo-tive routing problem at the Canadian national railways. Comput. Oper. Res. 139 (2022). 11. Su, R.Y., Zhou, L.S., Tang, J.J.: Locomotive schedule optimization for Da-Qin heavy Haul railway. Math. Probl. Eng. 2015, 1–14 (2015) 12. Zhang, F.Q., Wu, T.-Y., Wang, Y.O., Xiong, R., Ding, G.Y., Mei, P., Liu, L.Y.: Application of quantum genetic optimization of LVQ neural network in smart city traffic network prediction. IEEE Access 8, 104555–104564 (2020) 13. Kang, L.L., Chen, R.-S., Xiong, N.X., Chen, Y.-C., Hu Y.-X. Chen, C.-M.: Selecting hyperparameters of Gaussian process regression based on non-inertial particle swarm optimization in Internet of Thing. IEEE Access 7, 59504–59513 (2019)

Chapter 10

A Study of Optimization of Dynamic Freight Train Diagrams Based on Market-Orientation Meng Wang, Fangyu Shi, Ziqi Dong, and Hongxia Lu

Abstract The dynamic freight train diagram is the key link to realize the wholeprocess integrated transportation of goods and the key to realize the “According to the Timetable”. This paper studies the optimization of freight train diagram based on the basic diagram for dynamic route selection with market orientation as the core. Firstly, considering the quality of freight service and the satisfaction of cargo owners, taking the minimum transportation time consumption of each freight train as the optimization goal, starting from the characteristics of dynamic traffic flow, considering the connection time of traffic flow with the train at the station and the arrival period of the loaded goods, the dynamic route selection optimization model based on the basic diagram of the train diagram is established, and the solution strategy of the improved particle swarm algorithm based on binary coding is designed. Finally, a small road network case is constructed to verify the feasibility of the model and algorithm.

M. Wang · F. Shi · Z. Dong · H. Lu (B) Southwest Jiaotong University, Chengdu 610031, Sichuan, China e-mail: [email protected] F. Shi e-mail: [email protected] Z. Dong e-mail: [email protected] M. Wang China National Energy Group Railway Equipment Company, Beijing 100000, China F. Shi · Z. Dong · H. Lu National and Local Joint Engineering Laboratory of Comprehensive Intelligent Transportation, Southwest Jiaotong University, Chengdu 610031, China National Engineering Laboratory of Integrated Transportation Big Data Application Technology, Chengdu 610031, Sichuan, China © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_10

133

134

M. Wang et al.

10.1 Introduction With the continuous and steady development of Chinese economy, the demand for freight logistics is growing rapidly. Railway freight transportation is playing an increasingly important role in Chinese economic development. For a long time, Chinese railway freight transportation has implemented an organizational transportation organization model, which does not have an effective guiding role in the daily fluctuation of freight transportation demand under the current economic new normal. In order to further improve the adaptability of railway freight transportation and dynamic market, this paper studies the optimization theory and technology of market-oriented dynamic programming railway freight train scheduling, which is of great significance to enhance the competitiveness of railway freight transportation in freight market. Foreign scholars have different emphasis on the direction of optimization of the train diagram, and most of the research focused on the optimization of single-track and double-track train diagram. Caprara et al. [1] constructed a Lagrangian relaxed integer linear programming model based on directed multigraphs, and designed a heuristic algorithm to arrange routes for trains. With the minimum train running time as the optimization objective and transportation resources as the constraint condition, Zhou et al. [2] designed a branch and bound algorithm to optimize the train diagram of single-line section. Li et al. [3] considered the influence of actual freight volume on the freight train diagram, constructed the train route selection model with the constraints of traffic flow connection and freight delivery deadline, and designed the tabu search algorithm to solve it. Reisch et al. [4] designed a simulated annealing method based on Conflict Resolving to compile the annual train diagram of all freight trains in Germany. Chinese research on optimizing the train diagram has made substantial progress. Feng et al. [5] designed the earliest conflict optimization method for the compilation of single-track freight train lines, dividing the train diagram into different stages, and solving in a cyclic method with the constraint of train interval. Based on the background of the alternation of old and new trains, Miaomiao [6] proposed a method of compiling the train diagram based on different car bottoms and a car bottom optimization model, and used the ant colony algorithm in the swarm intelligence optimization algorithm to solve the problem. Hangtag [7] considered the constraints of train priority, train interval and station capacity, set up virtual vehicles and buffer time slots, and designed an improved particle swarm optimization algorithm to optimize the passenger and cargo train diagram. Jian et al. [8] considered the grade weights of different direct freight trains, and constructed a 0–1 programming model based on the line selection of the running line with the constraint of the connection optimization of the technical station. Kanga [9] proposes a non-inertial particle swarm optimization with elite mutation-Gaussian process regression (NIPSO-GPR). Compared with several frequently used algorithms of hyper-parameters optimization on linear and nonlinear time series sample data, experimental results indicate that GPR after hyper-parameters optimized by NIPSO-GPR has better fitting precision

10 A Study of Optimization of Dynamic Freight Train Diagrams Based …

135

and generalization ability. Xiaobing et al. [10] proposed that under the planned transportation organization mode, the arrival period of different goods should be fully considered, and the 0–1 planning model of train line selection was constructed with the goal of minimizing the running time of the train. In the research on the optimization of train diagram compilation, the attributes of the train are the factors that most scholars focus on, while the daily dynamic changes of the transportation market demand, the customer’s personalized freight demand, and the matching relationship between the train flow and the running line are often ignored in the research process. The lack of these key factors is the root cause of the poor quality of the train diagram compilation and the actual transportation demand. Based on this, this paper studies the optimization theory and technology of dynamic programming freight train diagram compilation based on market orientation. The dynamic market demand, delivery time and other factors are taken into account, and the whole process transportation plan from the time of carrying goods by the railway operation department to the time of delivery of goods is designed, so as to achieve the goal of effectively improving the efficiency of freight transportation and the satisfaction of shippers.

10.2 Problem Description Train diagram compilation is a NP-hard problem. Under the existing information technology level, Chinese huge and complex road network conditions make it very difficult to compile a train diagram that fully meets the dynamic traffic flow and market demand. In order to make the results of train diagram compilation in daily work closer to the ideal state, this paper integrates the characteristics of dynamic traffic flow transportation into the basic train diagram, and uses the method of dynamic line selection to determine the operation line of freight trains on the implementation day determined by the dynamic freight train operation plan. Figure 10.1 shows the abstract space–time diagram of train running lines, which is used to represent the relationship between the train running line and the station and time (one implementation day), which is embodied in the two aspects of the train running in the interval and staying at the station. The “train running arc” indicates that the train is running in the interval, that is, the train running line, and determines the train running path and the start and end time; “Train transit arc” indicates the transit connection of the train line at the technical station, that is, a station’s time difference between the arrival point of an arriving operation line to the departure point of a dispatching line. According to the content determined in the implementation of the dynamic freight train operation plan within the day, this paper aims to reduce the time consumption of the whole transportation process of each freight train as much as possible, and on the basis of meeting the requirements of the transit and connecting time of the traffic flow in technical stations, and the transportation conditions such as ensuring the delivery period of the goods and reducing the transportation delay, the optimization model of

136

M. Wang et al.

Fig. 10.1 Abstract space–time diagram of train running lines

the freight train diagram based on the dynamic line selection of the basic diagram is established. This article makes the following assumptions: 1. The train-related information (running route, type of operation, marshalling content, number of operations) determined by the dynamic freight train operation plan is known; 2. The full capacity train operation lines in each operation section are known, and each operation line contains basic information such as additional start-stop time; 3. The running path of the train in the station is not considered; 4. Each technical station has enough technical operation ability.

10.3 Model Construction 10.3.1 A Subsection Sample The symbols and descriptions used to build the model are as follows: 1. Collective description S: station set, s is the station index, s + 1 represents the front station of station s; F: freight train set, which represents all freight trains running between two technical stations, determined by the operation plan, f is the train index;

10 A Study of Optimization of Dynamic Freight Train Diagrams Based …

137

L: the set of running lines, L s represents the set of all running lines issued by station s. 2. Symbol description tis−1,s : the connection time between the running line i issued by station s − 1 and the j running line j issued by station s, which is expressed as the connection time between the above two running lines at station s; tris : the running time of the running line i issued by station s in the front section; s : the departure time of the running line i from station s; tdi tais−1,s : the arrival time of the running line i from station s − 1 to the station s in front; tw : the minimum connection time standard of freight train line in technical station without transfer; T flim : due date of freight train f . 3. Decision variables Set two sets of 0–1 decision variables x sf i and yfjs+1 , x sf i represents whether the train f chooses the running line i from the technical station s; yfjs+1 indicates that when x sf i is 1, whether train f chooses the running line j from the technical station s + 1 is defined as follows:  1, train f selects line i f r om technical station s s xfi = , ∀i, j ∈ S 0, other s  1, train f selects line i f r om technical station s + 1 s+1 yfj = , ∀i, j ∈ S 0, other s

10.3.2 Objective Function As a transportation product displayed by the railway operation department, the freight train diagram should fully consider the quality of freight service and the satisfaction of shippers. Therefore, the dynamic freight train line selection model constructed in this paper takes the minimum transportation time consumption of each freight train as the optimization goal. In the train diagram, the transportation time of the freight train includes the running time of the train in the interval and the transit staying time of the train at the station. Based on this, the objective function is expressed as follows:        s,s+1 x sf i y s+1 t + x sf i tris minZ = f j ij f ∈F

s∈S

i∈L s

j∈L s+1

f ∈F

s∈S

i∈L s

(10.1)

138

M. Wang et al.

10.3.3 Constraint Conditions 1. Operation line uniqueness constraint For a freight train f that operates within a day, if and only one operating line can be selected from all operating lines dispatched by station s, and station s+1 in front of station s should also follow this constraint, which is expressed as follows: 

 f ∈F

s∈S



 f ∈F

 

s∈S

x sf i = 1

(10.2)

y s+1 fj = 1

(10.3)

i∈L s

j∈L s

2. Connection constraint of operation lines The decision variables x sf i and y s+1 f j should select the same running line in all running lines dispatched by the same station, so as to ensure that the departure operation line selected by the train f at the technical station s is the same as the arrival operation line selected at the front station s + 1, and ensure the normal connection of the train f. s+1 ∀i = j, f ∈ F, s ∈ S, i, j ∈ L s x s+1 f i = yfj

(10.4)

3. Transfer time constraint of technical station The time required for the train to carry out various technical operations at the technical station can be expressed as the connection between the arrival of the station s and the departure of the operation line. Therefore, the stop time of the train at the technical station needs to meet the minimum time standard required for the technical operation without transfer.  t s − t s−1,s tds j − tais−1,s ≥ tw s−1,s = s d j s−1,sai s ∈ S, i ∈ L s−1 , j ∈ L s (10.5) ti j td j − tai + 1440 tds j − tais−1,s < tw 4. Freight transit period constraint The transportation time of each freight train, including the time of transportation on the way and the time of stay at the station, needs to meet the requirements of the freight transit period. s,s+1 + x sf i tris ≤ T flim f ∈ F, s ∈ S, i, j ∈ L s x sf i y s+1 f j ti j

5. Decision variable constraints Equation (10.7) represents the 0–1 integer constraint on decision variables.

(10.6)

10 A Study of Optimization of Dynamic Freight Train Diagrams Based …

139

x sf i , yfjs+1 ∈ {0, 1} f ∈ F, s ∈ S, i ∈ L s , j ∈ L s+1

(10.7)

10.4 Algorithm Design 10.4.1 Coding Strategy of Dynamic Line Selection Model Based on BPSO The dynamic line selection optimization model constructed in this paper is a nonlinear 0–1 integer programming problem, which is essentially to select the appropriate line for freight trains. Considering the independence of each train and each operation line, this paper uses the binary particle swarm optimization algorithm (BPSO) to encode the dynamic line selection optimization problem by binary coding. Let the particle swarm size be M, m be the particle index in the population, f be the train index, s be the station index, and i be the index of the operation line dispatched by station s. The dimension of each particle m is N f × Ns × Ni , denoted by R, where N f is the number of trains, Ns is the number of stations, and Ni is the number of lines from station s. The position X m of the particle m representing the dynamic line selection is expressed as follows: 

  X m = xm,1,1,1 , . . . , xm, f,s,i , . . . , xm,N f ,Ns ,Ni xm, f,s,i ∈ {0, 1}

(10.8)

In the formula, xm, f,s,i —if the train f of the particle m selects the operation line i from station s, the value is 1, otherwise it is 0. Similarly, the velocity Vm of particle m can be expressed as:   Vm = vm,1,1,1 , . . . , vm, f,s,i , . . . , vm,N f ,Ns ,Ni , vm, f,s,i ≤ Vmax

(10.9)

In the formula, vm, f,s,i —the velocity of particle m in the corresponding dimension, and should not exceed the limit of the maximum velocity Vmax . Correspondingly, the best position that each particle m has reached in  the particle swarm is expressed as Pm = pm,1,1,1 , . . . , pm, f,s,i , . . . , pm,N f ,Ns ,Ni , and the best have experienced is expressed as   position that all particles in the swarm Pg = pg,1,1,1 , . . . , pg, f,s,i , . . . , pg,N f ,Ns ,Ni . In the iteration g of the particle, the velocity update formula of each particle m in the r + 1 dimension (1 ≤ r + 1 ≤ R) is as follows:   r g,r +1 g,r r vm, f,s,i = wg · vm, f,s,i + c1 · rand() · pm, f,s,i − x m, f,s,i   r + c2 · rand() · prg, f,s,i − xm, f,s,i

(10.10)

140

M. Wang et al.

Fig. 10.2 Image schema of Sigmoid function

For the position of each particle m, since its value is 0 or 1, the Sigmoid function is introduced to ensure that the value of each particle m is limited to 0 or 1. The calculation formula of Sigmoid function used in this algorithm is as follows: Sigmoid(x) =

e x − e−x e x + e−x

(10.11)

The function image is as follows (Fig. 10.2): The position update formula for each particle m is expressed as follows:  r +1 xm, f,s,i

=

r +1 1, ξ < Sigmoid(vm, f,s,i ) 0, other s

(10.12)

In the formula, ξ is a random number with a range of [0,1]. For inertia weight w, in addition to setting the inertia weight as a constant, there are generally three adjustment strategies for inertia weight coefficients: linear adjustment strategy [11, 12], fuzzy adaptive adjustment strategy [13] and random inertia weight strategy [14]. Among them, the linear adjustment strategy is effective in optimizing static problems; the fuzzy adaptive adjustment strategy needs to use fuzzy rules to dynamically adjust the inertia weight. The rules are derived from a large number of weight selection experiences, and there are many limitations in practical applications. The random inertia weight strategy can maintain the vitality of particle exploration, and to a certain extent, and it can avoid the particle swarm falling into the local optimal solution prematurely, which is suitable for solving dynamic nonlinear problems.

10 A Study of Optimization of Dynamic Freight Train Diagrams Based …

141

Considering that the dynamic line selection optimization model constructed in this paper is a dynamic nonlinear problem, the random inertia weight strategy is adopted. The update formula of the inertia weight is as follows: wg = 0.5 +

rand( ) 2

(10.13)

wg —the value of inertia weight w in the iteration g of the population; rand( )—a uniformly distributed random number with a value interval of [0,1].

10.4.2 Particle Swarm Algorithm Process of Dynamic Line Selection Based on the coding strategy of the dynamic line selection model based on BPSO in Sect. 10.4.1, the optimization calculation is carried out according to the iterative steps of the particle swarm optimization algorithm, and the operation lines of each freight train on the implementation day are selected and optimized to determine the dynamic freight train diagram on the implementation day. The basic process of the algorithm is as follows: Step 1: Initialize the population parameters, set the size of the population particles, and set the values of the parameters w, c1 , c2 , Vmax , G max and random numbers; 1,r 1,r Step 2: Initialize the population size and the position xm, f,s,i and velocity vm, f,s,i of each particle m in the population; Step 3: According to the initial state of the particle swarm, set the historical best position Pm of each particle m and the historical best position Pg experienced by all particles in the group. Step 4: The inertia weight w is updated according to the formula (10.13), and the velocity and position of the particle m are iterated according to the velocity and position update formula (10.10) and (10.12) respectively. Step 5: The f itness(X m ) of each particle m is calculated, that is, the sum of the transit time of each freight train; Step 6: Comparing the fitness of particle m with its historical best position Pm and global historical best position Pg , if it is better than the historical or global historical best position, the value of the best position is updated; Step 7: If the maximum number of iterations G max or criteria is reached, the final scheme is output, otherwise go to Step 4.

10.5 Case Analysis A small section of the road network is constructed as shown in Fig. 10.3 below, with A-E 7 technical stations, A to E direction is specified as the downward direction, in

142

M. Wang et al.

Fig. 10.3 Small section road network diagram

which the intermediate station is ignored. Assume a basic train diagram based on historical freight data and plans. Assuming that on a certain implementation date, the dynamic freight train operation plan determines the operation of 10 pairs of freight trains, of which the up and down lines are 10 freight trains respectively. The minimum connection time standard for the given freight train operation line without transfer at the technical station is 30 min. The partial operation lines of each station are shown in Tables 10.1 and 10.2. The coding method and algorithm flow of the solution strategy based on particle swarm optimization algorithm in Sect. 10.2 are programmed on MATLAB commercial mathematical software. The particle swarm size is 120, the inertia weight is updated according to the formula, the acceleration constant and the value are 2, and the upper limit of the number of iterations is 100. The selection results of the running line of each freight train are shown in Tables 10.3 and 10.4. According to the line selection scheme, on the 61 operating lines included in the basic train diagram, where the operating lines are selected for 20 freight trains, Table 10.1 Operation line information dispatched downward to stations (Part) Station

Dispatch operation line number

Departure time

Station

Dispatch operation line number

Departure time

A

1

18:47

D

1

18:52

A

2

23:27

D

2

20:52

A

3

3:12

D

3

4:12

A

4

6:42

D

4

7:47

A

5

14:47

D

5

12:00

B

1

18:49

E

1

21:37

B

2

20:49

E

2

2:57

B

3

0:09

E

3

6:52

B

4

5:39

E

4

14:15

B

5

14:49

E

5

17:17

C

1

19:01

F

1

20:07

C

2

21:41

F

2

22:47

C

3

3:41

F

3

4:07

C

4

12:00

F

4

14:35

C

5

17:01

F

5

18:27

10 A Study of Optimization of Dynamic Freight Train Diagrams Based …

143

Table 10.2 Operation line information from stations upward (Part) Station

Dispatch operation line number

Departure time

Station

Dispatch operation line number

Departure time

G

1

19:15

D

1

18:56

G

2

21:55

D

2

22:24

G

3

2:35

D

3

1:04

G

4

8:16

D

4

6:24

G

5

15:45

D

5

14:35

F

1

19:09

C

1

21:37

F

2

22:29

C

2

2:57

F

3

2:29

C

3

6:52

F

4

5:49

C

4

14:15

F

5

15:08

C

5

17:17

E

1

19:39

B

1

20:07

E

2

22:19

B

2

22:47

E

3

2:59

B

3

4:07

E

4

6:19

B

4

14:35

E

5

15:37

B

5

18:27

Table 10.3 Table captions should be placed above the tables Serial

Train number

Operation line number Station A

Station B

Station C

Station D

Station E

Station F

1

22,001

19

20

22

24

25

27

2

22,003

13

14

16

18

19

21

3

22,005

22

23

25

27

28

30

4

22,007

1

2

4

6

7

9

5

22,009

23

24

26

28

29

31

6

22,011

29

30

32

2

3

5

7

22,013

31

32

2

4

5

7

8

22,015

14

15

17

19

20

22

9

22,017

2

3

5

7

8

10

10

22,019

11

12

14

16

17

19

they account for 32.8%. The 20 freight trains are through freight trains. There is no traffic reorganization on the way, and the transit time at each technical station meets the technical time standard of the minimum unscheduled transfer operation. The comparison between the travel time of each freight train on the implementation date and the corresponding time window of freight delivery deadline is shown in Fig. 10.4. It can be seen that each freight train can complete the transportation task

144

M. Wang et al.

Table 10.4 Selection scheme of upstream operation line Serial

Train number

Operation line number Station A

Station B

Station C

Station D

Station E

Station F

1

22,002

17

19

21

23

25

27

2

22,004

8

10

12

14

16

18

3

22,006

7

9

11

13

15

17

4

22,008

9

11

13

15

17

19

5

22,010

3

5

7

9

11

13

6

22,012

26

28

1

3

5

7

7

22,014

27

29

2

4

6

8

8

22,016

25

27

29

2

4

6

9

22,018

5

7

9

11

13

15

10

22,020

18

20

22

24

26

28

Fig. 10.4 Comparison of train travel time and arrival time window

within the specified delivery deadline, which is of great practical significance for improving the quality of railway freight service and market competitiveness.

10.6 Conclusion The dynamic freight train diagram is the key link to realize the integrated transportation of goods in the whole process, and is the fundamental document to realize “Riving According to the Diagram”. This paper takes the characteristics of dynamic

10 A Study of Optimization of Dynamic Freight Train Diagrams Based …

145

traffic flow as the starting point, fully considers the connection time of the traffic flow with the train in the station and the delivery period of the loaded goods, considers the quality of freight service and the satisfaction of the shipper, and takes the minimum consumption of the transit time of each freight train as the optimization goal. The dynamic route selection optimization model based on the basic diagram of the train diagram is constructed, and the solution strategy based on the binary particle swarm optimization algorithm is designed according to the characteristics of the model. Finally, a small road network case is designed to verify the model algorithm. Acknowledgements The authors would like to thank Prof. Shaoquan Ni and Prof. Dingjun Chen for their constructive comments. This research was supported by the National Natural Science Foundation of China (Project No. 52072314; 52172321; 52102391), China Shenhua Energy Co., Ltd. Science and Technology Program (Project No.: CJNY-20-02) Sichuan Science and Technology Program (Project No. 2020YJ0268; 2020YJ0256; 2022YFH0016; 2021YFQ0001; 2021YFH0175, 2022020; 2022YFQ0101). Key sceince and technology projects in the transportation industry of the Ministry of Transport (Project No.: 2022ZD7132).

References 1. Caprara, A., Fischetti, M., Toth, P.: Modeling and solving the train timetabling problem. Oper. Res. 50(5), 851–861 (2002) 2. Zhou, X., Zhong, M.: Single-track train timetabling with guaranteed optimality: branch-andbound algorithms with enhanced lower bounds. Transp. Res. Part B: Methodol. 41(3), 320–341 (2007) 3. Li, S., Lv, H., Xu, C., et al.: Optimized train path selection method for daily freight train scheduling. IEEE Access 8, 40777–40790 (2020) 4. Reisch, J., Großmann, P., Pöhle, D., et al.: Conflict resolving–A local search algorithm for solving large scale conflict graphs in freight railway timetabling. Eur. J. Oper. Res. 293(3), 1143–1154 (2021) 5. Shi, F., Li, X.H., Qin, J., Deng, L.B., et al.: A timing-cycle iterative optimizing method for drawing single-track railway train diagrams. J. China Railw. Soc. 27(1), 1–5 (2005) 6. Liu, M.M.: Optimization theory and method of compilation train operation diagram alternation periods between the current train diagram and the new one. Southwest Jiaotong University Doctor Degree Dissertation (2013) 7. Zhao, H.T.: Research on Optimized Establishment Method and Evaluation Method for Train Operation Plan Based on Alternative Graph Theory. Traffic Information Engineering and Control of China Railway Research Institute (2014) 8. Li, J., Lin, B.L., Tian, Y.M., et al.: Model and algorithm for optimizing train paths assignment of freight through train. J. Beijing Jiaotong Univ. 39(6), 15–20 (2015) 9. Kang, L., Chen, R.-S., Xiong, N., Chen, Y.-C., Hu, Y.-X., Chen, C.-M.: Selecting hyperparameters of gaussian process regression based on non-inertial particle swarm optimization in Internet of Things. IEEE Access 7, 59504–59513 (2019) 10. Zhang, X.B., Li, C.D., Liu, H.X., et al.: Study on optimization of train path selection based on freight due date. J. China Railw. Soc. 41(5), 10–15 (2019) 11. Kennedy, J., Eberhart, R.: Particle swarm optimization. IEEE (1995) 12. Wang, J.W., Wang, D.W.: Experiments and analysis on inertia weight in particle swarm optimization. J. Syst. Eng. 2, 194–198 (2005)

146

M. Wang et al.

13. Shi, Y., Eberhart, R.C.: Fuzzy adaptive particle swarm optimization. IEEE (2001) 14. Eberhart, R.C., Shi, Y.: Tracking and optimizing dynamic systems with particle swarms. IEEE (2001)

Chapter 11

Research on Equipment Management System of Railway Passenger Station Based on High-Precision Positioning Lexi Li, Bozhou Wang, Zhen Liu, and Shaoquan Ni

Abstract The equipment management informatization and intelligentization in railway passenger stations is a significant component of the construction of intelligent stations. The current equipment management of railway passenger stations is slowly developing and backward in its approach. This paper studied the weaknesses of equipment management in railway passenger stations, conducted demand analysis, and proposed the development direction of equipment management in railway passenger stations with the support of theoretical basis and technical basis. The technical architecture of the high-precision positioning-based railway passenger station equipment management system was built from the terminal layer, application layer, service layer, data layer and perception layer, and four functional modules were designed for equipment management: equipment real-time dynamic trackment, full life-cycle management, equipment fault responsibility traceability, and incident response. The key technologies required for the construction of the system were studied, including Bayesian fusion-based equipment positioning technology with high accuracy and neural network-based remaining life prediction technology.

L. Li · B. Wang · Z. Liu · S. Ni (B) School of Transportation and Logistics, Southwest Jiaotong University, Chengdu 610031, Sichuan, China e-mail: [email protected] L. Li e-mail: [email protected] B. Wang e-mail: [email protected] Z. Liu e-mail: [email protected] S. Ni National and Local Joint Engineering Laboratory of Comprehensive Intelligent Transportation, Southwest Jiaotong University, Chengdu 610031, China National Engineering Laboratory of Integrated Transportation Big Data Application Technology, Chengdu 610031, China © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_11

147

148

L. Li et al.

11.1 Introduction The railway passenger station is the link between the railway passenger transport department and the passengers. It has a complex structure and numerous equipment. However, the existing equipment management is mostly manual, with backward methods, low efficiency, and high security risks. Although the existing railway passenger station has introduced many advanced automation systems to improve the management level, such as monitoring systems, the decision-making, and processing parts of equipment management are still the responsibility of managers. Aiming at the weak parts of fixed and mobile equipment management in railway passenger station, this paper analyzes the management requirements and combines advanced positioning technology to study a railway passenger station equipment management system based on high precision positioning to improve the efficiency of passenger station equipment management.

11.2 Analysis of Equipment Management Requirements of Railway Passenger Station Railway passenger station equipment management refers to ensure that the equipment is in normal working condition by obtaining the real-time conditions of various equipment in the station, so as to avoid safety accidents caused by equipment failures. The equipment management requirements of railway passenger station are as follows: (1) Fixed equipment management There are a lot of large fixed equipment in railway passenger station, such as escalator, straight ladder, security inspection machine, automatic ticket machine, ticket gate, and so on. Passengers often conduct business at such facilities. Due to the large number of passengers in the station, once such equipment fails, it may delay the passenger’s check-in time, and at worst, the safety of passengers may be threatened due to equipment failure (such as escalators and elevators). Therefore, monitoring the working status of such equipment and ensuring its normal operation is the key to maintaining the safety and order in the station. (2) Mobile equipment management In the railway passenger station, there are a large number of loading and unloading trucks, trolleys, small mobile equipment shuttling across the platform. Such equipment has its own fixed path. Once the equipment deviates from the path and crosses the flow lines of passengers or operators, it is easy to cause accidents. Therefore, the location and operation status of such equipment in the passenger station are located and managed in real time.

11 Research on Equipment Management System of Railway Passenger …

149

(3) Operation equipment management In addition to the above two types of equipment, the railway passenger station also has some anti-slip equipment (such as iron shoes, etc.), maintenance tools, sewage suction equipment, and other operating equipment. Such equipment must be kept in good condition during operation and placed in the correct position, otherwise it will affect the normal operation of railway passenger station. Therefore, it is necessary to obtain its position and status during its work to avoid loss and damage, ensure the orderly operation in the station, and improve the efficiency of equipment management. At present, the equipment management of most railway passenger stations has some problems such as backward management mode, low efficiency, low degree of informatization and intelligence, management blind corners, and lacks scientific and effective management methods. Therefore, it is of great significance to develop a modern railway passenger station equipment management system by studying the equipment positioning technology in the station, combining data mining, electronic fences, deep learning, and other advanced technologies to promote the informatization and intelligent development of railway passenger station.

11.3 Research on Equipment Management Development Direction in Railway Passenger Stations 11.3.1 Full Lifecycle Management Theory of Equipment The basic idea of full life cycle management is to extend the scope of equipment management to the research, design, manufacture, procurement, installation and commissioning, use, maintenance, transformation, until the scrapping of equipment. It focuses on a series of processes from early design and production to later recycling of equipment, and realizes comprehensive management of equipment. Its basic purpose is to overall guarantee and improve the reliability, maintainability, and economy of the equipment through the whole process management. The full life cycle management is the development direction of modern equipment management, combining with the actual situation of equipment management in Chinese railway passenger station, and considering the scene of equipment management is railway passenger station, the full life cycle management scope of equipment in the station should be use, repair, maintenance, and scrap these four stages of management. The full life cycle management can be achieved by establishing electronic database of equipment management resources, dynamic tracking management of equipment, traceability and visualization of equipment, and establishing prediction model of remaining service life.

150

L. Li et al.

11.3.2 High-Precision Equipment Positioning Technology The development of positioning technology has greatly promoted the improvement of scientific management level. The earliest positioning technology uses GPS positioning technology to locate targets in outdoor scenes [1]. With the continuous development of technology, the existing emerging positioning technologies include radio frequency tag (RFID), Bluetooth (BT), ultra-wide band (UWB), inertial navigation (IMU), etc., and the positioning scene is also expanded from outdoor to indoor [1]. However, the positioning accuracy of single positioning technology is not high, and its application in complex indoor environment is not much [2]. In order to further improve the accuracy of indoor positioning, scholars have begun to study the fusion of two or more positioning technologies, such as UWB and IMU fusion [3–6], Wi-Fi and IMU fusion [7–9], Wi-Fi and BT fusion [10, 11], etc. Compared with single positioning technology, the accuracy of fusion positioning technology has been significantly improved. However, in the face of complex indoor activities, there are still problems such as low accuracy and poor real-time performance, and the application scope is limited. With the vigorous development of 5G technology, scholars have begun to study 5G high-precision fusion positioning technology [1, 12], which has good real-time performance and positioning accuracy up to centimeter level. The positioning technology with low delay and high precision provides more possibilities for the equipment management of railway passenger station. The comparison of various positioning methods is shown in Table 11.1. Table 11.1 Comparison of various positioning methods Positioning way

Positioning technology

Features

Single positioning

GPS, Wi-Fi, BT, UWB, IMU etc. High accuracy for outdoor scenes, low accuracy for indoor scenes, with a delay

Fusion positioning

UWB and IMU fusion, Wi-Fi and Compared with single IMU fusion, Wi-Fi and BT positioning the accuracy has fusion, other fusion methods been improved, not suitable for complex indoor scenes, with higher latency

5G high-precision fusion positioning

5G + multiple positioning technologies

5G assisted transmission, high precision, fast speed, large transmission volume, good real-time performance

11 Research on Equipment Management System of Railway Passenger …

151

11.3.3 Railway Passenger Station Equipment Management Development Direction Based on the theory of equipment full-cycle management and high-precision positioning technology, the development of railway passenger station equipment management can be considered from the following directions: (1) Adopt high-precision indoor positioning technology, real-time monitoring of all station equipment, obtain the precise coordinates of any positioning target at any time, and realize the dynamic tracking management of target; (2) Assign a unique code to the mobile equipment, so that the whole process of the equipment from taking out, moving, working, and putting it back to the location of placement matches the path of the staff who use the equipment, so as to facilitate the accountability for subsequent equipment loss, damage and other accidents; (3) Establish a railway passenger station equipment management database to store information such as the time of use, service times, overhaul times and maintenance times, etc., so as to facilitate the call, statistics and analysis at any time, and realize the full life cycle management of equipment; (4) Monitor the state of the equipment in real time. If the equipment fails and is not in the correct working position, the system automatically sends an alarm to the manager to quickly respond to the accident.

11.4 Design of Railway Passenger Station Equipment Management System Based on High-Precision Positioning Based on the demand and development direction of railway passenger station equipment management, combined with high-precision positioning technology, research on high-precision positioning-based railway passenger station equipment management system, using information technology to achieve the tracking of the whole process of fixed, mobile and operational equipment in railway passenger stations from, use, repair, maintenance and disposal, to enhance the modernization of railway passenger station equipment management.

11.4.1 System Target Based on high-precision positioning technology, the station equipment is precisely positioned in real time. On this basis, it is combined with Internet of Things technology, big data technology, deep learning technology and 5G technology to achieve

152

L. Li et al.

equipment real-time dynamic trackment, full lifecycle management, equipment fault responsibility tracing, and incident response.

11.4.2 Architecture Design The equipment management system of railway passenger station based on highprecision positioning consists of five layers: perception layer, data layer, service layer, application layer, and terminal layer. The technical architecture is shown in Fig. 11.1. The perception layer contains sensors and chips for positioning inside mobile and operational equipment, such as accelerometers, magnetic sensors, Wi-Fi signal transmitters, and Bluetooth sensors for positioning; components of circuit systems and network systems for monitoring the fixed equipment status; and cameras in key areas such as entrances, waiting halls, exit lanes, and ticket halls. The perception layer collects various basic data for sensing the location, equipment status, and video data to assist in identification, verification and decision-making. The data layer is used for storage of station electronic map data; equipment positioning and status base data; video data; equipment use, repair, maintenance, endof-life data and other full-cycle management data; operator use, maintenance and management equipment data, etc. The data in this layer can be called up anytime by the service layer for statistics, analysis, and model construction. The service layer is the core of the system, and the main service is data analysis service. Combining the basic positioning data and auxiliary video data provided by the data layer, through Bayesian, Kalman and other algorithms, calculations to obtain high-precision positioning coordinates of the target equipment; using data from all stages of the whole life cycle, such as equipment use, repair, maintenance and end-oflife, the use of neural networks and other algorithms to build models such as remaining life prediction, maintenance time prediction, and equipment health analysis. In addition, there are Web services using B/S architecture, providing standardised interfaces to interface with external systems for easy data invocation, and the service layer also contains other location service, clock service, early warning service, etc. The application layer can provide support, treatment, and feedback to requests received by the terminal layer interface based on a range of functions provided by the service layer. The location-based railway passenger station equipment management system can provide four types of functions: equipment real-time dynamic trackment, full lifecycle management, equipment fault responsibility traceability, and incident response. The terminal layer intends to achieve information interaction with the application layer, results display and decision treatment. It includes computer equipment terminals for managers, large screens in stations and receiving terminals for operators. Operators receive all instructions issued by the application layer. Through the terminal, the manager can obtain the status and location of the equipment in real time and make management decisions.

11 Research on Equipment Management System of Railway Passenger …

153

Fig. 11.1 System architecture figure

11.4.3 Function Design (1) Equipment Real-time Dynamic Trackment Module The module obtains real-time equipment location coordinates and working status at the station. Fixed equipment such as lifts, security checkers, and ticket machines can be judged by circuit and network information to determine whether they work

154

L. Li et al.

normally, while movable equipment such as loading and unloading vehicles and antiskid apparatus can be dynamically tracked by using various types of sensors on the equipment to collect their positioning information and determine whether they are in the correct working position. (2) Full Lifecycle Management Module The module collects, stores, and statistically analyses data generated by the production cycle of equipment through advanced algorithms such as big data, deep learning, and neural networks. A scientific and reasonable model is constructed to predict faults, analyse health and calculate life expectancy of equipment, providing a basis for subsequent equipment repair, maintenance, scrapping and renewal, realising full lifecycle management of equipment, and effectively avoiding safety hazards. (3) Equipment Fault Responsibility Traceability Module The module can be directly bound to the information of equipment users and maintenance personnel through equipment positioning data and equipment information, thus avoiding tedious links such as manual records and inspections. In addition, if the equipment is damaged or lost, it can be traced back to the relevant operator, which facilitates the implementation of responsibility and improves management efficiency. (4) Incident Response Module The module incorporates the results of the equipment’s operating status, if the equipment is not in normal working order, is not in the correct working position or deviates from the safe path of travel, the manager immediately receives a security alert command from the system. In addition, the system uses electronic fencing technology to circle the incident area and prevent other people from stepping in. The system automatically docks with maintenance personnel for faulty equipment and, in serious cases, firefighters and medical personnel, facilitating the fastest possible follow-up work, and reducing losses at passenger terminals.

11.4.4 Key Technology Realisation 11.4.4.1

High Precision Equipment Positioning Technology

A fusion of multiple position techniques is applied to position the device with high accuracy. Table 11.2 lists the accuracy, advantages, and disadvantages of several common position technologies. Considering the accuracy of device positioning, stability, and the cost of infrastructure deployment, as well as to improve the impact of error accumulation, it is proposed to use Bluetooth and inertial navigation technology for fusion and rely on the powerful transmission and decomputing capabilities of 5G technology, to address the problems of poor collaboration and insufficient self-adaptation in traditional multi-system fusion positioning, based on Bayesian theory, fusion positioning

11 Research on Equipment Management System of Railway Passenger …

155

Table 11.2 Comparison of various positioning techniques Positioning technology

Precision

Advantages

Disadvantages

UWB

Minute degree

High precision

High cost, technical immaturity, signal susceptible to interference

IMU

Minute degree

High precision, signal stability

Error accumulation

BT

Submeter degree

Low deployment costs, technical maturity

Low precision

WiFi

Meter degree

Low deployment costs, technical maturity

Low precision, signal susceptible to interference

GPS

Meter degree

Technical maturity

Large positioning errors in indoor environments

coordinate calculation, based on Bayesian The fusion algorithm consists of four core steps [13]: mixing\interaction, filtering, updating, and fusion output. The algorithm flow is as follows. (1) Collaborative system interaction Taking the Bluetooth positioning system as an example, the system establishes the state equation and the observation equation based on the movement pattern of the device at the station. State equation: x(k + 1) = f [k, x(k)] + G(k)ω(k)

(11.1)

Observation equation: Z (k) = h[k, x(k)] + ν(k)

(11.2)

f [k, x(k)] is the state transfer function, G(k) is the noise driven matrix, and ω(k), ν(k) are both Gaussian white noise. Initial mixing probability: μi| j (k|k + 1) =

pi j μi (k) cj

(11.3)

i is the INS system, k represents the moment, c j is the normalization constant, pi j is the system transfer probability, and μi (k) is the posterior probability of system i at k.

156

L. Li et al.

Mixed observation information: 0

Z j (k + 1|k + 1) = Z i (k + 1)μi| j (k|k + 1)

(11.4)

Z i is the inertial navigation and positioning system observation equation. Mixed noise variance: R 0j (k + 1|k + 1) = μi| j (k|k + 1)   Ri (k + 1) + [Z i (k + 1) − Z¯ 0j (k + 1|k + 1)][Z i (k + 1) − Z¯ 0j (k + 1|k + 1)]T (11.5) (2) Multi-system parallel filtering The mixed observation information and the mixed variance obtained in (11.1) are used as input and filtered using the extended Kalman filter for calculation. The difference between the model output frequency measurement and the real measurement is obtained. 0

ν j (k + 1) = Z j (k + 1) − h[x(k + 1)|k]

(11.6)

(3) System probability update The posterior probability μ j (k + 1) of the system j at the moment k + 1 is used for updating, and the likelihood functions and the weights occupied by the syste ms are calculated for the Bluetooth positioning system and the inertial navigation positioning system. The Bluetooth system is used as an example. Systematic probabilities:   μ j (k + 1) = P ρ j (k + 1)|X k+1

(11.7)

Systematic likelihood function:    j (k + 1) = P Z j (k + 1)|ρi (k + 1), X k

(11.8)

(4) System integration output At the output side, the fusion of device positioning results is carried out according to the weights occupied by the Bluetooth system and the inertial navigation system, combined with Bayesian theory. x(k + 1|k + 1) = x i (k + 1|k + 1)μi (k + 1) + x j (k + 1|k + 1)μ j

(11.9)

The flowchart is shown in Fig. 11.2. With 5G-based Bluetooth and inertial navigation fusion positioning technology, the output is able to meet the position requirements of railway passenger station equipment management.

11 Research on Equipment Management System of Railway Passenger …

157

Fig. 11.2 Bayesian-based fusion localization algorithm flowchart

11.4.4.2

Remaining Useful Life Projection

There are many equipment in railway passenger stations, and the amount of data related to the use, repair, maintenance, and end-of-life of the equipment that can be collected is huge. Neural network algorithms have good self-learning capability [14, 15], and this system constructs the equipment remaining life prediction model based on neural network with the following process. (1) The historical data of the equipment is counted, and the training data O = [o1 , o2 , . . . , on ] is composed of the equipment model (price, class, size, etc.), the number of hours of use, the number of repairs, and the degree of wear and tear (divided into classes for determination) as the input layer m neurons of the neural network, the number of hidden layer i neurons is set to 10, and the output layer k neurons represent the predicted remaining life parameters (hours, number of uses, etc.). (2) Take the input layer-hidden layer as an example. Set the input layer to hidden layer weight vector W = [w1 , w2 , . . . , wn ], initially set w1 = w2 = · · · = wn , and calculate the input value of a single neuron in the hidden layer.

158

L. Li et al.

NEin =

n 

oi wi

(11.10)

i=1

Transforming the input of a single neuron in the hidden layer into an output using the Sigmoid function. NEout = sigmoid(NEout )

(11.11)

Same for hidden layer—output layer. (3) The weights are updated using the gradient descent method. Firstly, the hidden layer—output layer connection weight matrix Wi,k is updated and the error between the output predicted remaining life parameter value and the actual remaining life parameter value is E k . The formula for updating the weight matrix is as follows. 

Wi,k = Wi,k + W i,k

(11.12)

W i,k is the updated value of the weights, The calculation formula is as follows. W i,k = α E i Ok (1 − Ok )Oi T

(11.13)

E i is the error vector between the actual and predicted values of the hidden layer, Ok is the output vector of the output layer, Oi is the output vector of the hidden layer and α is the learning rate to adjust the intensity of the weight change. The same applies to the input layer-hidden layer weights update. (4) Set the error threshold β. If the error of the model prediction result is greater than β, repeat iterations to update the weights, if the error is less than β, the model meets the requirements and training is completed. (5) The model calculates the values of the parameters related to the remaining life of the device by entering the values of the parameters related to the device to be predicted for testing. The flow chart is shown in Fig. 11.3.

11.5 Conclusion The continuous development of positioning technology has provided many ideas for the development of railway passenger stations equipment safety management. Combining the weaknesses of the existing equipment management in railway passenger stations to analyze the requirements, under the support of modern equipment management theories and advanced technologies, put forward the development

11 Research on Equipment Management System of Railway Passenger …

159

Fig. 11.3 A neural network-based residual life prediction model flowchart

direction and system objectives of railway passenger station equipment management. The multi-level system architecture integrating physical perception, data and application is developed, and the function design is carried out according to the requirements. The system can effectively improve the existing passenger station equipment management inefficiency, backward means, the existence of safety risks and other problems, and improve the level of intelligent equipment management. Acknowledgements This research was supported by the Science and Technology Plan of Sichuan province (Project NO. 2020YJ0268) and key science and technology projects in the transportation industry of the Ministry of Transport (2022-ZD7-132).

160

L. Li et al.

References 1. Chen, S., Wang, H., Chen, D.: 5G oriented high-precision fusion positioning architecture and key technologies. ZTE Technol. J. 1–9 (2018) 2. Yan, D.: Review of development status of indoor location technology in China. J. Navig. Position. 7(4) (2019) 3. Li, P., Li, X., Wang, R., et al.: A long short term memory (LSTM) indoor positioning algorithm based on fusion of UWB and inertial navigation. Telecommun. Eng. 61(02), 172–178 (2021) 4. You, W., Li, F., Liao, L., et al.: Data fusion of UWB and IMU based on unscented Kalman filter for indoor localization of quadrotor UAV. IEEE Access 8, 64971–64981 (2020) 5. Feng, D., Wang, C., He, C., et al.: Kalman filter based integration of IMU and UWB for high-accuracy indoor positioning and navigation. IEEE Internet Things J. PP(99), 1–1 (2021) 6. Kuang, B., Chen, F., Tian, C., et al.: Unit fusion positioning algorithm of ultra-wideband and inertial measurement based on improved particle filtering. Sci. Technol. Eng. 20(30), 12460–12466 (2020) 7. Chen, Z., Zou, H., Jiang, H., et al.: Fusion of WiFi, smartphone sensors and landmarks using the Kalman filter for indoor localization. Sensors (Basel) 15(1), 715–732 (2015) 8. Shen, H., Li, S.: Underground positioning method based on WiFi and PDR fusion. Coal Mine Mach. 41(10), 202–204 (2020) 9. Yu, C., Cheng, K.: WiFi and pedestrian dead reckoning adaptive unscented Kalman filter fusion positioning algorithm. Sci. Technol. Eng. 20(27), 11155–11160 (2020) 10. Wu, T., Wang, G., Dai, G., et al.: Research on weighted location algorithm of error region based on WiFi and Bluetooth fusion. J. Sichuan Univ. Sci. Eng. (Nat. Sci. Edn.) 32(03), 52–59 (2019) 11. Hua, H., Guan, W., Liu, Z., et al.: Indoor WiFi and Bluetooth fusion localization algorithm based on optimal bayes. Comput. Eng. 42(11), 114–119 (2016) 12. Shi, F.: Research on safety supervision system of railway passenger station based on 5G highprecision fusion positioning. Railw. Transp. Econ. 44(06), 84–91 (2022) 13. Ma, X., Deng, P., Chen, W., et al.: Multi-system positioning fusion algorithms based on Bayesian system. Sci. Technol. Eng. 19(26), 288–293 (2019) 14. Hu, X., Wang, Y., Sun, M., et al.: Indoor and outdoor scene recognition method based on the neural network. Sci. Technol. Eng. 21(03), 1091–1096 (2021) 15. Zhang, S.M., Su, X., Jiang, X.H., Chen, M.L., Wu, T.Y.: A traffic prediction method of bicyclesharing based on long and short term memory network. J. Netw. Intell. 4(2), 17–29 (2019) 16. Zhang, F., Wu, T.Y., Wang, Y., Xiong, R., Ding, G., Mei, P., Liu, L.: Application of quantum genetic optimization of LVQ neural network in smart city traffic network prediction. IEEE Access 8, 104555–104564 (2020) 17. Wu, J., Xu, M., Liu, F.F., Huang, M., Ma, L., Lu, Z.M.: Solar wireless sensor network routing algorithm based on multi-objective particle swarm optimization. J. Inf. Hiding Multimed. Signal Process. 12(1), 1–11 (2021) 18. Kang, L., Chen, R.S., Xiong, N., Chen, Y.C., Hu, Y.X., Chen, C.M.: Selecting hyper-parameters of gaussian process regression based on non-inertial particle swarm optimization in internet of things. IEEE Access 7, 59504–59513 (2019)

Chapter 12

Design of an Integrated System for the Train Working Diagram of Urban Rail Network Fan Gao, Xu Chen, Xiaoxu Zeng, Li Bai, Xiaohe Song, and Xuze Ye

Abstract In order to improve the quality, efficiency of urban rail transportation plan with the rapid development and networked operation of urban rail transportation in China, this paper designs an integrated system for train working diagram of urban rail transportation networks. The overall system architecture, network architecture, and functional structure are studied, focusing on the main functions such as network transportation plan evaluation, train working diagram preparation, and train connection evaluation.

12.1 Introduction Urban rail is an important component of urban public transport, a regional rail transport bearing core, but also the backbone of public transport, which undertakes the task of short and medium distance passenger transport within the city or suburban areas, promoting the digital and intelligent development of society and transportation, providing the development cornerstone of modern transportation service system for economic prosperity and social development. Urban rail transit has the characteristics of large passenger volume and operation scale, and with the increase of passenger F. Gao (B) · L. Bai China Academy of Railway Sciences, Institute of Computing Technology, Beijing, China e-mail: [email protected] X. Chen Tianjin Rail Transit Operation Group Co., Ltd., Tianjin, China X. Zeng Tianjin Rail Transit Network Management Co., Ltd., Tianjin, China X. Song Beijing Jingwei Information Technology Co., Ltd., Beijing, China X. Ye School of Transportation and Logistics, Southwest Jiaotong University, Chengdu 610031, Sichuan, China © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_12

161

162

F. Gao et al.

travel demand and diversification of route choice, urban rail transit system has the trend of diversification and networking. Therefore, there is an urgent need to build a highly intelligent and informative integrated system for train working diagram of urban rail transit network. In view of the above problems, this paper researches the line network transportation plan evaluation technology, constructs a single and comprehensive evaluation system for the line network transportation plan from the train working diagram technology, capacity and volume matching and train operation connection indexes, and provides a basis for transportation plan adjustment; the system is analyzed based on the preparation and evaluation theory to prepare and evaluate the urban rail line network transportation plan system, and at the same time, the functions that the system should have a detailed business description and functional analysis is conducted; the overall architecture of the system is designed, and C/S technology is used for network design to realize information sharing and collaborative work.

12.2 Literature Review The train working diagram is the link connecting the production activities of urban rail transit and determines the quality and level of urban rail transit organization to a large extent. Szpigel researches on train operation planning, treats train operation as workshop processing, proposes a workshop scheduling method to realize operation planning, and solves it by branch delimitation method [1]. Goossens et al. proposed an integer planning model to study the opening frequency aspects before having a branch cutting algorithm to solve [2]. Wang et al. developed a two-layer planning model to study the interchange, stopping and opening frequencies, considering users to balance the passenger flow distribution [3]. Chang et al. sequentially optimized the grouping, opening frequency and stopping scheme for the single line case, but did not consider the network case optimization [4]. Cronin and Taylor proposed a service performance index model to study the passenger experience evaluation of urban rail operations, which can complete the quantification of user perceived service quality [5]. Chenpeng et al. established a passenger service quality evaluation index system and used hierarchical analysis to solve the model [6]. Kunimatsu based on the passenger perspective on train operation and passenger travel choice simulation was conducted to evaluate the train time in terms of the number of interchanges, waiting time, and train congestion [7]. Through research on transport plan preparation methods and contents, transport plan evaluation contents and methods, and transport plan preparation systems, most of the multi-part transport plan preparation processes are prepared sequentially using a hierarchical and phased approach, with less consideration given to the links between modules, and the lack of a systematic, comprehensive, and holistic study of transport plans. The evaluation of transport plans mostly exists in the evaluation of train working diagram, and there is less research on the evaluation and analysis of the adaptability and matching of capacity and transportation, as well as the evaluation

12 Design of an Integrated System for the Train Working Diagram of Urban …

163

of train operation of network transport plans. Although China’s urban rail transit has widely adopted third-party charting software to overcome the shortcomings of different line charting software, the need for more charting personnel and the low level of automation in the past, it still faces major challenges and cannot support the unified preparation and management of train operation chart of urban rail transit network, and the synergy of train working diagram among lines is poor. The efficiency and quality of the train working diagrams among lines are poor and need to be improved. There is an urgent need to study and design an integrated system for the train working diagrams of urban rail transit line networks, so as to realize the unified preparation and management of train working diagrams of line networks based on the cooperative mode, which is conducive to the coordinated matching and connection of line network working diagrams and the improvement of the transportation organization level of urban rail transit line networks.

12.3 Overall System Design The integrated system for train working diagram of urban rail transit line network takes into account the technical conditions such as the travel pattern of passengers in the network, interchange and first and last train connections, and realizes the automated and coordinated preparation and management of train working diagrams of the network under the conditions of integration of urban rail transit network on the basis of unified management of data.

12.3.1 Overall System Architecture The overall architecture of this system adopts a centralized and distributed information sharing architecture. It can meet the need of preparing and managing urban rail transit train working diagrams under the condition of relatively independent preparation and management of each line, and at the same time, it can meet the need of unified preparation and management of urban rail transit train working diagrams under the condition of line network. The overall architecture of the system is as follows (Fig. 12.1). The application layer is a collection of business logic application systems of urban rail transit system and an application integration platform in the process of system construction, which can improve the sharing of each business data and simplify the business processing process. The core task of the data resource layer is to carry out the utilization and development of resources involved in urban rail transit operation maps. The construction of resources is carried out in accordance with the actual needs of the business system and under the guidance of the standard specification system.

164

F. Gao et al.

The applicaon layer

The applicaon system

...

The subsystem 6

...

external interface

...

Applicaon Integraon Plaorm

resource level Resource planning, resource classificaon, resource management, resource integraon, resource acquision

The plaorm layer

Safety and security system

The subsystem 5

...

The subsystem 4

The subsystem 3

The subsystem 2

The subsystem 1

Operaons management system

...

System soware/database plaorm Server and storage plaorm Network communicaon plaorm

standardized system Fig. 12.1 Overall system architecture

The computer network platform layer is mainly the software and hardware platform required for system operation. Depending on the application needs and resource requirements, it mainly includes the network communication platform, the server and storage platform, and the system software/database platform. The standard specification system, the business management system and the security assurance system provide security for all aspects of the system at all levels. These include various standards and norms, organization and management, security measures and security technology, etc. The “three systems” organically combine management forms and technical means to provide a solid foundation and a solid guarantee for the urban rail transit operation charting system (Fig. 12.2).

12 Design of an Integrated System for the Train Working Diagram of Urban … Fig. 12.2 Basic scheme of network architecture

165

The server YS

TUXEDO SERVER

YC

YC

YC

YC

YC

Operation diagram client (for each line/for each region/for each business)

12.3.2 System Network Architecture In order to adapt to the multi-user collaborative management and preparation of train operation diagrams of urban rail transit line network, C/S network architecture is adopted. To improve the network transmission reliability of the urban rail transit operation map preparation system under computer network conditions, this system uses TUXEDO middleware to complete the network communication. The network architecture scheme is shown in Fig. 12.3. Figure 12.3 shows the network architecture scheme, where YS (run chart server, same below) and TS (TUXEDO server, same below) can be deployed on the same server as the charting server of the Ministry of Railways, and each road bureau connects to TS through LAN or WAN. In this structure, TS and all TCs (TUXEDO clients, same below) are in the same TUXEDO domain, one YS corresponds to multiple YCs (run chart clients, same below), and both YSs and YCs act as TUXEDO clients.

12.4 System Function Structure The main business of the urban rail transit train working diagram preparation and adjustment function is the preparation, adjustment and scheme evaluation of urban

166

F. Gao et al. Load passenger flow data Passenger flow data statistics and generation

Passenger flow data

Passenger flow forecasting tool

Train Line planning

Passenger flow forecast data

Form OD

Operating data

Timetable technical indicators

Stop pattern Table of the number of trains operating

Transfer accessibility

Articulation assessment

Last bus transfer time

Optimiza tion

First and last bus transfer waiting time

Train routing scheme

Matching assessment

Passenger flow analysis

Passenger flow data display function

Capacity and volume matching indicators

Interchange station transfer flow

First and last train operation scheme Capacity allocation plan

Fig. 12.3 Functional structure diagram of the network transportation plan evaluation subsystem

rail transit train working diagram, which mainly includes three major functions such as line network transportation plan evaluation, network working diagram preparation, and train connection evaluation. The compilation of network working diagram is based on the processing of basic data to realize the automatic compilation and adjustment of train working diagram, and the statistics and calculation of relevant indicators of train working diagram, as well as the management of working diagram. The basic data is the basis of the timetabling. The operating unit shall provide the basic data required for the timetabling as required. The functional structure diagram of the line network working diagram preparation subsystem is as follows (Fig. 12.4). The train connection assessment is an important method to assess the train operation connection of each line of the network, providing support to passengers for route inquiry, retrieval, and route accessibility services. The functional structure of the train connection assessment subsystem is as follows (Fig. 12.5).

12.5 Model Building Network Transport Plan Assessment In total, the network transport plan evaluation includes four functional sections: passenger flow analysis, train operating schemes, technical indicators of the train working diagram, and capacity matching indicators.

12 Design of an Integrated System for the Train Working Diagram of Urban …

Train adjustments

Locomotive routing adjustment

167

timetable calibration

Timetable adjustment

Train timetable

Locomotive turnaround timetable

Initial scheme of train timetable

Basic data

Calculation of indicators

Train timetable framework

Network Information

Train operating scale

Locomotive routing

Station Information

Train operating parameters

Train line planning

Interval Information

Carriage Information

Train indicators

utilization

carriage utilization indicators Time indicators Operational mileage indicators Outbound and First and last bus time inbound indicators indicators

Speed indicators

Fig. 12.4 Functional structure diagram of the network transportation planning subsystem

(1) The system provides passenger flow analysis function. It can realize dynamic update of passenger flow data, import of historical data, and display of passenger flow data, specifically including the following. 1) View real-time passenger flow data: the system can automatically read the last 10 min (default time range) of passenger flow data from big data since the user is online, and re-query the big data system every 2 min to obtain the latest passenger flow data update and re-import; and provide the user with the operation of switching the real-time passenger flow viewing time range. 2) import historical traffic data: system can import different periods of history data, including recent 10, 15 min recently, recent half an hour, an hour and specify time recently, which refers to the big data according to specified time of the year, “month”, “day” the history of the three types of statistics to generate passenger flow data; 3) Passenger flow data interface display: six different types of passenger flow tables: OD passenger flow table, interval section passenger flow table, station up and down passenger flow table, line table, station interchange passenger flow table, line interchange passenger flow table can be displayed through the system interface, and it is possible to output the selected type of passenger flow table to the specified location.

168

F. Gao et al. Transfer passenger volume Transfer waiting time

Articulation assessment

Transfer service level assessment

Travel path information query

Regular Articulation Time Search

Transfer waiting time query

Line transfer connection scheme

First and last bus connection plan

Train Timetable data

Timetable preparation subsystem

loading rate Transfer passenger Data management volume

Fig. 12.5 Functional structure diagram of train connection evaluation subsystem

(2) The system provides urban rail transit train operation scheme query function. The system can display relevant information of different lines, such as rated passenger capacity, maximum running speed, number of groups, etc.; and provide specific train line plan, including: departure station, arrival station, departure time, arrival time, interval, train path and operation type for each period. (3) The system provides statistical and evaluation functions for technical indicators of train working diagram. The indicators can be automatically calculated and evaluated, and if there is an abnormal value of the indicator, the evaluation result will give the corresponding suggestion description. (4) The system provides capacity matching index statistics and evaluation functions. For different lines in the imported passenger flow data, it automatically calculates the capacity matching indicators such as average peak full load rate, average flat full load rate, average full load rate for the whole day (including different line histogram display), maximum section full load rate, and time

12 Design of an Integrated System for the Train Working Diagram of Urban …

169

ratio of full load rate higher than 100%. If the full load ratio is seriously over, over or under, the system will give explanations and hints in the evaluation results according to the calculation results. The system can display the capacity situation of different routes in different zones.

12.6 Network Train Working Diagram Preparation The line working diagram preparation functions mainly include: data management, diagramming, drawing output, schedule output, and operation simulation.

12.6.1 Data Management The data management module includes basic data and mapping data management functions. Data management is the basis for implementing the business functions of the system. The relevant data is required for the implementation of both map preparation and simulation functions. It includes line data information, station data information, basic time division, interchange definition, line plan and base map structure (displaying segment parameters), etc. (1) Line data management The line management module realizes line creation, insertion, deletion, name change, and dragging adjustment of line arrangement order. (2) Station data management The station management module realizes the creation, insertion, deletion, modification and movement of line stations and the management and maintenance of data such as topology of all line stations, layout form of yards, station plan structure, basic information of stations, and lines in the urban rail transit network: the CAD system of station plan is established, and the lines, turnouts, crossings, signals, and other equipment of stations can be laid out conveniently by means of graphical interface as required. The system can automatically generate various train approaches according to the connection between turnouts, crossings, signals, and other equipment. (3) Interval data management The interval management module realizes automatic generation, creation, insertion, deletion, modification, and movement of line intervals. (4) Train paths The train path management module can create, delete, generate opposite paths, modify the included intervals, and set the basic attributes of train paths, which will be used in priority in automatic mapping.

170

F. Gao et al.

(5) Interval operation type and time classification The management module enables the creation, insertion, deletion, modification, and movement of interval run types and time scores. The default operation types of interval scale are regular, high speed, low speed, and overrun. Including the setting of base time. (6) Train line plan The train operation plan management module realizes integrated management of all lines of urban rail transit network: it supports the preparation of train operation plan under complicated situations such as unbalanced transportation, multi-crossing operation (such as small and large crossings, Y-crossings, etc.), fast and slow trains, etc. The train operation plan management module realizes the creation, insertion, deletion, and modification of train operation plan. (7) Interval constraint The interval constraint management module realizes the management of various interval data for line stations and intervals. Line intervals are intervals that apply to all stations on the line. (8) Crossing definition The train crossing can be defined by the train running scheme, and the specific set of train travel stations can be defined by the aforementioned train path, which can meet the needs of complex line networks such as Y-type and common line type lines. The locomotive routings management module realizes the creation, deletion, and modification of successive trains of the locomotive routings. (9) Train management The train management module realizes the maintenance management of train data, including train number, origin station, destination station, destination code, train route, train characteristics, origin arrival point, origin departure point, arrival point, arrival departure point, and arrival point of the train at the passing station, departure point, occupied arrival and departure line, stopping time, and the running type, running time, and length of the passing interval, etc. (10) Base map structure The system manages the base map structure through the “Display segment parameters” management module, including base map station selection, station order selection, etc.

12 Design of an Integrated System for the Train Working Diagram of Urban …

171

12.6.2 Preparation and Adjustment 12.6.2.1

Automatic Preparation

Automatic preparation is based on the establishment of line, station and interval data, and the system automatically calculates and generates operating trains and incoming and outgoing trains according to the opening plan for each period. It realizes the integrated and automatic preparation of train running diagrams of all lines of urban rail transit network and the automatic preparation of train running diagrams of single lines, i.e. the automatic preparation of train running diagrams of one or more and all selected urban rail transit lines. To meet the needs of different users, the system has two modes of charting: according to the starting time specified in the traffic plan and according to the service frequency to meet passenger travel.

12.6.2.2

Human-Machine Interaction Adjustment Working Diagram

The system provides functions including train arrival and departure point adjustment, train panning, interval adjustment, modification of train number, modification of interval and station stopping time, modification of running scale, addition and deletion of trains, modification of platform usage plan, etc. It also provides functions such as “forward, reverse, two-way pushing line”, “line wiping”, “forward and reverse line rubbing”, and other train adjustment functions. It supports the adjustment operation of the set of running lines of selected lines under line network conditions and can realize the coordinated matching adjustment of train running diagrams between multiple lines based on line network coordination.

12.6.2.3

Locomotive Routing Inspection

Locomotive routings are prepared after the train operating diagram has been prepared. The system provides the means to adjust the intersection by creating, inserting and deleting locomotive routings, adding, inserting and deleting trains in the specified locomotives routings, establishing and deleting the carriage connection between two trains, deleting trains on the crossings, copying the crossings, pushing lines by crossings, and overall translation.

12.6.2.4

Full Working Diagram Check

The legality check of the train operation plan preparation results, including whether the departure and arrival times of trains at the passing stations are reasonable; whether the interval requirements are met, etc. When adjusting the train operation line by human-machine interaction, the system will automatically perform conflict checking

172

F. Gao et al.

and provide conflict indication, including displaying conflict information and train operation line indication in the information display window.

12.6.3 Working Diagram Output Working diagram output realizes multi-format and multi-format output styles to meet various needs. (1) Train schedule output, including EXCEL format output sorted by undercarriage, by station, format provided by the demander, etc. (2) Undercarriage crossing output, providing EXCEL format. (3) Interchange working diagram output, providing EXCEL format to meet the requirements of the demand side.

12.6.4 Train Working Diagram Indicators This function automatically calculates various indicator parameters required by the user. Including automatic calculation of the selected line train indicators, utilization rate, speed indicators, undercarriage utilization indicators, etc., and realize the EXCEL format output. The system provides Web Server standard interface to provide services to the public for the query statistics function.

12.6.5 Train Connection Assessment The main functions of train connection evaluation include: system first and last train connection query display, regular time connection query display, interchange service level query, travel interval query, path information query, and data management functions. It can provide comprehensive query and analysis of the connection situation of the line network and provide support for the evaluation of train operation diagrams, analysis of capacity and volume matching, and passenger inquiry services. (1) First and last train connection scheme The system will automatically calculate the feasible first and last train scheme for the line in the selected road network and display it on the interface. The contents include interchange station, source line name and line, destination line name and line, connecting train, arrival time, departure time, whether to connect, waiting time, and evaluation result. (2) Line interchange connection scheme The system will automatically calculate feasible interchange schemes for the specified date, line and interchange station, and display them on the interface. (3) Interchange Waiting Time Inquiry

12 Design of an Integrated System for the Train Working Diagram of Urban …

173

Provides information on the first train transfer waiting time, the last train transfer waiting time, the average waiting time during the peak period, and the average waiting time for a transfer during the average peak period. Specific information includes hub name, station name, line name, direction, interchange station name, interchange line name, interchange direction, and interchange waiting time. (4) Routine time connection query The system can display the arrival time, number of trains and transfer waiting time of trains in each transfer direction at the regular time point or regular time period at the interchange station, and the evaluation service personnel can analyze the connection of train operation at the interchange station according to the transfer waiting time. If trains at that point in time are not available for interchange, the latest available interchange time is given. (5) Travel route information query The system retrieves and displays travel route information based on the user-selected starting station, final destination station, travel time conditions, and travel strategy in the line network. (6) Evaluation of interchange service level The system shall be able to calculate and display the interchange volume and total interchange waiting time for each direction at the interchange station. (7) Train interval query The system calculates and displays the average interval time of trains in each direction at the interchange station during the specified time period. Enter the specified time period, and the system can automatically calculate the average train interval time based on the total number of trains departing during the time period.

12.7 Conclusion This paper has built a collaborative platform for metro train working diagram preparation management that integrates passenger flow analysis, capacity analysis and configuration, line network train working diagram preparation, train running map evaluation, and train working diagram information sharing and release, to realize line network train working diagram preparation that coordinates all lines in the scope of the whole network, rationalizes transportation, and efficiently relieves passenger flow in the whole network. Acknowledgements This research was supported by the Guangzhou Science and Technology Plan Project (202206030007).

174

F. Gao et al.

References 1. Szpigel, B.: Optimal train scheduling on a single track railway. Oper. Res. 343–352 (1973) 2. Goossens, J.-W., Van Hoesel, S., Kroon, L.: A branch-and-cut approach for solving railway line-planning problems. Transp. Sci. 38(3) (2004) 3. Wang, L., Jia, L.-M., Qin, Y., Xu, J., Mo, W.T.: A two-layer optimization model for high-speed railway line planning. J. Zhejiang Univ.-Sci. A 12(12) (2011) 4. Chang, Y.-H., Yeh, C.-H., Shen, C.-C.: A multiobjective model for passenger train services planning: application to Taiwan’s high-speed rail line. Pergamon 34(2) (2000) 5. Cronin, J.J., Taylor, S.A.: Measuring service quality: a reexamination and extension. J. Mark. 56(3) (1992) 6. Li, C., Yao, J., Zheng, Y., Wang, J.: Evaluation of passenger service quality in Urban Rail Transit: case study in Shanghai. Springer Singapore (2018) 7. Kunimatsu, T., Hirai, C., Tomii, N.: Train timetable evaluation from the viewpoint of passengers by microsimulation of train operation and passenger flow. Electr. Eng. Jpn. 181(4) (2012)

Chapter 13

Research on Optimization of Operation Organization of Transship Trains in Railway Hub Zongying Song, Yi Li, Mengyuan Yue, Kun Liu, and Miaomiao Lv

Abstract Taking the transship trains of the railway hub as the research object, a study was carried out on the optimization of the operation organization of the transship trains in the hub. With the goal of minimizing the running cost of transship trains, an optimization model of the operation organization of transship trains in railway hubs is constructed under the condition that the trains can be picked up and delivered on time. At the same time, the railway hub case is selected, the model is used to analyze, the improved SA algorithm is designed to solve it, and the model and algorithm are verified. The results show that the constructed model and algorithm can make full use of resources such as dispatching vehicles on the premise of realizing the timely pickup and delivery of trains in the hub, and obtain a suitable operation plan for transship trains in the hub, so as to minimize the overall operating cost, so as to achieve the optimization of the operation organization of transship trains in railway hubs improves the efficiency of cargo transportation.

Z. Song China Shenhua Energy Co., Ltd., Beijing 10011, China Y. Li · M. Yue · K. Liu · M. Lv (B) School of Transportation and Logistics, Southwest Jiaotong University, Chengdu 610031, China e-mail: [email protected] Y. Li e-mail: [email protected] National and Local Joint Engineering Laboratory of Comprehensive Intelligent Transportation, Southwest Jiaotong University, Chengdu 610031, China National Engineering Laboratory of Integrated Transportation Big Data Application Technology, Chengdu, China © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_13

175

176

Z. Song et al.

13.1 Introduction In recent years, my country’s transportation industry has developed rapidly. However, due to the problems of slow transportation speed, low timeliness, and complicated procedures for railway freight transportation, its freight volume growth rate is lower than that of other transportation methods. In view of this change in the overall freight transportation environment in my country, the railway freight system should improve the existing transportation organization method from the perspective of its own transportation organization. The rationality of the transportation organization of the railway hub is directly related to the smoothness of the entire railway network. Therefore, the research on the optimization technology of the flow organization of the transship trains in the railway hub has the advantages of improving the efficiency of freight transportation and changing the unfavorable position of railway freight in the current transportation market. Scholars at home and abroad have different research focuses on hub traffic organization. Foreign research mainly focuses on route selection and empty-car deployment. In terms of route selection, optimization is often considered in conjunction with the train marshalling plan [1, 2]; in terms of empty-car deployment, foreign experts and scholars have gradually shifted from static empty-car deployment optimization to optimization. Dynamic empty vehicle deployment optimization [3–5]. Domestic scholars have carried out some research on the optimization of railway hub traffic organization. Based on the analysis of the hub traffic flow, Yan Yusong established the transfer traffic flow, local traffic flow, and the hub transship train operation organization optimization model [6]; later the author proposed the basic idea of solving the decision-making problem of hub traffic scheduling that is to solve the problem dynamically and hierarchically [7]; Niu Huimin considers the operation process of railway hub through reorganization and local traffic flow and establishes a nonlinear 0–1 planning model based on hub capacity constraints [8]. After that, the author coordinated and optimized the heavy and empty traffic flow and established an optimization model with the goal of minimizing the cost [9]; Wang Zhengbin, with the goal of reducing the hours of vehicles staying in the hub, established the railway hub transfer traffic model and local traffic model in layers. And the transship train model, and design a genetic algorithm to solve it [10]. Considering the coordination and division of labor among multiple marshalling stations in the hub, Li Bing optimized the heavy and empty traffic flow organization in the hub and designed a two-stage comprehensive solution strategy [11]. After summarizing and sorting out the existing research status at home and abroad, it can be found that most of the problems of optimizing the operation organization of the transship trains in the hub are mostly attributed to the network flow problem. Based on the existing research, this paper comprehensively considers the service sequence of the transship trains in the hub at each freight station, the determination of the running path of the transship trains in the hub, and the connection between the transship trains and the large-running trains in the hub. With the goal of minimizing

13 Research on Optimization of Operation Organization of Transship …

177

the cost of train operation, a corresponding model is constructed to minimize the cost of pickup and delivery in the hub under the condition that the trains can be picked up and delivered on time.

13.2 Problem Analysis The optimization of the operation organization of transship trains in railway hubs aims to reasonably improve the corresponding workflow, so as to achieve the purpose of saving train turnaround time, reducing train running costs, and minimizing the running costs of transship trains. The running cost of the transship trains mainly includes the operation cost of adjusting the machine and the vehicle turnover cost. When the transship trains arrive earlier or later than the operation time window of the freight station, there will also be an early or late arrival fee. Therefore, this paper takes the comprehensive cost of running transship trains as the target to optimize the operation organization plan of the transship trains in the hub. The determination of the running path of the running train and the connection between the transship train and the large-running train at the hub.

13.3 Model Building 13.3.1 Conditional Assumptions To simplify the problem, the following assumptions are made: (1) It is assumed that there is only one bidirectional marshalling yard in the hub. (2) The lines connecting the stations in the hub satisfy the two-way operation of the trains, and it is assumed that the running time of the transship trains in the round-trip direction between the two stations is equal. (3) The transship trains of the hub start from the marshalling station and return to the marshalling station after completing the pickup and delivery operation at each designated station. (4) Freight stations with demand for pickup and delivery are only served once by the hub’s transship train, regardless of the situation of picking up and delivering in batches. (5) The transship trains at the hub must meet the constraints of the maximum number of trains, that is, the maximum number of trains is not allowed to exceed the maximum number of trains, but it is allowed to run under the shaft. (6) Consider the time limit of the service operation allowed by the freight station. The transship train arrives at the operation point before the specified operation time period, and it needs to wait until the operation time before it can work.

178

Z. Song et al.

If the transship train arrives at the operation point within or after the specified operation time, in order to ensure the time limit for the delivery of goods, it is assumed that it can be dumped immediately. Trailer work. (7) The pickup and delivery information of each freight station during the stage is known. (8) Assuming that there is a direct or indirect connection path between the two freight stations in the hub, when the transship train of the hub travels from one station to another station, the shortest path between the two freight stations is selected as the running path of the transship train.

13.3.2 Parameter Definition According to the above problem analysis, the relevant hub small operation train operation organization optimization model is established, and its parameters and variables are explained as follows, as shown in Tables 13.1 and 13.2.

13.3.3 Objective Function The operating costs of the transship trains at the hub mainly include the operation cost of adjusting the machine and the cost of vehicle turnover. At the same time, in order to ensure the timely pickup and delivery of goods and the satisfaction of the cargo owners, if the transship trains are earlier or later than i the operation time window [ei , f i ] of the freight station Arrivals will incur an early or late fee. This chapter will optimize the operation organization plan of transship trains at the hub with the goal of minimizing the comprehensive cost of running transship trains. The specific expressions are as follows: (1) Minimize the total operating cost of the machine: minZ 1 =

n  m n  

ti j xikj c1

(13.1)

i=0 j=0 k=1

(2) Minimize the total operating cost of the vehicle: minZ 2 =

n  m n  

ti j xikj Uikj c2

(13.2)

i=0 j=0 k=1

(3) The transship train of the hub arrives earlier than i the earliest allowable operation time of the freight station, and the total early arrival cost is minimized:

13 Research on Optimization of Operation Organization of Transship …

179

Table 13.1 Parameter description of the optimization model for the operation organization of transship trains in the hub Parameter

Illustrate

N

The collection of all stations in the hub, denoted as N = {0, 1, 2, ..., n}, among them, i, j ∈ N . At that time i, j = 0, it was recorded as a marshalling station; i, j = 0 at that time, it was recorded as a freight station

M

A collection of transship trains at the hub, M = {k|k = 1, 2, ..., m}

L

The set of all edges of the hub network, where l ∈ L

ti j

The average running time of ti j = ∞ transship trains at the hub from station i to station, j indicating that there is no connecting path between the two stations

t

The time required for the transship trains to complete the drop-and-retract operation at each freight station

b0

The marshalling station allows the latest return time of the transship trains at the hub

ei

Freight station i allows the earliest arrival time of the transship trains at the hub, i >0

fi

Freight station i allows the latest arrival time of the transship trains at the hub, i >0

ge

Early arrival fee factor

gf

Late fee factor

c1

Operating cost per minute

c2

Vehicle unit minute operating cost

Z

The maximum number of vehicles allowed to be carried by the transship trains at the hub

pj

Number of drop-offs of transship trains in the hub at the freight station j

qj

The number of trailers of transship trains in the hub at the freight station j

rl

Edge l Passability (column/phase plan)

Table 13.2 Description of variables in the optimization model for the operation organization of small-running trains in the hub Variable

Illustrate

s0k

The third k transship trains from the marshalling station to the freight station

ai

The moment when the transship train of the hub arrives at the freight station i

si

The actual departure time of the transship trains at the hub at the freight station i

xikj

The first k transship train runs from station i to station j, if so, xikj = 1; otherwise, xikj = 0

y kj

First k transship train drops the j trailer group at the freight station, if so, y kj = 1; otherwise, y kj = 0

z ikjl

The first k transship train i goes from station to station j, does it pass the edge? l

Uikj

Total number of vehicles assembled when the first k transship trains i traveled from station to station j

180

Z. Song et al.

minZ 3 =

n 

max[(ei − ai )g1 , 0]

(13.3)

i=1

(4) The transship trains at the hub arrive later than the latest allowable operation time of the freight station i, and the total late arrival fee is minimized: minZ 4 =

n 

max(ai − f i )g2 , 0]

(13.4)

i=1

Therefore, with the goal of minimizing the comprehensive operating cost of transship trains in the hub, the objective function is expressed as follows: min Z =

n  m n  

ti j xikj c1 +

i =0 j =0 k =1

+

n 

n  m n  

ti j xikj Uikj c2

i =0 j =0 k =1

max{[ei − ai ]g1 , 0} +

i =1

n 

max{[ai − f i ]g2 , 0}

(13.5)

i =1

13.3.4 Constraints Constraints are shown in Eqs. (13.6)–(13.19): (1)

Each freight station that needs to pick up and deliver vehicles is only served by one small hub train, and it is only served once at most: n m  

i =0 i = j

(2)

(13.6)

xikj ≤ 1, ∀i ∈ N

(13.7)

k=1

n m  

j =0 i = j

xikj ≤ 1, ∀ j ∈ N

k=1

Constraints of flow conservation. Indicate that a freight station is served by the same transship train at the hub: n  i=0,i= j

xikj =

n  i=0,i= j

x kji ∀ j ∈ N , k ∈ M

(13.8)

13 Research on Optimization of Operation Organization of Transship …

(3)

After completing the operation of each freight station along the way, each transship train of the hub needs to return to the marshalling station: n 

x0ik =

i=1

(4)

n 

x kj0 , ∀k ∈ M

Through ability constraints. When a transship train at a hub selects a certain running path, it needs to meet the passing capacity constraints of each section line passing on the path:

i∈N

j∈N i = j

xikjl ≤ rl , ∀l ∈ L

(13.10)

k∈M

The time continuation relationship between the arrival of the transship trains at a hub and the departure from the freight station: max{ai , ei } + t  = si i ∈ N &i = 0

(6)

(13.9)

j=1

  

(5)

181

(13.11)

The transship trains at the hub provide traffic sources for the large-running trains at the marshalling station and should return to the marshalling station before the latest return time allowed by the marshalling station: n 

k xi0 · (si + ti0 ) ≤ b0 ∀k ∈ M

(13.12)

i=1

(7)

The initial number of transship trains starting from the marshalling station satisfies the constraint of the maximum number of trains (allowing for undershafts): 0≤

n  n 

xijk p j ≤ Zi = j, ∀k ∈ M

(13.13)

i=0 j=1

(8)

Transship trains go from the marshalling station to each freight station. After the drop-and-hang operation is carried out on the way, the number of trains still meets the constraint of the maximum number of trains (allowing lack of axles): 0≤

n  n  i=0 j=1

xijk p j − y kj p j + y kj q j ≤ Z ∀k ∈ M, ∀ j ∈ N

(13.14)

182

Z. Song et al.

(9)

The connection between the departure time of the kth transship train from the marshalling station and the departure time of the freight station j that initially handles the decoupling operation: x0k j {s0k + t0 j + max[e j − (s0k + t0 j ), 0] − s j + t  } = 0∀ j ∈ N , ∀k ∈ M (13.15)

(10) The connection between the moment when the kth transship train departs from freight station i after the drop-off operation and the departure time after the next operation point of the next operation point j is completed: xikj {si + ti j + max[e j − (si + ti j ), 0] − s j + t  } = 0

(13.16)

∀i& j ∈ N , i = j, ∀k ∈ M (11) n 

xikj = y kj i = j, ∀ j ∈ N , ∀k ∈ M

(13.17)

i=0

(12) xikj ∈ {0, 1}i = j, ∀i& j ∈ N , ∀k ∈ M

(13.18)

y kj ∈ {0, 1}i = j, ∀i& j ∈ N , ∀k ∈ M

(13.19)

(13)

13.4 Solving Algorithm In this section, the SA algorithm will be used to solve the model. (1) Setting of the initial solution Since the quality of the initial solution has little effect on the solution result of the algorithm, this method can be used to obtain the initial solution. Arrange the m transship trains of the hub to m the route in the hub, select the freight stations that need to be served in the hub and add them to the route, respectively, and calculate the remaining station space that can be served by each route, and the number of stations that each train can serve at a certain station. The m time required to reach the next service station t, repeat the above process until all freight stations to be serviced have been added to each route. (2) State generation function

13 Research on Optimization of Operation Organization of Transship …

183

Neighborhood functions are applied to ensure that the search range of the algorithm covers as many feasible solutions as possible. For a current solution, there is always xi an adjacent solution x j that can be generated by exchanging a pair of basis variables and non-basic variables xi . In this chapter, the stochastic twotransformation is used to generate the neighborhood solution of the operation organization scheme of the transship trains in the hub. (3) State acceptance function According to the Metropolis criterion, the inferior solution should be accepted with a certain probability, that is, when the objective function value of the new solution is f (x j ) smaller than the objective function value f (xi ) of the current solution, if P >R, the new solution is used instead of the current solution, where f P = exp( t i j ), R = r and(0, 1). (4) Initial temperature setting The initial temperature t0 should try to ensure that the probabilities of the states in the stationary distribution are equal, that is, the following formula (13.20) is established: e f ji /t0 ≈ 1

(13.20)

An easy estimate is t0 = K λ, K for a sufficiently large number, λ = max{ f ( j)} − min{ f ( j)}. The experimental results of the t0 literature [12], as long as the selection can ensure that the initial acceptance probability is 0.9–0.98, satisfactory results can be obtained. Therefore, the initial temperature of the algorithm in this paper is set to 100. (5) Temperature control function In general, formula (13.16) can ensure that the algorithm gradually converges to the global minimum, but from the point of view of solving practical problems, using this function to control the temperature drop speed is slow and the calculation time is long. Therefore, this paper adopts the tk+1 = tk α decay function in the form of a commonly used linear function, where α ∈ (0.95, 0.99). (6) Algorithm Termination Principle 1) Set the end temperature. When the termination temperature is reached, the algorithm iteration stops and an approximate optimal solution is obtained. 2) After the algorithm reaches a certain number of iterations, and the objective function value is not improved, the iteration is stopped. Based on the above, the basic steps of the algorithm for solving the optimization model of the operation organization of the transship trains in the hub are as follows: STEP 1: initialization algorithm parameters, including initial temperature t0 , termination temperature tend , temperature decay coefficient α, Markov chain length , etc.; STEP 2: takes the initial feasible solution of the operation organization plan of the transship trains of the hub x0 ∈ X as the current solution at the current

184

Z. Song et al.

temperature tk = t0 , and calculates the initial objective function value Z 0 , and records the current optimal objective function value W = Z 0 ; STEP 3: performs STEP4–STEP6 at each temperature value tk until the Markov chain reaches a steady state distribution; STEP 4: uses the above random transformation method to generate the neighborhood solution of the transship trains operation organization scheme, that is, the order of two or three freight stations served in the running path of a transship trains in the current hub transship trains operation organization scheme is randomly exchanged. Or the sequence of two freight stations served in the running path of two different transship trains to generate a new running organization plan for transship trains; STEP 5: solves the objective function value under the neighborhood solution of the operation organization plan of the transship train of the hubZ j . IfZ j ≤ W , record the neighborhood solution as the optimal solution; if Z j > W , accept the neighborhood solution according to the Metropolis criterion; STEP 6: cools down to tk+1 , where tk+1 = tk α, if tk+1 < tend , the algorithm ends, and the optimal solution is output; otherwise. Go to STEP 3.

13.5 Case Analysis Figure 13.1 shows the plane layout of the marshalling station and each freight station in a railway hub (the numbers marked on each side of the network indicate the interval distance between the stations and the pure running time of the train). There are 8 freight stations and 1 marshalling station in the hub, of which the number 0 represents the marshalling station, and the numbers 1–8 represent the freight station. The additional time for starting and stopping of freight trains in the hub is taken as 2 m in. Taking the stage time 12:00–16:00 as an example, the specific information on picking up and delivering vehicles at each freight station and the time window for picking up and delivering vehicles at each freight station are known, as shown in Tables 13.3 and 13.4. The maximum number of transship trains at the hub is 52, and the drop and pull operation time at each freight station is uniformly 20 min. The maximum number of transship trains allowed is 5. The train sets provided by each freight station, as one of the traffic sources of the large-running trains at the marshalling station, need to ensure that the large-running trains depart on time and with full axles. In order to simplify the calculation, each station of the provided train group is used instead of each destination to express the time constraint, that is, the latest return time of the train group allowed by each freight station at the hub marshalling station is shown in Table 13.5. According to the data shown in Fig. 13.1, according to Dijkstra’s shortest path algorithm, the shortest running time of freight trains between any two stations in the hub is obtained. The specific calculation results are shown in Table 13.6. The simulated annealing algorithm designed in this paper is used to solve the calculation example. According to the relevant literature and books, the relevant

13 Research on Optimization of Operation Organization of Transship …

185

Fig. 13.1 Plane layout of marshalling yards and freight stations in a railway hub

Table 13.3 Information table of pickup and delivery of transship trains in the hub (unit: vehicle) Cargo station number

1

2

3

4

5

6

7

8

Number of dumped cars

25

8

15

8

18

14

22

18

Number of trailers

18

6

12

10

24

11

20

16

Table 13.4 Time window of service allowed for each freight station in the hub (unit: minute) Cargo station number

1

2

3

Earliest time window for acceptable service (starting at 30 60 70 12:00)

4

5

6

7

8

125 100 110 15 60

The latest time window for acceptable service (starting 60 90 100 150 130 135 45 90 at 12:00)

Table 13.5 The latest return time of the train group allowed by each freight station at the marshalling station (unit: minute) Cargo station number

1

2

3

4

5

6

7

8

Latest return time

15:00

14:45

15:00

15:30

None

15:00

13:30

None

parameters are set as follows: early e1 = 2 arrival cost coefficient, late arrival cost coefficient e2 = 8, operating cost per minute of dispatching machine, operating cost per minute per c1 = 15 vehicle c2 = 1.5, Initial temperature t0 = 100, cooling coefficient α = 0.96, termination temperature tend = 0.1, and Markov chain  = 80. Through multiple iterative solutions, the approximate optimal solution of the calculation example is obtained (Table 13.7).

186

Z. Song et al.

Table 13.6 The timetable of transship trains between stations in the hub i

0

1

2

3

4

5

6

7

8

0

0

16

27

28

19

14

26

18

10

1

16

0

15

27

31

26

14

21

13

2

27

15

0

16

23

18

25

36

24

3

28

27

16

0

13

18

37

42

34

4

19

31

23

13

0

9

41

33

25

5

14

26

18

18

9

0

36

28

20

6

26

14

25

37

41

36

0

14

22

7

18

21

36

42

33

28

14

0

12

8

10

13

24

34

25

20

22

12

10

Table 13.7 The operation and operation of transship trains in 10 hubs Train number Stations and operating conditions of transship trains I

0 (Depart at 12:18, group 47 cars) → 1 (Arrival at 12:32, drop 25 cars, hang up 18 cars) → 2 (Arrival at 13:05, drop 8 cars, put up 6 cars) → 1 (13:38 pass) → 6 (arrive at 13:50, throw 14 cars, hang 11 cars) → 1 (pass at 14:22) → 0 (return at 14:36, get back 40 cars)

II

0 (departure at 12:42, group 4 and 1 car) → 5 (pass at 12:54) → 4 (pass at 12:59) → 3 (arrive at 13:10, throw 15 cars, hang 12 cars) → 4 (pass at 13:41) → 5 (arrive at 13:48, drop 18 cars, hang up 23 cars) → 4 (arrive at 14:17, drop 8 cars, hang up 10 cars) → 5 (14: 44 pass) → 0 (14: 56 return, get back 4 5 cars)

III

0 (departure at 12:00, group 40 cars) → 8 (arrive at 12:08) → 7 (arrive at 12:18, drop 22 cars, hang up 20 cars) → 8 (arrive at 12:50, Throw 18 cars, hang 16 cars) → 0 (return at 13:30, get back 36 cars)

According to the above-mentioned transship trains operation plan, the total comprehensive operating cost is calculated to be Z = 8416.6 yuan, and the total time required for each hub transship trains to complete all the work is 342 min. Using the SA algorithm, the convergence process is shown in Fig. 13.2. It can be seen from the results that this method can make full use of resources such as dispatching vehicles on the premise of realizing the timely collection and delivery of vehicle groups in the hub, so as to minimize the overall operating cost.

13.6 Conclusions This paper analyzes the optimization of the operation organization of the transship trains in the hub from three aspects: the service sequence of the transship trains in the hub, the determination of the running paths of the transship trains, and the connection between the transship trains and the large-running trains. Taking the minimum

13 Research on Optimization of Operation Organization of Transship …

187

Fig. 13.2 Simulated annealing convergence process

comprehensive cost of running transship trains as the optimization objective, an optimization model of the running organization of transship trains in railway hubs is constructed under the condition that the trains can be picked up and delivered on time. At the same time, the case of railway hub is selected, the model is used to carry out analysis, and the improved SA algorithm is designed to solve it, which verifies the practicability and effectiveness of the model and algorithm. This paper takes the transship trains in the hub as the breakthrough point and studies the optimization of railway traffic flow organization. Funding This research was supported by the National Natural Science Foundation of China (Project No. 52172321; 52102391), China Shenhua Energy Co., Ltd. Science and Technology Program (Project No.: CJNY-20-02), and Sichuan Science and Technology Program (Project NO. 2020YJ0268; 2020YJ0256; 2022YFH0016; 2021YFQ0001; 2021YFH0175; 2022YFQ0101), Key science and technology projects in the transportation industry of the Ministry of Transport (2022-ZD7-132).

References 1. Assad, A.A.: Modelling of rail networks: toward a routing model. Transp. Res. B 14(1), 101–114 (1980) 2. Bo˙zejko, W., Grymin, R., Pempera, J.: Scheduling and routing algorithms for rail freight transportation. Procedia Eng. 178, 206–212 (2017) 3. Philip, C.E., Sussman, J.M.: Inventory model of the railroad empty car distribution process (1977)

188

Z. Song et al.

4. Lawley, M., Parmeshwaran, V., Richard, J.P., et al.: A time-space scheduling model for optimizing recurring bulk railcar deliveries. Transp. Res. Part B Methodol. 42(5), 438–454 (2008) 5. Rakhmangulov, A., Kolga, A., Osintcev, N., et al.: Mathematical model of optimal empty rail car distribution at railway transport nodes. Transp. Probl. Int. Sci. J. 9(3), 125 (2014) 6. Yan, Y., Jiang, G.: Research on traffic organization model of railway hub. In: National youth management science and system science proceedings, vol. 5, pp. 47–51. Nankai University Press, Tianjin (1999) 7. Yan, Y.: Research on intelligent railway system hub traffic scheduling decision and transship train organization model and its algorithm. Southwest Jiaotong University (2000) 8. Niu, H., Hu, A.: Nonlinear 0–1 programming model and algorithm of railway junction traffic organization. J. Railw. 23(3), 8–11 (2001) 9. Huimin, N.: Coordination optimization model and genetic algorithm of heavy and empty traffic organization in railway hubs. Railw. J. 23(1), 12–16 (2001) 10. Wang, Y.: Research on the traffic organization model of railway hub and its genetic algorithm. Southwest Jiaotong University (2001) 11. Li, B., Deng, S., Xuan, H.: Organization optimization of heavy and empty traffic flow in hub considering the coordination and division of labor among multiple marshalling stations. Control Decis. 1–9 (2021) 12. Lin, B., Zhu, S.: The nonlinear 0–1 programming model and simulated annealing algorithm for optimal marshalling plan (02) (1994)

Chapter 14

Optimization Principle of Freight Train Operation Plan for Shenhua Railway Meng Wang, Qiuqi Liu, Wenhui He, and Xiuyun Guo

Abstract At present, the freight train formation plan cannot dynamically reflect the changes in freight demand in the transportation market, which makes it difficult to guide the actual transportation production work and affects the economic benefits of enterprises. This paper takes Shenhua Heavy Duty Railway as the research object, studies the optimization technology of cargo train programming under dynamic planning type transportation organization mode, and proposes the optimization principle of dynamic cargo train programming of Shenhua Railway. By analyzing the characteristics of the cargo transportation organization of Shenhua Railway, the dynamic planning type transportation organization mode of Shenhua Railway is proposed. On the basis of the concept of the dynamic freight train operation scheme of Shenhua Railway, the feasibility of preparation optimization, influencing factors, and main principles, the idea of dynamic preparation optimization research from the perspective of synergistic consideration of heavy train operation scheme and empty train deployment scheme is also elaborated.

M. Wang Southwest Jiaotong University, Chengdu 610031, China e-mail: [email protected] China National Energy Group Railway Equipment Company, Beijing 100000, China Q. Liu · W. He · X. Guo (B) School of Transportation and Logistics, Southwest Jiaotong University, Chengdu 610031, China e-mail: [email protected] Q. Liu e-mail: [email protected] W. He e-mail: [email protected] National and Local Joint Engineering Laboratory of Comprehensive Intelligent Transportation, Southwest JiaoTong University, Chengdu 610031, China National Engineering Laboratory of Integrated Transportation Big Data Application Technology, Beijing, China © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_14

189

190

M. Wang et al.

14.1 Introduction The freight train operation plan is the main part of organizing cargo transportation, and it is also the key to influence the cargo owners’ loading demand and cargo arrival time, and its preparation is based on “driving by flow”. As an important railroad infrastructure of the national energy group, Shenhua Railway has a very important position in the whole national railroad transportation network and within the group, so this paper focuses on Shenhua Railway. When the traffic flow of each section is not closely coordinated, it may lead to many problems such as uneven arrival and departure of traffic at stations, difficulties in train succession, and inadequate utilization of running lines of each section and late trains,. Moreover, the heavy load transportation capacity of Shenhua Railway is relatively tight. Therefore, if we can prepare the operation plan according to the dynamic changes in cargo transportation demand in the Shenhua Railway market, it will greatly improve the efficiency of train operation. For the characteristics of the uneven spatial and temporal distribution of heavy and empty trains, Liu et al. [1, 2] based on the dynamic traffic flow of the decision cycle in order to prepare a freight train operation scheme that satisfies the dynamic heavy traffic flow. Wu et al. [3] proposed a dynamic traffic pooling and grouping strategy based on dynamic cargo flow; Li et al. [4] established a dynamic traffic flow-based operation programming model by constructing a spatio-temporal service network to describe the technical operation of wagons based on daily traffic flow; For the study of empty train deployment problem, Zheng et al. [5] established an empty train deployment optimization model based on Spatio-Temporal network; Wang [6] established a model for empty car deployment with different traction weights based on line traction weight differences. Zhang et al. [7] proposed the application of quantum optimization in the prediction of intelligent urban transportation networks. Kang et al. [8] used cache optimization methods in cloud-based communication systems to reduce network traffic. The current literature research can provide some theoretical guidance for this paper on the dynamic preparation and optimization technology research of the Shenhua Railway cargo train operation program, but the research is not yet comprehensive and there are certain shortcomings: (1) lack of research on the optimization method of heavy-duty railroad cargo train operation program preparation, and the current research focuses on the problem of cargo train operation program preparation under the road network conditions, while the research on the cargo train operation program of freight lines such as heavy-duty railroad. The current research focuses on the preparation of freight train operation program under the road network conditions, but not enough research on freight train operation program for heavy-duty railways and other freight lines. (2) Neglecting the influence of collaborative optimization of heavy and empty train organization on meeting the dynamic transportation needs of cargo owners, the traditional cargo train operation scheme cannot adapt to the dynamically changing cargo transportation needs, nor can it guarantee the time efficiency of cargo owners’ transportation. In order to meet the cargo owners’ transport

14 Optimization Principle of Freight Train Operation Plan for Shenhua …

191

demand and guarantee the arrival time, the synergistic preparation and optimization of the heavy train operation scheme and the empty train deployment scheme should be considered. Therefore, the purpose of this paper is to study the optimization principle of the dynamic freight train operation plan of Shenhua Railway based on the dynamic changing cargo transportation demand, to solve the problem of the dynamic organization of heavy trains and dynamic deployment of empty trains, to deeply explore the transportation capacity of Shenhua Railway, to protect the cargo owners’ transportation demand, and to improve the preparation efficiency and implementation efficiency of Shenhua Railway transportation plan.

14.2 Dynamic Planning Type Transportation Organization Model of Shenhua Railway Cargo transportation organization is an important element of railroad transportation organization, which organizes cargo transportation in the form of whole trains from the place of loading to the place of demand by means of corresponding transport organization theories, rules and methods, etc. When organizing cargo train transportation operations at each station, the train operation section is determined according to the grouping content, and the departure time is reasonably determined to organize the train operation on line.

14.2.1 Deficiencies of the Existing Transport Organization Model The existing transport organization model has weak adaptability to dynamic changes at the level of demand guarantee and cannot guarantee the shortage of the delivery period. Due to the large changes in the demand for goods in the cargo transportation market, the traditional transport organization plan has been unable to effectively guide the organization of cargo transportation for changes in cargo and traffic flow. Moreover, it is unable to effectively regulate the railroad transport plan from the level of changes in the freight market demand, which weakens the competitiveness of the railroad freight market. And due to the large span of the plan preparation level and weak guidance, the volatility of the cargo transportation market has a large impact on the number of starting and ending trains and the direction of the train grouping. Relying on on-site dispatching and command work alone, there is a large randomness in the number of trains and traffic succession, and the on-site traffic flow cannot be connected normally. At the level of organization and management, there are problems of poor professionalism in transportation planning and uncoordinated organization and management mode. National Energy Group controls the whole process of transportation

192

M. Wang et al.

organization at the macro level, and the current plans are basically prepared and implemented by the Group’s production command center, so the plans are not very professional and cannot be implemented to the level of train and train flow, which cannot ensure the reasonable connection of train operation among railroad companies. The information communication between production and dispatching departments at the grassroots level is not timely, and it is impossible to understand the operation process of each department in real time, resulting in the unreasonable matching of loading demand and loading equipment, which seriously affects the efficiency of transportation organization. At the level of operation business, there are also deficiencies of low integration of transportation plan and low utilization of resource capacity. At present, the quality of Shenhua Railway’s transportation plan preparation is not high, the integration of the railroad transportation plan is poor, the utilization of basic information such as traffic flow information and equipment capacity is weak, and the ability to coordinate cargo flow, traffic flow, and locomotives is lacking, which leads to the lack of fineness of transportation plan and the inability to use the plan to guide production operations and precisely guide transportation production work.

14.2.2 Characteristics of Shenhua Railway Transport Organization Operation As a heavy-duty railroad for coal transportation, Shenhua Railway has the following characteristics in its transportation organization operation. (1) The cargo category is relatively single and the road network structure is relatively simple. The upstream network integrates the scattered coal mining resources and transports them to power plants and ports along the downstream network, organically linking the western coal mining areas with the eastern coastal ports. (2) Fixed undercarriage circulation transportation and frequent empty and heavy trains succession. Shenhua Railway operates heavy-duty trains with “column” as the basic unit and adopts the way of fixed undercarriage circulation. After the empty cars are loaded in the loading area, they are transported in the form of heavy trains at the demarcation point—Shenchi South Station, and are gradually transported to the port and factories for unloading. Heavy trains enter the heavy transport channel in the opposite direction in the form of empty trains after unloading and are reasonably distributed to the loading station according to the loading demand. (3) The capacity of heavy-duty line is tight, and 10,000-ton trains are mainly operated. Due to the influence of the station arrival and departure line and other facilities and equipment capacity, there is a certain difference in the traction quota of trains on each line. Some stations of Shenhua Railway have technical conditions for loading and unloading, receiving, and dispatching of 10,000ton trains as well as combined decomposition operations, and mainly operate

14 Optimization Principle of Freight Train Operation Plan for Shenhua …

193

10,000-ton combined heavy-duty trains in order to maximize the use of line capacity. Although Shenhua Railway has adopted such methods as changing locomotive crossings and establishing direct transportation with fixed undercarriage circulation to strengthen transportation organization and management, the capacity mining in resource management is not comprehensive enough, and the use of its resource efficiency is not reasonable enough, resulting in a low utilization rate of railroad transportation capacity.

14.2.3 Dynamic Planning Type Transportation Organization Model of Shenhua Railway Traditional cargo transportation organization has the requirements of full axle and full length, and cargoes with high added value cannot meet the loading demand in a short time, resulting in delayed arrival of cargoes. Market-oriented dynamic planning type transport organization mode [9] is oriented to the transport market, serving transport customers as the goal, developing cargo transport marketing strategies for the whole process of cargo transport organization. Based on the train operation diagram, train loading operations are organized to ensure integrated services from loading, transportation to unloading of cargo and ensure reasonable connection of each operation in the transportation process. This provides shippers with accurate freight status and train tracks, ensuring safe and timely delivery of cargo transportation services. By analyzing the deficiencies of the existing transportation organization plan of Shenhua Railway in terms of the lack of prominent linkage between production, transportation, and marketing functions, the loose overall transportation organization process, and the inefficiency of transportation organization plan preparation, this section proposes a dynamic planning type freight train transportation organization model of Shenhua Railway, as shown in Fig. 14.1. Based on the basic transportation plan issued by the National Energy Group, this model prepares the dynamic planning transportation organization plan of Shenhua Railway based on the daily dynamic cargo transportation demand and capacity resources of Shenhua Railway and the actual transportation demand in the past and reflects the dynamically changing transportation demand on the cargo train operation plan and train operation chart in real time to guide the actual train operation and station dispatching command. This guides the actual train operation and station dispatching command. The traffic dispatcher and planning dispatcher adjust the implementation plan of cargo train operation plan and train operation chart by controlling the on-site dispatching command and actual train operation, so as to control the whole process of cargo in the transportation organization.

194

M. Wang et al.

non-coal integrative non-integrative daily plan of car loading

Statistics and analysis of transportation operations

train operation station operation

direct

Dynamic Freight Transport Demand

Dynamic planning freight train operation plan of Shenhua railway

Provide feedback information

dispatching management

perform

Combine the working plan of railway vehicle

Dynamic planning train diagram of Shenhua railway

perform

Fig. 14.1 Dynamic planning type transportation organization model of Shenhua railway

14.3 Optimization Principle of Freight Train Operation Plan for Shenhua Railway 14.3.1 Optimization Principles for the Preparation of Dynamic Cargo Train Operation Plan for Shenhua Railway The freight train operation plan is not only a key plan for organizing cargo transportation, but also an important plan that affects cargo owners’ loading demand and freight time frame. Based on the dynamic planning transportation organization mode of Shenhua Railway, this paper proposes the concept of Shenhua Railway’s cargo train operation plan, which is based on the daily loading plan issued by the group headquarters, combined with the daily changing dynamic freight demand, obtaining the real-time transportation production data of Shenhua Railway, summarizing and projecting the data of heavy and empty traffic flow at stations and in transit, and considering the train combination decomposition, traffic flow projection, and empty train deployment operation on this basis. On the basis of this, we consider train combination decomposition, traffic projection, and empty train deployment and collaborate to prepare and optimize Shenhua Railway’s dynamic heavy train operation plan and empty train deployment plan and determine the type of heavy and empty train operation, operation section, number of trains, and operation time. The factors influencing the preparation and optimization of the dynamic freight train operation plan of Shenhua Railway include cargo flow attributes, operation capacity, and the number of empty cars. Shenhua Railway has a single type of cargo, mainly self-produced coal, supplemented by purchased coal, and the direction of cargo flow is relatively single, and the direction of heavy and empty trains is clear,

14 Optimization Principle of Freight Train Operation Plan for Shenhua …

195

so the attributes of cargo flow and traffic flow are easier to grasp than those of ordinary railway networks. The daily cargo flow changes dynamically with the transport demand of the enterprise, and the cargo flow at each time of the day also shows unbalanced and dynamic characteristics. The Shenhua Railway network has a simple structure, but the number of loading points is large and widely distributed, and the number of trains from each loading point at each time is limited by the loading stations and line capacity. Moreover, Shenhua Railway mainly operates combination trains of 10,000 tons, and the number of combination trains is affected by the combination operation capacity of technical stations. Cargo train operation involves the transport organization operation of heavy and empty trains. If there is an imbalance in the supply of the number and type of empty trains between the loading demand and the unloading end, the heavy and empty trains cannot be connected normally, which affects the implementation of the basic cargo train operation plan and cannot be used to guide the actual transport production work. Considering the above factors, the following principles are proposed to optimize the preparation of the dynamic cargo train operation program of Shenhua Railway. (1) Meet freight demand. The primary principle of Shenhua Railway’s freight train operation plan is to meet the freight demand. Based on the changing freight demand, we can grasp the dynamic cargo flow and traffic data in real time, combine the current operating capacity of each line section and station, reasonably deploy empty cars according to the cargo transportation demand, and realize the dynamic preparation of cargo train operation program based on the change of cargo flow and empty car capacity resources to meet the cargo transportation demand of cargo owners. (2) Guaranteed arrival time. The preparation of Shenhua Railway’s cargo train operation plan should consider the requirements of different cargo owners’ arrival time and reduce the consumption of technical operation time during freight transportation by accelerating the turnover of cargo trains, so as to guarantee the arrival time of cargo owners. When arranging the operation of heavy and empty trains, if the line capacity is relatively tight, the utilization rate of line capacity is improved by operating combined 10,000-ton trains; otherwise, ordinary unit trains are operated to reduce the consumption of technical operation time and guarantee the arrival time of cargoes. (3) Improve economic efficiency. Based on the operation mode of integration of production, transportation, and marketing of National Energy Group, the positioning of the Shenhua Railway transportation segment in the integration chain is clarified, and the link function of Shenhua Railway is brought into play to ensure the smooth flow of production and marketing channels. By preparing a scientific and reasonable cargo train operation plan, we can ensure the completion of cargo transportation tasks with low cost and high efficiency and improve the economic efficiency of Shenhua Railway’s transportation.

196

M. Wang et al.

14.4 Dynamic Cargo Train Operation Programming Optimization Ideology for Shenhua Railway The cargo flow demand of Shenhua Railway changes dynamically on a daily basis, and its cargo flow and cargo flow demand points are uncertain in time, and a large amount of traffic is assembled upstream of the line, which seriously affects the succession operation of trains. The supply of empty cars is influenced by the loading and unloading operation plan and is closely related to the unloading of heavy cars at this stage and the quantity and status of empty cars carried over from the previous stage of decision-making, showing non-equilibrium and dynamic characteristics during the whole day. Due to the lack of empty car deployment plan, there is an unbalanced and mismatch between loading demand and empty car supply, and the unreasonable succession of heavy and empty cars leads to the cargo train operation plan that cannot guide the actual transportation production work. The preparation of a dynamic freight train operation plan requires that the contents of heavy train operation type, section, quantity, and time period match with the actual cargo flow, and at the same time, it needs to match with the available quantity and location of empty trains in the dynamic unbalanced period of preparation. The preparation of Shenhua Railway’s dynamic cargo train operation plan should reflect two major characteristics. First, it should ensure the dynamic and high timeliness of the cargo transportation market and grasp the dynamic cargo transportation information in time. By considering the daily dynamic change of cargo transportation demand, it can achieve accurate grasp of freight orders, cargo flow, traffic flow, and capacity resources in the decision-making cycle; secondly, it can improve the fulfillment rate of cargo train operation plan and play the guiding role of cargo train operation plan to the actual transportation work. In order to avoid the situation of “flow without cars”, it is necessary to coordinate the relationship between the loading demand and the deployment of empty cars and to consider the organization of heavy cars and the deployment of empty cars in cooperation. The preparation and optimization of the dynamic freight train operation plan are based on the dynamically changing freight transportation demand in the decisionmaking cycle market, timely grasping the dynamically changing cargo flow and traffic information, and preparing Shenhua Railway heavy train operation plan according to the dynamically changing transportation demand. On the basis of dynamic mastering of transportation supply empty car quantity, transportation capacity, and other capacity resources, it determines the type, quantity, and time period of empty car demand at each loading station according to the heavy car operation scheme, makes dynamic allocation of empty cars according to the loading demand at the time and space level, and prepares empty car allocation scheme. And according to the actual demand changes and the matching relationship between heavy and empty trains, it cooperates to optimize the empty and heavy train operation plan, adjusts the empty and heavy train operation mode, and ensures the reasonable succession of empty and heavy trains, so as to adapt to the dynamics of transportation market and guarantee the arrival time of cargo owners.

14 Optimization Principle of Freight Train Operation Plan for Shenhua …

197

14.4.1 Optimization Process of Dynamic Freight Train Operation Plan for Shenhua Railway The freight train operation plan is prepared with the starting point of meeting the dynamic cargo flow demand and can be dynamically adjusted according to the changes in cargo flow demand and capacity resources to ensure that the operation plan meets the cargo flow demand. In addition, the heavy train operation plan should be prepared and optimized in cooperation with the empty train deployment plan, and the empty train should be dynamically deployed to meet the loading demand at the time and space level, and the heavy and empty train should be reasonably connected by dynamically adjusting the operation plan, so as to ensure the implementability of the freight train operation plan. Therefore, according to the technical requirements for the preparation and optimization of dynamic freight train operation plan, a freight train operation plan can be formulated to meet the characteristics of Shenhua Railway’s transportation organization. The following steps should be followed in the preparation and optimization of dynamic freight train operation scheme of Shenhua Railway, and the specific process is shown in Fig. 14.2. First, based on the daily dynamic cargo transportation demand and loading day plan, after loading according to the station operation plan, the dynamic projection of heavy train flow information from the loading station at the collection end, according to the constraints of loading capacity, line transportation capacity, and combined station operation capacity, train categories are determined and vehicles are matched, and the general direct train operation plan and combined 10,000-ton train operation plan are prepared, respectively. Then, based on the heavy train operation plan, determine the loading station loading vehicle category, quantity, and time requirements, project the empty train situation at each station of the line, including the source, type, quantity, and locomotive information, combine the transport capacity and loading station loading demand, dynamically determine the type, destination, and quantity of empty train operation at each time and the combined decomposition operation technology station, and prepare the empty train deployment plan. Finally, the empty train deployment plan in the decision cycle is matched with the initial heavy train operation plan, and the heavy train operation plan is optimized according to the succession of heavy and empty trains to form a new freight train operation plan, and the matching test between the heavy train operation plan and the empty train deployment plan is conducted again. The specific heavy and empty train scheme matching principle is shown in Fig. 14.3. The following situations may occur in the process of matching the heavy train operation scheme and the empty train deployment scheme based on the dynamic cargo flow preparation: firstly, the change of cargo flow affects the section of heavy train operation; secondly, the quantity, time period, and type of the deployed empty train affect the type, quantity, and time period of the heavy train operation and other contents. Therefore, based on the matching result of heavy and empty car demand, the cargo train operation plan should be optimized from the perspective of cargo

198

M. Wang et al. start

Obtain the initial freight train operation plan

Master dynamic daily cargo flow

Whether the empty cars stored at the station are matched

Load according to the loading plan

N

Calculation of departure flow of loading station

Calculate empty car flows

Whether to combine and decompose?

Whether to combine and decompose?

N

Direct loaded car

Y

Combined decomposition loaded car

N

Y

Combined decomposition empty car

Direct empty car

Assign empty cars according to loading requirements Determine the train operation section

Compile the initial operation plan

Determine the type of train operation Determine the number of train operation Determine the train operation period

Y

Compile empty car allocation plan

Whether the allocation of empty cars meets the demand N

Optimize the initial freight train operation plan

Y

Implement the initial freight train operation plan

Fig. 14.2 Shenhua Railway dynamic cargo train operation plan preparation optimization process

flow demand and the deployment of empty cars, and the corresponding optimization countermeasures are as follows: (1) adjusting the train operation section according to the change of cargo flow direction; (2) adjusting the train operation quantity and time period according to the change of cargo flow time and flow; (3) when the arrival quantity of empty cars does not meet the overall loading demand, according to the loading plan, (3) when the number of empty cars arriving does not meet the overall loading demand, the loading plan is adjusted according to the approval order of the loading plan and the arrival period of the cargo owner, and the number of trains running is adjusted; (4) when the time and type of empty cars arriving do not meet the loading demand, the time and type of trains running are adjusted.

14 Optimization Principle of Freight Train Operation Plan for Shenhua …

operation plan of loaded car

Train operation section

Flow flow direction flow duration

Type of train operation

Number of empty cars arriving

Number of train operation

Empty car arrival time

Train operation period

Type of empty car arrival

199

cargo flow

Empty car allocation scheme

Fig. 14.3 Matching Principle of Shenhua Railway Cargo train operation plan

14.5 Conclusion This paper focuses on the optimization principle of dynamic cargo train operation programming of Shenhua Railway. Firstly, it analyzes the characteristics of the cargo transportation organization operation of Shenhua Railway and proposes the dynamic planning type transportation organization mode of Shenhua Railway in combination with the market-oriented dynamic planning type transportation organization mode. On this basis, the concept of dynamic freight train operation plan for Shenhua Railway is proposed. Through analyzing the feasibility, key factors, and main principles of dynamic preparation and optimization of the freight train operation plan of Shenhua Railway, the idea and process of preparation and optimization of heavy train operation plan and empty train deployment plan are proposed. Funding This research was supported by the National Natural Science Foundation of China (Project No. 52072314; 52172321;52102391), Sichuan Science and Technology Program (Project NO. 2020YJ0268; 2020YJ0256; 2022YFH0016; 2021YFQ0001; 2021YFH0175), Science and Technology Plan of China Railway Corporation (Project No.: 2019F002), China Shenhua Energy Co., Ltd. Science and Technology Program (Project No.: CJNY-20-02), China Railway Beijing Bureau Group Co., Ltd. Science and Technology Program (2021BY02, 2020AY02), and Key science and technology projects in the transportation industry of the Ministry of Transport (2022-ZD7-132).

References 1. Xiaowei, L., Qipeng, Y., Shaoquan, N., et al.: Railway freight transportation service network design problem with empty car distribution. J. Transp. Syst. Eng. Inf. Techno. 21(02), 180– 188+204 (2021) 2. Xiaowei, L., Ge, Q., Shaoquan, N., et al.: Freight transportation service network design problem with unbalanced spatiotemporal distribution of Wagon-flow. J. Transp. Syst. Eng. Inf. Technol. 19(02), 137–145+159 (2019)

200

M. Wang et al.

3. Xinyang, W.: Study on the operation scheme of railway freight train based on dynamic flow. Southwest Jiaotong University, Chengdu (2018) 4. Shengdong, L., Hongxia, L., Miaomiao, L., et al.: Daily dynamic freight train service optimization. J. Transp. Syst. Eng. Inf. Technol. 20(05), 177–184 (2020) 5. Kexin, Z., Rui, S., Guangye, L.: An optimization model of dynamic allocation of empty railway cars based on time-space network. J. Transp. Inf. Safety 39(2), 145–152 (2021) 6. Yong, W.: Research on empty car distribution of enterprise coal transportation channel. Beijing Jiaotong University, Beijing (2014) 7. Fuquan, Z., Tsu-Yang, W., Yiou W., et al.: Application of quantum genetic optimization of LVQ neural network in smart city traffic network prediction. IEEE Access 8, 104555–104564 (2020) 8. Lanlan, K., Ruey-Shun, C., Yeh-Cheng, C., et al.: Using cache optimization method to reduce network traffic in communication systems based on cloud computing. IEEE Access 7, 124397– 124409 (2019) 9. Wei, C.: Research on the theory and technology of market orientation dynamic freight train diagram. Southwest Jiaotong University, Chengdu (2020)

Part II

Smart Vehicular Electronics, Networks, and Communications

Chapter 15

Resource Recovery Vehicle Picking Up Resource Recovery Bin Robot Arm Structure Design Yu-Yang Yuan, Yi-Jui Chiu, Wen-Qi Yang, and Yung-Hui Shih

Abstract In recent years, domestic waste is increasing day by day, which puts forward higher requirements for waste recycling capacity. In order to make the garbage collection process more efficient, the structure of the garbage collection manipulator is optimized. Firstly, the simulation software UG nx12.0 is used for three-dimensional modeling, and then the virtual simulation software ADAMS is used for kinematic simulation analysis. Finally, the finite element static analysis and modal analysis are carried out for the whole manipulator. The results show that the optimized manipulator has no resonance in the recovery process and operates stably.

15.1 Introduction With the rapid development of automation in vehicles [1, 2], the recovery efficiency of a resource recovery bin is closely related to the structural performance of the recovery robotic arm and its movement trajectory. A well-designed robotic arm structure and a reasonably planned movement trajectory can significantly reduce the labor intensity of workers and improve the recovery efficiency. Lee et al. [3] proposed a virtual spring damper (VSD) based on the hydraulic robotic arm to improve the positioning accuracy and used a dual VSD controller to realize the position of the robotic arm control. The results showed that the positioning accuracy of the robotic Y.-Y. Yuan · Y.-J. Chiu (B) · W.-Q. Yang School of Mechanical and Automotive Engineering, Xiamen University of Technology, No. 600, Ligong Rd, Xiamen 361024, Fujian Province, China e-mail: [email protected] Y.-Y. Yuan e-mail: [email protected] W.-Q. Yang e-mail: [email protected] Y.-H. Shih Department of Materials Science and Engineering, I-Shou University, Kaohsiung 840, Taiwan e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_15

203

204

Y.-Y. Yuan et al.

arm was effectively improved and verified. Kim et al. [4] proposed a three-degreeof-freedom balancing mechanism based on a double parallelogram mechanism for the gravitational torque of the robotic arm itself and the output payload, which effectively reduced the torque required to support the weight of the robot. Chen et al. [5] proposed a lightweight design method for the truss structure. The parameters related to the structural components of the robot arm were optimized by discrete optimization. Then, the material distribution of the truss structure was optimized by topology optimization under multiple working conditions. Finally, the thickness of the truss structural members is optimized by discrete optimization under multiple working conditions. The optimization results achieved the performance optimization objective of the intelligent sanitation vehicle and proved the feasibility of the proposed lightweight design method. Choi et al. [6] presented a system with a linear output tracking control trajectory planning method. The results show effective validation of the robotic arm trajectory derived from the proposed method through simulation and experimental studies using a single link. Bian et al. [7] proposed a method for vibration control through redundant decomposition for self-motion that satisfies the damping requirements, revealing its additional optimization capability. Simulation results demonstrate the effectiveness of the strategy. Dinh et al. [8] proposed a robust control method based on a disturbance observer for position and torque tracking control subject to external disturbances and parameter uncertainties. The effectiveness of the designed control algorithm was verified by numerical simulations. Zhang [9] proposed a master-slave flow limiting control scheme hydraulic manipulator for planning end actuator speed to meet the dynamic limitation. The results show that the control scheme reduces the velocity error, avoids manipulator vibration, and improves performance. In this paper, UG was used to design the recovery robotic arm structure and ADAMS was used to simulate the dynamics, and then the finite element module of UG NX12.0 was used to perform static and modal analysis. Finally, the designed robotic arm is calculated and analyzed to meet the requirements of the actual working conditions.

15.2 Robotic Solution Development and Working Principle Analysis In the process of garbage collection and transportation, the manipulator is the key part of the process, so the structure design of the manipulator has a significant impact on garbage collection and transportation. One of the key points of the design is to improve the efficiency of garbage collection and transportation, minimize the time required for the manipulator to aim at the garbage can, and achieve a stable combination of the garbage can and the manipulator.

15 Resource Recovery Vehicle Picking Up Resource Recovery Bin Robot …

205

Fig. 15.1 Mechanical arm structure diagram

15.2.1 Structure Design of Manipulator Considering the structure of the garbage truck box and the position of the garbage dump entrance on the roof, combined with the overall functional requirements of the garbage collection and transportation manipulator, a four-degree-of-freedom mechanical arm structure is proposed. The mechanism diagram is shown in Fig. 15.1. The lifting arm can move under the drive of the lifting hydraulic cylinder, complete the action of the manipulator contacting the garbage bin and lifting the garbage bin, and create working conditions for the garbage dumping link. Under the action of the oil motor, the oil sleeper on the movable slide rail can be moved by chain transmission or gear rack transmission to realize the movement of the sleeper, so as to reach the effect of the extension and recovery of the sleeper. The small motor of the manipulator realizes the small range movement of the manipulator claw along the direction of the lifting arm in order to carry out subtle adjustments in the clamping.

15.2.2 Design of End Grab Scheme The function of the mechanical claw at the end of the manipulator is mainly to grasp the garbage bucket, clamping and carrying, and complete the garbage transfer and collection process. By comparing and analyzing the advantages and disadvantages of the two manipulator design schemes, as shown in Table 15.1, the end gripper of the claw manipulator can be connected with the four-degree-of-freedom manipulator, which has good usability. Therefore, the claw scheme is adopted.

206

Y.-Y. Yuan et al.

Table 15.1 Comparison of advantages and disadvantages between hanging bucket type and claw type Grab selection

Merit

Defect

Hanging bucket manipulator Simple structure, low failure rate, easy operation, and low manufacturing cost

Claw-holding manipulator

The relative position of the vehicle and the bucket is highly required, and it is not easy to coordinate the manipulator with the garbage bucket

High degree of freedom, good The structure is complex and the adaptability, and easy to production design cost is high garbage bin and manipulator alignment

15.3 Kinematics Analysis of Manipulator In this chapter, the dynamics and kinematics simulation of the manipulator are carried out based on the Adams virtual prototype platform to meet the analysis of the motion law of the manipulator under three working conditions, which provides an important basis for the selection and model determination of hydraulic cylinders and rotary cylinders and other equipment, and also provides important load data for the static characteristics analysis of the manipulator.

15.3.1 Basic Theory of Kinematics 1. The degrees of freedom of components can generally be calculated by the following methods: F = 6n −

n  i=1

pi −

x 

qi −



Rk

(15.1)

j=1

where n is the number of moving components, pi is the number of constraints of the ith kinematic pair, m is the total number of kinematic pairs, q j is the number of constraints of the jth prime mover, x is the total number of prime movers, and Rk is the number of other constraints. 2. The position vector of a point P in the local coordinate system in the ground coordinate system can be determined by the following formula: r p = r + As 



p

(15.2)

where s p is the coordinate constant vector of point P in the local coordinate system and A is the direction cosine matrix of the local coordinate system with the ground coordinate system as the reference. By deriving the above formula

15 Resource Recovery Vehicle Picking Up Resource Recovery Bin Robot …

207

Fig. 15.2 External load of manipulator virtual prototype model

with time as a variable, the velocity calculation formula can be obtained: r



p





=r +As



p

∼



= r + w As



p

(15.3)

where w˜ is the angular velocity of the local coordinate system. The acceleration calculation method can be obtained by deriving the above formula again with time as the variable: r



p





=r +A r



P

(15.4)

15.3.2 Solid Modeling of Manipulator In terms of modeling, professional modeling software (such as SW, UG, or Pro/E) is used to model the parts of the manipulator and assemble the manipulator. The assembly diagram is saved in Parasolid (*.x _ t) format and imported into Adams. The results are shown in Fig. 15.2.

15.3.3 Motion Analysis of Manipulator Gripper Clamping Condition During the clamping process of the manipulator, the left and right mechanical grippers are clamped inside under the action of the clamping hydraulic cylinder. The working stroke of the hydraulic cylinder is set to be 20 mm, and the total clamping time is

208

Y.-Y. Yuan et al.

5 s. The motion trajectory of the left and right grippers in the clamping condition of the manipulator is simulated. Since the manipulator does not contact the trash can in the clamping condition, the manipulator until it receives the influence of gravity is in the direction of −Z-axis. After processing according to the simulation results, the centroid motion displacement and velocity curves of the two grippers are shown in the following four Figs. 15.3, 15.4, 15.5, and15.6. From the curves of displacement and velocity, it can be seen that the displacement and velocity curves of the left and right grips in the X-axis direction are roughly

Fig. 15.3 X-Direction displacement curve of center of mass of left and right grip of manipulator

Fig. 15.4 Y-Direction displacement curve of center of mass of left and right grip of manipulator

Fig. 15.5 X-Direction velocity curve of center of mass of left and right grip of manipulator

15 Resource Recovery Vehicle Picking Up Resource Recovery Bin Robot …

209

Fig. 15.6 Y-Direction velocity curve of center of mass of left and right grip of manipulator

coincident, and the displacement and velocity in the Y-direction are equal, and the direction is opposite. The change of the whole clamping speed is relatively uniform, and the movement is relatively stable.

15.3.4 Motion Analysis of Manipulator Lifting Condition When the manipulator claw clamps the garbage bucket, the clamping condition ends, the manipulator turns to the lifting condition, the lifting hydraulic cylinder begins to shrink, the lifting arm moves upward, and the garbage bucket is lifted from the ground to the specified height. In order to facilitate the study of the stability in the lifting process, the manipulator grabs and the lifting arms are Boolean summation, and the centroid coordinates of the new components formed after the summation are found by the quality summation command. The motion of the centroid is observed by post-processing the simulation result data, as shown in Fig. 15.7. As can be seen from Fig. 15.7, in the lifting process, due to the existence of the lifting guide device, the front and rear left and right displacement of the flip arm and the manipulator is small in the XOY-plane relative to the displacement of the Z-direction. The main motion is to drive the lifting arm upward along the Z-axis

Fig. 15.7 Y-Displacement curve of centroid of manipulator lifting arm in all directions

210

Y.-Y. Yuan et al.

Fig. 15.8 Velocity acceleration curves of hoisting arm center of mass in different directions

under the action of the lifting hydraulic cylinder. The acceleration image is shown in Fig. 15.8. It can be concluded from Fig. 15.8 that the maximum acceleration in the lifting process is 225 mm/s2 , which occurs at the moment when the hydraulic cylinder starts and stops. The speed of the lifting process reaches the maximum value of 285 mm/s in the middle of the process, and the direction is Z-axis. The movement in other directions is small, and the lifting process is relatively stable.

15.3.5 Kinematics Analysis of Manipulator Turnover When the end of the manipulator turns over, the steering hydraulic cylinder contracts and the steering arm rotates. The lifting arm, mechanical gripper, and garbage can be rotated at a certain angle together. The centroid coordinate of the manipulator is set as the mark point for applying force along the −z-axis, with a size of 1300 N. The motion curve of the point in the X-, y-, and Z-directions is shown in Fig. 15.9. It can also be seen from Fig. 15.9 that the working range of the manipulator is 3250 mm in the Z-axis direction and 300–1100 mm in the X-axis direction. Then we simulate the speed of the centroid point of the mechanical gripper in three directions

Fig. 15.9 Displacement curve of centroid flip of manipulator claw

15 Resource Recovery Vehicle Picking Up Resource Recovery Bin Robot …

211

Fig. 15.10 Speed curve of centroid turnover of manipulator claw

Fig. 15.11 Acceleration curve of centroid turnover of manipulator claw

and the rotation angle of the manipulator. The results are shown in Figs. 15.10 and 15.11. It can be seen from Fig. 15.11 that there is no acceleration mutation in the center of mass of the manipulator, and the flip process is relatively stable. Next, we study the relationship between the central angular velocity and the angular acceleration of the flip arm, as shown in Fig. 15.12. As can be seen from Fig. 15.12, since the flip arm is driven by a hydraulic cylinder, the angular velocity is relatively fluctuating. The angular velocity increases first and then decreases, reaching the maximum value of around 1.5 s. At this time, the velocity

Fig. 15.12 Angular velocity and angular acceleration curve of flip arm

212

Y.-Y. Yuan et al.

Fig. 15.13 Speed acceleration curve of turnover hydraulic cylinder

of the manipulator in the Z-axis direction also reaches the maximum value. After 1.5 s, the flip speed slows down until the end of the flip condition. However, the acceleration changes abruptly at the beginning and at the end, which is a normal shock phenomenon, which is also related to the set driving function. The acceleration state of the flip hydraulic cylinder is similar to that of the lifting hydraulic cylinder, which is accelerated first and then decelerated. The velocity acceleration curve of the flip hydraulic cylinder is shown in Fig. 15.13.

15.4 Static and Modal Analysis of Mechanical Arm of Garbage Truck In order to ensure the rigor and reliability of the analysis, finite element analysis was carried out on the robot arm under three working conditions, respectively, so as to determine the deformation diagram and stress cloud diagram of the robot arm under different forms of action, to find out the danger areas within the robot arm, and to determine whether the structure meets the design requirements in conjunction with modal analysis.

15.4.1 Manipulator in Bucket Holding Condition The deformation and stress of the manipulator when clamping the bucket are shown in Fig. 15.14. It can be seen from Fig. 15.14a that the maximum deformation value of the manipulator is 23.51 mm when the garbage bin is just picked up. The maximum deformation position is at the left claw tip of the manipulator gripper structure, and the stress direction is -Z direction. At the same time, it can be found that the deformation of the manipulator component at the rotation center of the manipulator flip arm is small. Figure 15.14b shows that the maximum stress of the manipulator is 85.10 MPa, which is located at the root of the lifting arm guide device. The structure is easy to form stress concentration.

15 Resource Recovery Vehicle Picking Up Resource Recovery Bin Robot …

213

Fig. 15.14 Deformation nephogram and stress nephogram of bucket holding condition

15.4.2 Manipulator Lifting Condition The finite element method is used to carry out the static analysis of the working condition when the manipulator lifts to the highest point, and the deformation and stress nephogram under this working condition are obtained, as shown in Fig. 15.15. It can be seen from Fig. 15.15a that the maximum deformation of the manipulator is 2.689 mm in the lifting condition, and the force direction is −Z at the end of the left claw of the manipulator. At the same time, it can be found that the deformation of the manipulator component at the rotation center of the flip arm is small. Figure 15.15b shows that the maximum stress of the manipulator bucket is 23.50 MPa, and the position is at the root of the guide device of the lifting arm. The reason for this phenomenon is that the L-shaped structure is formed, which is easier to form stress concentration.

Fig. 15.15 Deformation nephogram and stress nephogram of lifting condition

214

Y.-Y. Yuan et al.

Fig. 15.16 Deformation nephogram and stress nephogram of dumping condition

15.4.3 Manipulator Dumping Condition The deformation and stress contours under dumping conditions are obtained by finite element analysis, as shown in Fig. 15.16. Under the dumping condition, the maximum deformation of the manipulator is 25.10 mm, and the maximum stress at the end of the left claw of the manipulator is 208.54 MPa, which is located at the root of the lifting arm guiding device.

15.4.4 Simulation Analysis of Manipulator According to the finite element analysis of the manipulator, the maximum stress value and distribution of the deformation of the manipulator under various working conditions can be obtained by the analysis results, and the position of the maximum stress and deformation point in each working condition can be found, as shown in Table 15.2. Table 15.2 Maximum stress and maximum deformation of manipulator in three working conditions Working condition

Maximum stress (MPa)

Position

Maximum deflection (mm)

Position

Holding

166.9

The corner of the lifting arm

19.00

Front end of left grip

Lifting

168.7

The corner of the lifting arm

15.09

Front end of left grip

Dumping

146.8

The corner of the lifting arm

12.29

Front end of left grip

15 Resource Recovery Vehicle Picking Up Resource Recovery Bin Robot …

215

15.4.5 Modal Analysis of Manipulator Structure 15.4.5.1

Modal Analysis Theory

The natural frequency (NF) and mode shape (MS) of the structure can be obtained by modal analysis. The motion control equation is ..

.

[M]{U } + [C]{U } + [K ]{U } = {F(t) }

(15.5)

where [M] is the mass matrix, [C] is the damping matrix, [K] is the stiffness matrix, {U} is the displacement matrix, and {F (t) } is the force matrix. The derivative form of the equation in its physical sense is m·

dx d2x +c· + k · x = F(t) 2 d t dt

(15.6)

The eigenvalues and eigenvectors of this equation are the NFs and MSs.

15.4.5.2

Structure Modal Analysis

In practical work, the resonant frequency of the structure is generally low-order natural frequency. Therefore, the modal analysis module of UG is used to analyze the first six modal characteristics of the manipulator structure. The first six modal frequency characteristics are listed in Table 15.3. The vibration characteristics of each order are shown in Fig. 15.17. It can be seen from Table 15.3 and Fig. 15.17 that under conventional working conditions, the manipulator structure is higher than that under external excitation, so the resonance phenomenon of the structure is avoided. The maximum deformation of the manipulator is mostly in the front of the manipulator claw, and there will be a certain rotation phenomenon. Table 15.3 1–6 mode vibration characteristics of manipulator model Mode

Frequency (Hz)

Maximum deformation (mm)

Vibration characteristic

1

24.338

0.503

Left and right grip curled inward

2

26.602

0.555

Left and right grip roll outward

3

29.199

0.083

Outward bending of the chute

4

41.860

0.399

Left and right grip curled inward

5

42.878

0.340

Left and right grip roll outward

6

24.338

0.197

The lifting rod bends outward

216

Y.-Y. Yuan et al.

1 st

2 nd

3 rd

4 th

5 th

6 th

Fig. 15.17 The first ten natural frequencies and vibration modes of manipulators

15.5 Conclusion The main research contents of this paper are the mechanical arm and the claw structure of the side-mounted garbage truck, and the modeling analysis is carried out. Considering the actual work situation, the force of the mechanical arm in the actual situation is analyzed, and the kinematics characteristics of the manipulator grasping the mechanical arm in the movement process are studied. The static finite element analysis and modal analysis are carried out for the moment of the manipulator under several working conditions, so that the manipulator can better complete the actual working conditions. Acknowledgements This project is sustained by the Graduate Technology Innovation Project of the Xiamen University of Technology NO.YKJCX2021030.

15 Resource Recovery Vehicle Picking Up Resource Recovery Bin Robot …

217

References 1. Li, Z., Miao, Q., Chaudhry, S.A., et al.: A provably secure and lightweight mutual authentication protocol in fog-enabled social Internet of vehicles. Int. J. Distrib. Sens. Netw. 18(6) (2022) 2. Chen, C.M., Xiang, B., Liu, Y., et al.: A secure authentication protocol for internet of vehicles. IEEE Access 7, 12047–12057 (2019) 3. Lee, J., Kim, J.T., Kim, H., et al.: Method for improving the position precision of a hydraulic robot arm: dual virtual spring-damper controller. Intel. Serv. Robot. 9(2), 93–99 (2016) 4. Kim, H.S., Song, J.B.: Multi-DOF counterbalance mechanism for a service robot arm. IEEEAsme Trans. Mechatron. 19(6), 1756–1763 (2014) 5. Chen, X.: On the lightweight truss structure for the trash can-handling robot. Actuators 10, 1–17 (2021) 6. Choi, Y., Cheong, J., Moon, H.: A trajectory planning method for output tracking of linear flexible systems using exact equilibrium manifolds. IEEE/ASME Trans. Mechatron. 15(5), 819–826 (2009) 7. Bian, Y., Gao, Z., Yun, C.: Study on vibration reduction and mobility improvement for the flexible manipulator via redundancy resolution. Nonlinear Dyn. 65(4), 359–368 (2011) 8. Dinh, T.X., Thien, T.D., Anh, T.H.V., et al.: Disturbance observer based finite time trajectory tracking control for a 3 DOF hydraulic manipulator including actuator dynamics. IEEE Access 6, 36798–36809 (2018) 9. Zhang, F., Zhang, J., Cheng, M., et al.: A flow-limited rate control scheme for the master-slave hydraulic manipulator. IEEE Trans. Ind. Electron. 69(5), 4988–4998 (2021)

Chapter 16

Reconfigurable Multibody Space Systems Based on Magnetic Flux Pinning Lifeng Zhao, Qingyun Mao, Bo Zhang, Pei Wang, Jun Tao, Haige Qi, Jin Jiang, Yong Zhang, and Yong Zhao

Abstract In this paper, a space satellite cluster manipulation device based on the interaction of high-temperature superconductors (HTS) and permanent magnets (PM) is designed. The device can maintain the mutual spatial positions and attitudes of different satellites in a self-stabilizing manner without fuel consumption. It can also be manipulated to change their mutual position to achieve arbitrary changes in the satellite cluster configuration.

L. Zhao (B) · P. Wang · J. Tao · H. Qi · J. Jiang · Y. Zhang · Y. Zhao The School of electrical Engineering, Southwest Jiaotong University, Chengdu 610031, China e-mail: [email protected] J. Jiang e-mail: [email protected] Y. Zhang e-mail: [email protected] Y. Zhao e-mail: [email protected] L. Zhao · J. Jiang · Y. Zhang · Y. Zhao Key Laboratory of Maglev Train and Maglev Technology of Ministry of Education, Southwest Jiaotong University, Chengdu 610031, China Q. Mao (B) · B. Zhang Innovation Academy for Microsatellites of CAS, Shanghai 201203, China e-mail: [email protected] B. Zhang e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_16

219

220

L. Zhao et al.

16.1 Introduction Microsatellite cluster flights have promising applications in synthetic aperture telescopes, space communications, etc. [1–3]. In the past 30 years, the development of mobile communications has profoundly changed the world. Combined with artificial intelligence, the development and maturity of 5G technology has made it possible to connect everything intelligently [4–6]. Currently, the combination of terrestrial mobile and communication satellites is the main trend of future communication technology development [7, 8]. As the current satellite communication has the problem of small bandwidth and capacity, using multiple small satellite clusters to increase the bandwidth and signal reception is a possible solution. Such applications require the ability to maintain mutual spatial positions among satellite clusters autonomously, and also the ability to regulate the configuration of microsatellite cluster assemblies. In the conventional way, the adjustment of the satellite configuration is achieved by consuming the fuel carried by the satellite itself, which causes more fuel consumption. Due to the limited fuel-carrying capacity of stars, new techniques for maintaining and manipulating the configuration of microsatellite clusters need to be found in order to extend the lifetime of stars in the orbit as much as possible. Due to the flux pinning effect between the high-temperature superconductor and the permanent magnet, a flux pinning force is generated between them. It has the nature of reversion force, which can make the high-temperature superconductor and the magnet achieve self-stabilizing interaction, that is, when the HTS is field cooled and transformed into the superconducting state, the interaction force between the HTS and PM is zero, which is the initial state; when the distance between them increases from the initial state, the interaction force between them is expressed as an attractive force, and the attractive force will rapidly decrease with the further increase of distance. However, when the spacing between them decreases from the initial state, the interaction force between them is repulsive, and the repulsive force increases almost exponentially with the decrease of distance. This property can maintain the dynamic stability of the mutual spatial position of the system [9–14]. However, the basic configuration change for two stars to rotate 180° relatively to each other cannot be effectively achieved yet by using HTS and PM interaction interfaces at home and abroad. Achieving this configuration change implies the ability to achieve arbitrary combinations between multiple stars. In this paper, we provide a solution to realize this technique and demonstrate its implementation on two airfloating platforms.

16.2 Device Structure and Principle The device includes a PM carrier and a HTS carrier suspended side by side to simulate two stars in orbit, as shown in Fig. 16.1. The PM carrier possesses two pinning PM with the size of ∅30 × 15 mm and a PM shaft with the size of ∅60 × 60 mm, and

16 Reconfigurable Multibody Space Systems Based on Magnetic Flux Pinning

221

Fig. 16.1 a Top view of the simulator for manipulation of microsatellite configuration based on the interaction force between HTS and PM. b Front view of the simulation device. 1, PM carrier; 11, PM shaft; 2, HTS carrier; 21, HTS shaft; 22, liquid nitrogen container; 3, variable magnetic rod; 4, flywheel 1; 5, flywheel 2; 6, initial positioning assembly; 61, pinning PM 1; 62, pinning HTS bulk 1; 63, liquid nitrogen container 1; 71, pinning PM 2; 72, pinning HTS bulk 2; 73, liquid nitrogen container 2; 81, angular limiting device; 9, air floating platform

the superconducting carrier possesses two pinning HTS bulks with the size of ∅30 × 15 mm and a rotary HTS shaft piled up with six HTS bulks in their liquid nitrogen container. For the initial assembly position in Fig. 16.1, the interaction force between the pinning PM 1 and the pinning HTS bulk 1, and the force between the PM shaft and the HTS shaft keep the system self-stable. The PM carrier and the HTS carrier are independently driven to rotate by their flywheels. Figure 16.2 shows the rotation of the HTS carrier on the axis of the HTS shaft at 90° from the initial position in Fig. 16.1, and the rotation of the HTS carrier on the axis of the PM shaft at 90° is shown Fig. 16.3, respectively. During rotation, the interaction force between the PM shaft and HTS shaft keeps the PM carrier and HTS carrier in a self-stabilizing connection. At End-of-rotation position (180°) of the superconducting carrier with respect to the permanent magnet carrier, the interaction force between the pinning PM 2 and the pinning HTS bulk 2 is presented.

16.3 Pinning Force Since the force between the pinning PM and the pinning HTS is smaller than the force between the HTS shaft and the PM shaft. Therefore, the self-stabilizing interaction between the HTS carrier and the PM carrier mainly depends on the interaction effect between the pinning PM and the pinning HTS bulk. The device shown in Fig. 16.4

222

L. Zhao et al.

Fig. 16.1 (continued)

Fig. 16.2 90° rotation of the superconducting carrier with respect to the PM carrier

is thus designed to test the interaction force between the HTS bulk and the PM to evaluate the self-stabilizing performance of this system. As shown in Fig. 16.4, a stepper motor drives the HTS bulk to move close to and far away from the PM at the field-cooled position to test the effect of distance variation on the interaction force. As shown in Fig. 16.4, the superconducting bulk is located at the field cooling positions of 80 mm and 130 mm away from the PM, respectively, and takes the field cooling position as the center, making ±20 mm continuous movement close to and

16 Reconfigurable Multibody Space Systems Based on Magnetic Flux Pinning

223

Fig. 16.3 End-of-rotation position (180°) of the superconducting carrier with respect to the PM carrier 5

4

3

2

1

Fig. 16.4 The device for interaction force measurement

224

L. Zhao et al.

Force (N)

far from the PM at a speed of 1 mm/s to test the influence of the distance variation between the HTS bulk and the PM on the interaction force. The experimental results of the interaction force varying with distance are shown in Fig. 16.5. When the superconductor moves close to the PM from the field cooling position, the force between them is repulsive, and its value is negative; On the contrary, when the superconductor passes through the equilibrium point and moves far away from the PM, the force between them is adsorptive, and its value is positive. Although the acting force at 80 mm field cooling position is obviously greater than that at 130 mm, it is still very small. For space weightless environment, the force stiffness required to maintain the system stability shall not be less than 0.4 N/m. In this regard, considering the data acquisition frequency, we take the derivative of the value in Fig. 16.5 to obtain the result of force stiffness varying with distance. The results for the stiffness of the interaction force are shown in Fig. 16.6. At the field cooled position of 80 mm, the interaction force stiffness values in most data ranges are greater than 0.4 N/m, which means that the system consisted of the pinning HTS and the pinning PM possesses good relative stability. For the 130 mm field cooling position, only the data near the extreme values of the interaction force stiffness are greater than 0.4 N/m, which means that the system will not be separated 0.04 0.02 0.00 -0.02 -0.04 -0.06 -0.08 -0.10 -0.12 -0.14

80mm

0

2000

4000

Time (s)

6000

Force (N)

0.01

8000

130mm

0.00 -0.01 -0.02 0

4000

8000

Time (s)

12000

16000

Fig. 16.5 Results for the interaction force variation with the HTS movement at the field cooling positions of 80 and 130 mm

16 Reconfigurable Multibody Space Systems Based on Magnetic Flux Pinning

80mm

4

Stiffness (N/m)

225

2 0 -2 -4 0

1000

2000

3000

4000

5000

Stiffness (N/m)

Time (s) 130mm

0.5

0.0

-0.5 0

5000

10000

15000

20000

Time (s) Fig. 16.6 Results for the stiffness of the interaction force varying with the HTS movement at the field cooling positions of 80 and 130 mm

from the interaction, but the ability to maintain its attitude is relatively weak. For this disadvantage, the flywheel can be used to make up for it. This is described in more detail below.

16.4 Magnetic Field Design and Rotation Angle Control The magnetic field presented by the PM shaft is partially changed by two variable magnetic rods. The structure of the PM shaft and the variable magnetic rods is shown in Figs. 16.7 and 16.8. The variable magnetic rod consisted of two copper rods and an iron rod assembled with screw. Typical Z component distributions of the magnetic field for the points from W0 to W5 are presented in Fig. 16.9. At the Z of 100, as the intensity of the magnetic field increases from W0 to W5, it increases the revolution resistance of the HTS shaft around the PM shaft. This causes the revolution resistance to be greater than the rotation resistance, so that the rotation of the HTS carrier platform around the HTS shaft is prior to the revolution of the HTS carrier around the PM shaft to the set rotation angle. Then, by continuing to use the

226

L. Zhao et al.

torque generated by the rotation of the flywheel assembly to overcome the revolution resistance, control and adjust the revolution angle of the HTS carrier around the PM shaft, so as to accurately control the revolution angle of the HTS carrier. The maximum angle for the HTS carrier rotating around HTS shaft is set with an angular limiting device. The angular limiting device includes two screw bolts fixed beside the HTS shaft, as shown in Fig. 16.1. During the rotation of the HTS carrier Fig. 16.7 Top view of the PM shaft

Fig. 16.8 Cross-sectional view in the direction of A-A in Fig. 16.4. 31. Iron; 32. Copper

16 Reconfigurable Multibody Space Systems Based on Magnetic Flux Pinning W0 W1 W2 W3 W4 W5

0.16 0.14 0.12

H (T)

227

0.10 0.08 0.06 0.04 0.02 0

50

100

Z (mm)

150

200

Fig. 16.9 Schematic diagram of the variation of magnetic field distribution along the Z-axis direction at each point from W0 to W5 in Figs. 16.4 and 16.5

Fig. 16.10 The experimental demonstration of the device

relative to the HTS shaft, the end of the screw bolt can be respectively butted with its adjacent wall of the liquid nitrogen container for the HTS shaft. Actually, the maximum rotation angle is set as 90° here. The experimental demonstration of the device is shown in Fig. 16.10. The HTS carrier and PM carrier are suspended over two air-floating platforms.

16.5 Summary The flywheels provide the driving force, and the HTS carrier can rotate along the radial magnetic field provided by the PM shaft. By setting the limit of the rotation angle of the HTS carrier around the HTS shaft, and through the interaction between the pinning PM and the pinning HTS, the collision between the PM platform and

228

L. Zhao et al.

the HTS platform during the rotation process is prevented. In addition, at the initial position and the final position, when the flywheel assemblies stop working, the selfstability of the system state is maintained through the interaction of the HTS and PM. By fixing the system to different micro-satellites, the micro satellite configuration can be controlled. In addition, through the dynamic control of the flywheel, the HTS platform can also stably stay at any angle within 0–180° relative to the PM platform. This means that installing the device in a multi satellite cluster can realize the control of any configuration of the multi-satellite cluster.

References 1. Woffinden, D., Geller, D.: Navigating the road to autonomous orbital rendezvous. J. Spacecr. Rocket. 44, 898–909 (2007) 2. Sabol, C., Burns, R., McLaughlin, C.: Satellite formation flying design and evolution. J. Spacecr. Rocket. 38, 270–278 (2001) 3. Hu, Y., Ng, A.: Robust control of spacecraft formation flying. J. Aerosp. Eng. 2021, 209–214 (2007) 4. Wu, T.-Y., Guo, X., Chen, Y.-C., Kumari, S., Chen, C.-M.: Amassing the security: an enhanced authentication protocol for drone communications over 5G networks. Drones 6(1), 10 (2022) 5. Yang, L., Chen, Y.-C., Wu, T.-Y.: Provably secure client-server key management scheme in 5G networks. Wirel. Commun. Mob. Comput. 21, 4083199 (2021) 6. Wu, T.-Y., Lee, Z., Obaidat, M.S., Kumari, S., Chen, C.-M.: An authenticated key exchange protocol for multi-server architecture in 5G networks. IEEE Access 8, 28096–28108 (2020) 7. Lv, W., Lv, W., Li, J., Wang, X.: High-resolution satellite multi-class cloud detection based on improved AlexNet. J. Netw. Intell. 6(2), 189–205 (2021) 8. Altaf, I., Saleem, M.A., Mahmood, K., Kumari, S., Chaudhary, P., Chen, C.-M.: A lightweight key agreement and authentication scheme for satellite-communication systems. IEEE Access 8, 46278–46287 (2020) 9. Zhang, Y.-W., Zhu, Y.-W., Zhu, H.-K.: Spacecraft self- and soft-docking via coupled actuation of magnetic fields. In: Proceedings of the 39th Chinese Control Conference, pp. 6721–6725 (2020) 10. Shawn, B.M., Romano, M.: Ground and space testing of multiple spacecraft control during close-proximity operations. In: AIAA Guidance, Navigation and Control Conference, p. 6664 (2008) 11. McCamish, S.B.: Distributed autonomous control of multiple spacecraft during close proximity operations. Ph.D. dissertation, Electrical Engineering Dept., Naval Postgraduate School, Monterey, CA (2007) 12. Shoer, J.P., Peck, M.A.: Flux-pinned interfaces for the assembly, manipulation, and reconfiguration of modular space systems. J. Astronaut. Sci. 57, 667–688 (2009) 13. Bai, M., Yang, W., Liao, D., Song, D., Tang, H.: Effect of composite superconductor structure on dynamic properties of superconducting interface and formation in space application. J. Supercond. Novel Magn. 33, 599–607 (2020) 14. Shoer, J.P.: Dynamics of reconfigurable multibody space systems connected by magnetic flux pinning. Ph.D. dissertation, Faculty of the Graduate School, Cornell University (2008)

Chapter 17

Research on Supply Chain Financing Mode of New Energy Vehicle Industry Cheng-Xiao Ju, Hui-Jun Xiao, and Mei-Feng Chen

Abstract As new energy vehicles can minimize energy use and environmental harm, both the government and the public have expressed interest in and support for the new energy vehicle industry, which is developing significantly. The market for new energy vehicles is growing and prospering, but it is also struggling with capital limitations. The traditional auto financial service system places restrictions on the financing of the new energy vehicle industry. Additionally, the financing mode is unable to support the requirements of the supply chain for the manufacture of new energy vehicles, and the channel for transmitting supply chain information is constrained. The new energy vehicle industry should aggressively create a financing mode, adopt confirming storage financing mode, accounts receivable financing mode, private equity fund, and financial leasing financing mode to raise funds in order to address the issue of supply chain financing, and support the industry’s wholesome and long-term growth.

17.1 Introduction 17.1.1 Background and Significance of the Research To gradually replace conventional gasoline and diesel automobiles that harm the environment, new energy vehicles employ clean energy batteries as their main power plant. Due to its ability to lessen environmental impact and energy consumption, the new energy vehicle industry has gained the interest and support of government organizations and the public in numerous countries. As a result, it is growing swiftly, which raises the demand for capital. The new energy vehicle industry is now the longest single category in the market, and the majority of the businesses

C.-X. Ju (B) · H.-J. Xiao · M.-F. Chen Bussiness School of Dongguan City University, Dongguan 523419, China e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_17

229

230

C.-X. Ju et al.

are startup businesses. Every component of the supply chain for new energy vehicles is significantly impacted by the capital flow situation. However, given the new energy vehicle industry’s current state of development, financing issues in the supply chain are preventing the promotion of many new energy vehicles. Specifically, in the new energy vehicle supply chain, suppliers, vehicle manufacturers, vehicle sellers, and end users are all subject to financial constraints, with the financing of battery suppliers posing the biggest issue. A new financing mode for the supply chain that is supportive of the new energy vehicle industry’s sustainable development must be found in order to further boost the industry’s growth. This requires a detailed analysis of the challenges the new energy vehicle industry’s financing methods confront.

17.1.2 Research Status According to chain financing, Deng and Zeng [1] supply chain financing refer to a financial business model where banks no longer concentrate on individual firms but rather combine financing firms with core supply chain firms from the perspective of the supply chain and then offer credit support to supply chain firms. Chang et al. [2] looked at the issues with supply chain company credit and economic order quantity under cash discounts and delayed payments. Wesley and Farris [3] argue that the overall profitability of the supply chain can be improved by taking advantage of the cost of capital of supply chain partners and influencing the delayed payment period of accounts through a cash flow cycle strategy. Kunter [4] points out that previous research on supply chain management only considers supply chain synergy and the majority of studies do not take into account the synergy between capital flow and logistics in the case of financial constraints. In these circumstances, supply chain member firms can optimize their interests through supply chain financing and other means. The researchers mentioned above have discussed supply chain financing in depth, primarily from the standpoint of supply chain financing optimization. According to chain financing of automobile industry: Yang and Xie [5] proposal regarding supply chain financing in the automobile industry, order financing, accounts receivable pledge financing, and factoring financing are the primary financing modes for new energy vehicle suppliers, while prepayment financing and confirming storage financing are the primary financing modes for new energy vehicle dealers. Based on a three-stage DEA model, Wang [6] investigated the effectiveness of supply chain finance in the new energy vehicle industry. Tan [7] believed that there are risks associated with the supply chain for new energy vehicles, including credit risk, operational risk, default risk, and structural risk, but these risks can generally be managed. The opinions of the scholars mentioned above demonstrate that supply chain financing generally has advantages over traditional financing modes and can alleviate the financing problems to some extent in the development of the new energy vehicle industry, making it worthwhile to further investigate and employ. Foreign scholars have also conducted research in related fields. About transportation, Chen et al. [8] studied an improved honey badger algorithm for electric vehicle

17 Research on Supply Chain Financing Mode of New Energy Vehicle …

231

charge orderly planning. Zhang et al. [9] studied a traffic prediction method of bicycle-sharing based on long- and short-term memory network. About Vehicles, Kumar et al. [10] studied robust authentication protocol for RFID-based vehicular cloud computing. According to the development status of new energy vehicle digital technology, the Internet of Vehicles (IOV) is a major application of data in the automotive supply chain. Dozens of sensors are installed on the vehicle to connect every change of the vehicle to the access network. Chen et al. [11] believe that through the Internet of Vehicles, financial institutions can locate and remotely control vehicles in real time and obtain data such as vehicle mileage and battery power. From the perspective of risk prevention, the Internet of Vehicles will greatly protect the control right of financial institutions over vehicles and solve the problem of information asymmetry prevailing in supply chain finance. Wu et al. [12] and Chen et al. [13] discussed the cost and security of the Internet of Vehicles in terms of encryption algorithms and authentication protocols.

17.2 Problems of the Traditional Supply Chain Financing Mode in New Energy Vehicle Industry 17.2.1 Financing Sources Are Limited by the Traditional Auto Financial Service System Currently, the traditional auto financial service system still serves as the primary means of addressing the capital requirements for the development of the new energy vehicle industry. The primary channels for doing so are still bank loans or advance sales. The funding method used by producers of new energy vehicles is rather simple, and the framework cannot be maintained. But given the nature of the new energy vehicle manufacturing industry, it is clear that substantial capital expenditure and a protracted period of investment return are necessary. Due to the fact that commercial banks provide the majority of the credit loans used for enterprise production, it is likely that these commercial banks have some influence over the future of new energy vehicle manufacturers. This lack of independence places significant restrictions on manufacturers of new energy vehicles.

17.2.2 The Supply Chain for New Energy Vehicle Manufacture Cannot Be Supported by the Financing Mode The entire new energy vehicle supply chain cannot be financed through bank loans. The single loan mode cannot satisfy the needs of all new energy vehicle manufacturers

232

C.-X. Ju et al.

in the industrial supply chain or maximize the overall benefits of the new energy vehicle supply chain. Although the status of the financial markets is improving and reforms are being strengthened by the government, a standard and complete market model has not yet developed, which in some cases can lead to moral hazard behavior and distort credit relationships in the financial markets. The peculiar occurrence that many new energy vehicle producers are unable to meet their commitments in accordance with the legislation would unavoidably result if the financial information is substantially distorted, and the pressure to lose trust is considerably less than the pressure to be honest. Many of the tasks faced by manufacturers of new energy vehicles simply cannot be accomplished, and advanced financial products will be invalid.

17.2.3 Limited Supply Chain Information Transmission Channels in the New Energy Vehicle Industry The complexity of automotive component parts, as compared to industries like steel and coal, results in a sizable number of participating businesses in the supply chain. Many upstream and downstream small and medium-sized enterprises, in addition to certain large businesses, lack a high level of information technology, and many of their accounts are missing and their deviations are significant, which increases the information asymmetry between financial institutions and financing businesses, and some enterprises lack of convention-implementing awareness and are tardy in disclosing information on enterprises engaged in the manufacture of new energy vehicles. This results in a breakdown in the flow of information between the two sides, and it is unable to resolve the issue of information asymmetry between the bank and the manufacturer of new energy vehicles. Instead, it can only ensure planning for the development of the new energy vehicle manufacturer, and the new energy vehicle manufacturer’s information is unclear and does not play a significant role in the information flow. It increases the threat of the development of new energy vehicle manufacturers.

17.3 Innovative New Energy Vehicle Industry Supply Chain Financing Model In order to compensate for the lack of traditional financing sources, the new energy vehicle industry can develop creative supply chain financing strategies, such as prepayment financing, also known as confirming storage financing, accounts receivable financing, private equity fund financing, and financial leasing financing.

17 Research on Supply Chain Financing Mode of New Energy Vehicle …

233

17.3.1 Prepayment Financing, i.e., Confirming Storage Financing Confirming storage business is a kind of supply chain financing model with repurchase guarantee, it is a particular bill business service model used by the accepting bank and the dealer (accepting applicant, buyer), and the supplier (seller) through a three-party cooperation contract, with reference to the preservation warehouse method, under the premise of the seller’s commitment to repurchase, and with the control of the real right in trade (such as repurchase guarantee, etc.) as a guarantee measure. According to the confirming storage mode, the risky upstream core enterprises (suppliers) promise to purchase the backlog of goods at the end of the period, meaning that the risk of a backlog caused by a downstream order quantity exceeding the demand will be fully or partially transferred to the upstream enterprise. It makes the supply chain risk sharing possible and encourages the small and medium distributors downstream to place larger orders, which can boost the sales of core enterprises and thus increase channel profits. The power battery system is the most essential component of the new energy vehicle power system, and it presently costs roughly 50% of the total cost of the vehicle. Electric drive system manufacturers must spend a lot of money on research and development because the technology is still in its infancy. Therefore, the new energy vehicle industry will not be able to see considerable development without finding a solution to the financial issues facing power battery system manufacturers. Power battery system producers, who occupy a significant position in the supply chain for new energy vehicles, can work with automakers to conduct confirming storage business and ease capital strain due to their superior negotiating ability. Under the confirming storage mode, the bank receives a portion of the deposit from the vehicle manufacturer, gives a banker’s acceptance bill to the producer of the power battery systems, and retains custody of the products. Based on the amount of the deposit paid by the vehicle manufacturer, the bank issues a bill of lading to the vehicle manufacturer for the supply of the car’s electric drive system. The power battery system manufacturer supplies the automaker in accordance with the information on the bill of lading, and the aforementioned procedure is repeated. The automaker must, however, pay the bank the difference between the acceptance bill amount and the deposit amount once the bill of exchange has expired. If the automaker is unable to pay the bank the difference between the acceptance bill amount and the deposit amount, the bank will require the power battery system producer to make the payment. Figure 17.1 depicts the confirming storage financing mode’s transaction flow. The confirming storage financing model can successfully reduce the financial strain on new energy vehicle power battery system manufacturers and automobile manufacturers, as well as increase their market share and resolve the inventory issue, creating a situation that is mutually beneficial and win–win.

234

C.-X. Ju et al.

Fig. 17.1 Confirming storage financing mode transaction flow

17.3.2 Accounts Receivable Financing Mode Due to the extremely high cost of power batteries used in the production of new energy vehicles, power battery producers typically have sizable accounts receivable from vehicle manufacturers. If these amounts are not refilled promptly, it will undoubtedly interfere with the power battery manufacturer’s regular operations and impose a great deal of financial strain on it. In response to such a situation, accounts receivable finance can be used to renew the bond market, effectively resolving the capital issue. Accounts receivable assignment financing is also known as factoring business conducted by banks and factoring agencies. In the conventional auto business, the automobile manufacturer holds a dominant position throughout the entire supply chain and has a better credit rating than the other participants. Therefore, factoring parties are more likely to favor receivables from upstream suppliers to automobile manufacturers as a result of business transactions. Despite the fact that the manufacturer of new energy vehicles’ leading position has deteriorated, due to its strong strength and positive reputation in the new energy vehicle supply chain model, it is easy for them to manage the factoring business for the producer of power battery (Fig. 17.2). The financing model of accounts receivable assignment and repurchase has been gradually established in recent years by trusts, fund subsidiaries, and banks’ investment divisions. In accordance with this financing strategy, companies that make power battery systems sign contracts with banks and other financial organizations to assign their accounts receivable from companies that make new energy vehicles with the promise to repurchase the accounts receivable in a predetermined amount of time. Banks, trusts, and other financial institutions are supportive of this model, which adds the credit of the new energy vehicle power battery system manufacturer in comparison to the conventional factoring business. In this way, the issue of the

17 Research on Supply Chain Financing Mode of New Energy Vehicle …

235

Fig. 17.2 Accounts receivable financing mode transaction flow

power battery manufacturer’s accounts receivable being challenging to apply in the event that the new energy vehicle manufacturer is unable to pay the accounts receivable on time can be effectively solved, offering a solid financial guarantee for the power battery manufacturer’s future growth.

17.3.3 Private Equity Fund Financing Mode The new energy vehicle business can pursue alternative financing strategies in addition to the ones mentioned above to meet its financial needs. Manufacturers of power battery systems and new energy vehicles can collaborate with private equity funds, which can help the auto industry achieve integration and relieve capital pressure by not only lending money to businesses but also forming equity financing through equity and even establishing industrial funds.

17.3.4 Financing Leasing Financing Mode Due to the extensive industrial policy assistance it receives, the energy vehicle industry can also collaborate with financial leasing firms. According to the conventional paradigm, manufacturers of new energy vehicles collaborate with leasing firms. Power battery system manufacturers, new energy vehicle manufacturers, and leasing companies can explore and try out the battery leasing model under the new model,

236

C.-X. Ju et al.

combined with the structure and cost characteristics of new energy vehicles, in order for power battery system manufacturers and new energy vehicle manufacturers to solve the capital problem.

17.4 Conclusion In order to address the financing needs of businesses in the supply chain, this paper combines the current development requirements of the new energy vehicle industry with a number of financing modes. The modes include prepayment, also known as confirming storage financing mode, accounts receivable financing mode, and other approaches like private equity, private equity fund, and financial leasing. Supply chain financing for new energy vehicle manufacturers enables the automotive industry to reasonably control financial costs and develop faster business. It also contributes to strengthening information sharing between banks and enterprises, giving banks access to more first-hand information and lowering costs. Additionally, automobile manufacturers can look for the best distribution of resources in this industry, secure bank loan limits, eventually achieve self-development, eliminate financing, and market barriers, and support the long-term growth of the complete industry chain for new energy vehicles. Remarks Project fund: Research on New Energy Vehicle Industry in GuangdongHong Kong-Macao Greater Bay area under the background of Double Carbon (2022WTSCX186) by 2022 Characteristic Innovation Projects of Colleges and Universities in Guangdong Province.

References 1. Deng, K., Zeng, H.: The financial constraints in China. Econ. Res. J. 2, 47–60 (2014) 2. Chang, C.T., Ouyang, L.Y., Teng, J.T.: An EOQ model for deteriorating items under supplier credits linked to ordering quantity. Appl. Math. Model. 12, 27 (2003) 3. Randall, W.S., Farris II, T.: Supply chain financing: using cash to-cash variables to strengthen the supply chain. Int. J. Phys. Distrib. Logist. Manag. 8, 39 (2009) 4. Kunter, M.: Coordination via cost and revenue sharing in manufacturer-retailer channels. Eur. J. Oper. Res. 2, 216 (2012) 5. Yang, G., Xie, J.: Research on supply chain finance model of new energy vehicles. Mod. Manag. 36, 21–23 (2016) 6. Wang, Q.: Research on the efficiency of supply chain finance in new energy vehicle industry based on three-stage DEA model. Hebei Enterp. 10, 106–109 (2021) 7. Tan, Y.: Research on the application mode of supply chain finance in the field of new energy vehicles. Times Financ. 35, 36–37 (2018) 8. Chen, R.-F., Luo, H., Huang, K.-C., Nguyen, T.-T., Pan, J.-S.: An improved honey badger algorithm for electric vehicle charge orderly planning. J. Netw. Intell. 7(2), 332–346 (2022)

17 Research on Supply Chain Financing Mode of New Energy Vehicle …

237

9. Zhang, S.-M., Su, X., Jiang, X.-H., Chen, M.-L., Wu, T.-Y.: A traffic prediction method of bicycle-sharing based on long and short term memory network. J. Netw. Intell. 4(2), 17–29 (2019) 10. Kumar, V., Kumar, R., Kumar, V., Kumari, A., Kumari, S.: RAVCC: robust authentication protocol for RFID based vehicular cloud computing. J. Netw. Intell. 7(3), 526–543 (2022) 11. Zhang, W., Chen, R.-S., Chen, Y.-C., Lu, S.-Y., Xiong, N., Chen, C.-M.: An effective digital system for intelligent financial environments. IEEE Access 7, 155965–155976 (2019) 12. Wu, T.-Y., Guo, X., Chen, Y.-C., Kumari, S., Chen, C.-M.: SGXAP: SGX-based authentication protocol in IoV-enabled fog computing. Symmetry 14(7), 1393 (2022) 13. Chen, C.-M., Chen, L., Huang, Y., Kumar, S., Wu, J.M.-T.: Lightweight authentication protocol in edge-based smart grid environment. EURASIP J. Wirel. Commun. Netw. 2021, 68 (2021)

Chapter 18

Design of Intelligent Baby Walker Yong Wu, Yi-Jui Chiu, Tian-Hang Deng, and Yung-Hui Shih

Abstract As an important tool for parenting, baby walkers (BW) need adequate safety and functionality. At present, many problems in the safety and functionality of BW on the market have affected the personal safety of infants and young children. This paper aims to design a BW that can realize automatic braking and alarm function when overtime use. The strength and size of the BW are checked by the finite element analysis method, the comfort is analyzed according to the modal analysis theory and the ergonomics theory, and the function is realized by the control component with the single-chip as the core. The experimental results show that the design meets the set objectives.

18.1 Introduction In the context of high-quality parenting, the advent of BWs has eased the burden of childcare. However, existing BWs cannot predict and prevent emergencies. With the development of computer for object recognition [1, 2], it provides us with rich selection methods for the prediction of obstacles. In this paper, the structure of the BW is improved according to the physiological structure of the baby. The functions of automatic braking and alarming due to prolonged use of the BW are realized by installing intelligent components and braking systems on the BW. Chagas et al. [3] used a three-dimensional motion analysis system to evaluate the gait parameters and kinematics of infants and toddlers. The collected parameters and evaluation data were then processed, and it was concluded that there was no age delay in gait acquisition, but there were differences in kinematics. Santos et al. [4] developed a braking system Y. Wu · Y.-J. Chiu (B) · T.-H. Deng School of Mechanical and Automotive Engineering, Xiamen University of Technology, No. 600, Ligong Road, Xiamen 361024, Fujian Province, China e-mail: [email protected] Y.-H. Shih Department of Materials Science and Engineering, I-Shou University, Kaohsiung 840, Taiwan e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_18

239

240

Y. Wu et al.

consisting of rim brakes, which were applied to wheelchairs. The system is able to stop the wheelchair in 3.26 s with a braking distance of 1.09 m. Fan et al. [5] use a single-chip computer and a variety of sensors to achieve a smart baby car with anticollision, automatic follow-up, baby loss alarm, and baby urine wet alarm functions. Okajima et al. [6] analyzed the vibration frequency when the stroller passed over the protrusions in a single-wheeled and dual-wheeled manner, and clarified the vibration characteristics of infants. Siddicky et al. [7] recorded the EMG signal and muscle activity time of the infant in different accommodation devices. These data, after t-tests, indicate that prolonged constraints adversely affect the development of the baby’s spine. Rahim et al. [8] designed a smart stroller equipped with ultrasonic sensors and a motor that drives the wheels. When the obstacle is detected by the ultrasonic sensor, the stroller will stop moving. The structure design and finite element analysis of the BW are realized by UG, the calculation results are simulated in MATLAB, and the control circuit and software are simulated by Keil and Proteus.

18.2 Structural Design of BW In order to simplify the braking system and prevent the rollover caused by emergency braking of the baby walker, it is installed on the rear wheel, as shown in Fig. 18.1. The rear component box designed between the two rear wheels is mainly used to place intelligent control components such as motors, speed sensors, microcontrollers, and batteries. As shown in Fig. 18.2. The center of gravity of the infant’s body is lower than the upper frame of the BW, and it is easy to fall out. If the depth of the seat is too deep, it will affect the infant’s vision. Therefore, the seat depth of the BW is more suitable between 210 and 240 mm. In order to fully disperse the pressure on the infant’s crotch, the width of the seat bottom of the baby walker is 60 mm. In order to ensure that the baby can move normally and prevent the BW from tipping over, the width of the upper part

(a) Structure at the rear wheel Fig. 18.1 Structure of the rear wheel

(b) Assembly effect drawing of rear wheel

18 Design of Intelligent Baby Walker

241

Fig. 18.2 Tail assembly box

Fig. 18.3 Structural design of the seat

Fig. 18.4 Complete vehicle assembly model of BW

of the seat is taken between 1.1 and 1.2 times the shoulder width of a 12-month-old child, that is, 258 mm. According to the above dimensional data, draw a sketch of the BW seat and establish a three-dimensional model of the seat, as shown in Fig. 18.3. The overall assembly diagram of the BW is shown in Fig. 18.4.

18.3 Structural Design of Brake System The braking system includes a motor, a worm gearbox, a pulley, and a brake caliper. The motor adopts ZGB37RG DC motor from ZHENGKE Company. The worm gear box prevents the motor from being driven back, as shown in Fig. 18.5. The brake caliper using the brake pad made of rubber as the actuator of the braking system brings the advantages of easy installation and good effect, as shown in Fig. 18.5. The

242

Y. Wu et al.

Fig. 18.5 Brake caliper assembly drawing

Fig. 18.6 Pulleys, worm gearboxes, and cables

braking process is shown in Fig. 18.6: the worm gearbox driven by the motor pulls the wire cable on the brake caliper by driving the pulley, so that the brake caliper clamps the rim to realize the braking function.

18.4 Vibration Analysis The human body is most sensitive to vibration in the vibration frequency range of 0.5–12.5 Hz. Among them, when the vertical vibration is 4–8 Hz, the human internal organs resonate, the human spinal system resonates at 8–12.5 Hz, and the human body resonates when the horizontal vibration is 0.5–2 Hz.

18 Design of Intelligent Baby Walker

243

18.4.1 Modal Analysis Theory The system equation followed by the modal analysis of the baby walker in this paper is   [M] X¨ + [K ]{X } = 0

(18.1)

Defining the position vector {X } as []{u}, where [] is the modal matrix, Eq. (18.1) can be rewritten as ¨ + [A]{u} = 0 [I ]{u}

(18.2)

In which: ⎡

⎤ 1 0 ... 0 ⎢ . ⎥ ⎢ 0 1 . . . .. ⎥ T ⎢ ⎥ [] [M][] = [I ] = ⎢ . ⎥ ⎣ .. 0 . . . 0 ⎦ 0 0 ... 1 ⎤ ⎡ ω22 0 . . . 0 . ⎥ ⎢ ⎢ 0 ω22 . . . .. ⎥ []T [K ][] = [A] = ⎢ ⎥ ⎣ 0 0 ... 0 ⎦

(18.3)

(18.4)

0 0 . . . ω22 So the natural frequency of the system is ωn =

ωn EI ρ AL 4

n = 1, 2, 3 . . . . . .

(18.5)

18.4.2 Finite Element Modal Analysis The modal simulation analysis of the three-dimensional model of the BW is carried out with UG. Then the vibration results of the first six orders are analyzed. As shown in Fig. 18.7. The NFs, MSs, and maximum amplitudes of each order are obtained from Fig. 18.7, as shown in Table 18.1. The reason that only the results of the 1storder modal analysis are analyzed is that its NF is within the human body resonant frequency range. In Fig. 18.7a, the vibration frequency is 9.303 Hz, and the vibration type is the horizontal swing of the rear of the frame at the bottom of the BW. The

244

Y. Wu et al.

(a) 1st

(b) 2nd

(c) 3rd

(d) 4th

(e) 5th

(f) 6th

Fig. 18.7 First 6 vibration modes of BW structure

amplitude reaches a maximum value of 1.191 mm at the far-left end of the chassis and continues to decay to the right. Although 9.303 Hz is in the sensitive range of human body resonance, the vibration is only generated in the bottom frame and is not transmitted to the upper frame and seat. That is, the vibration will not affect the human body. Table 18.1 First to sixth order vibration characteristics of BW model Order

Max. deformation (mm)

Mode shape characteristics

1

NF (Hz) 9.303

1.191

The rear of the chassis swings horizontally

2

15.537

1.262

The chassis twists up and down around the center of mass

3

28.449

1.290

The chassis swings to the right

4

38.732

1.846

Bottom frame twists inward of frame

5

48.563

1.614

The chassis twists to the right

6

49.062

1.786

The upper frame column swings out

18 Design of Intelligent Baby Walker

245

18.5 Collision Analysis Check the frame size of the BW according to the relevant standards in China. The distance a from the rigid wall to the frame should be greater than 120 mm, and the distance b should be greater than 13 mm. The initial position of the baby walker is shown in Fig. 18.8, a1 is 166.4 mm, b1 is 48.9 mm. The frame material of the baby walker is ABS plastic, which is replaced by 40 steels with similar properties, and the seat material is nylon. Apply a force of 90 N toward the wall at the rear of the chassis. The settlement result is shown in Fig. 18.9. The deformation of the model in the x direction is 0.93 mm, so the distances a and b from the upper frame to the rigid wall can be calculated as a = 166.4mm − 0.93mm = 165.47mm

Fig. 18.8 Static position of the BW

Fig. 18.9 Deformation finite element analysis of BW model

246

Y. Wu et al.

b = 48.9mm − 0.93mm = 47.97mm The crash results a and b of the baby walker meet the national standard requirements.

18.6 Strength Analysis China’s strength standard for BWs is to place a 30 kg weight in the center of the seat without damaging the BW. A vertical force of 294 N is applied to the center of the seat of the BW, and the settlement result is shown in Fig. 18.10. As can be seen in Fig. 18.11, the stress of the BW is concentrated on the axle of the front wheel. Draw the stress distribution curve of the front wheel axle, as shown in Fig. 18.12. The maximum stress on the front axle is 27.47 MPa, which is much smaller than the 60 MPa in Table 18.2. The structural strength of the BW meets the requirements. Fig. 18.10 Stress analysis after loading

Fig. 18.11 Stress distribution at the front axle

18 Design of Intelligent Baby Walker

247

Fig. 18.12 Front wheel axle stress distribution curve

Table 18.2 ABS plastic mechanical properties ABS plastic

Density

Shrinkage

Elastic modulus

Poisson’s ratio

Yield strength

1.05118 g/m3

0.4–0.9%

2 GPa

0.394

60 MPa

18.7 Control System The system selects AT89C51 to process various signals. The non-contact laser speed sensor can meet the accuracy and save space. It can transmit and receive laser signals through a small hole in the rear assembly box. Its signal processing unit converts the received laser signal into the speed of the BW and transmits it to the AT89C51. A pressure sensor mounted on the bottom of the seat transmits the received pressure to the AT89C51. When AT89C51 finds that the speed exceeds 1 m/s, it will send a braking signal to the motor. When the AT89C51 detects that the pressure signal exceeds 100 Pa, it starts timing, and if the timing reaches 30 min, it drives the buzzer to give an alarm sound. Use Proteus to simulate the circuit and meet the expected effect, as shown in Fig. 18.13.

18.8 Conclusion The braking distance of the braking system designed in this paper is 0.396 m, and the maximum deceleration when braking is 1.45 m/s2 . The size of the seat is designed according to the size of the baby’s body, with a depth of 230 mm, a bottom width of 60 mm, and an upper width of 258 mm, which satisfies comfort. Vibration analysis, collision analysis, and strength analysis are carried out on the baby walker, all of

248

Y. Wu et al.

Fig. 18.13 Simulation of control circuits

which meet the requirements. The control circuit with AT89C51 as the core can complete precise control. Acknowledgements This project is sustained by Graduate Technology Innovation Project of Xiamen University of Technology No. YKJCX2021030.

References 1. Wang, E.K., Wang, F., Kumari, S., et al.: Intelligent monitor for typhoon in IoT system of smart city. J. Supercomput. 77(3), 3024–3043 (2021) 2. Wang, E.K., Xu, S.P., Chen, C.M., et al.: Neural-architecture-search-based multi objective cognitive automation system. IEEE Syst. J. 15(2), 2918–2925 (2020) 3. Chagas, P.S.C., Fonseca, S.T., Santos, T.R.T., et al.: Effects of baby walker use on the development of gait by typically developing toddlers. Gait & Posture 76, 231–237 (2020) 4. Santos, D., Machado, J., Varela, M.L.: Automatic brake system for manual wheelchairs. Rom. Rev. Precis. Mech., Opt. Mechatron. 48, 98 (2015) 5. Fan, Z., Zhao, H., Liang, Z.: Design of smart baby carriage based on MCU. In: Proceedings of the International Symposium on Big Data and Artificial Intelligence, pp. 286–289 (2018) 6. Okajima, H., Ota, S., Ota, R.: Dynamic characteristics of infants riding on stroller. In: ASME International Mechanical Engineering Congress and Exposition. American Society of Mechanical Engineers (2020)

18 Design of Intelligent Baby Walker

249

7. Siddicky, S.F., Bumpass, D.B., Krishnan, A., et al.: Positioning and baby devices impact infant spinal muscle activity. J. Biomech. 104, 109741 (2020) 8. Rahim, N., Aziz, W., Abd Rashid, W.N., et al.: Intelligent self-propelling baby stroller with obstacle avoidance. In: 2018 4th International Conference on Electrical, Electronics and System Engineering (ICEESE), pp. 96–101. IEEE (2018)

Chapter 19

Research on the Method of Handling Missing ETC Transaction Data Songyang Wu, Fumin Zou, Feng Guo, Qiqin Cai, and Yongyu Luo

Abstract A model based on the random forest is constructed to repair the missing trade times in ETC transaction data. The driving speed and traffic volume characteristics of the vehicles in the ETC transaction data are analyzed, while the driving speed of the missed transaction vehicles, and the distance of the road section where they are located, are combined as input features to repair the missing transaction time. A one-day transaction data of a province is used to test. The analysis results show that the random forest model has a better restoration effect and has a smaller mean absolute error and root mean square error; its mean absolute error is 2.71 s, the highest accuracy among the compared methods, and the data are more accurate and complete after interpolation using the random forest model. This paper suggests that the research based on ETC transaction data should first adopt the processing method in this paper to repair the missing trade time in the transaction data to improve the integrity of the data used and ensure the validity and accuracy of the relevant calculation results.

S. Wu (B) · F. Zou · Y. Luo Fujian Key Laboratory of Automotive Electronics and Electric Drive, Fujian University of Technology, Fuzhou 350108, China e-mail: [email protected] F. Zou e-mail: [email protected] Y. Luo e-mail: [email protected] F. Guo College of Computer and Data Science, Fuzhou University, Fuzhou 350108, China e-mail: [email protected] Q. Cai School of Mechanical Engineering and Automation, Huaqiao University, Xiamen 361021, China © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_19

251

252

S. Wu et al.

19.1 Introduction By February 2021, a total of 26,600 ETC gantry systems were built nationwide. As these ETC gantry systems are put into use, the ETC transaction data they generate provides a huge amount of basic data for in-depth research on intelligent transportation. These transaction data are not only the basic data for highway traffic management by related departments, but also for research on highway traffic flow. Researchers and scholars at home and abroad have gradually launched research on smart highways based on ETC transaction data, including speed limit analysis of highway vehicles [1], traffic operation characteristics analysis [2], travel time estimation [3], vehicle speed prediction [4], vehicle trajectory matching [5], etc. However, because ETC gantries on highways are affected by the climatic environment, surrounding vehicles, etc., gantry transaction data can generate a large amount of missing data. The lack of transaction data not only affects the integrity of ETC data, but also is not conducive to the subsequent traffic safety management of the highway. And it has a certain impact on the toll collection of expressways and the subsequent data mining research on Expressway transaction data. There have been many studies on missing traffic data, for example, Wang et al. [6] used 3D shape function combined with time interval, distance, and time lag parameters for repairing high speed km traffic flow fault data. Li et al. [7] fused multi-source data and constructed a spatio-temporal correlation feature extraction method based on deep learning models to repair the missing data. Pei et al. [8] proposed XGBoostbased predictive repair algorithm for multidimensional data to repair anomalies in highway toll data. Lu et al. [9] proposed an optimal random forest-based model for the interpolation of GPS data from floating vehicles. Lu et al. [10] proposed a traffic data restoration method based on spatio-temporal characteristics and gray residuals. Zhao et al. [11] proposed the inverse distance weight interpolation method to repair the missing data. Guo et al. [12] used the improved DTW algorithm to quickly detect anomalies in the transaction trajectory generated by ETC data. These methods do not mine the data properly and do not take into account factors such as vehicle and roadway characteristics. When fixing missed trade times, the time a vehicle spends in the roadway is an important figure, and it is related to many factors. Therefore, when repairing the ETC missed trade time data, we need to analyze the ETC transaction data first, study the correlation between its passage time and vehicle speed, vehicle type, road section distance, and other factors, and then combine these features to build a missed trade time repair model. In the second chapter, the ETC transaction data characteristics are analyzed, mainly from speed characteristics and traffic flow characteristics. Then the missing data types and reasons for missing data are analyzed. In the third chapter, a model based on random forest is constructed to repair the missing trade time. In the fourth chapter, the conclusion is given.

19 Research on the Method of Handling Missing ETC Transaction Data

253

19.2 Transaction Data Analysis 19.2.1 Transaction Data Characterization To fix the missing trade time data, it is necessary to know the passage time of the vehicle in the section, which is related to the vehicle’s travel speed. Statistical principles show that random phenomena will show certain regularity under the analysis based on a large amount of statistical data. Therefore, it is necessary to conduct statistical analysis on the ETC transaction data to extract the speed characteristics and traffic flow characteristics of different vehicles. According to the vehicle classification proposed by Xu et al. [13], the models in the general roadway can be divided into four categories, class I: a class of passenger cars; class II: two to four types of passenger cars; class III: a class of trucks; class IV: two to six types of trucks. The feature extraction of vehicle travel speed is performed on the transaction data according to these four categories of vehicle classification [14]. Since the traffic flow is different in different road sections and different time periods, we need to analyze the traffic flow characteristics from different road sections and different time periods. The flowchart of the extracted features algorithm is shown in Fig. 19.1. The ETC transaction data feature extraction algorithm is shown in Table 19.1. The algorithm first extracts the transaction data of the desired gantry from the ETC transaction data to build a dataset and classifies the vehicle types contained in the dataset according to Class I, Class II, Class III, and Class IV vehicles. Then extracts the time dimension where the transaction data is located and use the average speed formula to calculate the average speed of each vehicle. After that, we traverse each vehicle, put the same type of vehicle data together, count the frequency distribution of different types of vehicle speed, and the average speed of different time dimensions, calculate the quartiles of the average speed of different types of vehicles as speed features; then count the number of different vehicles passing in different time periods as traffic flow features. Finally, the extracted feature results are output as Feature. Fig. 19.1 Flowchart of feature recognition algorithm

254 Table 19.1 ETC transaction data driving speed and traffic flow feature extraction algorithm

S. Wu et al. Algorithm 1: ETC transaction data driving speed and traffic flow feature extraction algorithm Input: E Data Output: Featur ev , Featur et f 1. DataSet = getDataSet(EData[Flagid i ]) // Filter the transaction dataset containing the selected section 2. vehClassSet = getVehClass(DataSet) // Extracting vehicle type characteristics from transaction datasets 3. timeSliceSet = getTimeSlice(DataSet) // Extracting the time dimension of transaction data 4. Dis = get_dis(Flagid i , Flagid i+1 ) // Get distance between gantries 5. passtime = (T radetimei+1 − T radetimei ) 6. Vi = Dis/( passtime/3600) // Calculate the speed of the vehicle 7. speedSet = getSpeed(DataSet) // Speed characteristics of extracted transaction data 8. numSet = getNum(DataSet) 9. FOR each data IN DataSet 10. Feature.append (speed,Set,num, Set,timeslice) // Extracting vehicle speed and traffic flow characteristics 13. END FOR 14. Return Feature

19.2.2 Analysis of Abnormal Transaction Data When the vehicle quickly passes through the ETC lane, abnormal transaction data may occur due to the stability of the equipment, large vehicle shading, wireless crosstalk, etc. [15]. After analyzing the data, the following three main types of problems were found in the ETC toll data, as shown in Table 19.2. (1) Missing transaction data. Data is not available at the moment when it should have been collected, usually due to vehicle density, equipment failure, etc. At the same time this type of problem data appears most frequently among these three types of data and is also the main research object of this paper. (2) Incorrect transaction data. Non-conforming transaction records generated by misdetection of the opposite side of the gantry to the highway, and also including some of the transaction data in an irregular format. (3) Duplicate transaction data. Two identical transaction records due to network or system failures or software bugs.

19.2.3 Analysis of the Missing Situation Select the transaction data of some road sections for data missing analysis. First, we construct the standard trajectory of the sections. Then, compare the trajectories

19 Research on the Method of Handling Missing ETC Transaction Data

255

Table 19.2 Part of the vehicle anomaly trajectory Vehicle identification

Trading time

Trading gantry ID

Abnormalities

Plate 1

2021-05-11 02: 07: 50

351A0F

2021-05-11 02: 11: 50

351A0B

Leak transaction gantry 351A0D

2021-05-11 02: 18: 48

351A09

2021-05-11 02: 23: 21

351A07

2021-05-11 09: 25: 22

340835

2021-05-11 09: 35: 56

340837

2021-05-11 03: 35: 57

350837

Plate 2

Plate 3

2021-05-11 03: 40: 42

340839

2021-05-11 11: 11: 24

350803

2021-05-11 11: 12: 22

350801

2021-05-11 11: 25: 30

350202

2021-05-11 11: 25: 30

350202

Detected opposite side gantry 350837

Repeat transaction gantry 350202

extracted from the transaction data with the standard trajectories, and analyze the problems of the transaction trajectories. The transaction data statistics of the selected road sections are shown in Fig. 19.2, and the analysis of the selected road sections containing gantries shows that the gantries in the selected road sections are all solid gantries, and all three gantries have missing data, in addition to one gantry’s data also occurs abnormally. The average percentage of deficiencies is about 18.67% for section 1, 17.84% for section 2, and 19.79% for section 3. The missing data rate of the selected road sections is shown in Fig. 19.3. The percentage of missing data in these three sections is more than 15%, and directly using the transaction data of the three sections to perform data mining work will make the results less accurate. The missing rate is calculated by the following formula.

7

total_num

6

abnormal_num

45000 trade_num

40000

leak_num

35000

5

30000

4

25000

3

20000

2

15000 10000

1

5000

0 secon_1

secon_2

secon_3

0 secon_1

Fig. 19.2 Number of gantry abnormalities and transaction tracks

secon_2

secon_3

256

S. Wu et al. 30.00%

Fig. 19.3 Missing data rate for different road sections Deleon rate

25.00% 20.00% 15.00% 10.00% 5.00% 0.00% 10

11

12

13

14

15

16

days secon_1

Deletionrate =

secon_2

Losenum × 100% Totalnum

secon_3

(19.1)

19.3 Missing Data Repair Method 19.3.1 Random Forest-Based Restoration Method The random forest algorithm is one of the most powerful and commonly used supervised algorithms. As an extended variant of the parallel integration algorithm Bagging, it has the capability of solving both classification and regression problems. The random forest model has a stable performance, is not easily overfitted, is relatively simple to implement, and is suitable for parallel computing, and it has performed at an excellent level in many tasks. The random forest model also performs well with very high-dimensional data and does not require feature selection in the process, and also evaluates the importance of each feature in the classification problem. The basic steps of the random forest algorithm are as follows: (1) Random sampling with put-back of the data set to obtain multiple training subsets. (2) Using each training subset to build the corresponding decision tree, k features are randomly selected from the candidate features, and for each decision tree no pruning is performed to allow the tree to grow as much as possible. (3) Combine all the base learners, the combination strategies are averaging and voting, and the final output estimation results. According to the feature analysis of each part of the ETC transaction data in Chap. 2, the vehicle type, the number of hours, the average speed of the vehicle on the section, and the distance of the section are used as feature vectors. The specific meaning and classification of feature vectors are shown in Table 19.3. After obtaining the feature vector, the passage time of the road section is estimated using the random forest algorithm, and the missing trade time of the repair is subsequently obtained by summing the passage time of the vehicle and the trade time of the vehicle through the front gantry.

19 Research on the Method of Handling Missing ETC Transaction Data Table 19.3 Feature vectors and their meaning

257

Feature vector

Meaning and classification

Car t ype

Cartype (class I, class II, class III, class IV)

H our

Time period (0:00, …, 23:00)

vi

Average speed of vehicle in this section

Dis i

Distance of section

The algorithm first obtains the trade time before and after the missing gantry, then calculates the passage time ti of the vehicle in the section, and then obtains the distance Dis i of the section according to the distance calculation function, and calculates the average speed vi of the missing transaction vehicle in the section by using the average speed calculation formula. Then the feature vectors of the vehicles with missing data are constructed by combining the time period in which the vehicles are traded and the type of vehicles, and the feature vector sets Mi of different vehicles are input into the random forest model. The output result is the set of predicted passage times ti for the missing vehicles on the roadway, and the front door plus trade time for each vehicle is added to the predicted passage time to be the repaired missing transaction time Ti (Table 19.4). Table 19.4 ETC missed transaction time data repair algorithm

Algorithm 2: ETC missed transaction time data repair algorithm Input: E Data_ pr e, E Data_next Output: E Data_loset 1. E Data_next ti = E Data_next t [tradetime] 2. E Data_ pr eti = E Data_ pr et [tradetime] 3. ti = E Data_next ti − E Data_ pr eti 4. Dis i = get_dis(E Data_ pr e, E Data_next) // get_dis is a function to get the distance of the road 5. vi = Dis i /(ti /3600) 6. H our = [hour 1 , hour 2 , . . . , hour i ] // Get the vector of time periods where the missing vehicles are located 7. Car t ype = [car t ype1 , car t ype2 , . . . , car t ypei ] // Get the vehicle type vector of the missing vehicle   8. Mi = H our i , Car t ypei , Dis i , vi // Construction of feature vectors for missing vehicles 9. ti = R F(Mi ) // Inputting feature vectors into the random forest model 10. T = (t 1 , t2 , . . . , ti ) // Output the predicted travel time ti of the missing vehicle on the section 11. Ti = E Data_ pr eti + ti 12. Return Ti

258

S. Wu et al.

19.4 Experimental Analysis and Results 19.4.1 Feature Analysis Results The driving speed data of different cars of the road section were subjected to frequency statistics to obtain histograms of the frequency distribution of the speed of different road sections [16], as shown in Fig. 19.4, and the main findings are as follows. The number of Class I cars is mostly concentrated in the speed interval of 100– 110 km/h, which is much larger than the speed of the most frequent interval of the rest of the classes. From the distribution of the speed of the Class II vehicles, we can see that it is more concentrated in the maximum speed, the speed value is about 93 km/h, and the frequency of more than 93 km/h is significantly reduced, most of the Class II vehicles are operating buses, due to the existence of operating vehicle speed limits, so fewer vehicles will exceed the speed limit value, and the performance is concentrated in a maximum value. Class III and Class IV vehicles are all trucks, and their overall driving speed is lower than other classes, which is also in line with the speed limit for trucks at highway. Table 19.5 shows the speed characteristic values of different types of vehicles on different road sections, which can be used as a reference for the determination of speed limit values in highway speed management, and also as a reference for the characteristic values of different types of vehicles on that road section when repairing missing data [17]. The distribution of traffic flow for different time periods and the distribution of the average speed of vehicles were analyzed [18], and the specific distribution is shown in Fig. 19.5, with the following main findings. Class I, Class II, and Class III vehicles at 0:00–5:00 time period in the number of vehicles significantly reduced at 7:00–10:00 traffic flow gradually increased, at 10:00–12:00 traffic flow is declining trend, at 13:00–17:00 traffic flow is increasing trend and reached the maximum value of traffic flow in a day. The traffic volume

Fig. 19.4 Speed distribution of different types of vehicles on section 1 and section 2

19 Research on the Method of Handling Missing ETC Transaction Data

259

Table 19.5 Speed characteristic values for different road sections Section

Section 1

Car type

Class I

Class II

Class III

Class IV

Class I

Section 2 Class II

Class III

Class IV

75th

107.29

91.26

86.66

83.12

111.08

91.97

91.97

83.68

50th

102.16

87.82

83.53

78.42

106.22

88.86

85.03

78.67

25th

96.95

84.21

77.35

73.37

100.79

85.49

78.67

73.36

AVG

101.82

87.02

83.55

77.94

105.86

88.30

85.35

78.16

S.D.

8.23

8.42

9.91

7.34

8.33

7.46

10.48

7.84

Note Unit: km/h

Fig. 19.5 Distribution of traffic volume and speed at different time periods

of the Class IV vehicles in the early morning hours far exceeded the remaining types of vehicles, but the overall nighttime traffic volume was much smaller than the daytime traffic volume. At the same time, the average speed of the Class IV vehicles at different times did not change significantly and was at a stable stage. The average speed of Class I vehicles at night is greater than during the day, and its speed gradually decreases with the increase in traffic flow; Class II vehicles at night are significantly reduced, because Class II vehicles contain most of the operating buses, due to the existence of operating vehicle speed limits, so at night its speed drops significantly; Class III vehicles at different times of the speed are also in a steady state without a large change.

19.4.2 Results of Data Repair To examine the advantages and disadvantages of data repair methods on large-scale missing data sets, it is necessary to construct corresponding data sets and calculate the performance of various missing value processing methods on different evaluation

260

S. Wu et al.

metrics [19]. Therefore, this paper generates the dataset by the following simulation scheme. The scheme includes four steps: (1) selecting the road section containing three gantries or more; (2) screening the data containing the above gantries; (3) filter out the gantry ID, transaction time, vehicle identification (license plate) and PASSID in the filtered data, and construct the transaction time data set of vehicles in the section; (4) extracting the gantries that may produce missing, marking the data of the gantries according to different missing proportions, and recording the missing data. Through the above steps, based on the missing data from the previous statistics, and simulating the real value missing mechanism, this paper generates two datasets, one for the original dataset A, and one for dataset B containing five data with missing values of 10, 15, 20, 25, and 30%, numbered B1 to B5. In this paper, we use different methods to repair missing values for data set B and evaluate the advantages and disadvantages of the methods by examining the accuracy of restoring data set A. The evaluation metrics used in this paper are mean absolute error and root mean square error, where the MAE and RMSE are calculated by the following formula. 1 |yi − yi | n i=1   m 1 

2  RMSE = yi − yi m i=1 n

MAE =



(19.2)



(19.3)

Analysis of the Repair Results In this paper, mean interpolation method, random forest method, and SVR method are selected for experimental comparison to simulate the performance of RMSE and MAE under the real absence mechanism. The results of various methods are shown in Table 19.6. It can be seen that the missing rate has an effect on the data restoration results, and the error of data restoration increases accordingly as the missing rate increases. Random forest works the best among the three methods and performs well under both evaluation metrics, where the minimum error is 2.71 s and the maximum error is 3.35 s. In addition, the RMSE values of random forest are both much smaller than the remaining two methods, indicating that the prediction error of random forest is more stable. SVR is less effective compared to random forest, but better than the mean interpolation method. And the mean interpolation method has the worst effect, and its error is around 40 s.

19.5 Conclusion Data integrity is a prerequisite for data applications and a concern for big data applications. Missing data problem is one of the common data quality problems. Missing

19 Research on the Method of Handling Missing ETC Transaction Data

261

Table 19.6 Results of different models Models

Evaluation metrics

B1

B2

B3

B4

B5

Mean interpolation

MAE

39.71

37.60

40.07

40.14

39.31

RMSE

6.85

6.55

6.96

6.96

6.88

SVR

MAE

7.97

8.49

8.31

8.27

9.39

RMSE

1.07

1.17

1.22

1.24

1.28

MAE

2.71

2.90

2.87

2.89

3.35

RMSE

0.48

0.50

0.50

0.51

0.52

RF

data problem is one of the common data quality problems. Since the ETC transaction data is representative of each vehicle’s passage, using only the common method of mean interpolation to deal with missing data is much less effective, while simply using the method of deleting missing data will have an impact on the integrity of the data. And the popular machine learning interpolation method has better results in dealing with complex, high-dimensional missing data for ETC missing data with good results. With the deep mining of ETC transaction data, its data integrity is becoming more and more important, therefore, the detection and repair work for the abnormal data of different cases of ETC transaction data will be an important research direction, and the integration of statistical interpolation method and machine learning interpolation method to deal with different types of ETC missing data will be called mainstream. At the same time, considering the correlation law between ETC transaction data will also be the development direction of ETC missing data processing methods.

References 1. Zou, F., Guo, F., Tian, J., Luo, S., Yu, X., Gu, Q., Liao, L.: The method of dynamic identification of the maximum speed limit of expressway based on electronic toll collection data. Sci. Program. 2021, 15 (2021) 2. Lai, J.-H., Qi, Y., Wang, Y., Han, Y., Huang, L.-H., Zhao, Y.-F.: Estimation method of traffic state parameters based on toll data. China J. Highw. Transp. 35(03), 205–215 (2022) 3. Luo, S., Zou, F., Zhang, C., Tian, J., Guo, F., Liao, L.: Multi-view travel time prediction based on electronic toll collection data. Entropy 24, 1050 (2022) 4. Zou, F.-M., Ren, Q., Tian, J.-S., et al.: Expressway speed prediction based on electronic toll collection data. Electronics 11(10), 1613 (2022) 5. Kang, A.-Z., Li, Y., Han, W.-Y., Fu, X.-J., Zhao, J.-D.: Vehicle trajectory matching based on ETC data. Sci. Technol. Eng. 22(13), 5481–5487 (2022) 6. Wang, W., Cheng, Z.-Y., Liu, M.-Y., Yang, Z.-S.: Repair method for traffic flow fault data based on spatial-temporal correlation. J. Zhejiang Univ. (Eng. Ed.) 51(09), 1727–1734 (2017) 7. Li, L.-C., Qu, X., Zhang, J., Wang, Y.-G., Li, H.-C., Ran, B.: Missing value imputation method for heterogeneous traffic flow data based on feature fusion. J. Southeast Univ. (Nat. Sci. Ed.) 51(09), 1727–1734 (2018) 8. Pei, L.-L., Sun, Z.-Y., Han, Y.-X., Li, W., Hu, Y.-J.: Algorithm for repairing abnormal toll data of expressway based on SSC and XGBoost. J. Jilin Univ. (Eng. Technol. Ed.) 1–9 (2022)

262

S. Wu et al.

9. Lu, Q.-X., Guo, D.-D., Li, X., Zhao, L.: GPS data interpolation model of floating car based on optimized random forest algorithm. Sci. Technol. Eng. 22(04), 1656–1661 (2022) 10. Lu, B.-C., Zhang, D.-M., Shu, Q., Li, Y.-L.: Diagnosis and repair of traffic fault data based on spatio-temporal characteristics and grey residual. J. Chongqing Jiaotong Univ. (Nat. Sci.) 39(09), 8–16 (2020) 11. Zhao, S.-X., Qu, R.-T., Wang, J.-W.: A vehicle trajectory reconstruction method based on improved inverse distance weighted interpolation. J. Highw. Transp. Res. Dev. 35(10), 133–139 (2018) 12. Guo, F., Zou, F., Luo, S., Liao, L., Wu, J., Yu, X., Zhang, C.: The fast detection of abnormal ETC data based on an improved DTW algorithm. Electronics 11(13), 1981 (2022) 13. Xu, J., Yang, Z.-M., Chen, Q., Chen, Z.-W.: Research on speed distribution and vehicle type classification of mountain expressway based on electronic toll collection data. J. Transp. Syst. Eng. Inf. Technol. 1–12 (2022) 14. Chen, J.-N., Huang, Z.-J., Zhou, Y.-P., Zou, F.-M., Chen, C.-M., Wu, J.M.-T., Wu, T.-Y.: Efficient certificate-based aggregate signature scheme for vehicular ad hoc networks. IET Netw. 9(6), 290–297 (2020) 15. Wu, T.-Y., Guo, X., Yang, L., Meng, Q., Chen, C.-M.: A lightweight authenticated key agreement protocol using fog nodes in Social Internet of vehicles. Mob. Inf. Syst. 2021, 3277113 (2021) 16. Wu, T.-Y., Guo, X., Chen, Y.-C., Kumari, S., Chen, C.-M.: SGXAP: SGX-based authentication protocol in IoV-enabled fog computing. Symmetry 14(7), 1393 (2022) 17. Chen, C.-M., Chen, L., Gan, W., Qiu, L., Ding, W.: Discovering high utility-occupancy patterns from uncertain data. Inf. Sci. 546, 1208–1229 (2021) 18. Kumar, V., Kumar, R., Kumar, V., Kumari, A., Kumari, S.: RAVCC: robust authentication protocol for RFID based vehicular cloud computing. J. Netw. Intell. 7(3), 526–543 (2022) 19. Gan, W., Chen, L., Wan, S., Chen, J., Chen, C.-M.: Anomaly rule detection in sequence data. IEEE Trans. Knowl. Data Eng. (2021)

Chapter 20

Highway Traffic Volume Prediction Based on GRU and Attention by ETC Data Shibin Huang, Fumin Zou, Feng Guo, and Qiang Ren

Abstract Highway is facing the pressure of increasing transportation demand, Intelligent Transportation System (ITS) can improve the traffic efficiency and prevent the occurrence of congestion, ETC data provides the data foundation for the construction of ITS. Traffic flow prediction as an important part of ITS, and it has important research meaning. To address the problem that existing traffic flow prediction models cannot identify the dynamic changes of spatial–temporal dependency in traffic flow sequence data, this paper proposes a model based on the combination of self-attentive mechanism and GRU to predict the highway traffic volume. This paper first analyzes the different features affecting the changes in highway traffic flow and then describes the way of combining the attention mechanism and GRU. The GRU layer processes the input of the temporal sequence, and the self-attention layer extracts features that are important to the results in the traffic flow time series data. Finally, the experimental validation is performed by using the actual data generated from the ETC gantries of the highway in Fujian Province, China. The results show that the performance of the model proposed in this paper is better than other models at different time intervals.

S. Huang (B) · F. Zou · F. Guo · Q. Ren Fujian Key Laboratory of Automotive Electronics and Electric Drive, Fujian University of Technology, Fuzhou, China e-mail: [email protected] F. Zou e-mail: [email protected] F. Guo e-mail: [email protected] Q. Ren e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_20

263

264

S. Huang et al.

20.1 Introduction With the rapid development of social economy in recent years, the number of automobiles has increased rapidly, by the end of 2021, China’s vehicle ownership reached 290 million. Facing such a huge traffic load, traffic congestion and accidents occur frequently, bringing great inconvenience to people’s lives, and also becoming an important factor restricting urban development. Intelligent Transportation System (ITS) is an advanced application system that can improve the efficiency of road traffic and reduce the travel time of travelers. The construction of ITS can alleviate the problems in modern transportation effectively. As one of the major modes of transportation between cities, highways are another important road system besides the urban road system. By the end of 2021, the total mileage of highways in China has reached 169,100 km, ranking the first in the world. In recent years, the construction of Electronic Toll Collection (ETC) system has provided the data foundation for the highway intelligent transportation system. The ETC gantry is a special system and supporting facilities with functions such as toll segment calculation and license plate image recognition built on the highway section [1]. Whenever a vehicle passes through the ETC gantry, it generates data and uploads it to the corresponding data center, and the ETC gantry generates a large amount of data, which provides data support for the highway ITS. The schematic diagram of the ETC gantry on the highway and the distribution of the ETC gantry on the highway are shown in Fig. 20.1a, b. By the end of February 2021, China has built 26,600 sets of ETC gantries, which provide a large amount of road network sensing data, and it is an urgent problem for highway managers to consider and solve how to use the multiple and large amount of sensing data provided by the gantries to support the development of highway network operation management and services [2, 3]. Traffic flow prediction is an important part of ITS, which can provide traffic managers and travelers with accurate and efficient prediction of traffic flow information in future periods, and this information can formulate appropriate traffic management methods for traffic management departments; travelers can make travel plans

Fig. 20.1 a Schematic diagram of highway gantry; b schematic diagram of the distribution of highway gantry, in which the red dot is the gantry on the highway

20 Highway Traffic Volume Prediction Based on GRU and Attention …

265

in advance to effectively avoid traffic congestion and traffic accidents. Combining the above advantages, the study of traffic flow forecasting is of great significance. Traffic flow prediction can be defined as the prediction of traffic flow information for a future period given the historical information of traffic flow. The difficulty lies in the nonlinear variation and dynamic changes of traffic flow, which are caused by the changes in traffic demand and traffic capacity over time. To address these problems, many researchers have conducted research on traffic flow forecasting in recent years. According to the length of prediction time, traffic flow prediction can be divided into two types short-term prediction and long-term prediction, generally, the prediction length of short-term traffic flow prediction is concentrated in 5, 10, and 15 min [4]. Short-term traffic flow forecasts have smaller traffic variations and higher prediction accuracy compared to long-term traffic flow forecasts, and the results of shortterm traffic flow forecasts can provide real-time and effective information to travelers to help them make better path choices and achieve path guidance to shorten travel time and reduce traffic congestion [5, 6]. According to the prediction methods, they can be divided into statistical-based methods and machine learning-based methods. Many early studies used historical averaging (HA), ARIMIA [7] and Seasonal ARIMIA [8, 9] are statistical-based methods. Statistical-based methods are simple and fast to compute, but they cannot deal with unexpected traffic conditions and their accuracy is low. These methods are linear models and cannot handle nonlinear variations in traffic flow. Machine learning-based neural network methods have higher prediction accuracy and better prediction capability and have also been experimentally proven to perform better in recent years. Machine learning-based methods have become more popular among researchers in recent years. Machine learning-based neural networks are mainly classified into convolutional neural networks (CNN) and recurrent neural networks (RNN). RNN can use its recurrent mechanism to deal with the relationship between front and behind data in sequential data, therefor recurrent neural networks are more suitable for predicting time-dependent sequential data such as traffic flow data [10, 11]. Long short-term memory (LSTM) and Gate recurrent unit (GRU) are two common variants of convolutional RNN networks [12, 13], and both models have achieved good results in the task of predicting sequential data. While GRU benefits from a simpler structure and smaller time complexity, GRU is relatively better than LSTM [14, 15]. When traffic flow prediction is performed, besides the temporal dependency, the traffic flow of target area is also affected by neighboring areas, which is the spatial dependency [16]. How to consider the feature of spatial dimension in traffic flow prediction models is also a hot topic of research nowadays. Zhao et al. combined GRU and graph convolutional neural network (GCN) to capture both temporal and spatial dependencies in traffic flow, where GCN is responsible for capturing spatial features and GRU is responsible for capturing temporal features [16]. However, the current research only captures the static spatial–temporal dependence on traffic flow, the spatial–temporal dependence of traffic flow in reality is more complex and time-varying, and the impact of different times and spaces on traffic flow changes over time; for example, the traffic flow changes less during midnight, while the traffic flow changes more during the daytime peak hours, and the traffic

266

S. Huang et al.

flow during midnight hours and peak hours have different impacts on future time periods. Existing traffic flow prediction studies have used GPS data or road sensor data, which have the disadvantages of incomplete sample data and high sparsity, while ETC data have the advantages of full sample size, accuracy, time series, and good data quality [2]. The potential of using ETC data for traffic estimation is currently explored by Kim et al. They used a dynamic OD estimation model proposed in a previous study and input additional ETC data to improve the accuracy of the estimation [17], and the results of the study indicate that ETC data can help to understand traffic demand and its variability. Zou et al. use ETC data to predict the section speed by combining the GCN and GRU models [18]. Tian et al. proposed a random forest algorithm using ETC data for highway traffic flow prediction [19], which constructed a spatial–temporal feature vector model that was input to the random forest algorithm for highway traffic flow prediction. In order to address the above problems, this paper proposes a model based on ETC data, which combines the attention mechanism and GRU, called ATT_GRU, and uses the attention mechanism to extract dynamic changing spatial–temporal features in traffic flow. The model is trained and tested using the data generated from the ETC gantries of highways between two major cities in Fujian province, China; the ETC data contains a large amount of vehicle traffic records, which makes it easy to calculate the traffic volume at different time periods. The data set is obtained through the calculation, and comparison experiments are conducted, which show that the model proposed in this paper outperforms other models, and the experimental analysis shows that the model proposed in this paper can improve the shortcomings of traditional RNN models. The following chapters are organized as follows; Chap. 2 describes the highway traffic flow prediction model proposed in this paper. Chapter 3 presents the experimental analysis of the model proposed in this paper, and Chap. 4 concludes the paper.

20.2 Methodology 20.2.1 Problem Definition Based on the highway traffic volume forecasting objectives in this paper, the following definitions are given: Definition 1 Highway section: The section of the highway between two gantries is defined as a section. For example, the section between gantry A and gantry B in Fig. 20.2 is called section AB. Definition 2 Unit time section traffic volume: the total number of vehicles passing through a particular section in a unit time period.

20 Highway Traffic Volume Prediction Based on GRU and Attention …

267

Fig. 20.2 a Model structure; b structural diagram of the GRU

Problem definition: The objective of this paper is a highway traffic volume forecasting method, i.e., given information on the historical characteristics of a highway segment for k time intervals before a certain moment t, as shown in Eq. 20.1: X (xt−k , xt−k+1 , xt−k+2 , ...xt−1 )

(20.1)

The purpose of the model is to predict the traffic volume in a time interval after time t, represented by yt . It can be regarded as to learn a mapping function f from historical traffic information X to the traffic volume yt , as shown in Eq. 20.2: yt = f (X (xt−k , xt−k+1 , xt−k+2 , ...xt−1 ); θ )

(20.2)

where yt is the traffic volume in a time interval after time t of the target section, X is the feature matrix containing k feature vectors from time t − k to time t − 1, and θ is the parameter to be learned by the model.

20.2.2 Model Structure The model proposed in this paper consists of the following three parts, as shown in Fig. 20.2a. The first part is a GRU layer consisting of GRU units, which receives the input X (xt−k , xt−k+1 , xt−k+2 , . . . xt−1 ) for k time steps of the model, and xt−k is the feature at the moment t − k, where the specific input features will be introduced in the next subsection. It outputs the hidden states E(et−k , . . . , et−2 , et−1 ) for k time steps; the second part is the self-attention layer, which receives the k hidden states E(et−k , . . . , et−2 , et−1 ) output by the GRU layer, it obtains the weights of the different time steps by computing the similarity between the k hidden states and outputs the vectors C(ct−k , . . . , ct−2 , ct−1 ) that take into account the weight information between

268

S. Huang et al.

the different time steps; then the third part is the GRU layer composed of GRU units, which receives the output C from the self-attention layer and calculates the time dependence of the vector C. Finally, the prediction result of the model is output.

20.2.3 Model Input To obtain accurate traffic forecasts, it is necessary to consider the feature inputs to the model. This section analyzes the features of the highway traffic flow and describes how the features are input to the model. The variation in traffic flow at ten-minute intervals throughout the week on the highway is shown in Fig. 20.3, which shows that there is a clear cyclical and timedependent flow of traffic. The chart shows that traffic flow has two daily peak of 300–400 vehicles between 8 and 10 pm and between 2 and 4 pm, which traffic flow reaching a minimum at 4 am. The graph shows that June 6–7, 2020 falls on a weekend, and it can be seen that its peak is higher than the weekday peak. The relationship between the traffic volume in a section at a given moment t and the traffic volume 10 min ago is shown in Fig. 20.4a, which shows that there is an obvious linear relationship between them. Therefore, this paper uses the traffic flow of k time intervals before the target moment as a temporal feature for traffic prediction.

Fig. 20.3 Weekly traffic flow chart of the expressway

20 Highway Traffic Volume Prediction Based on GRU and Attention …

269

Fig. 20.4 a The relationship between the traffic flow at time t and its traffic flow ten minutes ago; b relationship between the traffic flow of a gantry and its neighboring gantry on highway

The relationship between the traffic flow of neighboring gantries is shown in Fig. 20.4b, with the x and y axes for gantry A and gantry B, respectively. The points on the graph indicate the traffic flow of gantry A and gantry B at a same moment, with each point indicating a moment. From this graph, it can be seen that there is an obvious linear relationship between the traffic flow of neighboring gantries. As analyzed above, highway traffic volume is cyclical and correlates with the traffic flow at past moments, as well as with the traffic flow in neighboring section, so this paper uses the above feature to forecast traffic flow. As described in the model structure diagram in the previous section, the input of the model in this paper is vector X which is a k-dimensional vector (xt−k , . . . xt−2 , x1 ) and xt−k as a feature vector at moment t − k. This paper uses five features for prediction, i.e., xt−k is a vector of length 5. As shown in Fig. 20.2, suppose the objective is to predict the traffic flow of gantry B. The five features are the traffic flow of gantry B at moment t − k; the A C and vt−k , traffic flow of front and rear gantry A and C corresponding to moment vt−k and the number of hours corresponding to moment t − k and the number of days of the week corresponding to date thour and tweekday .

20.2.4 GRU The current moment of traffic flow on the highway and the previous k time interval traffic flow, and the impact of historical traffic flow condition on present traffic volume is dynamic changes, this characteristic is known as dynamic temporal dependence. Capturing the temporal dependence of traffic flow changes is the key to accurate traffic flow forecasting. The relationship between highway traffic flow and time is shown in Fig. 20.4a, b. Most of the current research uses recurrent neural networks (RNN) to model the temporal dependence of traffic flow, but RNNs face the problems of gradient disappearance and gradient explosion. LSTM and GRU are improved upgrades of RNN, they solve the aforementioned problems. GRU model is simpler and has low time complexity compared with LSTM model, and its effect is comparable to LSTM, therefore, researchers have conducted research based on GRU model in recent years. The GRU’s structure is shown in Fig. 20.2b, where h t−1 is the hidden state at time t − 1, xt denotes the input at time t, and rt denotes the reset gate, whose

270

S. Huang et al.

role is to control how much information from past moments is ignored. z t denotes the update gate, whose role is to control how much past information is added to the state at the present moment; h t denotes the memory content saved by the model at moment t; and finally, h t denotes the output of the model at moment t. The reset and upgrade gates in the GRU model are calculated by Eqs. 20.3–20.6 respectively, as follows: 

r t = σ (W r x t + U r h t−1 ))

(20.3)

z t = σ (W z x t + U z h t−1 )

(20.4)

h t = φh (Wh xt + Uh (rt  h t−1 ))

(20.5)





h t = z t  h t + (1 − z t )  h t−1

(20.6)

where W r , W z and U r , U z are all weight matrices,  denotes Hadamard Product, rt is the state of the reset gate at moment t and z t is the state of the update gate at moment t. The values of these two gates will determine how much past information is ignored and how much new information is added to the future moment. σ function converts the values to the interval 0–1 as a signal for the gating unit, with a value of 0 indicating that the gate is completely closed and a value of 1 indicating that the gate is completely open. The GRU model takes the hidden unit at moment t − 1 and the traffic information at moment t as inputs to the model through three gating units, i.e., the historical traffic information and the traffic information at the present moment, and finally obtains the traffic information at moment t. Once the traffic information at time t is obtained, the GRU model retains the information of historical traffic flow as the input at moment t + 1, thus achieving the ability to capture the temporal dependence. The traffic volume of highways is similarly spatially dependent, i.e., the traffic volume of a section in a highway is correlated with the traffic volume of neighboring section. In this paper, we use the traffic volume of the neighboring gantries before and after the predicted target section as the spatial feature input to model to capture the spatial dependency of traffic flow.

20.2.5 Self-attention The attention mechanism was originally proposed for tasks with sequential data such as text translation, motivated by the fact that when a neural network is doing a sequence-based prediction task, different parts of the sequence have different effects on the prediction result, some parts of the sequential data may have a greater impact on the result than others and others may have a lighter impact on the results. The temporal data of traffic flow can be considered as a kind of sequential data and

20 Highway Traffic Volume Prediction Based on GRU and Attention …

271

the information at different moments has different degrees of effects on the final result, so naturally we thought of introducing an attention mechanism for traffic flow prediction. The self-attention layer in this model accepts the k hidden states output from the GRU layer, calculates the similarity between each value to get the similarity score, and then gets the representation vector C based on the similarity score. The representation vector C is a vector that takes into account the importance of different time steps. The attention mechanism is calculated as in Eqs. 20.7–20.9: scores = Q · K T

(20.7)

distribution = SoftMax(scores)

(20.8)

C = distribution · V

(20.9)

where Q and K are the Query value and Key value in the attention mechanism [20], respectively, corresponding to the output and hidden state of the GRU layer of k time steps in the model. SoftMax is the activation function. SoftMax maps the input to a real number between 0 and 1, indicating the probability of taking the corresponding item; V and K generally take the same value. C is the output of the attention mechanism, which is a vector of length k after considering the different time steps of the output after weights is a vector of length k.

20.3 Experiment and Verification 20.3.1 Experimental Comparison The model proposed in this paper will be used to evaluate the models against traditional models, including ARIMA, SVR, GRU, and bidirectional GRU (Bi-GRU).

20.3.2 Data Description For the experimental validation part, we use the data generated by the ETC gantries on the highway sections between two major cities in Fujian Province from June 1st to 23rd, 2020, with approximately 11.5 million pieces of ETC data. When a vehicle passes through the gantry to communicate with it, the data generated by the gantry is called ETC transaction data, which mainly includes fields such as transaction time, transaction gantry id, vehicle license plate, etc. The specific fields are shown in Table 20.1. This section will use the data cleared of abnormal data for verification, abnormal data including license plate miss, duplicate records, and so on. After the

272

S. Huang et al.

Table 20.1 ETC data example Field name

Example

Description

Tradetime

20200601 00:00:00

Time for vehicles to pass through the gantry

FlagID

350xxx

The unique id of gantry

EnStation

9901EN

The unique ID of toll station

Entime

20200602 00:00:00

Time for vehicles to enter the toll station

completion of the abnormal data cleaning, we calculate the traffic volume of the section according to the time interval of 5, 10, and 15 min interval, and the results are used as the experimental data set. The lengths of the datasets obtained from the different time intervals were, respectively, 6624 items for 5-min intervals, 3312 items for 10-min intervals, and 2208 items for 15-min intervals. The data were normalized using maximum-minimum normalization, which is given in Eq. 20.10. The purpose of normalization is to eliminate the influence of the magnitude between different indicators, and normalization can speed up the training of neural network models. The model proposed in this paper was used to experiment with three different time intervals, of which 80% of the data was used as the training set and 20% as the test set of the model, and finally the prediction results of the model were compared and analyzed.



x =

x − min(x) max(x) − min(x)

(20.10)

20.3.3 Evaluation Metrics In order to compare and evaluate the prediction results of the models, the following metrics are used in this paper to evaluate the performance of the models: root mean square error (RMSE) and R 2 coefficient of determination, which are calculated as in Eqs. 20.11 and 20.12:   m 1  2  yi − yi RMSE =  m i=1 

m  R =1− 2

i=1

m

i=1

(20.11)

2



yi − yi 

yi − y

2

2

(20.12)



where m is the sample size, yi is the actual value and yi is the predicted value of the model. RMSE represents the sample standard deviation of the difference between

20 Highway Traffic Volume Prediction Based on GRU and Attention …

273

the predicted and observed values, which is the smaller the better. R 2 represents the degree to which the predicted and true values match, with the closer to 1 indicating a better result.

20.3.4 Model Hyperparameter Settings The hyperparameters of the model in the neural network model include batchsize, epoch, learning rate, and the choice of optimizer. In this paper, we use the Adam optimizer as the optimizer, which has the advantage of being insensitive to the learning rate, so we use an initial learning rate of 0.02. After our experiments, we have verified that the model works best when the other parameters are chosen as follows, with a batchsize of 32 and an epoch of 300, so we use the above parameters in this paper.

20.3.5 Experiment The experimental results comparing ARIMA, SVR, GRU, bidirectional GRU and the model proposed in this paper for three different time intervals are shown in Table 20.2, bolded numbers indicate the best results (Fig. 20.5). From the results of the experiments, it can be seen that the model has a larger improvement on accuracy than the other models on the 10 and 15-min time intervals and a smaller improvement on the 5-min time intervals. The reason for this is that the change in traffic flow within 5 min is smaller, which corresponds to a smaller change in the prediction result of the model, so the variation of RMSE values is also smaller. Among the compared model, SVR and ARIMA are the worst performers because they can only capture linear relationships, GRU and Bi-GRU take into account temporal dependence, and they outperform SVR and ARIMA. The RMSE of ATT_GRU at three different time intervals is better than that of the compared conventional models, and the stability of the model is also better than that of the compared model. When the time interval is 10 min, the prediction results of the proposed model are shown in Table 20.2 Experiment result Time interval (min)

Metric

SVR

ARIMA

GRU

Bi-GRU

ATT_GRU

5

RMSE

13.7890

13.1792

11.4440

11.4250

11.1293

5

R2

10

RMSE

10

R2

15

RMSE

15

R2

0.9465

0.9517

0.9636

0.9637

0.9655

21.1021

21.5887

21.1152

19.9662

19.3393

0.9679

0.9671

0.9681

0.9715

0.9733

28.8583

31.582993

27.048099

27.103354

25.903758

0.9729534

0.9684742

0.9763254

0.9762286

0.9782862

274

S. Huang et al.

Fig. 20.5 Diagram of the predicted results (10 min interval)

Fig. 20.5, which shows that the prediction results of the model basically fit the actual values. The above analysis shows that the model proposed in this paper can effectively improve the accuracy of traffic flow prediction, and the application of attention mechanism can improve the deficiency of the RNN neural network in capturing static spatial–temporal features.

20.4 Conclusion In this paper, we propose a method for predicting highway traffic flow. Based on traditional recurrent neural networks, we incorporate an attention mechanism to improve the shortcomings of RNN neural networks which can only extract static traffic flow spatial–temporal features. The attention mechanism is able to identify the important parts of the traffic flow spatial–temporal data for the results, which in turn improves the accuracy of the model prediction. The experiment on actual data shows that the ATT_GRU model outperforms the traditional model at different time intervals, indicating that the use of the attention mechanism can improve the shortcomings of the RNN neural network for traffic flow prediction and help to improve the prediction of traffic flow on highways.

References 1. Li, Y.: The application and research of forecast analysis based on expressway networking toll data. Master’s thesis, Beijing Jiaotong University (2017)

20 Highway Traffic Volume Prediction Based on GRU and Attention …

275

2. Liu, Q., Yang, Z., Cai, L.: Predicting short-term traffic on expressway based on ETC gantry system data. J. Highw. Transp. Res. Dev. 04, 123–130 (2022) 3. Wu, T.-Y., Guo, X., Chen, Y.-C., Kumari, S., Chen, C.-M.: SGXAP: SGX-based authentication protocol in IoV-enabled fog computing. Symmetry 14(7), 1393 (2022) 4. Sun, P., AlJeri, N., Boukerche, A.: A fast vehicular traffic flow prediction scheme based on Fourier and wavelet analysis. In: 2018 IEEE Global Communications Conference (GLOBECOM), pp. 1–6. IEEE (2018, December) 5. Liu, J., Guan, W.: A summary of traffic flow forecasting methods. J. Highw. Transp. Res. Dev. 21(3), 82–85 (2004) 6. Chen, J.-N., Huang, Z.-J., Zhou, Y.-P., Zou, F.-M., Chen, C.-M., Wu, J.M.-T., Wu, T.-Y.: Efficient certificate-based aggregate signature scheme for vehicular ad hoc networks. IET Netw. 9(6), 290–297 (2020) 7. Ahmed, M.S., Cook, A.R.: Analysis of freeway traffic time-series data by using Box-Jenkins techniques. No. 722 (1979) 8. Hernandez, C., Giral, D., Martinez, F.: Radioelectric spectrum prediction based in ARIMA and SARIMA time series models. Int. J. Appl. Eng. Res. 13(22), 15688–15695 (2018) 9. Liao, L., Lin, J., Zhu, Y., Bi, S., Lin, Y.: A bi-direction LSTM attention fusion model for the missing POI identification. J. Netw. Intell. 7(1), 161–174 (2022) 10. Zhang, S.-M., Su, X., Jiang, X.-H., Chen, M.-L., Wu, T.-Y.: A traffic prediction method of bicycle-sharing based on long and short term memory network. J. Netw. Intell. 4(2), 17–29 (2019) 11. Yin, W., Kann, K., Yu, M., Schütze, H.: Comparative study of CNN and RNN for natural language processing. arXiv preprint arXiv:1702.01923 (2021) 12. Yu, Y., Si, X., Hu, C., Zhang, J.: A review of recurrent neural networks: LSTM cells and network architectures. Neural Comput. 31(7), 1235–1270 (2019) 13. Yang, S., Yu, X., Zhou, Y.: LSTM and GRU neural network performance comparison study: taking yelp review dataset as an example. In: 2020 International Workshop on Electronic Communication and Artificial Intelligence (IWECAI), pp. 98–101. IEEE (2020, June) 14. Shewalkar, A.: Performance evaluation of deep neural networks applied to speech recognition: RNN, LSTM and GRU. J. Artif. Intell. Soft Comput. Res. 9(4), 235–245 (2019) 15. Chen, R.-F., Luo, H., Huang, K.-C., Nguyen, T.-T., Pan, J.-S.: An improved honey badger algorithm for electric vehicle charge orderly planning. J. Netw. Intell. 7(2), 332–346 (2022) 16. Zhao, L., Song, Y., Zhang, C., Liu, Y., Wang, P., Lin, T., ... Li, H.: T-GCN: a temporal graph convolutional network for traffic prediction. IEEE Trans. Intell. Transp. Syst. 21(9), 3848–3858 (2019) 17. Kim, J., Kurauchi, F., Uno, N., Hagihara, T., Daito, T.: Using electronic toll collection data to understand traffic demand. J. Intell. Transp. Syst. 18(2), 190–203 (2014) 18. Zou, F., Ren, Q., Tian, J., et al.: Expressway speed prediction based on electronic toll collection data. Electronics 11(10), 1613 (2022) 19. Tian, J., Zou, F., Guo, F., Gu, Q., Ren, Q., Xu, G.: Expressway traffic flow forecasting based on SF-RF model via ETC data. In: International Conference on Frontiers of Electronics, Information and Computation Technologies, pp. 1–7 (2021, May) 20. Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., ... Polosukhin, I.: Attention is all you need. In: Advances in Neural Information Processing Systems, vol. 30 (2017)

Chapter 21

Traffic Flow Prediction of Expressway Toll Station Exit Based on ETC Gantry Data and Attention Mechanism Haolin Wang, Fumin Zou, and Feng Guo

Abstract Comprehending the variation in traffic flow is critical to alleviating traffic congestion at expressway toll station exits. Despite the fact that various traffic flow forecasting models have been proposed, most of them make predictions based on the entry traffic in the area near the target toll station. For origin–destination data like from entry to exit, these methods can hardly capture information on vehicles in transit. In this work, we suggest for the first time predicting toll station exit flows based on expressway gantry data. Moreover, in order to obtain the contribution of multiple gantry series to the exit traffic flow, a recurrent neural network incorporating spatiotemporal attention mechanism is proposed. The proposal not only predicts effectively but also improves the interpretability of the model. Comparative experiments were conducted using data from the gantry system and toll station data of the expressway in Fujian Province, China. The experimental results show that the proposed model performs better than other baseline methods.

21.1 Introduction Toll stations, as the unique channel for vehicles to enter and exit expressways, significantly impact the traffic network in terms of their operation [1–3, 22]. The vehicles leaving the expressways can bring a tremendous amount of traffic input to the ordinary roadway when there is a high traffic volume at toll station exits [4, 23]. This inhomogeneity of traffic flow can lead to toll stations becoming bottlenecks on expressways H. Wang (B) · F. Zou Fujian Key Laboratory of Automotive Electronics and Electric Drive, Fujian University of Technology, Fuzhou 350118, Fujian, China e-mail: [email protected] F. Zou e-mail: [email protected] F. Guo College of Mathematics and Computer Science, Fuzhou University, Fuzhou 350118, Fujian, China e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_21

277

278

H. Wang et al.

[5, 24]. Consequently, robust and accurate predictions of traffic flow at toll station exit assist the expressway management in perceiving congestion in advance and in dredging traffic. ETC (Electronic Toll Collection) gantry system all-weather collects data on vehicles passing through mainline gantries and toll stations, such as trading time, vehicle types, and speeds [6]. Different from MTC (Manual Toll Collection), ETC can record vehicle in-transit information in real time, which facilitates monitoring and evaluating the operational status of the expressway network [7, 8]. Considering the propagation of flows along the expressways network, the exit flows at the target toll station derive from vehicles entering via toll stations in other areas [9]. Existing approaches predict the exit traffic flow of toll stations using the entrance traffic flow of the relevant stations. Based on origin–destination (OD) data like from entry to exit, however, these methods may not yield high accuracy, mainly due to insufficient observations of traffic flow in transit. Inspired by this, this work presents an approach for toll station exit flow prediction based on ETC gantry data and captures the traffic flow contribution of correlated gantries on the exit flow of the target toll station using an attention mechanism. The main contributions of this work are presented as follows: 1. We propose to incorporate ETC gantry data into the exit traffic flow prediction of toll stations. To the best of our knowledge, this is the first study to attempt to select ETC gantry data as relevant input series to make predictions, 2. Spatio-temporal attention mechanisms are integrated within LSTM (Long ShortTerm Memory) network to determine the contribution of the input features across all time steps. In this way, optimal combined weights can be assigned to traffic flows of multiple relevant gantries at different historical periods.

21.2 Related Works Wang et al. [10] earlier proposed to leverage expressway toll data to predict exit traffic flow. They constructed features using toll data and presented a hybrid model of SVM and KNN for prediction. Lin et al. [11] proposed a traffic flow prediction model that combines the series features of exit traffic flow and the spatial–temporal features with the relevant stations. With the coexistence of two types of toll systems on China’s expressway, Wu et al. [12] used LSTM model to predict the exit station’s traffic flow for three different OD traffic data (i.e., Manual Toll Collection, Electronic Toll Collection, and the mix of both). In subsequent work [9] by the group, the entrance station data was split and predicts the target exit station traffic flow using multi-split traffic flows. Shuai et al. [13] presented a hybrid model by stacking full connection layer and LSTM networks to improve the ability of LSTM to capture the spatial correlation of traffic flow. And evaluate the effectiveness of the proposed method using traffic data from 51 toll stations (15% of all toll stations) in Guizhou, China. All of the aforementioned expressway exit flow forecasting studies are based on OD traffic data like the station-to-station. In addition, when multiple station traffic series

21 Traffic Flow Prediction of Expressway Toll Station Exit Based on ETC …

279

are input, recurrent neural networks, a deep neural network designed for sequence data, cannot explicitly select relevant series for prediction. With the large-scale construction and application of ETC gantry systems on expressways, the transaction records generated by the gantry provide plenty of information on vehicles in transit [25, 26]. Motivated by this, we suggest to use the gantry data to predict the toll station exit flow. And utilize recurrent neural networks incorporating attention mechanisms to obtain the traffic contribution of different gantries.

21.3 Methodology 21.3.1 Notation and Problem Definition Definition Given a target toll station, we use X ∈ R N ×T to represent the historical traffic flows passing by N gantries highly correlated with the target toll station, where  T is the length of time series with a 15-min interval. x k = x1k , x2k , . . . , x Tk ∈ R T denotes the traffic volumes series of gantry K , and xt = xt1 , xt2 , . . . , xtN ∈ R N denotes the traffic volumes vector of N gantries at time t. Y = (y1 , y2 , . . . , yT ) ∈ R T represents the historical exit traffic volumes of the target toll station. 

Problem The model aims to predict the most likely exit traffic volume y T +1 in the next time interval T + 1 at the target toll station, i.e.,



y T +1 = Fθ (x1 , x2 , . . . , x T )

(21.1)

where F(·) is prediction function with parameter θ .

21.3.2 Model The historical traffic flow of relevant gantries affects the exit traffic flow of the target toll station, i.e., a typical spatio-temporal correlation of traffic [11]. Therefore, we propose a spatiotemporal attention-based LSTM network for exit traffic flow prediction. The proposed model consists of two layers of LSTM networks. A spatial attention mechanism is embedded in the first layer, where traffic series of highly relevant gantries are assigned higher weights. A temporal attention mechanism is embedded into the second layer in which hidden states across the historical time steps are attributed to different weights. A graphical illustration of the proposed model is shown in Fig. 21.1.

280

H. Wang et al.

Fig. 21.1 Graphical illustration of the proposed model

Spatial attention. Given the time series X = (x1 , x2 , · · · x T ) with xt ∈ R N , the standard LSTM cell [14] maps xt to h t with i t = σ (Wxi xt + Whi h t−1 + bi ),   f t = σ Wx f xt + Wh f h t−1 + b f , ∼

ct = tanh(Wx c˜ xt + Wh c˜ h t−1 + bc˜ ), ∼

ct = f t  ct−1 + i t  ct , ot = σ (Wxo xt + Who h t−1 + bo ), h t = ot  tanh(ct )

(21.2)

where h t and ct denote the hidden and cell states (at time step t). W and b are weights to learn. σ and tanh are activation functions. In order to find highly relevant gantries from xt , an additive attention [15], i.e., a linear layer, is constructed before the first LSTM layer with     Stk = VS tanh W S h t−1 , ct−1 + U S xtk αtk

   k ex p Stk = so f tmax St =  N  i i=1 ex p St

(21.3)

(21.4)

where Stk represents attention score, in the case of x k as key, h t−1 and ct−1 as query, [·, ·] is a concatenation of h t−1 and ct−1 . Vs ∈ R T , Ws ∈ R T ×2m and Us ∈ R T ×T are parametersthat can be jointly  trained with the LSTM network, where m is the hidden size. αt = at1 , at2 , ..., αtN is the attention probability vector that is normalized with the Softmax function. Based on these attention weights, the weighted sequence inputs

21 Traffic Flow Prediction of Expressway Toll Station Exit Based on ETC …

281

of the gantries updated as ∼

xt =

N

αk x k k=1 t t

(21.5)

Then the cell state and hidden state at time t are updated according to Eq. (21.3) with x˜t . Ulteriorly, the hidden state is also the input series for temporal attention module. With the aforementioned spatial attention, the LSTM network pays more attention to the gantries with high traffic contributions. Temporal attention. In order for traffic series to be assigned appropriate attention at all time steps, temporal attention is introduced to selectively focus on the hidden states of the LSTM network across historical periods. The attention weights of the hidden states of the previous LSTM network are calculated with the same score function:      St k = VST tanh W S  h  t−1 , c t−1 + U S  h kt βtk 





= softmax St

k



   exp St k = T  k  i=1 exp St

(21.6)

(21.7)



where h t−1 and ct−1 are the hidden states and cell states of the second LSTM layer. Vs  ∈ R m , Ws  ∈ R m×2 p and Us  ∈ R m×m are parameters that p is the hidden size. Each hidden state h i is mapped to a temporal component for the input of the second LSTM layer: ∼

hk =

T

β k hk t=1 t t

(21.8)

  Then h t−1 and ct−1 are updated according to Eq. (21.3) with h˜ k . The final prediction result is generated by a fully connected layer. Training procedure. We train the model with the Adam optimizer [16] and leverage the mean squared error (MSE) as the loss function to learn parameters:

Loss =

2 1 n  yi − yi , i=1 n 



where n is the number of samples, y i and yi are prediction and ground truth.

(21.9)

282

H. Wang et al.

21.4 Experiments 21.4.1 Datasets Description The real transaction data collected by the ETC gantry system of the Fujian Province highway are used to evaluate the performance of the proposed model. The dataset is provided by Fujian Provincial Expressway Information Technology Co., Ltd. The Xiangqian toll station, one of the highest throughput toll stations in Fujian, is chosen as the target station for exit traffic flow prediction. The relevant gantries that contribute to the target station traffic flow are selected based on historical data and road network topology. The gantry data were sampled in real time, covering the period from 1 to 5 June 2021. In the experiment, the first 80% of the data was used as the training set and the following data as the test set, and was smoothed with a 15-min interval.

21.4.2 Relevant Gantry Selection The relevant gantries were chosen according to OD travel time and Pearson’s coefficient considering time delay [17]. The pseudo-code is shown in Algorithm 1. In the experiments, the threshold is equal to 0.8. To validate the proposed idea, the relevant stations that the exit traffic flow of the target station derives from their entry traffic flow are selected in the same way. The location of the target toll station, the selected gantries, and stations are marked on the map in Fig. 21.2.

Fig. 21.2 Geographical location of the target toll station, relevant gantries, and station on the map. The red dot represents the target station, the green dots represent relevant gantries and the yellow dots represent relevant stations (Map is from https://wprd01.is.autonavi.com)

21 Traffic Flow Prediction of Expressway Toll Station Exit Based on ETC …

283

Algorithm 1. Algorithm for choosing gantries associated with target toll stations. Input The target station and its traffic flow series ; and their traffic flow series The Candidate gantries ; The OD data that the gantry is origin and the station is destination ; The correlation coefficient threshold: ; ; Output The relevant gantries 1)Calculate OD travel time 2)for i in [1, M] do 3) for j in [1, len(ODG1)]1 do 4) TO, TD ← ODGj // Extract timestamps from OD data. 5) Tdiff = TD – TO // Calculate OD travel time of a vehicle. 6) T = T {Tdiff} //// The set of travel time from the gantry to the station . 7) end for 8) Tmean = stats.dist.fit(T).mean() //Calculate the mean travel time by fitting the optimal distribution function. 9) di = Tmean 10) d = d {di}//The set of time delay for each OD. 11)end for 12)Calculate correlation coefficient considering time delay 13) for i in [1, M] do 14)

numerator =

15)

denominator =

16) = numerator/ denominator > then 17) if 18) G = G Gi 19) end if 20)end for 21)Return

21.4.3 Evaluation Metrics and Parameter Settings Three error metrics were considered to evaluate the accuracy of prediction results: Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Square Error (RMSE). Lower error means better performance. The equations are the following:

284

H. Wang et al.

 1 n  yi − yi  i=1 n   1 n  yi − yi  × 100% M AP E = i=1  n yi  2 1 n  RMSE = yi − yi i=1 n 

M AE =

(21.10)



(21.11)



(21.12)

For a fair comparison, the hyperparameters of all baseline approaches are set as follows. We set the learning rate to 0.001, the training epoch to 1000, the batch size to 128, and the number of time steps to 10. In addition, the hidden size of the recurrent neural network is set to 128, i.e., m = p = 128.

21.4.4 Baseline Approaches The proposed model is compared to representative baseline approaches that include the parametric models that presuppose the regression function (ARIMA), nonparametric models (SVR), deep neural network models considering temporal dependence only (LSTM, GRU), and considering spatio-temporal features (CNN-LSTM). Specifically, ARIMA is the Autoregressive Integrated Moving Average Model [18], which builds a regression function to predict traffic flow. SVR (support vector regression) [19, 27, 28] learns the statistical regularity without assumption. LSTM and GRU [20, 29, 30], RNN (recurrent neural network) variants, are designed to learn temporal dependence for time series prediction. For time series data, CNN-LSTM [21] takes into account both the spatial correlation of the surrounding area and the temporal correlation. In this work, all RNN-based models are set up as two layers for a fair comparison.

21.4.5 Results and Analysis Overall Performance. Table 21.1 summarizes the performance of different methods for exit traffic flow prediction at the Xiangqian toll station. The comparison covers input series based on gantries traffic flow and stations entry traffic flow. For all deeplearning-based models (i.e., CNN-LSTM, Double-GRU, Double-LSTM), 10 times independent tests are conducted and the average and standard deviations of the results are reported. As can be seen from Table 21.1, the prediction errors of our model are generally lower than other baselines. ARIMA is the least effective because only the target toll station traffic series is considered and the relevant gantries or stations cannot be

0.81 0.097 ± 0.008 0.105 ± 0.008 0.11 ± 0.01 0.095 ± 0.01

55.31

19.02 ± 1.53

20.37 ± 1.72

21.43 ± 3.16

17.76 ± 1.46

SVR

CNN-LSTM

Double-GRU

Double-LSTM

The proposal

RMSE

23.68 ± 1.67

29.1 ± 4.43

27.36 ± 2.28

25.35 ± 2.01

69.61

123.16

19.91 ± 1.04

24.00 ± 1.68

21.80 ± 0.83

20.68 ± 1.16

52.5

–

0.106 ± 0.007

0.115 ± 0.005

0.108 ± 0.004

0.106 ± 0.008

0.82

–

MAPE

MAE

0.97

MAPE

MAE

105.28

Based on entry traffic flows

Based on gantry traffic flows

ARIMA

Methods

Table 21.1 Prediction results of different methods for exit traffic flow at the Xiangqian toll station

27.50 ± 1.47

34.55 ± 2.56

30.76 ± 0.99

28.95 ± 1.76

67.7

–

RMSE

21 Traffic Flow Prediction of Expressway Toll Station Exit Based on ETC … 285

286

H. Wang et al.

Fig. 21.3 Line plot of predicted flows at Xiangqian toll station exit on June 5

considered. SVR, while more accurate than ARIMA, has higher errors than RNNbased models due to the difficulty of mapping a high-dimensional feature space, such as the input series of multiple relevant gantry traffic flows. For deep learning networks, our approach achieves the best results, verifying the effectiveness of recurrent networks that integrate spatial and temporal attention mechanisms. Figure 21.3 visualizes the predicted result of our model and the Double-LSTM based on gantry traffic flows, which shows that the proposal better fits the ground truth. Spatio-temporal attention validation. The evolution of the spatial attention weights assigned to the relevant toll stations during the 3000 iterations is shown in Fig. 21.4. The weights produce significant differences at about the 500th iteration and gradually smooth at the 2000th iteration. To further verify the effectiveness of the spatial attention mechanism, non-relevant series is incorporated into the original input series. Specifically, in the gantry traffic data, a noisy series is generated by randomly permuting a gantry traffic flow. The weight evolution is shown in the figure, where the noisy series is assigned the lowest weight. This indicates that the spatial attention mechanism adaptively selects the relevant gantry series and restrains the noisy series. For the temporal attention mechanism, the evolution of temporal weights of the proposal with different time steps is presented in Fig. 21.5. Different from spatial attention weights, the temporal attention weights have an earlier divergence and faster convergence rate, smoothing out at roughly 500 iterations. There is a common phenomenon in the figure, where the recent time step is assigned the highest weight in the case of time steps varies from 3, 5, 10 to 15. The reason for this may be that the evolution trend of recent traffic flow will continue in the future with high probability.

21 Traffic Flow Prediction of Expressway Toll Station Exit Based on ETC …

287

Fig. 21.4 Evolution of spatial attention weights on gantry dataset. The 21 weights on 21 original input series (top). The 22 weights on 21 original input series and a noisy series where 0 denotes the noisy series (bottom)

Dataset comparison. From Table 21.1 we can also observe that the experimental performance based on the gantry dataset is better than that based on the station dataset with the same prediction method. We figure that the gantry systems can detect vehicles in transit on the expressway, differing from ‘black box’ process like stationto-station. Furthermore, there is ambiguity in route choice of vehicles entering the expressway from toll stations, while the vehicle trajectories recorded by the gantry system are unidirectional. Especially in the key sections of the expressway, it is probable that most vehicles have the same destination. According to pre-analysis, we found that the number of vehicles from the relevant stations to the target station only accounts for 24.96% of the total number of vehicles entering the relevant stations. And for the five gantries circled in Fig. 21.2, the proportions of the number of vehicles passing through them to reach the target station exceeds 50% of the total number, which has a more robust temporal and spatial correlation. From the above analysis, the prediction of toll station exits traffic flow based on gantry data performs better than that based on entrance traffic data.

21.5 Conclusion This work is the first to leverage the gantry data to predict the exit traffic flow of toll stations, which provides new ideas for future work. And in order to capture the contribution of different gantries to the toll station exit traffic, a spatio-temporal attention mechanism is introduced on the LSTM network to select the more relevant

288

H. Wang et al.

Fig. 21.5 Evolution of temporal attention weights on gantry dataset. The evolution line plots from top to bottom correspond to the length of time steps 3, 5, 10, and 15, respectively

gantry traffic series. Comparative experiments on the gantry dataset and the station dataset show that the proposed model outperforms the other methods for traffic flow forecasting and that the prediction results based on gantry data are more accurate. The expressway gantry system records the section flow (i.e., the number of vehicles passing each gantry), which cannot provide insights on interconnections between gantries. Based on the gantry topology of the expressway network, the intact vehicle trajectories can be constructed as a complement to the spatio-temporal modeling.

21 Traffic Flow Prediction of Expressway Toll Station Exit Based on ETC …

289

Therefore, incorporating trajectory data furnish a promising direction for expressway toll station exit traffic flow prediction.

References 1. Navandar, Y.V., Dhamaniya, A., Patel, D., Chandra, S.: Traffic flow analysis at manual tollbooth operation under mixed traffic conditions. J. Transp. Eng., Part A: Syst. 145(6), 04019023 (2019) 2. Huang, Y., Chen, L., Xia, Y., Qiu, X., et al.: Design of expressway toll station based on neural network and traffic flow. Am. J. Oper. Res. 8(03), 221 (2018) 3. Pang, G.: The forecast and analysis of toll channel scale of expressway toll station. In: E3S Web of Conferences 283 (2021) 4. Huang, Z., Lin, P., Lin, X., Zhou, C., Huang, T.: Spatiotemporal attention mechanism-based multistep traffic volume prediction model for highway toll stations. Arch. Transp. 61(1), 21–38 (2022) 5. Neuhold, R., Garolla, F., Sidla, O., Fellendorf, M.: Predicting and optimizing traffic flow at toll plazas. Transp. Res. Procedia 37, 330–337 (2019) 6. Chang, Z., Li, H., Kang, N.: Research on freeway on-ramp traffic control model based on etc gantry data. In: Fifth International Conference on Traffic Engineering and Transportation System (ICTETS 2021), 12058, 1350–1363 (2021) 7. Guo, F., Zou, F., Luo, S., Liao, L., Wu, J., Yu, X., Zhang, C.: The fast detection of abnormal etc data based on an improved dtw algorithm. Electronics 11(13), 1981 (2022) 8. Zou, F., Guo, F., Tian, J., Luo, S., Yu, X., Gu, Q., Liao, L.: The method of dynamic identification of the maximum speed limit of expressway based on electronic toll collection data. Sci. Program. 2021, 1350–1363 (2021) 9. H. Ruan, B. Wu, B. Li, Z. Chen, and W. Yun: Expressway exit station short-term traffic flow prediction with split traffic flows according originating entry stations. IEEE Access 9, 86 285–86 299 (2021) 10. Wang, X., An, K., Tang, L., Chen, X.: Short term prediction of freeway exiting volume based on svm and knn. Int. J. Transp. Sci. Technol., Part A: Syst. 4(3), 337–352 (2015) 11. Lin, Y., Wang, R., Zhu, R., Li, T., Wang, Z., Chen, M.: The short-term exit traffic prediction of a toll station based on lstm. In: International Conference on Knowledge Science, Engineering and Management, 462–471 (2020) 12. Chen, Z., Wu, B., Li, B., Ruan, H.: Expressway exit traffic flow prediction for etc and mtc charging system based on entry traffic flows and lstm model. IEEE Access 9, 54 613–54 624 (2021) 13. Shuai, C., Wang, W., Xu, G., He, M., Lee, J.: Short-term traffic flow prediction of expressway considering spatial influences. J. Transp. Eng., Part A: Syst. 148(6), 04022026 (2022) 14. Yu, Y., Si, X., Hu, C., Zhang, J.: A review of recurrent neural networks: Lstm cells and network architectures. Neural Comput. 31(7), 1235–1270 (2019) 15. Chorowski, J. K., Bahdanau, D., Serdyuk, D., Cho, K., Bengio, Y.: Attention-based models for speech recognition. Advances in neural information processing systems 28, (2015) 16. Cho, K., Van Merrienboer, B., Bahdanau, D., Bengio, Y.: On the properties of neural machine translation: Encoder-decoder approaches. arXiv preprint, arXiv:1409.1259 (2014) 17. Xia, Y., Liang, Z., Wang, G.: Research of short-term traffic flow forecasting model based on spatio-temporal analysis. J. Nanjing Univ. (Nat. Sci.) 46(05), 552–560 (2010) 18. Kumar, S.V., Vanajakshi, L.: Short-term traffic flow prediction using seasonal arima model with limited input data. Eur. Transp. Res. Rev. 7(3), 1–9 (2015) 19. Wu, C.-H., Ho, J.-M., Lee, D.-T.: Travel-time prediction with support vector regression. IEEE Trans. Intell. Transp. Syst. 5(4), 276–281 (2004) 20. Fu, R., Zhang, Z., Li, L.: Using lstm and gru neural network methods for traffic flow prediction. In: 2016 31st Youth Academic Annual Conference of Chinese Association of Automation (YAC), 324–328 (2016)

290

H. Wang et al.

21. Shi, X., Chen, Z., Wang, H., Yeung, D.-Y., Wong, W.-K., Woo, W.-c.: Convolutional lstm network: A machine learning approach for precipitation nowcasting. Adv. Neural Inf. Process. Syst. 28 (2015) 22. Wu., Tsu-Yang, Lee, Z., Yang, L., Chen, C.-M.: A provably secure authentication and key exchange protocol in vehicular ad hoc networks. Secur. Commun. Netw. 2021, 9944460 (2021) 23. Wu., Tsu-Yang, Lee, Z., Yang, L., Luo, J.-N., Tso, R.: Provably secure authentication key exchange scheme using fog nodes in vehicular ad-hoc networks. J. Supercomput. 77, 6992–7020 (2021) 24. Kumar, V., Kumar, R., Kumar, V., Kumari, A., Kumari, S.: RAVCC: robust authentication protocol for RFID based vehicular cloud computing. J. Netw. Intell. 7(3), 526–543 (2022) 25. Chen, C.-M., Chen, L., Gan, W., Qiu, L., Ding, W.: Discovering high utility-occupancy patterns from uncertain data. Inf. Sci. 546, 1208–1229 (2021) 26. Gan, W., Chen, L., Wan, S., Chen, J., Chen, C.-M.: Anomaly Rule Detection in Sequence Data. In: IEEE Transactions on Knowledge and Data Engineering (2021) 27. Zhang, F., Wu., Tsu-Yang, Pan, J.-S., Ding, G., Li, Z.: Human motion recognition based on SVMin VR art media interaction environment. HCIS 9, 40 (2019) 28. Wu, Q., Zang, B.-Y., Zhang, Y., Qi, Z.-x.: Wavelet Kernel twin support vector machine. J. Inf. Hiding Multimed. Signal Process. 12(2), 93-101 (2021) 29. Zhang, S.-M., Su., Xin, Jiang, X.-H., Chen, M.-L., Wu., Tsu-Yang: A traffic prediction method of bicycle-sharing based on long and short term memory network. J. Netw. Intell. 4(2), 17–29 (2019) 30. Liao, L., Lin, J., Zhu, Y., Bi, S., Lin, Y.: A Bi-directionLSTM attention fusion model for the missing POI identification. J. Netw. Intell. 7(1), 161–174 (2022)

Chapter 22

Expressway Short-Term Traffic Flow Forecasting Considering Spatio-Temporal Features of ETC Gantry Gen Xu, Fumin Zou, Junshan Tian, Feng Guo, and Qiqin Cai Abstract Through the application and expansion of expressway ETC gantry transaction data, we propose a short-term traffic flow forecasting of expressway based on the Kalman Filtering (KF) and Random Forest (RF) model, which not only takes into account the basic external features and periodic features but also considers the spatio-temporal correlation relationship in the road section, so as to construct the spatial correlation features and temporal correlation features. In this paper, we use the ETC gantry transaction data of Fuzhou–Xiamen section of the expressway to forecast and verify in Fujian Province, China, the final results show that: When the rolling window is 20 min, compared with the results before and after Kalman Filtering algorithm processing traffic flow data, the performance indicators is greatly improved, which verifies the positive effect of Kalman Filtering algorithm; it is also verified that the constructed features have a great influence on traffic flow forecasting and play a positive role in improving forecasting accuracy; and it is also verified that the RF model has better forecasting effect than the baseline models.

G. Xu (B) · F. Zou Fujian Key Laboratory of Automotive Electronics and Electric Drive, Fujian University of Technology, Fuzhou 350108, China e-mail: [email protected] F. Zou e-mail: [email protected] J. Tian Fujian Expressway Science & Technology Innovation Research Institute Co., Ltd, Fuzhou 350001, China e-mail: [email protected] F. Guo Fujian Key Laboratory of Automotive Electronics and Electric Drive, Fuzhou University, Fuzhou 350108, China e-mail: [email protected] Q. Cai Fujian Key Laboratory of Automotive Electronics and Electric Drive, Huaqiao University, Xiamen 361021, China © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_22

291

292

G. Xu et al.

22.1 Introduction With the continuous increase in car ownership, there will be a phenomenon that the increase of expressway mileage cannot meet the current growth rate of vehicles, resulting in increasingly serious congestion, especially during holidays and rush hours, expressway congestion has become the norm, restricting the efficient operation of traffic system. In the face of increasingly severe traffic conditions, the construction of an Intelligent Transport System (ITS) can effectively relieve road traffic pressure [1–5]. ITS can interpret traffic conditions accurately in real time and create new traffic management models, thus enabling the rational allocation of traffic resources to meet the growing demand for travel. Traffic flow forecasting is an important function of ITS, and it is one of the key technologies to ensure ITS reliability. Traffic flow forecasting is based on historical traffic flow data to forecast the traffic flow condition of the road section in the future period by exploring the inherent laws of traffic flow. Traffic flow forecasting includes long-term traffic flow forecasting and short-term traffic flow forecasting, compared with long-time traffic flow forecasting, short-term traffic flow forecasting is more highly nonlinear and the data is also unstable and random, so it is very difficult and challenging to forecast short-term traffic flow accurately in real time. Real-time and accurate expressway short-term traffic flow forecasting is an important basis for traffic guidance, which can alleviate traffic congestion, reduce traffic accidents and reduce traffic pressure effectively, and is of great significance to improve expressway transportation efficiency and service level [6–8]. Therefore, how to improve the accuracy, effectiveness, and stability of short-term traffic flow forecasting on the expressway has become one of the focuses and hotspots of domestic and foreign scholars. In the research of short-term traffic flow, many domestic and foreign scholars are based on the condition that the traffic state of expressway remains unchanged, and by identifying the inherent mode of traffic flow to forecast traffic flow [9]. In the early research stage, the traditional statistical method model should be the most extensive, mainly including Historical Average forecasting method (HA) [10], time series model, etc. However, the statistical method models are relatively simple, which can only capture the linear features in the traffic flow data and cannot capture nonlinear features, moreover, most of them are only suitable for the situation of smooth traffic flow, when the traffic flow is affected by external conditions and changes drastically, these models are inapplicable, which will have a great impact on the traffic flow forecasting, and resulting in a low accuracy of the forecasting. In order to further improve the accuracy of traffic flow forecasting, many other models are applied to traffic flow forecasting, such as K-Nearest Neighbor model (KNN) [11], Support Vector Machine model (SVM) [12, 13], Decision Tree model (DT) [14], Random Forest model (RF) [15], Gradient Boosting Decision Tree model (GBDT) [16], Long Short-Term Memory, LSTM [17–20], etc., these machine learning models have good results in dealing with the nonlinear features and some external features of traffic flow, which can improve the accuracy of traffic flow forecasting greatly. However, due to the complexity and instability of traffic flow, it is difficult for a

22 Expressway Short-Term Traffic Flow Forecasting Considering …

293

single model to adapt to all scenarios sometimes, at the same time, in the process of collecting the original traffic flow data, it may be affected by spatio-temporal characteristics, such as noise and outliers, etc., these influencing factors will reduce the effect of traffic flow forecasting. In this paper, we propose a hybrid model based on Kalman smoothing model and Random Forest (KF-RF) to forecast traffic flow, so as to improve the accuracy, effectiveness, and stability of expressway short-term traffic flow forecasting. The main work of this paper is as follows: 1. The non-stationary and nonlinear problems of traffic flow are mainly caused by the noise and error in the data. In view of these situations, we propose the Kalman Filtering algorithm to carry out reasonable noise reduction and smoothing on the traffic flow data to solve the non-stationary and nonlinear problems of traffic flow, and finally, get the consistent traffic flow data; 2. In order to obtain better traffic flow forecasting results, a variety of traffic flow features are constructed to reflect the changing laws of traffic flow, including periodic features, temporal correlation features, spatial correlation features, and external features; 3. In order to make full use of the constructed traffic flow features, this paper uses the Random Forest algorithm for traffic flow forecasting and comparing the forecasting results of the baseline model.

22.2 Methodology In this section, we mainly describe some definitions of short-term traffic flow forecasting and algorithm in the whole experiment process, including Kalman Filtering algorithm and Random Forest algorithm, as well as the features we constructed. Figure 22.1 is the basic process framework of forecasting algorithm in this paper.

22.2.1 Related Definitions ETC gantry transaction data has rich vehicle traffic and charging information, the main task of this paper is to fully mine the data characteristics to make more accurate traffic flow forecasting based on these data. In order to express more clearly, it is necessary to describe the necessary symbols and some key definitions: Definition 1: Each ETC gantry on the expressway is called N ode, and two adjacent gantries form an expressway section, referred to as Q D: Q D i =

(22.1)

where N odei is the i-th gantry and the starting point of section Q D i , N odei+1 is the end point of section Q D i , and N odei+1 is also the starting point of section Q D i+1 .

294

G. Xu et al.

Fig. 22.1 The basic framework of forecasting model (Light blue represents the main steps)

Definition 2: Expressway sections by adjacent gantry nodes form road section, referred to as RS: RS =

(22.2)

The schematic diagram of road section is shown in Fig. 22.2. Definition 3: Rolling window is a given unit time length that can frame the time period to calculate the number of vehicles in the framed time period, it is equivalent to a slider with a specified length moving on the scale, and the statistical data in the slider can be fed back every time it moves. Rolling interval is the fixed length of time that the rolling window moves in the time period, which is equivalent to the

Fig. 22.2 The schematic diagram of road section(R S)

22 Expressway Short-Term Traffic Flow Forecasting Considering …

295

Fig. 22.3 The schematic diagram of rolling

distance that the slider moves once on the scale. Rolling window is represented by the symbol , rolling interval is represented by the symbol σ, and the schematic diagram of rolling is shown in Fig. 22.3.

22.2.2 Data Transformation Under the given rolling window  and rolling interval σ, count the number of vehicles passing through the expressway ETC gantry and recorded as traffic flow. Traffic flow data passed through a gantry can be represented by X i : T  X i = Z 1i Z 2i Z 3i · · · Z ki · · · Z ni

(22.3)

For the entire expressway road section, the traffic flow data can be represented as XM: T  XM = X1 X2 X3 · · · Xi · · · Xm

(22.4)

296

G. Xu et al.

22.2.3 Kalman Filtering Algorithm The Kalman Filtering (KF) algorithm is a statistical estimation method initially proposed by Kalman in the 1960s [21], with the research development of many scholars over the years, it has been widely applied in many different research fields. The smoothing method based on Kalman Filtering can process a series of actual measurement statistics data with errors, so as to obtain the optimal estimation, it is an effective method for data processing with noise and can effectively improve the accuracy of data. The original data will be affected by many factors in the process of collection, such as the failure of ETC gantry results in failing to collect the vehicle passing time or the false detection of the gantry, which will lead to data errors in the statistics of traffic flow data. In order to ensure the rationality of the data, it is necessary to smooth the original traffic flow data. The state equation of traffic flow error is expressed as 

i + BUki + Wki Z ki = AZ k−1 i Wk ∼ N (0, Q)

(22.5)

i where Z ki and Z k−1 represent the system state variables in k and k − 1 states, which are the real traffic flow; A represents the state transition matrix; Uki represents the control input variable in k state Uki ; B represents the optional control input gain, but in most cases, there is no control input gain; Wki represents the system process noise, which is the traffic flow noise in k state; and Q represents the covariance of the system noise. In k state, the traffic flow system measurement equation is



Yki = H Z ki + Vki Vki ∼ N (0, R)

(22.6)

where Yki represents the observed variable of k state; H represents the gain of the state variable Z ki to the observed variable Yki ; Vki represents the measurement noise; and R represents the covariance of the measurement noise. Then, updating the measurement variables: Estimating the result Z ki in k state i in k − 1 state, the updating equation based on the known system state variables Z k−1 is ⎧ i i ⎪ Z k/k−1 = AZ k−1/k−1 + BU ik ⎪ ⎪ ⎨ Pi i T k/k−1 = A Pk−1/k−1 A + Q (22.7) i i i ⎪ Z k/k = Z k/k−1 + K g ik [Yki − H Z k/k−1 ] ⎪ ⎪ ⎩ i K g ik = Pki H T /[H Pk/k−1 H T + R] i where Z k/k−1 represents the priori state estimate in k state, which is an estimated i i value based on Z k−1/k−1 ; Z k−1/k−1 represents the posteriori state estimate in k − 1

22 Expressway Short-Term Traffic Flow Forecasting Considering …

297

i state, which is the optimal estimate and the output traffic flow result; Pk/k−1 represents i the priori state estimate covariance in k state; Pk−1/k−1 represents the posteriori state i estimate covariance in k − 1 state; Z k/k represents the posteriori state estimate in k state, which is the optimal estimate; and K g ik represents Kalman gain. Q and R are input to the Kalman Filtering as conditions and are the main parameters of the experiment, which is difficult to determine in the Kalman Filtering and mainly through experiments to find the optimal value. i of traffic flow has been obtained in k state, but the The optimal estimate Z k/k Kalman Filtering will continue to operate until the end to obtain the final optimal estimate of the traffic flow. Then, updating time status:



i i Pk/k = [1 − K g ik H ]Pk/k−1 i i Pk/k = Pk+1/k

(22.8)

i where Pk/k represents the posteriori state estimate covariance in k state; when the i i state is updated to k + 1 state, Pk/k is the a priori estimate covariance Pk+1/k in k + 1 state. The traffic flow data of one gantry after Kalman Filtering can be represented by Ki:

T

i i i i i K i = Z 1/1 · · · Z k/k Z 2/2 Z 3/3 · · · Z n/n

(22.9)

For expressway road section (RS), the traffic flow data after Kalman Filtering can be expressed by as K M : T  KM = K 1 K 2 K 3 · · · Ki · · · K m

(22.10)

The schematic diagram of Kalman Filtering is shown in Fig. 22.4.

22.2.4 Feature Construction In order to obtain better traffic flow forecasting results, a variety of traffic flow features are constructed to reflect the changing laws of traffic flow, including periodic features, temporal correlation features, spatial correlation features, and external features. Periodic features It is found that under a certain time scale, the daily traffic flow and weekly traffic flow show periodic changes by analyzing the expressway traffic flow data, that is, the traffic flow shows a certain repeatability every time period in an expressway section. therefore, the traffic flow data are decomposed into time series with 1 and 1 week as cycles, respectively, and the periodic component Z iS is regarded as an independent

298

G. Xu et al.

Fig. 22.4 The schematic diagram of Kalman Filtering

variable, which is the periodic features of traffic flow in an expressway section, the periodic features of traffic flow in expressway road section are represented by Z S : i i i − Tt/t − et/t Z iS = Z t/t

(22.11)

where the periodic component Z iS represents the orderly fluctuation of traffic flow in i reflects a a certain section, which has a strong regularity; the trend component Tt/t trend or state of continuous development and change of traffic flow data over a long i represents the influence of many period of time; and the residual component et/t accidental factors on the time series, which has randomness. Temporal correlation features In the time dimension, the traffic flow under the current rolling window can be regarded as the continuation of the traffic flow under the previous rolling window when the vehicle is driving on the expressway, the traffic flow under the next rolling window can be seen as the future trend of the traffic flow under the current rolling window, in large-scale time, the traffic flow conditions of many days ago may have a profound impact on the current traffic flow, therefore, traffic flow will be restricted and influenced by historical traffic flow state with the change of time. The trend of current traffic flow changes has a strong correlation with the trend of historical traffic flow changes. As shown in Fig. 22.5, under the current section Q D i , the traffic flow in k state is the continuation of the traffic flow in k − 1 state, and also the large-scale continuation of the traffic flow in k − x state; meanwhile, the traffic flow in k + x state

22 Expressway Short-Term Traffic Flow Forecasting Considering …

299

Fig. 22.5 Road section spatio-temporal characteristics

is also a large-scale continuation of the traffic flow in k state. Therefore, under the current rolling window, the traffic flow under the previous rolling window, the traffic flow under the previous two rolling windows, the traffic flow under the rolling window of the previous day, and the traffic flow under the rolling window of the previous 2 days are constructed to construct the temporal correlation feature, respectively, i , the temporal correlation features of the current section are represented by Z time the temporal correlation features of the expressway road section are represented by Z time . Spatial correlation features In the spatial dimension, the node locations on the expressway are different and affected by the actual road conditions, the change law and flow size of traffic flow in different expressway sections are different. However, the expressway section has continuity, the traffic flow potential of the current section will affect the traffic flow potential of the next section and the traffic flow potential of the next section will also be transferred to the previous section through the fluctuation of traffic flow, the traffic flow of the current section can be regarded as the continuation of the traffic flow of the previous section and the traffic flow of the next section can also be regarded as the continuation of the traffic flow of the current section, at the same time, the current section will not only affect the traffic flow potential of adjacent sections but also affect the traffic flow potential of non-adjacent sections. The traffic flow between sections has a high correlation, and taking the spatial relationship between sections as a feature has a great reference value. As shown in Fig. 22.5, in k state, the traffic flow status of current section Q D i is the continuation of Q D i−1 and the traffic flow status of section Q D i+1 is the continuation of Q D i ; meanwhile, the current traffic flow in section Q D i is also correlated with section Q D i−y and section Q D i+y in k state. Therefore, taking the traffic flow between sections under the same rolling window as the expression form, the traffic flow of the upper section, the traffic flow of the upper two sections, the traffic flow of the next section, and the traffic flow

300

G. Xu et al.

of the next two sections are constructed to construct the spatial correlation features, respectively, the spatial correlation features of the current section are represented by i , the spatial correlation features of the expressway road section are represented Z topo by Z topo . External features Vehicle travel also has strong regularity, such as: What day is the day of travel, whether it is a working day, whether the vehicle travel time is at the peak of the day, and whether the vehicle travel time is in the morning or afternoon, all these external factors may have an impact on traffic flow forecasting. Therefore, it is necessary to extract these external factors to construct external features, the external features of the current section are represented by Z ei , the external features of the expressway road section are represented by Z e . The above features are obtained by mining the internal information of expressway traffic flow, which provides solid data support for improving the results of traffic flow forecasting.

22.2.5 Random Forest Random Forest [22] was proposed by Tin Kam Ho of Bell LABS in 1995, this model can process high-dimensional data, make full use of features without feature selection, and has strong generalization ability and fast operation speed. Random Forest mainly adopts bootstrap resampling to conduct random sampling on the dataset, assuming that the dataset is composed of data for S objects, R sub-datasets composed of Sr objects are randomly extracted from the dataset based on bootstrap resampling, because there is a put-back takeout, these sub-datasets may have data duplication or even the same data, but each sub-dataset is independent of each other, and these sub-datasets make up the training dataset of the decision tree. The data of each subdataset contain all the constructed features, and the features are randomly selected from all the features, the number of randomly selected features should not be more than all the constructed features, forming R sub-data training set sr , then, a single decision tree model is trained on the R sub-datasets to form a “forest” containing R decision trees. Each decision tree will give its own forecasting results according to the selected features of the sub-data set, and the forecasting results of Random Forest are averaged according to the forecasting results of all decision trees. The schematic diagram is shown in Fig. 22.6.

22.3 Experiment In this section, we mainly describe some definitions of short-term traffic flow forecasting and algorithm in the whole experiment process, including Kalman Filtering

22 Expressway Short-Term Traffic Flow Forecasting Considering …

301

Fig. 22.6 The schematic diagram of Random Forest

algorithm and Random Forest algorithm, as well as the features we constructed. Figure 22.1 is the basic process framework of forecasting algorithm in this paper.

22.3.1 Data Description and Evaluating Indicators All the data of this experiment are provided by Fujian Expressway Information Technology Co., Ltd, and uses the vehicle transaction data of the Fuzhou–Xiamen section of the Fujian Expressway ETC gantry system from May 3 to May 31, 2021, including ETC vehicle data and CPC vehicle data, after removing part of the error data, including incomplete timestamp and the false detection of the gantry, etc., which 18,703,163 ETC vehicle data and 6,012,788 CPC vehicle data. The total amount of data is 24715951, involving 22 ETC gantries of expressway in total, the main attributes of vehicle transaction data are shown in Table 22.1. In order to better compare and evaluate the forecasting results of the model, the Root Mean Square Error (R M S E), Mean Absolute Error (M AE), and Coefficient of Determination (R 2 ) are used to evaluate the accuracy of the forecasting method, where R M S E and M AE represent the degree to which the traffic flow forecasting results deviate from the true results, namely, the actual size of the forecasting error. Table 22.1 Main attributes of transaction data of ETC gantry system

Attribute name

Examples

Attribute name

Examples

TradedID

G0015…0001

OBUPlate

********

TradeTime

2020–05-16 00:00:15

VehClass

1

FlagID

34023F

FlagIndex

1

FlagType

0

OBUID

0******5

302

G. Xu et al.

The smaller the result is, the smaller the forecasting model error is and the better the forecasting effect is; R 2 represents the correlation between the traffic flow forecasting result and the true result, and the closer it is to 1, the better the forecasting effect is, the specific formulas are as follows: Root Mean Square Error (R M S E)

N

1 2 (y − yi ) RMSE =  N i=1 

(22.12)

Mean Absolute Error (M AE) M AE =

N  1  y − yi  N i=1 

(22.13)

Coefficient of Determination (R 2 ) N 

R2 = 1 −

i=0 N 



2



2

(y − yi )

(22.14) (y − yi )

i=0 

where yi is the real value of traffic flow, y is the forecasted value of traffic flow, N is the sample  N size of 2traffic flow, and yi is the average of the real value of traffic flow, yi = N1 i=1 yi , R ∈ [0, 1].

22.3.2 Selection of Kalman Filtering Parameters Using Kalman Filtering to smooth the original traffic flow data, only guarantee the result value is unbiased and variance is small, can the smoothing effect be better, therefore, the parameters selection of the covariance of system noise Q and the covariance of the measurement noise R is critical, their quality will directly affect the effect of smoothing. In the selection process of Q and R, the principle to be followed: the D-value of traffic flow data before and after Kalman Filtering smoothing cannot be too large or too small, if the D-value is too large, it will lead to data distortion, if the D-value is too small, the effect is not obvious, and the algorithm will lose its effect, therefore, only finding appropriate parameters can ensure the accuracy of traffic flow forecasting. When adjusting parameters, the way is to fix one parameter and change the other. When  = 20 min, taking the vehicle traffic flow data passing through the 34023F gantry as an example. The experiment is carried out by fixing a parameter and changing a parameter, respectively, the results show that when Q

22 Expressway Short-Term Traffic Flow Forecasting Considering …

303

Fig. 22.7 Comparison of Kalman Filtering results (a)

= 1e−3 and R = 1e−3, the D-value of traffic flow data before and after Kalman Filtering smoothing is within a reasonable range, which is the appropriate parameter value, the comparison results are shown in Figs. 22.7 and 22.8:

22.3.3 Comparison of Results In this study, Table 22.2 expresses the comparison results between the Random Forest model and other baseline models, when  = 20 min, compared with five baseline models, including Historical Average (HA), Decision Tree model (DT), AdaBoost, K-Nearest Neighbor model (KNN), and XGBoost, from the results, the Random Forest model has higher forecasting accuracy than the five baseline models, R M S E increases by 5936%, 12.0%, 62.4%, 180.6%, and 4.3%, respectively, M AE increased by 8166%, 25.6%, 110.4%, 212.3%, and 8.8%, respectively, indicating that Random Forest model has greater advantages in this study. After the original traffic flow data is smoothed by Kalman filtering, when  = 20 min, R M S E and M AE increased by 34.4% and 40.4%, respectively, which shows that Kalman Filtering has a positive effect on traffic flow forecasting, and also makes traffic. The flow forecasting has higher accuracy, the result is shown in Fig. 22.9.

304

G. Xu et al.

Fig. 22.8 Comparison of Kalman Filtering results (b) Table 22.2 Comparison with other baselines under different rolling window Rolling window () 20 min

Evaluating indicators

HA

DT

ADaBoost

KNN

XGBoost

RF

RMSE

155.36

25.09

36.39

62.85

23.37

22.40

M AE

124.659

17.08

28.62

42.47

14.79

13.6

R2

Models

0.811

0.993

0.985

0.954

0.994

0.995

Evaluating Indicators

Fig. 22.9 Comparison before and after Kalman smoothing

30.1 19.1 0.99

R2_score

13.6

40 20

22.4

0.995

0

Original Smoothing data data

RMSE MAE R2_score

22 Expressway Short-Term Traffic Flow Forecasting Considering …

305

22.3.4 Feature Influence Different features have different effects on the results, in order to verify the effects of different features on traffic flow forecasting, multiple groups of experiments are set up to verify. Experiments are shown in Table 22.3. Among them, the external features are the basic features, which need to be considered in each group of experiments; combining other features with date features to set up multiple groups of experiments, the results show that, the forecasting effect of the model is very poor when only considering external features; after adding periodic features, temporal correlation features, and spatial correlation features, respectively, the forecasting effects of the model have increased greatly; when multiple features are combined separately, it is found that there is a certain promotion effect, and the construction of spatial correlation features makes the model more perfect; when all features are input into the model, the forecasting effect is the best, indicating that the constructed features have a great impact on traffic flow forecasting and a positive promotion on improving the forecasting accuracy, the results are shown in Table 22.4. Table 22.3 Features setting model

R Fe R F e, p R F e,s R F e,t R F e, p,s R F e, p,t R F e,t,s R F e, p,t,s

External features Ze

Periodic feature Z S

√ √

Temporal correlation feature Z time

√

√

√

√

√

√

√

√

√

√ √

√ √

Spatial correlation features Z topo

√ √

√

√

Table 22.4 Result of features setting Rolling window ()

model

RMSE

M AE

R2

20 min

R Fe

169.4

122.7

0.670

R F e, p

121.3

79.6

0.830

R F e,s

87.6

50.5

0.912

R F e,t

25.5

14.8

0.992

R F e, p,s

82.8

48.2

0.921

R F e, p,t

25.6

15.2

0.992

R F e,t,s

25.0

14.9

0.993

R F e, p,t,s

22.4

13.6

0.994

306

G. Xu et al.

Fig. 22.10 Fitting diagram of Traffic flow forecasting results (8 sections)

Figure 22.10 shows the comparison results of traffic flow data and forecasted data for 7 days from May 24 to May 30, 2021, according to the specific conditions of traffic flow in eight sections when  = 20 min, the overall change trend of traffic flow is similar, indicating that Kalman Filtering–Random Forest model has high reference value, it can provide strong support for expressway decision-making.

22.4 Conclusion Accurately forecasting the short-term traffic flow of expressway can not only provide decision-making assistance for expressway managers but also alleviate traffic congestion, reduce traffic accidents, and reduce traffic pressure effectively, in this paper, we propose a short-term traffic flow forecasting based on Kalman Filtering–Random Forest model to improve the accuracy of traffic flow forecasting. The experimental results show that: compared with several baseline models, the Random Forest model has higher forecasting accuracy; in addition, it is also found that there is a strong correlation between sections, the traffic flow between sections affects each other and mining the features of traffic flow has a great impact on traffic flow forecasting. The research of this paper can spread to the entire expressway network eventually, providing strong support for the management decision-making of relevant departments. In the future, we will further increase the research on the overall expressway network, not only considering the correlation between adjacent sections but also

22 Expressway Short-Term Traffic Flow Forecasting Considering …

307

discussing the correlation between further sections, and also considering the traffic flow conditions of toll stations, so as to forecast the traffic flow of expressway better.

References 1. Macheret, P.D., Savchuk, R.R., Shkuratov, G.I.: Intelligent transport systems: analysis of the current state and prospects of development. In: 2021 international conference on quality management, transport and information security, information technologies (IT&QM&IS). IEEE, pp. 234–237 (2021) 2. Wu., Tsu-Yang, Lee, Z., Yang, L., Chen, C.-M.: A provably secure authentication and key exchange protocol in vehicular ad hoc networks. Secur. Commun. Netw. 2021, 9944460 (2021) 3. Wu., Tsu-Yang, Lee, Z., Yang, L., Luo, J.-N., Tso, R.: Provably secure authentication key exchange scheme using fog nodes in vehicular ad-hoc networks. J. Supercomput. 77, 6992–7020 (2021) 4. Kumar, V., Kumar, R., Kumar, V., Kumari, A., Kumari, S.: RAVCC: robust authentication protocol for RFID based vehicular cloud computing. J. Netw. Intell. 7(3), 526–543 (2022) 5. Chen, C.-M., Chen, L., Gan, W., Qiu, L., Ding, W.: Discovering high utility-occupancy patterns from uncertain data. Inf. Sci. 546, 1208–1229 (2021) 6. Liu, L.: A short-term traffic flow prediction method based on svr. In: 2021 2nd international conference on urban engineering and management science (ICUEMS), IEEE, pp. 1–4 (2021) 7. Chen, J.-N., Huang, Z.-J., Zhou, Y.-P., Zou, F.-M., Chen, C.-M., Wu, J.M.-T., Wu, T.-Y.: Efficient certificate-based aggregate signature scheme for vehicular ad hoc networks. IET Netw. 9(6), 290–297 (2020) 8. Chen, R.-F., Luo, H., Huang, K.-C., Nguyen, T.-T., Pan, J.-S.: An improved honey badger algorithm for electric vehicle charge orderly planning. J. Netw. Intell. 7(2), 332–346 (2022) 9. Chen, X., Chen, H., Yang, Y., et al.: Traffic flow prediction by an ensemble framework with data denoising and deep learning model. Phys. A 565, 125574 (2021) 10. Liu, J., Guan, W.: A summary of traffic flow forecasting methods. J. Highw. Transp. Res. Dev. 21(3), 82–85 (2004) 11. Cai, L., Yu, Y., Zhang, S., Song, Y., Xiong, Z., Zhou, T.: A sample-rebalanced outlier-rejected k-nearest neighbor regression model for short-term traffic flow forecasting. IEEE Access 8, 22686–22696 (2020) 12. Zhang, F., Wu., Tsu-Yang, Pan, J.-S., Ding, G., Li, Z.: Human motion recognition based on SVM in VR art media interaction environment. HCIS 9, 40 (2019) 13. Wu, Q., Zang, B.-y., Zhang, Y., Qi, Z.-x.: Wavelet Kernel twin support vector machine. J. Inf. Hiding Multimed. Signal Process. 12(2), 93–101 (2021) 14. Meena, G., Sharma, D., Mahrishi, M.: Traffic prediction for intelligent transportation system using machine learning. In: 2020 3rd international conference on emerging technologies in computer engineering: machine learning and internet of things (ICETCE), 145–148, 2020 15. Zhang, L., Alharbe, N.R., Luo, G., Yao, Z., Li, Y.: A hybrid forecasting framework based on support vector regression with a modified genetic algorithm and a random forest for traffic flow prediction. Tsinghua Sci. Technol. 23(4), 479–492 (2018) 16. Liu, Y., Zhang, N., Luo, X., Yang, M.: Traffic flow forecasting analysis based on two methods. J. Phys: Conf. Ser. 1861(1), 012042 (2021) 17. Zhang, S.-M., Su., Xin, Jiang, X.-H., Chen, M.-L., Wu., Tsu-Yang: A traffic prediction method of bicycle-sharing based on long and short term memory network. J. Netw. Intell. 4(2), 17–29 (2019) 18. Liao, L., Lin, J., Zhu, Y., Bi, S., Lin, Y.: A bi-direction LSTM attention fusion model for the missing POI identification. J. Netw. Intell. 7(1), 161–174 (2022) 19. Wang, J.-N., Cui, J.-F., Chen, C.-L.: A prediction method of consumer buying behavior based on attention mechanism and CNN-BiLSTM. J. Netw. Intell. 7(2), 375–385 (2022)

308

G. Xu et al.

20. Kumar, S., Damaraju, A., Kumar, A., Kumari, S., Chen, C.-M.: LSTM network for transportation mode detection. J. Internet Technol. 22(4), 891–902 (2021) 21. Kalman, R.E.: A new approach to linear Filtering and prediction problems. ASME. J. Basic Eng. 82(1), 35–45 (1960) 22. Ho, T. K.: Random decision forests. In: Proceedings of 3rd international conference on document analysis and recognition, 1, pp. 278–282 (1995)

Part III

Artificial Intelligence—Innovation Technologies

Chapter 23

Objectionable Image Content Classification Using CNN-Based Semi-supervised Learning Shukla Mondal, Arup Kumar Pal, SK Hafizul Islam, and Debabrata Samanta

Abstract Due to the increased online activity, people of all ages, especially adolescents, may get exposed to objectionable image content such as internet pornography. These images are spread quickly and widely over the internet, which causes serious social problems. Many researchers have proposed objectionable image content classification models by utilizing deep neural networks to prevent such images from being retrieved while surfing the web. The performance of such models can be enhanced by the semi-supervised learning method by effectively utilizing the image data from an overwhelming number of unlabeled objectionable samples. For many such unlabeled objectionable images, this paper proposes a semi-supervised image content classification framework using a balanced sample inclusion mechanism based on a higher class probability outcome to include the pseudo labels effectively in the existing model. The proposed framework fully utilizes semi-supervised learning and gradually improves model classification accuracy and reliability.

S. Mondal · A. K. Pal Department of Computer Science and Engineering, Indian Institute of Technology (ISM) Dhanbad, Jharkhand 826004, India e-mail: [email protected] A. K. Pal e-mail: [email protected] S. Mondal · S. H. Islam (B) Department of Computer Science and Engineering, Indian Institute of Information Technology Kalyani, Kanchrapara, West Bengal 741235, India e-mail: [email protected] D. Samanta Department of Computing and Information Technologies, Rochester Institute of Technology, RIT Global Campus, Dr. Shpetim Robaj, n.n., Germia Campus, Prishtina 10000, Kosovo e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_23

311

312

S. Mondal et al.

23.1 Introduction People spend most of their time online due to the easy availability of tools and devices that have made data sharing more convenient. Color images that add meaningful information to express our daily lives are shared widely over the Internet. Also, the increased online activity makes us vulnerable to easy exposure to objectionable image content as it spreads quickly and widely over the Internet. Children are also vulnerable to such objectionable content as they get used to online activities more post-COVID19 circumstances. The objectionable image content, especially internet pornography, is exposed easily to people of all ages, which leads to dangerous interactions with potentially unidentifiable sources [1–5]. Moreover, many of these degrade mental and physical safety leading to various social problems, such as sexual abuse, child pornography, domestic violence, and other moral issues [6]. To prevent objectionable image content from being retrieved while browsing the Internet, recognizing and filtering out such images is vital for creating a healthy online platform for children and grown-ups, especially at the workplace. In the past decades, many approaches have been proposed to prevent objectionable and pornographic image retrieval on the Internet. Traditionally, many hand-crafted feature-based techniques, such as skin color-based feature descriptors, are used as solutions to filter out such images. As the large area of the most objectionable and pornographic images consists of human skin, it is evident that most researchers have used skin color and structural-based features [7–13]. Recently, deep learning architectures have developed rapidly for many different domains, especially in the areas of computer vision and image recognition task [14–19]. The deep convolutional features-based methods have also effectively solved various objectionable image filtering problems. Yan [20] introduced a demonize system to classify and filter out pornographic images as a post-classification problem using pre-trained CNN models based on transfer learning. Cheng et al. [21] proposed a deep feature-based approach to filter obscene images by taking the global context of an entire image and the meaningful local context of the input images. Their proposed network comprises three deep feature-based networks derived from low-level discriminative visual characteristics, a sensitive region in adult images, and fusion for high-level feature generation. Nian et al. [22] presented a deep convolutional neural network-based pornographic image filtering method using an improved sliding window method and some data augmentation methods. Shen et al. [23] developed an ensemble framework to recognize pornographic images using an uncertain inference engine based on a Bayesian network. The presence of visual objects in the images represented as local semantic features improved the identification of pornography in images based on the computed confidence. Lin et al. [24] proposed a method for pornographic image identification and classification problem by using extracted fuse features from various pre-trained models with similar network structures, which achieved higher accuracy compared to a single model approach. Mostly, previous researchers used labeled data to train their models for objectionable image recognition problems. With the huge growth of objectionable image

23 Objectionable Image Content Classification Using CNN-Based …

313

content and the lack of supervision of the image sources, a semi-supervised classification framework can make good use of many such unlabeled samples and save staffing. Semi-supervised learning includes unlabeled samples to the training set to enhance the predictive performance of learning algorithms [25–28]. Many semi-supervised methods such as teacher-student network [29] or GAN [30] generate pseudo labels from the unlabeled samples, which often require a complicated approach for training two models which fail to achieve re-usability characteristics of semi-supervised learning. This paper proposes a semi-supervised objectionable image content classifier using a balanced sample inclusion approach for each pseudo-class label based on their higher probabilistic outcome. Two significant advantages of our proposed framework for objectionable image content classification are as follows. (i) Our model trains unlabeled objectionable images from only one existing model and flexibly generates pseudo-labels to include in the existing model effectively. (ii) This framework gradually improves the classification accuracy and reliability of the proposed model by including unlabeled objectionable samples based on their high-class probabilities with a balanced inclusion mechanism. The rest of this paper is structured as follows. Section 23.2 describes the proposed framework for objectionable image content classification via semi-supervised learning. Section 23.3 presents the experimental results of the proposed framework based on various evaluation metrics. Finally, Sect. 23.4 provides some valuable conclusions, and future directives.

23.2 Proposed Methodology This section discusses the proposed semi-supervised objectionable image content classification framework as shown in Fig. 23.1. The objectionable images in an unlabeled image set are predicted with a pre-trained supervised classifier and obtain the confidence score based on probability outcomes for each class label. The predicted images are sorted in descending order, and a balanced sample inclusion mechanism is introduced for each class sample to be included in the semi-supervised training.

23.2.1 Framework for Semi-supervised Objectionable Image Content Classification The proposed semi-supervised classification uses DenseNet [31] architecture as the initial trained model which has shown significant performance in case of objectionNt are used to fine-tune the able image detection [24]. The labeled samples {xi , li }i=1   Nt(u) initial classifier to obtain O(θ ) model. The pseudo-labeled samples xi , li i=1 are

314

S. Mondal et al.

Fig. 23.1 Overview of the proposed framework N (u)

t generated using the model O(θ ) for the unlabeled samples {xi }i=1 . The proposed semi-supervised model includes the pseudo-samples by using a balanced sample inclusion mechanism in the training process and updates the model to reduce classification error with the increasing train iteration of the network.

23.2.2 Balanced Sample Inclusion Mechanism The confidence score for each of the class samples predicted by the model O(θ ) N (u)

t for the unlabeled samples {xi }i=1 are selected in descending order by setting a high confidence threshold value T . For each of the class label the selected pseudo-samples   Nt(u) xi , li i=1 are added in the training set while maintaining class balances as described in the Algorithm 1. The inclusion of each pseudo-sample is increased with λ while adding to the training set and updating the model O(θ ) by training iteration.

23.3 Experiments 23.3.1 Dataset and Implementation Details In this paper, the NSFW1 dataset is considered for evaluating the proposed framework. We have selected 5000 image samples with five classes labeled as “porn”, “sexy”, “hentai”, “neutral”, and “drawings” to train the primary supervised classifier based on densenet121 architecture. The system configuration for training and evalu1

https://github.com/alex000kim/nsfw_data_scraper.

23 Objectionable Image Content Classification Using CNN-Based …

315

Algorithm 1: Balanced sample inclusion N

(u)

Nt t Input: Labeled samples {xi , li }i=1 , unlabeled samples {xi }i=1 , and trained objectionable image detection model O(θ)   Nt(u) Output: Selected pseudo-labeled samples xi , li i=1 N

(u)

t Input the unlabeled samples {xi }i=1 into the pre-trained objectionable image detection model O(θ)   Nt(u)   Nt(u) 2 Get the predicted label li i=1 and their probability score pi , li i=1 3 Set the confidence threshold T  Nt(u)  4 Select the unlabeled samples where pi , li i=1 > T , and sort them in descending order to    , l , · · · l where N is the cases with class label Ci (i = 1, 2, · · · Q) get arrays li,1 C i i,2 i,NC

1

i

for i ← 1to Qdo    , l , · · · l 6 L i = li(0) , where li(0) = l0 , and Si = li,1 i,2 i,NC 5

7 8 9 10 11 12 13

for i ← 1 to NCi do  from S such that s = arg min {l (m) × λ} Select li,s i i i (m)

(m)

 and l Let li = li,s i  Remove li,s from Si end end   Nt(u) Return xi , li i=1

i

to the end of the list L i

ating the classification performance of the proposed semi-supervised objectionable image content classifier consists of a CPU of Core i9-10900K and dedicated graphics of NVIDIA Quadro P1000 with 64GB of RAM. We have also partially used the Colab [32] platform for executing the performance of the proposed semi-supervised model.

23.3.2 Results and Analysis We have analyzed our proposed semi-supervised image detection model in this segment and compared the performance with the supervised deep learning model. The evaluation metrics used in this paper are accuracy, precision, recall, and F1 score. Each of the metrics is evaluated for a test data of size 1000 containing five NSFW classes mentioned earlier. The confusion matrix for each class label is shown in Fig. 23.2, which analyzes the training evaluation where the proposed objectionable semi-supervised classifier gradually obtains higher efficiency with the balanced inclusion mechanism of unlabeled sample image data when T ≥ 0.8 compared to supervised deep learning. Table 23.1 describes the results for each class label for the proposed semisupervised objectionable image detection model compared with supervised deep

316

S. Mondal et al.

(a)

(c)

(b)

(d)

Fig. 23.2 Training evaluation for objectionable image content classifiers: confusion matrix for a supervised classification, b proposed semi-supervised classification with 10% unlabeled samples, c proposed semi-supervised classification with 20% unlabeled samples, and d proposed semisupervised classification with 30% unlabeled samples

learning architecture. The precision result for the proposed semi-supervised model is significantly higher where the class label is “porn”, “sexy”, and “drawings”, indicating higher actual correct predictions than the supervised deep learning model. The recall result is higher where the class label is “porn”, “sexy”, “hentai”, and “neutral” with the proposed semi-supervised model that indicates the most correct predictions out of these specific class labels compared to the supervised deep learning model. The f1-score for all the class labels for the proposed semi-supervised objectionable image

23 Objectionable Image Content Classification Using CNN-Based …

317

Table 23.1 Comparison results of supervised and semi-supervised objectionable image content classifiers on test data Methods Evaluation Drawings Hentai Neutral Porn Sexy Supervised

Proposed semisupervised

Precision Recall F1 score Precision

0.77 0.82 0.80 0.86

0.91 0.78 0.84 0.88

0.85 0.85 0.85 0.84

0.90 0.90 0.90 0.92

0.84 0.91 0.87 0.88

Recall F1 score

0.77 0.81

0.87 0.87

0.91 0.87

0.92 0.92

0.93 0.90

Fig. 23.3 Classification results on test data between supervised and proposed semi-supervised methods

88

88

88 87.6

87.4

(%)

87

86 85.4 85

85.2

85

Accuracy

Precision

Recall

Proposed semi-supervised

85.2

F1-score Supervised

detection is higher compared to the supervised model indicating better balances in each of the predicted class labels. The average classification accuracy, precision, recall, and F1 score on the test data for the proposed semi-supervised objectionable image detection model is compared with the supervised deep learning model in Fig. 23.3. The average performance has shown significantly better for all the evaluation metrics for our proposed semisupervised objectionable image detection model.

318

S. Mondal et al.

23.4 Conclusion The paper proposed a semi-supervised objectionable image content classification framework using CNN-based architecture that obtains better classification results than the existing supervised methods. The proposed approach utilizes the characteristic of semi-supervised learning by including a balanced sample inclusion mechanism based on higher class probability outcomes to provide flexibility over class distribution mismatch, effectively improving objectionable image classification accuracy. In the future, local objects segmented from such images may be utilized to improve the model performance results further to filter out objectionable images effectively. Acknowledgements The work was supported by the Council of Scientific and Industrial Research (CSIR), Ministry of Science and Technology, Govt. of India (Sanction No.: 22(0836)/20/EMR-II).

References 1. Jevremovic, A., Veinovic, M., Cabarkapa, M., Krstic, M., Chorbev, I., Dimitrovski, I., Garcia, N., Pombo, N., Stojmenovic, M.: Keeping children safe online with limited resources: analyzing what is seen and heard. IEEE Access 9, 132, 723–132, 732 (2021). https://doi.org/10.1109/ ACCESS.2021.3114389 2. Chen, J., Liang, G., He, W., Xu, C., Yang, J., Liu, R.: A pornographic images recognition model based on deep one-class classification with visual attention mechanism. IEEE Access 8, 122, 709–122, 721 (2020). https://doi.org/10.1109/ACCESS.2020.2988736 3. Brown, J.D., L’Engle, K.L.: X-Rated: Sexual attitudes and behaviors associated with U.S. early adolescents’ exposure to sexually explicit media. Commun. Res. 36(1), 129–151 (2009). https://doi.org/10.1177/0093650208326465 4. Ybarra, M.L., Strasburger, V.C., Mitchell, K.J.: Sexual media exposure, sexual behavior, and sexual violence victimization in adolescence. Clin. Pediatr. 53(13), 1239–1247 (2014). https:// doi.org/10.1177/0009922814538700 5. Cohen-Almagor, R.: Online child sex offenders: challenges and counter-measures. Howard J. Crim. Justice 52(2), 190–215 (2013). https://doi.org/10.1111/hojo.12006 6. Ybarra, M.L., Mitchell, K.J., Hamburger, M., Diener-West, M., Leaf, P.J.: X-rated material and perpetration of sexually aggressive behavior among children and adolescents: is there a link? Aggressive Behavior 37(1) (2011). 10.1002/ab.20367 7. Lee, J.S., Kuo, Y.M., Chung, P.C., Chen, E.L.: Naked image detection based on adaptive and extensible skin color model. Pattern Recognit. 40(8), 2261–2270 (2007). https://doi.org/10. 1016/j.patcog.2006.11.016 8. Zhu, H., Zhou, S., Wang, J., Yin, Z.: An algorithm of pornographic image detection. In: Fourth International Conference on Image and Graphics (ICIG 2007), pp. 801–804 (2007). https:// doi.org/10.1109/ICIG.2007.29 9. Yan, C.C., Liu, Y., Xie, H., Liao, Z., Yin, J.: Extracting salient region for pornographic image detection. J. Vis. Commun. Image Represent. 25(5), 1130–1135 (2014). https://doi.org/10. 1016/j.jvcir.2014.03.005 10. Srisaan, C.: A classification of internet pornographic images. Int. J. Electron. Commer. Stud. 7(1), 95–104 (2016). https://doi.org/10.7903/ijecs.1408 11. Zhuo, L., Geng, Z., Zhang, J., Li, X.G.: ORB feature based web pornographic image recognition. Neurocomputing 173, 511–517 (2016). https://doi.org/10.1016/j.neucom.2015.06.055

23 Objectionable Image Content Classification Using CNN-Based …

319

12. Zhang, F., Wu, T.Y., Zheng, G.: Video salient region detection model based on wavelet transform and feature comparison. EURASIP J. Image Vid. Process. 2019(1), 58 (2019). https://doi.org/ 10.1186/s13640-019-0455-2 13. Zhang, Y.J., Chen, J.Y., Lu, Z.M.: Face anti-spoofing detection based on color texture structure analysis. J. Netw. Intell. 7(2), 319–331 (2022) 14. Huu, P.N., Tien, D.N., Manh, K.N.: Action recognition application using artificial intelligence for smart social surveillance system. J. Inf. Hiding Multimed. Signal Process. 13(1), 1–11 (2022) 15. Kumar, S., Damaraju, A., Kumar, A., Kumari, S., Chen, C.M.: LSTM network for transportation mode detection. J. Internet Technol. 22(4), 891–902 (2021) 16. Gao, J., Zou, H., Zhang, F., Wu, T.Y.: An intelligent stage light-based actor identification and positioning system. Int. J. Inf. Comput. Secur. 18(1–2), 204–218 (2022). https://doi.org/10. 1504/IJICS.2022.122920 17. Zhang, F., Wu, T.Y., Pan, J.S., Ding, G., Li, Z.: Human motion recognition based on SVM in VR art media interaction environment. Hum.-Centric Comput. Inf. Sci. 9(1), 40 (2019). https:// doi.org/10.1186/s13673-019-0203-8 18. Wang, R.B., An, Z.W., Wang, W.F., Yin, S., Xu, L.: A multi-stage data augmentation approach for imbalanced samples in image recognition. J. Netw. Intell. 6(1), 94–106 (2021) 19. Tawfeeq, L.A., Hussein, S.S.: Predication of most significant features in medical image by utilized CNN and heatmap. J. Inf. Hiding Multimed. Signal Process. 12(4), 217–225 (2021) 20. Yan, H.: Detect and depornize pornographic images using pre-trained CNN models. In: 2020 International Conference on Computing and Data Science (CDS), pp. 48–51 (2020). https:// doi.org/10.1109/CDS49703.2020.00017 21. Cheng, F., Wang, S.L., Wang, X.Z., Liew, A.W.C., Liu, G.S.: A global and local context integration DCNN for adult image classification. Pattern Recognit. 96, 106, 983 (2019). https:// doi.org/10.1016/j.patcog.2019.106983 22. Nian, F., Li, T., Wang, Y., Xu, M., Wu, J.: Pornographic image detection utilizing deep convolutional neural networks. Neurocomputing 210, 283–293 (2016). https://doi.org/10.1016/j. neucom.2015.09.135 23. Shen, R., Zou, F., Song, J., Yan, K., Zhou, K.: EFUI: an ensemble framework using uncertain inference for pornographic image recognition. Neurocomputing 322, 166–176 (2018). https:// doi.org/10.1016/j.neucom.2018.08.080 24. Lin, X., Qin, F., Peng, Y., Shao, Y.: Fine-grained pornographic image recognition with multiple feature fusion transfer learning. Int. J. Mach. Learn. Cybern. 12(1), 73–86 (2021). https://doi. org/10.1007/s13042-020-01157-9 25. Dong, Z., Chen, Z.: Semi-supervised cell classification based on deep learning. In: The 6th International Conference on Control Engineering and Artificial Intelligence, CCEAI 2022, pp. 49–52. Association for Computing Machinery, New York, NY, USA (2022). https://doi.org/ 10.1145/3522749.3523086 26. Wu, M.E., Syu, J.H., Chen, C.M.: Kelly-based options trading strategies on settlement date via supervised learning algorithms. Comput. Econ. 59(4), 1627–1644 (2022). https://doi.org/10. 1007/s10614-021-10226-2 27. Zeng, X., Martinez, T.R.: Distribution-balanced stratified cross-validation for accuracy estimation. J. Exp. Theor. Artif. Intell. 12(1), 1–12 (2000). https://doi.org/10.1080/095281300146272 28. Tseng, K.K., Zhang, R., Chen, C.M., Hassan, M.M.: DNetUnet: a semi-supervised CNN of medical image segmentation for super-computing AI service. J. Supercomput. 77(4), 3594– 3615 (2021). https://doi.org/10.1007/s11227-020-03407-7 29. Yu, G., Sun, K., Xu, C., Shi, X.H., Wu, C., Xie, T., Meng, R.Q., Meng, X.H., Wang, K.S., Xiao, H.M., Deng, H.W.: Accurate recognition of colorectal cancer with semi-supervised deep learning on pathological images. Nat. Commun. 12(1), 6311 (2021). https://doi.org/10.1038/ s41467-021-26643-8 30. Rubin, M., Stein, O., Turko, N.A., Nygate, Y., Roitshtain, D., Karako, L., Barnea, I., Giryes, R., Shaked, N.T.: TOP-GAN: stain-free cancer cell classification using deep learning with a small training set. Med. Image Anal. 57, 176–185 (2019). https://doi.org/10.1016/j.media.2019.06. 014

320

S. Mondal et al.

31. Huang, G., Liu, Z., Van Der Maaten, L., Weinberger, K.Q.: Densely connected convolutional networks. In: 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 2261–2269 (2017). https://doi.org/10.1109/CVPR.2017.243 32. Bisong, E.: Google Colaboratory. Building Machine Learning and Deep Learning Models on Google Cloud Platform, pp. 59–64 (2019). https://doi.org/10.1007/978-1-4842-4470-8_7

Chapter 24

Software and Hardware Cooperative Implementation of the Rafflesia Optimization Algorithm Zonglin Fu, Jeng-Shyang Pan, Yundong Guo, and Václav Snášel

Abstract With the development of edge technology in the fields of transportation, wireless sensor networks, and the internet of things, more and more intelligent optimization algorithms are implemented in hardware structures. The Rafflesia optimization algorithm (ROA) is a novel intelligent optimization algorithm proposed recently. To observe its performance on hardware, this paper implements the ROA algorithm in a cooperative way of software and hardware, referred to as the FROA algorithm. To convenient testing and accelerated computing, the initialization module, fitness module, and update module of the FROA algorithm are deployed on the advanced RISC machine (ARM) platform and the field programmable gate array (FPGA) platform, respectively. In the experimental part, we test the FROA algorithm on nine different benchmark functions. The results are compared with those implemented in software. The experimental results demonstrate the effectiveness and superiority of the FROA algorithm.

24.1 Introduction Intelligent optimization algorithms provide good solutions for optimization problems in many fields such as wireless sensor networks [1–3], information hiding [4], transportation [5–7], and so on. Typically, intelligent optimization algorithms are Z. Fu · J.-S. Pan (B) · Y. Guo College of Computer Science and Engineering, Shandong University of Science and Technology, Qingdao 266590, China e-mail: [email protected] Z. Fu e-mail: [email protected] Y. Guo e-mail: [email protected] V. Snášel Faculty of Electrical Engineering and Computer Science, VŠB-Technical University of Ostrava, 70032 Ostrava, Czech Republic e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_24

321

322

Z. Fu et al.

deployed and executed in software [8–11]. However, with the development of science and technology such as the internet of things and automation, more and more algorithms are deployed on edge devices and hardware systems [12–14]. Hardware platforms generally have faster processing speeds. It can speed up the operation of the algorithm. At present, some intelligent optimization algorithms have been implemented on the hardware platform, such as the particle swarm optimization (PSO) algorithm [15–17], the bat algorithm (BA) [18, 19], the genetic algorithm (GA) [20, 21], and so on. In the hardware implementation of these algorithms, most of them are implemented in pure hardware structure [22, 23]. Although pure hardware design can maximize the execution speed of the algorithm, it has a serious drawback. Once the pure hardware design of the algorithm is complete, it cannot be modified. Therefore, it can only deal with one optimization problem, missing the generality of the algorithm [24]. Another hardware implementation of intelligent optimization algorithms is software–hardware co-design. Although it is less parallel than a pure hardware design, it can increase the speed while maintaining the flexibility of the program [25, 26]. The ROA algorithm is a novel intelligent optimization algorithm that we propose in another paper [27]. Its software results on the benchmark test set demonstrate the optimized performance of the algorithm. In order to further understand its effect on hardware, we implemented the ROA algorithm in the way of hardware and software co-design in this paper, named the FROA algorithm. According to the characteristics of the ROA algorithm, the update module with more computation is deployed on the field programmable gate array (FPGA). Some performances of the FPGA platform can improve the execution efficiency of the algorithm, such as the parallelism of the platform itself and strong computing power. The initialization module and fitness module are deployed to the advanced RISC machine (ARM). This facilitates the replacement of test functions. The development kit used in this paper is the Vivado design suite [28]. It contains three development tools: Vivado high-level synthesis (HLS) [29], Vivado, and Xilinx software development kit (SDK). In the experimental part, the FROA algorithm is tested on nine benchmark functions [30, 31] and compared with the software results. The comparison results demonstrate the superiority of the FROA algorithm in terms of convergence value and execution time. The rest of the paper is structured as follows: Sect. 24.2 introduces the ROA algorithm in detail. Section 24.3 expounds the details of the software–hardware coimplementation of the FROA algorithm. Section 24.4 verifies the effectiveness of the FROA algorithm through experiments. Section 24.5 summarizes the paper.

24.2 Rafflesia Optimization Algorithm The ROA algorithm is a recently proposed intelligent optimization algorithm. It is inspired by the growth properties of the Rafflesia plant. The ROA algorithm consists of three phases: pollination (attracting insects), fruiting (devouring insects), and sowing.

24 Software and Hardware Cooperative Implementation of the Rafflesia …

323

24.2.1 Pollination Phase This phase corresponds to the exploitation stage of the algorithm. Rafflesia plants emit a scent as they grow to attract some scent-chasing insects. These insects can pollinate Rafflesia. The population of insects flying toward the target Rafflesia is not fixed. Individuals with poor fitness will be replaced by new individuals. Their numbers are one-third the size of the population. Their positions are updated using strategy 1. The positions of the remaining two-thirds of individuals are updated using strategy 2.

24.2.1.1

Strategy 1

We map the individual dimensions k (k = 1, 2, . . . , D) into three-dimensional space to solve. This three-dimensional space consists of k − 1, k, and k + 1, as shown in ) represents the Fig. 24.1. X best represents the best individual. X i (i = 1, 2, . . . , NP 3 newly added individual. NP is the population size. We assume that the distance from X i to X best is equal to the distance from random individual X R to X best . Based on the above conditions, the position update equation of the newly added individual is X ik = X bestk + d × sin βk cos γk

(24.1)

−−−−→ Among them, βk is the angle between X best X i and k + 1 dimension. Its value range is − − − − → (0, π2 ). X best X i is a vector composed of X i and X best . γk represents the angle between −−−−→ the k dimension and the projection of X best X i on the plane formed by k dimension and k + 1 dimension. Its value range is (0, π ). d is the distance between X i and X best .

Fig. 24.1 Example diagram in strategy 1

k+1

Xi XR d d Xbest

k

k-1

324

Z. Fu et al.

  D  d =  (X Rk − X bestk )2

(24.2)

k=1

After that, individuals with poor fitness are replaced by new individuals. X worsti = X i

24.2.1.2

(24.3)

Strategy 2

In the process of insect flapping flight, the flight speed is usually represented by translational speed and rotational speed. Among them, the calculation equation of translation velocity is  ω0 v1 = A2 sin2 (ω0 t + θ ) + B 2 cos2 (ω1 t + θ ) (24.4) 2 The update equation for the rotational speed of the insect is v2 = v2 ω0 cos(ω0 t + θ + ϕ)

(24.5)

where A is the amplitude of the insect wings, with a value of 2.5. B is the lateral offset with a value of 0.1. ω0 is the period of the flapping frequency. ω1 is the frequency period of the lateral flapping frequency. The values of ω0 and ω1 are both 0.025. θ is the phase, the value is (2, 2π ). ϕ is the phase difference between translation and rotation, and the value is –0.78545. t is time and takes the value 1. The speed of insect flapping flight is the superposition of translational speed and rotational speed, namely: v = v1 + v2

(24.6)

The position update of the insect individual is influenced by the optimal individual and the previous state. Its update equation is X i = X i + C × v × t + (X best − X i ) × (1 − C) × rand

(24.7)

) represents the current update individual. C is the impact where X i (i = 1, 2, ..., 2NP 3 factor, and its value interval is [–1,1]. The (1 − C) × rand in the third term reduces the influence of the optimal individual on the current individual and prevents the premature convergence of the algorithm.

24 Software and Hardware Cooperative Implementation of the Rafflesia …

325

24.2.2 Fruiting Phase Rafflesia flowers that have been pollinated will bear fruit. At the same time, Rafflesia’s unique flower chamber structure traps some pollinating insects. According to this characteristic, the ROA algorithm reduces an individual every certain number of iterations. The total number of reductions is about one-third of the initial population size.

24.2.3 Sowing Phase This phase corresponds to the exploration stage of the algorithm. Ripe fruits contain many seeds. They are scattered randomly all over the place in different ways. Their position update equation is   iter − 1 × sign(rand − 0.5) (24.8) X ik = X bestk + r d × exp Max_iter where X i (i = 1, 2, . . . , NP) is the current update individual. iter and Max_iter repre

iter sent the current and the maximum iteration number, respectively. exp Max_iter − 1 is the influence factor that varies with the number of iterations. r d = rand × (ub − lb) + lb

(24.9)

where rand is a random number between (0, 1). ub is the maximum boundary. lb is the minimum boundary.

24.3 Implementation of FROA This section describes the detailed design and implementation of the FROA algorithm. The implementation of the FROA algorithm consists of two parts: FPGA part and ARM part. The FPGA is responsible for the update module of the algorithm to achieve its hardware acceleration. ARM is responsible for the initialization module and fitness module to increase the flexibility of the overall structure. The programming language in both FPGA and ARM is C++ language. The random number of the FPGA part is generated using a 32-bit linear feedback shift register (LFSR) [32, 33]. The feedback function is f (x) = x 32 + x 22 + x 2 + x + 1. The random numbers on the ARM side are generated using the rand() function in the C++ library. First, the program of the FROA algorithm is processed in Vivado HLS. The Vivado HLS can convert C/C++ code to VHDL/Verilog code. Its use shortens the FPGA development cycle. In the Vavido HLS, the data types we use are mainly floating-

326

Z. Fu et al.

Fig. 24.2 The layout of the FROA algorithm in Block Design

point data. We use the advanced eXtensible interface (AXI) bus to transfer data between the FPGA part and the ARM part. The Vivado HLS allows users to add optimization directives to functions and loops. Therefore, we add the “pipeline” instruction to the loop statement in the FROA algorithm. It can make the code in the loop body execute in parallel to reduce the delay of the program. The rules for adding “pipeline” directives in loops are as follows: (1) The “pipeline” instruction is directly added to the single-layer loop. (2) If the current loop is a double-layer loop and there are no statements between the inner and outer layers, the “pipeline” instruction is added to the inner loop. Otherwise, the “pipeline” instruction is added to the outer loop. After adding optimization instructions, the Vivado HLS synthesizes and verifies the program. The verified program is packaged in the IP core. The IP core is exported from the Vivado HLS. After that, it is loaded into Vivado’s Block Design. Figure 24.2 shows the layout of the IP core in Block Design. The core modules of the layout diagram are the ZYNQ core and the IP core. Among them, the IP core encapsulates the main structure of the algorithm. It corresponds to the program on the FPGA side. The ZYNQ core executes programs on the ARM side and controls the overall scheduling of the algorithm. The AXI bus is used to transfer data between the ZYNQ core and the IP core. First, the data in the ZYNQ core is outgoing from the M_AXI_HPM0_LPD interface. After the lightweight AXI_Lite bus transmission, the data arrives at the IP core. At the same time, the IP core receives the command from the ZYNQ to start working. The computing task is completed when the state of the IP core becomes done. After that, the updated data is transmitted from the m_axi_gmem interface of the IP core. After the AXI_master bus transmission, the data arrives at the ZYNQ core. The ZYNQ re-evaluates the updated data before passing it to the IP core. This is repeated until the stopping condition of the algorithm is reached. After that, the correctness of the layout and routing diagram of the algorithm is verified. After successful verification, the design source files are generated as external product and bitstream files. They are loaded into the Xilinx SDK along with driver files, etc. Finally, the platform is deployed on the development board to complete the overall operation of the program.

24 Software and Hardware Cooperative Implementation of the Rafflesia …

327

24.4 Experimental Results In this section, we prove the performance of the FROA algorithm through experiments. The model of the development board we are using is the Xilinx UltraScale MPsoc AXU3EG. We use nine different types of benchmark functions to evaluate the fitness of the population. The two-dimensional (2D) benchmark functions are dropwave function, Shubert function, three-hump camel function, levy n.13 function, and Goldstein-price function. The ten-dimensional (10D) benchmark functions are sphere function, Rastrigin function, sum squares function, and Styblinski-Tang function. The number of evaluations of the algorithm is set to 200*NP. NP is the population size, and the initial value is 30.

24.4.1 Comparison of Results with and Without Instruction Optimization To understand the effect of the algorithm after adding the “pipeline” instruction, we conduct experiments in terms of convergence value, latency, execution time, and resource occupancy. Table 24.1 records the resource occupancy and latency in 2D and 10D. Table 24.2 shows the convergence values and execution times of the FROA algorithm on nine functions. In Table 24.1, “Latency” represents the delay from input to output of the algorithm. “Total” represents the total number of resources on the board. “Y” and “N”, respectively, indicate whether a pipeline instruction is added to the algorithm. As can be seen from the table, the optimized algorithm has a significant improvement in latency. In particular, the “latency” of the program optimized on 10D is 2.13 times lower than the original. In addition, the optimized algorithm occupies less DSP48E and LUT resources. This is because the parallel execution of the program makes it unnecessary to repeat some multiplication and logical judgments in the FROA algorithm. But the resource occupancy of the FF has increased. It should be noted that the increase or decrease of a certain resource is not absolute, it is determined by the features of the algorithm and the HLS instructions added by the user.

Table 24.1 The resource occupancy and latency of the FROA algorithm Total 2D 10D N Y N Latency BRAM DSP48E FF LUT

– 432 360 141120 70560

11025 8 76 9867 20037

6762 9 57 10324 18192

39413 9 76 9149 19445

Y 12586 9 57 9581 17526

328

Z. Fu et al.

Table 24.2 The convergence value and running time of the FROA algorithm Target value Y N Value Time Value Drop-wave Shubert Three-hump camel Levy n.13 Goldsteinprice Sphere Rastrigin Sum squares Styblinskitang

Time

–1 –186.7309 0

–0.97449 –186.605 2.99E–02

8 15 11

–0.97449 –186.605 2.99E–02

11 18 14.1

0 0

1.72E–06 19.19991

9 7

1.72E–06 19.19991

12 10

0 0 0 –391.6599

6.26E–04 24.27915 4.00E–02 –344.98

16 24.7 15 40

6.26E–04 24.27915 4.00E–02 –344.98

36 44.8 35 60

In Table 24.2, “value” represents the convergence value of the algorithm. “time” represents the execution time of the algorithm in milliseconds. “Target value” is the target optimal value of the benchmark function. The optimization instruction does not change the code and data of the algorithm. Therefore, no matter whether the algorithm adds the “pipeline” instruction or not, its convergence value will not change. In addition, we can know from the table that the execution time of the FROA algorithm with the “pipeline” instruction is reduced. In particular, the optimized algorithm has a more obvious acceleration effect on high-dimensional functions.

24.4.2 Comparison with Software Results The software platform is ARM with 666MHz running memory. Table 24.3 records the comparison results of the FROA algorithm and the software-implemented ROA algorithm. The convergence value of the FROA algorithm is better than the software result of the ROA algorithm on the eight benchmark functions. Only on the Goldstein function, the convergence value of the FROA algorithm is worse than that of the software-implemented ROA algorithm. Additionally, the execution time of the FROA algorithm is faster than the software results in all test functions. Especially for the sphere and sum squares functions, the execution speed of the FROA algorithm is about three times that of the software-implemented ROA algorithm.

24 Software and Hardware Cooperative Implementation of the Rafflesia …

329

Table 24.3 Comparison with software results in convergence value and running time Target value FROA ROA Value Time Value Time Drop-wave Shubert Three-hump camel Levy n.13 Goldsteinprice Sphere Rastrigin Sum squares Styblinskitang

–1 –186.7309 0

–0.97449 –186.605 2.99E–02

8 15 11

–0.9212 –147.597 1.20E–01

19 27.4 13.3

0 0

1.72E–06 19.19991

9 7

1.22E–02 3.031202

17 14

0 0 0 –391.6599

6.26E–04 24.27915 4.00E–02 –344.98

16 24.7 15 40

13.72556 68.76995 71.60998 –310.178

47.4 61.7 47.3 50

24.5 Conclusion In this paper, the ROA algorithm is implemented by the design scheme of software and hardware cooperation, named the FROA algorithm. The computationally intensive update module is executed on the FPGA to achieve hardware acceleration. The initialization module and fitness module are executed on the ARM to facilitate the replacement of test functions. The experimental results of the algorithm on nine benchmark functions demonstrate its effectiveness and superiority. To test the effect of the FROA algorithm in practical problems, we plan to apply it to problems such as path planning in the future.

References 1. Chai, Q.W., Chu, S.C., Pan, J.S., Zheng, W.M.: Applying adaptive and self assessment fish migration optimization on localization of wireless sensor network on 3-d Te rrain. J. Inf. Hiding Multim. Signal Process. 11(2), 90–102 (2020) 2. Chu, S.C., Du, Z.G., Pan, J.S.: Symbiotic organism search algorithm with multi-group quantumbehavior communication scheme applied in wireless sensor networks. Appl. Sci. 10(3), 930 (2020) 3. Pan, J.S., Dao, T.K., Pan, T.S., Nguyen, T.T., Chu, S.C., Roddick, J.F.: An improvement of flower pollination algorithm for node localization optimization in WSN. J. Inf. Hiding Multim. Signal Process. 8(2), 486–499 (2017) 4. Chu, S.C., Huang, H.C., Shi, Y., Wu, S.Y., Shieh, C.S.: Genetic watermarking for Zerotreebased applications. Circuits Syst. Signal Process. 27(2), 171–182 (2008) 5. Pan, J.S., Yang, Q., Shieh, C.S., Chu, S.C.: Tumbleweed optimization algorithm and its application in vehicle path planning in smart city. J. Internet Technol. 23(5), 927–945 (2022)

330

Z. Fu et al.

6. Zhang, F., Wu, T.Y., Wang, Y., Xiong, R., Ding, G., Mei, P., Liu, L.: Application of quantum genetic optimization of LVQ neural network in smart city traffic network prediction. IEEE Access 8, 104555–104564 (2020) 7. Cai, Z.M., Lu, J., Ling, Y.F., Li, T.J., Xu, L.: GSGC: An improved path planning optimization method using guided sampling and gradual cutting. J. Netw. Intell. 7, 84–100 (2022) 8. Chai, Q.W., Chu, S.C., Pan, J.S., Hu, P., Zheng, W.M.: A parallel WOA with two communication strategies applied in dv-hop localization method. EURASIP J. Wirel. Commun. Netw. 2020(1), 1–10 (2020) 9. Xue, X., Yang, H., Zhang, J.: Using population-based incremental learning algorithm for matching class diagrams. Data Sci. Pattern Recognit. 3(1), 1–8 (2019) 10. Chiang, S., Chu, S.C., Hsin, Y.C., Wang, M.H.: Genetic distance measure for k-modes algorithm. Int. J. Innov. Comput. Inf. Control 2(1), 33–40 (2006) 11. Xi, J., Chen, Y., Liu, X., Chen, X.: Whale Optimization Algorithm Based on Nonlinear Adjustment and Random Walk Strategy (2022) 12. Mustafa, E.M., Elshafey, M.A., Fouad, M.M.: Enhancing CNN-based image steganalysis on GPUs. J. Inf. Hiding Multim. Signal Process. 11(3), 138–150 (2020) 13. Kang, L., Chen, R.S., Chen, Y.C., Wang, C.C., Li, X., Wu, T.Y.: Using cache optimization method to reduce network traffic in communication systems based on cloud computing. IEEE Access 7, 124397–124409 (2019) 14. Kumari, A., Kumar, V., Abbasi, M.Y., Kumari, S., Chaudhary, P., Chen, C.M.: Csef: cloudbased secure and efficient framework for smart medical system using ECC. IEEE Access 8, 107838–107852 (2020) 15. Pan, T.S., Dao, T.K., Chu, S.C., et al.: Optimal base station locations in heterogeneous wireless sensor network based on hybrid particle swarm optimization with bat algorithm. J. Comput. 25(4), 14–25 (2015) 16. Cavuslu, M.A., Karakuzu, C., Karakaya, F.: Neural identification of dynamic systems on FPGA with improved PSO learning. Appl. Soft Comput. 12(9), 2707–2718 (2012) 17. Kang, L., Chen, R.S., Xiong, N., Chen, Y.C., Hu, Y.X., Chen, C.M.: Selecting hyper-parameters of gaussian process regression based on non-inertial particle swarm optimization in internet of things. IEEE Access 7, 59504–59513 (2019) 18. Dao, T.K., Pan, T.S., Chu, S.C., et al.: Evolved bat algorithm for solving the economic load dispatch problem. In: Genetic and Evolutionary Computing, pp. 109–119. Springer (2015) 19. Ameur, M.S.B., Sakly, A., Mtibaa, A.: A hardware optimization of bat algorithms implemented on fpga. In: 2015 16th International Conference on Sciences and Techniques of Automatic Control and Computer Engineering (STA), pp. 146–151. IEEE (2015) 20. Lei, T., Ming-Cheng, Z., Jing-Xia, W.: The hardware implementation of a genetic algorithm model with FPGA. In: 2002 IEEE International Conference on Field-Programmable Technology (FPT). Proceedings, pp. 374–377. IEEE (2002) 21. Torquato, M.F., Fernandes, M.A.: High-performance parallel implementation of genetic algorithm on FPGA. Circuits Syst. Signal Process. 38(9), 4014–4039 (2019) 22. Zou, X., Wang, L., Tang, Y., Liu, Y., Zhan, S., Tao, F.: Parallel design of intelligent optimization algorithm based on FPGA. Int. J. Adv. Manuf. Technol. 94(9), 3399–3412 (2018) 23. Juang, C.F., Lu, C.M., Lo, C., Wang, C.Y.: Ant colony optimization algorithm for fuzzy controller design and its FPGA implementation. IEEE Trans. Ind. Electron. 55(3), 1453–1462 (2008) 24. Becker, J., Hubner, M., Hettich, G., Constapel, R., Eisenmann, J., Luka, J.: Dynamic and partial FPGA exploitation. Proc. IEEE 95(2), 438–452 (2007) 25. Jiang, Q., Guo, Y., Yang, Z., Wang, Z., Yang, D., Zhou, X.: Improving the performance of whale optimization algorithm through OpenCL-based FPGA accelerator. Complexity 2020 (2020) 26. Wolf, W.H.: Hardware-software co-design of embedded systems. Proc. IEEE 82(7), 967–989 (1994) 27. Pan, J.S., Fu, Z., Hu, C.C., Tsai, P.W., Chu, S.C.: Rafflesia optimization algorithm applied in the logistics distribution centers location problem. J. Internet Technol., 1–17 (2022) 28. Feist, T.: Vivado design suite. White Paper 5, 30 (2012)

24 Software and Hardware Cooperative Implementation of the Rafflesia …

331

29. Nane, R., Sima, V.M., Pilato, C., Choi, J., Fort, B., Canis, A., Chen, Y.T., Hsiao, H., Brown, S., Ferrandi, F., et al.: A survey and evaluation of FPGA high-level synthesis tools. IEEE Trans. Comput.-Aided Des. Integr. Circuits Syst. 35(10), 1591–1604 (2015) 30. Molga, M., Smutnicki, C.: Test functions for optimization needs. Test Funct. Optim. Needs 101, 48 (2005) 31. Ali, M.M., Khompatraporn, C., Zabinsky, Z.B.: A numerical evaluation of several stochastic algorithms on selected continuous global optimization test problems. J. Glob. Optim. 31(4), 635–672 (2005) 32. Wang, L.T., McCluskey, E.J.: Linear feedback shift register design using cyclic codes. IEEE Trans. Comput. 37(10), 1302–1306 (1988) 33. Savir, J., McAnney, W.H.: A multiple seed linear feedback shift register. In: Proceedings. International Test Conference, pp. 657–659. IEEE (1990)

Chapter 25

A Hybrid Orthogonal Learning and QUATRE Algorithm Based on PPE Algorithm Lulu Liang, Shu-Chuan Chu, Tien-Szu Pan, and Tsu-Yang Wu

Abstract Combining the characteristics of PPE, QUATRE algorithm, and orthogonal learning, this paper proposes a hybrid Orthogonal Learning and QUATRE algorithm based on the PPE algorithm (OLQTPPE). This algorithm takes the PPE algorithm as the main body and uses the QUATRE algorithm to search deeper. The purpose of using the QUATRE algorithm is to prevent the algorithm from falling into local optimization. After this, orthogonal learning is used to optimize the whole, to find better particles in a small area. The algorithm is tested on CEC2014 and compared with PSO, PPSO, BA, and PPE. The results show that the algorithm is superior to the four algorithms. In particular, it is the proposed OLQTPPE algorithm that has high performance and effectiveness compared to PSO, PPSO, and BA algorithms.

L. Liang · S.-C. Chu (B) · T.-Y. Wu College of Computer Science and Engineering, Shandong University of Science and Technology, Qingdao 266590, China e-mail: [email protected] L. Liang e-mail: [email protected] T.-S. Pan Department of Electronic Engineering, National Kaohsiung University of Science and Technology, Kaohsiung 807, Taiwan e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_25

333

334

L. Liang et al.

25.1 Introduction In recent years, meta-heuristic algorithms have become popular, and more and more meta-heuristic algorithms have been created. A meta-heuristic algorithm is a search on a certain search space to find the optimal approximate solution. The core ideas are exploration and exploitation. Exploration is to explore the whole search space to find the possible locations of the optimal solution. Exploitation is to develop the global optimal solution as much as possible after exploration. From the birth of meta-heuristics to the present, in addition to classical algorithms such as Particle Swarm Optimization (PSO) [5, 20, 21, 28] and Genetic Algorithm (GA) [7, 8, 27], there are many more popular emerging algorithms, for example, Butterfly Optimization Algorithm (BOA) [2, 19], Artificial Bee Colony (ABC) [6], Cat Swarm Optimization [1, 9], Ant Colony Optimization (ACO) [16], and other algorithms [23]. The PSO algorithm was first proposed in 1995. The GA first dates back to the 1960 s and was inspired by natural selection and genetics in Darwin’s theory of biological evolution. Since the meta-heuristic algorithm combines the characteristics of stochastic algorithm and local search, it is widely used in different fields, such as wireless sensor networks [3, 12, 24, 26], remote sensing image comparison [18, 25], artificial neural network tuning [15], and so on. However, with the progress of technology, the single meta-heuristic algorithm can no longer solve practical problems well. Scholars have proposed various types of optimization algorithms containing optimization strategies: For example, the Parallel Particle Swarm Optimization [4]. The orthogonal learning method is used in [14, 22] to optimize the pigeon algorithm and is applied to the photovoltaic system. In [10], the orthogonal learning method is used to optimize the cuckoo search algorithm [13] and is effective. The above algorithms show that the orthogonal learning method works well for meta-heuristic algorithms. And from all kinds of literature, it can be concluded that the QUATRE algorithm has a remarkable effect. This paper tries to apply orthogonal learning and the QUATRE algorithm to the Phasmatodea Population Evolution (PPE) [11], expecting to get better results. The rest of this paper is organized as follows. Section 25.2 presents the details of the PPE algorithm and the QUATRE algorithm. Section 25.3 provides a detailed description of the algorithm. Section 25.4 shows the experimental results. Section 25.5 gives conclusions and an outlook for future work based on the results of this paper.

25.2 Related Work 25.2.1 PPE The algorithm mainly consists of initialization and update. The initialization is similar to other evolutionary algorithms, generating N p solutions randomly, and the formula is shown in Eq. 25.1. The k historical optimal solutions are used in the PPE algorithm,

25 A Hybrid Orthogonal Learning and QUATRE Algorithm Based on PPE Algorithm

335

and the formula is shown in Eq. 25.2. The population location is updated as shown in Eq. 25.3: 1 pi = . (25.1) Np k = log(N p) + 1 .

(25.2)

x t+1 = x t + ev .

(25.3)

There are three main types of updates in this section. The first two are determined by the autonomous decisions of the populations. The last one is determined by interpopulation effects. The formula for updating the population size for both the first and the second is Eq. 25.4. If the updated population is better than the one before the update, the first update is performed, i.e., the update formula for ev is Eq. 25.5. Conversely, if the updated population is not as good as the previous one, the second update is performed, i.e., the update formula for ev is Eq. 25.6: p t+1 = a t+1 p t (1 − p t ) .

(25.4)

ev t+1 = (1 − p t+1 )A + P t+1 (ev t + m) .

(25.5)

ev t+1 = rand · A + st B .

(25.6)

In the last type of update, G is to determine whether competition will occur between two populations, and its formula is shown in Eq. 25.9. If the update condition is met, the following update is performed. The population size was updated in Eq. 25.7, and the evolutionary trend ev was updated in the way of Eq. 25.8: pi = pi + ai pi (1 − pi − ev t+1 = ev t+1 +

f (x j ) pj) . f (xi )

(25.7)

f (x j ) − f (xi ) (x j − xi ) . f (x j )

(25.8)

Maxgen + 1 − t . Maxgen

(25.9)

G = 0.1 × (U B − L B)

The pseudo-code of the PPE algorithm is shown in Algorithm 1.

336

L. Liang et al.

Algorithm 1 PPE Algorithm Require : N p populations,Dim dimension; Ensure : gbest and f (gbest); 1: ev, p, k are obtained by Eqs. 25.1 and 25.2; 2: Calculate f (x), gbest; 3: for t = 2 to Maxgen do 4: Implementation Eq. 25.3; 5: Calculate f (newx); 6: Update gbest; 7: for i = 1 to N p do 8: if f (newx) ≤ f (x) then 9: x = newx and f (x) = f (newx); 10: pi is obtained by Eq. 25.4; 11: evi is obtained by Eq. 25.5; 12: else 13: if rand ≤ pi then 14: x = newx and f (x) = f (newx); 15: pi is obtained by Eq. 25.4; 16: end if 17: evi is obtained by Eq. 25.6; 18: end if 19: Choose a random population x j ,( j = i); 20: if dist (xi , x j ) < G then 21: pi is obtained by Eq. 25.7; 22: evi is obtained by Eq. 25.8 23: end if 24: end for 25: end for

25.2.2 QUATRE The QUATRE algorithm used in this paper is very similar to the DE algorithm but more superficial than the DE algorithm. There are three control parameters in the DE algorithm, while in the QUATRE algorithm, there is only one control parameter, i.e., the population size. This algorithm generates particle candidate solutions as shown in Eq. 25.10: (25.10) X → M ⊗ X + M ⊗ B , where ⊗ denotes multiplication, M denotes the selection matrix consisting of 0 and 1, and M denotes the binary inverse matrix of M. In this paper, we mainly use the initialization of M when P S = Dim. A random ordering of the row and column vectors of Mtmp gives M. An example of generating M is shown in Eq. 25.11: ⎡

Mtmp

1 ⎢1 =⎢ ⎣1 1

0 1 1 1

⎤ ⎡ 00 11 ⎢1 1 0 0⎥ ⎥M =⎢ ⎣1 1 1 0⎦ 11 10

1 1 0 0

⎤ 0 1⎥ ⎥ . 0⎦ 1

(25.11)

25 A Hybrid Orthogonal Learning and QUATRE Algorithm Based on PPE Algorithm

337

The pseudo-code of the QUATRE algorithm is shown in Algorithm 2. Algorithm 2 QUATRE Algorithm Require : Dim dimension, searching space V , population size ps, the coordinates matrix X and find the global best particle X gbest ; Ensure : The global optima X gbest , f (X gbest ); 1: for t = 2 to Max do 2: generate matrix M; 3: B = X gbest,G + F(X r 1,G − X r 2,G ) 4: X → M ⊗ X + M ⊗ B 5: for i = 1 to ps do 6: if f (X tmp (i)) > f (X i ) then 7: Update X i = X tmp (i) 8: Update X gbest = X i 9: end if 10: end for 11: end for

25.3 OLQTPPE Algorithm This paper proposes a hybrid Orthogonal Learning and QUATRE (QT) algorithm based on the PPE algorithm (OLQTPPE). The PPE algorithm is combined with the QUATRE algorithm and then optimized using orthogonal learning. First, the PPE algorithm is used to perform the optimization search. Second, each particle is numbered, and two particles are randomly selected. Third, determine whether the two particles chosen meet the requirements. Fourth, the QUATRE algorithm and orthogonal learning method are performed if the conditions are met. Only an orthogonal learning strategy is performed if the requirements are not met. The pseudo-code of the whole algorithm is shown below, and its pseudo-code is shown in Algorithm 3. The strategy used in this paper is orthogonal learning. In orthogonal learning, the most important is the orthogonal table. The following is an introduction to orthogonal tables. An orthogonal table is a specially designed table, generally denoted by L n (m k ). L denotes an orthogonal table. n denotes n rows, i.e., n experiments to be performed. k denotes k columns, i.e., there are at most k factors. m denotes that there are m types of numbers, i.e., factors with level 1, level 2, . . ., level m. In an orthogonal table, the numbers in each column appear an equal number of times. In any two columns whose horizontal pairs of numbers form, each number pair appears an equal number of times. The purpose of using orthogonal learning is to find the best combination while reducing the number of experiments. In this paper, two particles are randomly selected for orthogonal learning to get the best combination. In this paper, the QUATRE algorithm is performed when the serial number of the selected particles is a multiple of 4. The PPE algorithm is first run once, after which it determines whether the particle chosen is a multiple of 4. If it is a multiple of 4, the QUATRE algorithm is performed to find the optimal value. And the optimal

338

L. Liang et al.

Algorithm 3 OLQTPPE Algorithm Require : N p populations, Dim dimension; Ensure : gbest and f (gbest); 1: for gen = 1 to Maxgen do 2: Select two particles a and b; 3: if a or b then 4: Execute Algorithm 2 to obtain the gbestq and f (gbestq ); 5: if f (gbestq ) < (gbest) then 6: gbest = gbestq and f (gbest) = f (gbestq ); 7: end if 8: end if 9: Execute Orthogonal Learning to obtain the gbestol and f (gbestol ); 10: if f (gbestol ) < f (gbest) then 11: gbest = gbestol and f (gbest) = f (gbestol ); 12: end if 13: end for

value found is compared with the optimal value found by the PPE algorithm, and the better particle is taken as the final result. The selected particles are then optimized using orthogonal learning and compared with the optimal values obtained above to arrive at the final optimal values. In order to better use the QUATRE algorithm and orthogonal learning method, the number of particles and dimensionality are set to 30 in this paper. In this algorithm, the optimal value is adjusted using the QUATRE algorithm for each iteration of the PPE algorithm to prevent the overall from falling into a local optimum. Local adjustment is performed using an orthogonal learning strategy to find the optimal value for the current position.

25.4 Experiment To verify the algorithm, PSO, PPSO, BA, and PPE algorithms [11, 17] are selected for comparison in this paper. For fairness, the number of particles is set to 30 for all algorithms with a dimension of 30. In PSO and PPSO, the values of c1 and c2 are both 2, and the value of weight w is 0.9. In this paper, the CEC2014 benchmark test suite was selected for testing. The group suite function contains a variety of functions such as single-peak and multi-peak, which is very representative. It has been recognized as authoritative by many scholars in the international arena. In this paper, all algorithms have 2000 iterations. Each algorithm is run independently 10 times, and its average value is obtained. The average values obtained are shown in Table 25.1. The comparison results of the five algorithms are shown in Table 25.1, and bold in the table indicates good results. Table 25.1 demonstrates the effectiveness and efficiency of the proposed OLQTPPE algorithm. Only the five algorithms of F1, F4, F12, F18, and F24 cannot beat PPE. Moreover, over F4 and F12, the OLQTPPE algorithm is close to the optimal value of the PPE algorithm. In

25 A Hybrid Orthogonal Learning and QUATRE Algorithm Based on PPE Algorithm Table 25.1 Experimental results of algorithm comparison OLQTPPE PSO PPE F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15 F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30

9097425.967 194956.2816 275.9405733 111.5536972 20.03749325 12.53839867 0.141928236 33.47595498 69.24739426 656.1577919 3284.915368 0.399158958 0.276805143 0.20534889 12.22782482 11.0986148 143928.3321 2877.74345 10.35909562 332.9354006 42564.64168 46.07417963 315.4136602 229.6043347 209.4194185 100.263216 409.8505147 960.8009967 3249.690865 4130.100204

102250411.8 12846679098 147974.7758 731.0877568 20.97944259 29.39863386 133.1061458 185.7184453 206.9955103 4394.685026 5084.58707 2.701212286 3.393297521 50.90269967 11303.58224 12.69633219 3002300.237 375846480.1 43.63616806 42558.99688 3377846.055 920.619324 371.4754135 265.4769468 217.5708833 106.0055563 1183.748958 2198.803806 41122001.23 604113.4348

7921078.361 207912.7594 1931.721409 98.05181043 20.0504001 25.31329444 0.177619359 82.08499352 161.7932562 1482.579629 3426.77309 0.386181227 0.369188096 0.22507119 20.351034 12.52930205 393493.5408 1815.113464 26.18620202 4250.000221 213188.4368 538.4429875 315.4284736 228.4439858 214.6789295 200.1301416 678.1560652 3109.955055 4070.597984 6198.197652

339

PPSO

BA

39319115.82 842882005.5 77506.88472 291.2111233 20.73915044 28.83061898 31.81943065 139.2418597 181.5583981 4106.809671 5569.128867 1.537569744 0.837389262 11.41986448 87.27600268 12.75252924 2574666.232 828891.196 43.46535826 17530.83097 956719.8285 779.9652063 344.4951305 200.000000 214.0809185 103.2069702 864.4233242 2097.455356 34888296.17 238309.538

511371352.7 35110398476 246102.4786 3599.401966 21.14052608 41.22950019 270.9759582 245.530769 262.4981491 4659.20536 5417.257021 2.088291938 4.751289336 105.8816421 19427.0045 13.50192864 16152492.58 624086.7617 193.9944427 946064.5053 13062421.75 1377.12587 445.8238536 324.819873 251.95207 129.7111218 1788.599078 5119.771118 23306967.13 1336659.352

particular, the optimal value of PPE and the optimal value of the OLQTPPE algorithm are almost equal on F12. Compared with the PPSO algorithm, the OLQPPE algorithm only occupies a disadvantageous position over F24. Compared with PSO and BA algorithms, the OLQTPPE algorithm ultimately wins. Due to the limited space, this paper only shows the convergence curves on 8 test functions, as shown in Fig. 25.1. These eight test functions are F3, F6, F8, F10, F11, F17, F27, and F30. All kinds of test functions are included. Figure 25.1 shows that OLQTPPE excels other algorithms in terms of convergence speed and optimal

340

Fig. 25.1 Comparison of OLQTPPE, PPE, PPSO, PSO, and BA algorithms

L. Liang et al.

25 A Hybrid Orthogonal Learning and QUATRE Algorithm Based on PPE Algorithm

341

value. On the simple multimodal functions F8 and F10, when the iterations are close to 2000 times, the proposed algorithm once again finds a better value. We have reason to believe that when there are enough iterations, the OLQTPPE algorithm will converge to a better value. On the simple multimodal function F11, the convergence speed of the PPE algorithm and the OLQTPPE algorithm is very similar. However, in the end, the OLQTPPE algorithm still converges to a better value. On the synthesis functions F27 and F30, the five algorithms easily fall into local optimization.

25.5 Summary and Prospects After the experimental test, this work can draw the following conclusions. First, the proposed OLQTPPE algorithm has well-defined performance on the whole. Second, since the algorithm includes the QUATRE algorithm, it ensures that it will not fall into the local optimum to the greatest extent. However, it will still fall into the local optimum in the composite function. Therefore, the algorithm also has significant room for improvement. Third, the algorithm uses orthogonal learning to fine-tune the algorithm and develop better particles as much as possible. Finally, this paper does not apply the algorithm to practical problems, hoping that the algorithm can be used for practical engineering problems.

References 1. Ahmed, A.M., Rashid, T.A., Saeed, S.A.M.: Cat swarm optimization algorithm: a survey and performance evaluation. Comput. Intell. Neurosci. (2020) 2. Arora, S., Singh, S.: Butterfly optimization algorithm: a novel approach for global optimization. Soft Comput. 23(3), 715–734 (2019) 3. Kandris, D., Nakas, C., Vomvas, D., Koulouras, G.: Applications of wireless sensor networks: an up-to-date survey. Appl. Syst. Innov. 3(1), 14 (2020) 4. Kang, L., Chen, R.S., Chen, Y.C., Wang, C.C., Li, X., Wu, T.Y.: Using cache optimization method to reduce network traffic in communication systems based on cloud computing. IEEE Access 7, 124397–124409 (2019) 5. Kang, L., Chen, R.S., Xiong, N., Chen, Y.C., Hu, Y.X., Chen, C.M.: Selecting hyper-parameters of gaussian process regression based on non-inertial particle swarm optimization in internet of things. IEEE Access 7, 59504–59513 (2019) 6. Karaboga, D., Basturk, B.: A powerful and efficient algorithm for numerical function optimization: artificial bee colony (ABC) algorithm. J. Glob. Optim. 39(3), 459–471 (2007) 7. Katoch, S., Chauhan, S.S., Kumar, V.: A review on genetic algorithm: past, present, and future. Multimed. Tools Appl. 80(5), 8091–8126 (2021) 8. Kong, L., Pan, J.S., Snášel, V., Tsai, P.W., Sung, T.W.: An energy-aware routing protocol for wireless sensor network based on genetic algorithm. Telecommun. Syst. 67(3), 451–463 (2018) 9. Kong, L., Pan, J.S., Tsai, P.W., Vaclav, S., Ho, J.H.: A balanced power consumption algorithm based on enhanced parallel cat swarm optimization for wireless sensor network. Int. J. Distrib. Sens. Netw. 11(3), 729680 (2015) 10. Li, J., Li, Y.x., Tian, S.S., Zou, J.: Dynamic cuckoo search algorithm based on Taguchi opposition-based search. Int. J. Bio-Inspired Comput. 13(1), 59–69 (2019)

342

L. Liang et al.

11. Liang, L.L., Du, Z.G., Shieh, C.S., Hu, C.C., Chu, S.C., Feng, Q.: A new PPE algorithm based on parallel communication strategy. In: Advances in Intelligent Information Hiding and Multimedia Signal Processing, pp. 289–298. Springer (2022) 12. Pan, J.S., Dao, T.K., Pan, T.S., Nguyen, T.T., Chu, S.C., Roddick, J.F.: An improvement of flower pollination algorithm for node localization optimization in WSN. J. Inf. Hiding Multim. Signal Process. 8(2), 486–499 (2017) 13. Pan, J.S., Song, P.C., Chu, S.C., Peng, Y.J.: Improved compact cuckoo search algorithm applied to location of drone logistics hub. Mathematics 8(3), 333 (2020) 14. Pan, J.S., Tian, A.Q., Snášel, V., Kong, L., Chu, S.C.: Maximum power point tracking and parameter estimation for multiple-photovoltaic arrays based on enhanced pigeon-inspired optimization with Taguchi method. Energy 251, 123863 (2022) 15. Qi, C., Fourie, A., Chen, Q.: Neural network and particle swarm optimization for predicting the unconfined compressive strength of cemented paste backfill. Constr. Build. Mater. 159, 473–478 (2018) 16. Tang, L., Ma, K., Li, Z.: A new scheduling algorithm based on ant colony algorithm and cloud load balancing. J. Inf. Hiding Multim. Signal Process. 8(1), 191–199 (2017) 17. Tsai, P.W., Pan, J.S., Liao, B.Y., Tsai, M.J., Istanda, V.: Bat algorithm inspired algorithm for solving numerical optimization problems. Appl. Mech. Mater. 148, 134–137 (2012); Trans Tech Publ 18. Tu, T.N., Hoang, H.N., Van, T.V., Van, D.D., Phuong, A.H.T.: Enhancing remote sensing image contrast based on combination of fuzzy logic and local approach. J. Inf. Hiding Multimedia Sig. Process 10(4), 488–499 (2019) 19. Tubishat, M., Alswaitti, M., Mirjalili, S., Al-Garadi, M.A., Rana, T.A., et al.: Dynamic butterfly optimization algorithm for feature selection. IEEE Access 8, 194303–194314 (2020) 20. Wang, H., Sun, H., Li, C., Rahnamayan, S., Pan, J.S.: Diversity enhanced particle swarm optimization with neighborhood search. Inf. Sci. 223, 119–135 (2013) 21. Wu, J., Xu, M., Liu, F.F., Huang, M., Ma, L., Lu, Z.M.: Solar wireless sensor network routing algorithm based on multi-objective particle swarm optimization. J. Inf. Hiding Multim. Signal Process. 12(1), 1–11 (2021) 22. Wu, M.E., Syu, J.H., Chen, C.M.: Kelly-based options trading strategies on settlement date via supervised learning algorithms. Comput. Econ. 59(4), 1627–1644 (2022) 23. Xi, J., Chen, Y., Liu, X., Chen, X.: Whale optimization algorithm based on nonlinear adjustment and random walk strategy. J. Netw. Intell. 7(2), 306–318 (2022) 24. Xiankang He, Lijun Yan, S.C.C.S.J.L., Pan, J.S.: A dynamic parallel Harris hawks optimization based WSN node localization algorithm. J. Netw. Intell. 6(4), 688–703 (2021) 25. Xu, L., Li, T.J., Ling, Y.F., Lu, J., Cai, Z.M.: GSGC: an improved path planning optimization. J. Netw. Intell. 7(1), 84–100 (2022) 26. Xue, X., Pan, J.S.: A compact co-evolutionary algorithm for sensor ontology meta-matching. Knowl. Inf. Syst. 56(2), 335–353 (2018) 27. Zhang, F., Wu, T.Y., Wang, Y., Xiong, R., Ding, G., Mei, P., Liu, L.: Application of quantum genetic optimization of LVG neural network in smart city traffic network prediction. IEEE Access 8, 104555–104564 (2020) 28. Zhao, M., Pan, J.S., Chen, S.T.: Entropy-based audio watermarking via the point of view on the compact particle swarm optimization. J. Internet Technol. 16(3), 485–493 (2015)

Chapter 26

Research on Gannet Optimization Algorithm and Its Application in Traveling Salesman Problem Jeng-Shyang Pan, Fei-Fei Liu, Jie Wu, Tien-Szu Pan, and Shu-Chuan Chu

Abstract With the high level of information technology in modern society, a series of intelligent optimization algorithms have emerged to solve classic multicombinatorial optimization applications. The origin of intelligent algorithms is the intelligent behavior and physical phenomenon of biological communities in nature, and a large number of intelligent optimization algorithms are widely used in various combinatorial optimization problems. Gannet optimization algorithm (GOA) is a newly proposed intelligent optimization algorithm, which is applied to largescale constrained optimization problems with the advantages of high convergence and high-quality solutions. For the traveling salesman optimization problem (TSP), the original traditional way is very difficult to calculate. The calculation difficulty increases exponentially with the increase in the number of cities and is rarely used in real life. In this paper, we use the GOA to optimize the TSP problem. Experiments are carried out through two TSP instances, it can be seen from the experimental results that GOA can find a better solution with less computation time.

J.-S. Pan · F.-F. Liu · S.-C. Chu College of Computer Science and Engineering, Shandong University of Science and Technology, Qingdao 266590, China e-mail: [email protected] J. Wu School of Electrical and Information Engineering, Zhengzhou University of Light Industry, Zhengzhou 450002, China e-mail: [email protected] T.-S. Pan Department of Electronic Engineering, National Kaohsiung University of Science and Technology, Kaohsiung 807, Taiwan e-mail: [email protected] S.-C. Chu (B) College of Science and Engineering, Flinders University Sturt Rd, Bedford Park, SA 5042, South Australia e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_26

343

344

J.-S. Pan et al.

26.1 Introduction With the high level of information technology in modern society, the optimal design is always infiltrated in many practical application fields such as logistics and transportation, scientific research, and engineering applications [1–4]. The optimal design can also be used to solve many complex combinatorial optimization problems, such as common electronic maps, electrical wiring, and the traveling salesman problem (TSP) problems [5, 6]. The solution to the TSP problem has always been the focus of research and has been widely used in real life [7, 8]. The first method used to solve the TSP problem is the exhaustive method, which has a simple idea and can quickly find the shortest route with fewer city nodes. However, when solving large-scale cities, the computational load is too large, the operation efficiency is low, and it takes a long time, which has a large defect. Then a series of exact algorithms appeared to solve TSP problems, such as bound method [9–11] and dynamic programming method [12–14]. Although this type of algorithm can obtain accurate solutions, an excessively large search space range will lead to excessive computation time and greatly increase the time complexity, and it is rarely used in practical applications. In later research, intelligent optimization algorithms have attracted the attention of a wide range of scholars. The intelligent optimization algorithms use the principle of bionics to continuously adjust itself during the calculation process. For N-P difficult problems like traveling salesman, the intelligent optimization algorithm can solve such problems faster and can obtain approximate solutions in a relatively short time. At present, this kind of algorithm is a promising way to solve the traveling salesman problem, it is also the research direction of many researchers. Common intelligent optimization algorithms include Particle Swarm Optimization (PSO) [15–19], Cat Swarm Optimization (CSO) [20, 21], Ant Colony Optimization (ACO) [22, 23], Pigeon Inspired Optimization (PIO) [24], Dragonfly Algorithm (DA) [25, 26], Whale Optimization Algorithm (WOA) [27, 28], Colony Predation Algorithm (CPA) [29, 30], Fish Migration Optimization (FMO) [31, 32], and Phasmatodea population evolution algorithm (PPE) [33].

26.2 Gannet Optimization Algorithm Recently, an intelligent gannet optimization algorithm based on the population has been proposed [34]. Gannets mainly live in lakes and the seaside, they often hunt fish on the beach. The algorithm simulates a series of predation behaviors of gannets and conducts mathematical modeling. The GOA algorithm first initializes the population position matrix X . Each xi, j in the population position matrix X represents the position of individual i in the population of dimension j. The calculation of the position matrix is shown in Eq. 26.1.

26 Research on Gannet Optimization Algorithm and Its Application in Traveling …

xi, j = r × (ub j − lb j ) + lb j , [i = 1, 2, . . . N , j = 1, 2, . . . dim]

345

(26.1)

where ub j and lb j both represent the boundaries of the population exploration space, N is the total number of individuals, and dim is the specific dimension of the search space. r ∈ (0, 1). In addition to the initial position matrix X is established, the algorithm also defines a memory matrix N X , and assigns the value of the matrix X to N X during initialization. As the iterative process progresses, the position update of the gannet is stored in N X . Judging by the fitness value evaluation, if the fitness value of individual N X i is better than the value of individual X i , N X i will replace X i , otherwise, it will not be replaced. An important phase is found during predation: exploration. Once the gannet finds prey in the air, it will automatically adjust the diving model according to the depth of the prey. Diving patterns are divided into a deep and long u-shape and a shallow and short v-shape. Two diving modes ensure that gannets explore more possible areas and capture better prey in more search space. The U-diving and V-diving patterns are described by Eqs. 26.3–26.4. t =1−

I ter Max I ter

(26.2)

c = 2 ∗ cos (2 ∗ π ∗ r1 ) ∗ t

(26.3)

d = 2 ∗ v (2 ∗ π ∗ r2 ) ∗ t

(26.4)

v(y) =

 1 − π ∗ y + 1, y (0, π ) 1 ∗ y − 1, y (π, 2π ) π

(26.5)

where I ter represents the number of iterations of the I ter th generation in the algorithm, and Max I ter represents the maximum number of iterations run. r1 , r2  (0, 1). After the diving mode of the gannet is determined, the position of the gannet needs to be updated. Since the probability of the two diving modes being selected is basically the same, using a as the probability number, make a random selection of two dive modes. The position vector equation of the gannet is shown in Eqs. 26.6a–26.6b.  N X i (t + 1) =

X i (t) + u 1 + u 2 , a ≥ 0.5 X i (t) + v1 + v2 , a ≤ 0.5

u 2 = C ∗ (X i (t) − X r (t))

(26.6a) (26.6b)

(26.7)

346

J.-S. Pan et al.

v2 = D ∗ (X i (t) − X a (t))

(26.8)

C = (2 ∗ r3 − 1) ∗ c

(26.9)

D = (2 ∗ r4 − 1) ∗ D

(26.10)

where X i (t) represents the position vector of the individual i when the current iteration number is t. r3 , r4  (0, 1), u 1  (−c, c), v1  (−d, d). X r (t) represents the position of an individual for random selection, and X a (t) represents the average position of all individual positions. The calculation of X a (t) is given in the Eq. 26.11. N X i (t) (26.11) X a (t) = i=1 N Gannets prey after jumping into the water, while fish in the water escape the gannets by suddenly shifting direction. Gannets must consume their own energy to prey on escaped fish. We describe this behavior by defining a capture factor ω, as in the Eq. 26.13. As the gannet’s energy is gradually depleted, it will be unable to catch fish. ω=

t1 = 1 +

R=

1 R ∗ t1

(26.12)

I ter Max I ter

(26.13)

m ∗ vel 2 l

l = 0.2 + (2 − 0.2) ∗ r5

(26.14)

(26.15)

where r5  (0, 1), here we set the mass of the gannet to m = 2.5 Kg, ignore some resistance factors in the water, and set the speed of the gannet to vel = 1.5 m/s. If the prey is within the capture ability range of the gannet, the position of the gannet will be updated with the sudden turn of the fish; Otherwise, the gannet fails to capture the prey and randomly search for the prey through the Levy method, which is shown in Eqs. 26.16a–26.16b.

N X i (t + 1) =

 t ∗  ∗ (X i (t) − X best (t)) + X i (t) , ω ≥ b

(26.16a)

X best (t) − (X i (t) − X best (t)) ∗ L ∗ t, ω < b (26.16b)

26 Research on Gannet Optimization Algorithm and Its Application in Traveling …

347

 = ω ∗ |X i (t) − X best (t)|

(26.17)

L = Levy (dim)

(26.18)

where X best (t) is the optimal individual position vector found in the entire population. The b value of 0.2 is obtained through several experimental tests, and it belongs to an empirical value. Levy() denotes for the Levy flight, as shown in Eq. 26.19. Levy (dim) = 0.01 ×

μ×σ 1

|v| β

(26.19)

⎛

  ⎞ β1  (1 + β) × sin πβ 2 ⎜ ⎟  ⎠ σ =⎝   β−1 1+β 2  2 ×β ×2

(26.20)

where β always takes a value of 1.5, μ, σ  (0, 1). The specific process of the GOA algorithm is shown in Pseudo-code 1.

26.3 Traveling Salesman Problems TSP is a well-known multi-combinatorial optimization application. It can be described in this way, assuming there are n cities, a traveling salesman starts from a certain city, passes through the remaining n − 1 cities, and finally returns to the starting city [35–37]. Each city can only be visited once, our goal is to find the shortest path length among all possible paths [38–40]. In this problem, from the perspective of graph theory, the input of the TSP problem is a complete graph with sideband weights, and the goal is to find a hamiltonian circuit with the smallest weight and minimum. Traverse all nodes of graph G = (v, e, c) at a time, and return to the starting node. Connections are established sequentially, and

these connected city nodes have the least cost, where v = ν1 , ν2 , . . . , ν j , . . . νn ( j = 1, 2, . . . , n) represents the city set, n represents the total number of city nodes, ν j denotes the jth city node. e = {(r1 , s1 ) : r, sv} represents the set formed by the full arrangement of all city nodes, and c = xr1 ,s1 : r1 , s1 == v is the weight value of connections between different city nodes (generally, the distance between cities is used as the standard). The computational cost of the path is represented by Eq. 26.21. f (x) =

n−1  j=1

x ( j), ( j+1) + x (1), (n)

(26.21)

348

J.-S. Pan et al.

Algorithm 1 GOA algorithm Pseudo-code display Input: Population number: N ; search dimension: dim; maximum number of iterations: Max I ter . Output: The position vector of the optimal individual and its corresponding fitness value. 1: Randomly initialize the population position matrix X by Eq. 26.1. r, a (0, 1). 2: Initialize memory matrix N X 3: Evaluate the fitness value of the position vector X . 4: while the iteration termination condition is not met do 5: if rand > 0.5 then 6: for N X i do 7: if a ≥ 0.5 then 8: Update the position vector of the individual by Eq. 26.6a. 9: else 10: Update the position vector of the individual by Eq. 26.6b. 11: end if 12: end for 13: else 14: for N X i do 15: if b ≥ 0.5 then 16: Update the position vector of the individual by Eq. 26.16a. 17: else 18: Update the position vector of the individual by Eq. 26.16b. 19: end if 20: end for 21: end if 22: for N X i do 23: Evaluate the fitness function value of memory matrix N X i . 24: If the evaluation value of N X i is better than the value of X i , replace X i with N X i , otherwise, do not replace. 25: end for 26: end while

where f (x) is the cost of the shortest path of connecting between cities, denotes a sequence of arranged nodes. j is the city node traversed in the jth step. x ( j), ( j+1) is the distance between city j and city j+1 .

26.4 Experimental Results and Analysis In this paper, GOA is applied to TSP problem to solve the shortest path, the computational complexity is greatly reduced, and the solution speed is also accelerated. This paper selects two urban cases, the first case is the ulysses16 datasets, which are commonly used in the TSP dataset. The number of city nodes in these two datasets is 16, which are low-dimensional problems. The second case is a randomly generated 70 city nodes, which belong to the middle and high latitude problem. In the two cases, the number of individuals is set to 50, and iterations are set to 500. The experiment runs on a personal desktop computer with Windows 10 operating system, 24 GB

26 Research on Gannet Optimization Algorithm and Its Application in Traveling …

349

Fig. 26.1 First case: ulysses16 dataset

memory, Intel(R) Core(TM) i5-8500 CPU @ 3.00 GHz 3.00 GHz, and the software environment is Matlab2020b. Figures 26.1 and 26.2 show the experimental results of the GOA algorithm on these two urban cases. Subgraph (a) represent the scatterplot coordinates of the two urban cases, respectively. Subgraph (b) is a fully arranged path graph between the nodes of each case city. Subgraph (c) shows the shortest path graph obtained by the GOA. Subgraph (d) shows the iterative curve of the shortest path distance obtained by the GOA algorithm on these two urban cases. According to the iterative convergence curve, it can be seen that on the ulysses16 dataset, the distance of the shortest path obtained by the GOA is 97.0544. On the randomly generated 70 city nodes, the distance of the shortest path obtained by the GOA is 3081.2733. If the number of problem dimensions (city nodes) increases, there are more connections between all city nodes, and the GOA algorithm can still find a shorter path among many paths. The optimization ability of the GOA algorithm does not weaken with the increase of the dimension, and the convergence performance can also maintain a relatively excellent effect.

350

J.-S. Pan et al.

Fig. 26.2 Second case: 70 randomly generated city nodes

26.5 Conclusions In this paper, the gannet optimization algorithm (GOA) is introduced in detail. According to the predation habits of gannets, the corresponding mathematical model is established. Mathematical modeling of the two diving modes of gannets u and v is helpful for gannets to fully explore the area. During the predation period, two strategies of sudden turning and random walking were used for small-scale exploitation, which also improved the predation ability of gannets. The GOA algorithm has good optimization ability, and We apply the GOA algorithm to solve the TSP problem. According to the experimental results, as the number of city nodes increases, GOA can solve the TSP problem faster and can obtain approximate solutions in a relatively short time. Its performance of finding the shortest path is also quite satisfactory.

26 Research on Gannet Optimization Algorithm and Its Application in Traveling …

351

References 1. Zhang, F., Wu, T.Y., Wang, Y., Xiong, R., Ding, G., Mei, P., Liu, L.: Application of quantum genetic optimization of LVQ neural network in smart city traffic network prediction. IEEE Access 8, 104555–104564 (2020) 2. Kang, L., Chen, R.S., Chen, Y.C., Wang, C.C., Li, X., Wu, T.Y.: Using cache optimization method to reduce network traffic in communication systems based on cloud computing. IEEE Access 7, 124397–124409 (2019) 3. Wang, E.K., Zhang, X., Wang, F., Wu, T.Y., Chen, C.M.: Multilayer dense attention model for image caption. IEEE Access 7, 66358–66368 (2019) 4. Xi, J., Chen, Y., Liu, X., Chen, X.: Whale optimization algorithm based on nonlinear adjustment and random walk strategy. J. Netw. Intell. 7(2), 306–318 (2022) 5. Schrijver, A.: On the history of combinatorial optimization (till 1960). Handb. Oper. Res. Manag. Sci. 12, 1–68 (2005) 6. Ibarra, O.H., Kim, C.E.: Heuristic algorithms for scheduling independent tasks on nonidentical processors. J. ACM (JACM) 24(2), 280–289 (1977) 7. Yang, J., Shi, X., Marchese, M., Liang, Y.: An ant colony optimization method for generalized TSP problem. Prog. Nat. Sci. 18(11), 1417–1422 (2008) 8. Tao, Z.: TSP Problem solution based on improved Genetic Algorithm. In: 2008 Fourth International Conference on Natural Computation, vol. 1, pp. 686–690 (2008) 9. Böckenhauer, H.J., Hromkoviˇc, J., Klasing, R., Seibert, S., Unger, W.: An improved lower bound on the approximability of metric TSP and approximation algorithms for the TSP with sharpened triangle inequality. In: Annual Symposium on Theoretical Aspects of Computer Science, pp. 382–394 (2000) 10. Azar, Yossi: Lower bounds for insertion methods for TSP. Comb. Probab. Comput. 3(3), 285– 292 (1994) 11. Shmoys, D.B., Williamson, D.P.: Analyzing the Held-Karp TSP bound: a monotonicity property with application. Inf. Process. Lett. 35(6), 281–285 (1990) 12. Chauhan, C., Gupta, R., Pathak, K.: Survey of methods of solving tsp along with its implementation using dynamic programming approach. Int. J. Comput. Appl. 52(4), 0975–8887 (2012) 13. Bellman, R.: Dynamic programming treatment of the travelling salesman problem. J. ACM (JACM) 9(1), 61–63 (1962) 14. Bouman, P., Agatz, N., Schmidt, M.: Dynamic programming approaches for the traveling salesman problem with drone. Networks 72(4), 528–542 (2018) 15. Xue, X., Tsai, P.W.: Integrating Energy Smart Grid’s ontologies through multi-objective particle swarm optimization algorithm with competitive mechanism. Sustainable Energy Technologies and Assessments 53, 102442 (2022) 16. Luo, Y., Liu, J., Xue, X., Hu, R., Wang, Z.: The Experimental analysis on transfer function of binary particle swarm optimization. In: International Conference on Swarm Intelligence, vol. 12689, pp. 254–264 (2021) 17. Song, W., Huang, C.: Mining high average-utility item sets based on particle swarm optimization. Data Sci. Pattern Recognit. 4(2), 19–32 (2020) 18. Shi, Y.: Particle swarm optimization: developments, applications and resources. In: Proceedings of the 2001 Congress on Evolutionary Computation, vol. 1, pp. 81–86 (2001) 19. Wu, J., Xu, M., Liu, F.F., Huang, M., Ma, L., Lu, Z.M.: Solar wireless sensor network routing algorithm based on multi-objective particle swarm optimization. J. Inf. Hiding Multimed. Signal Process. 12(1), 1–11 (2021) 20. Chu, S.C., Tsai, P.W., Pan, J.S.: Cat swarm optimization. In: Pacific Rim International Conference on Artificial Intelligence, vol. 4099, pp. 854–858 (2006) 21. Bahrami, M., Bozorg-Haddad, O., Chu, X.: Cat swarm optimization (CSO) algorithm. Adv. Optim. Nat.-Inspired Algorithms 720, 9–18 (2018) 22. Chu, S.C., Roddick, J.F., Pan, J.S.: Ant colony system with communication strategies. Inf. Sci. 167(1–4), 63–76 (2004)

352

J.-S. Pan et al.

23. Blum, C.: Ant colony optimization: introduction and recent trends. Phys. Life Rev. 2(4), 353– 373 (2005) 24. Pan, J.S., Tian, A.Q., Snášel, V., Kong, L., Chu, S.C.: Maximum power point tracking and parameter estimation for multiple-photovoltaic arrays based on enhanced pigeon-inspired optimization with Taguchi method. Energy 251, 123863 (2022) 25. Mirjalili, S.: Dragonfly algorithm: a new meta-heuristic optimization technique for solving single-objective, discrete, and multi-objective problems. Neural Comput. Appl. 27(4), 1053– 1073 (2016) 26. Liu, F.F., Chu, S.C., Wang, X., Pan, J.S.: A collaborative dragonfly algorithm with novel communication strategy and application for multi-thresholding color image segmentation. J. Internet Technol. 23(1), 45–62 (2022) 27. Mirjalili, S., Lewis, A.: The whale optimization algorithm. Adv. Eng. Softw. 95, 51–67 (2016) 28. Chen, H., Xu, Y., Wang, M., Zhao, X.: A balanced whale optimization algorithm for constrained engineering design problems. Appl. Math. Model. 71, 45–59 (2019) 29. Tu, J., Chen, H., Wang, M., Gandomi, A.H.: The colony predation algorithm. J. Bionic Eng. 18(3), 674–710 (2021) 30. Shi, B., Ye, H., Zheng, L., Lyu, J., Chen, C., Heidari, A.A., Wu, P.: Evolutionary warning system for COVID-19 severity: Colony predation algorithm enhanced extreme learning machine. Comput. Biol. Med. 136, 104698 (2021) 31. Chu, S.C., Du, Z.G., Pan, J.S.: Discrete fish migration optimization for traveling salesman problem. Data Sci. Pattern Recognit. 4(2), 1–18 (2020) 32. Liu, Y., Zheng, W.M., Liu, S., Chai, Q.W.: Gaussian-based adaptive fish migration optimization applied to optimization localization error of mobile sensor networks. Entropy 24(8), 1109 (2022) 33. Liang, L.L., Du, Z.G., Shieh, C.S., Hu, C.C., Chu, S.C., Feng, Q.: A new PPE algorithm based on parallel communication strategy. Adv. Intell. Inf. Hiding Multimed. Signal Process. 277, 289–298 (2022) 34. Pan, J.S., Zhang, L.G., Wang, R.B., Snášel, V., Chu, S.C.: Gannet optimization algorithm: a new metaheuristic algorithm for solving engineering optimization problems. Math. Comput. Simul. 202, 343–373 (2022) 35. Zhou, A., Kang, L., Yan, Z.: Solving dynamic TSP with evolutionary approach in real time. In: The 2003 Congress on Evolutionary Computation, vol. 2, pp. 951–957 (2003) 36. Luo, X. H., Yang, Y., Li, X.: Solving TSP with shuffled frog-leaping algorithm. In: 2008 Eighth International Conference on Intelligent Systems Design and Applications, vol. 3, pp. 228–232 (2008) 37. Fang, L., Chen, P., Liu, S.: Particle swarm optimization with simulated annealing for TSP. In: Proceedings of the 6th WSEAS International Conference on Artificial Intelligence, Knowledge Engineering and Data Bases, pp. 206–210 (2007) 38. Xu, L., Li, T.J., Ling, Y.F., Lu, J., Cai, Z.M.: GSGC: an improved path planning optimization method using guided sampling and gradual cutting. J. Netw. Intell. 7(1), 84–100 (2022) 39. Nguyen, Trong-The., Dong-Nguyen, Trinh, Nguyen, Vinh-Tiep.: An optimizing pulse coupled neural network based on golden eagle optimizer for automatic image segmentation. J. Inf. Hiding Multimed. Signal Process. 13(3), 155–164 (2022) 40. Chen, C.M., Xiang, B., Wang, K.H., Yeh, K.H., Wu, T.Y.: A robust mutual authentication with a key agreement scheme for session initiation protocol. Appl. Sci. 8(10), 1789 (2018)

Chapter 27

Artificial Hummingbird Algorithm with Parallel Compact Strategy Shu-Chuan Chu, Zhi-Yuan Shao, Chin-Shiuh Shieh, and Xiaoqing Zhang

Abstract This paper improves the Artificial Hummingbird Algorithm (AHA). First, we introduce a compact scheme to reduce computer storage capacity and speed up computation. Second, the parallel strategy is added to improve the optimization ability of the algorithm. Third, we improve the original algorithm’s territorial and migration foraging strategies. The purpose of enhancing the territorial foraging strategy is to optimize the algorithm to be more directional. We removed the migration-foraging strategy, which is more suitable for combining with the compact scheme. Finally, we tested the improved algorithm on the cec2013 test set, which showed good performance.

27.1 Introduction In real life, various problems must be optimized, such as wireless sensor network design problems, aircraft optimal path problems, reducer design problems, smart city traffic network prediction problems [1], cloud computing traffic design problems [2], IoT design problems [3], smart medical and path planning problems [4, 5], etc. Two main algorithms to solve these problems are a deterministic algorithm and a meta-heuristic algorithm. Deterministic algorithms always work mechanically and produce the same result given a specific input. When solving some nonlinear problems, deterministic algorithms can effectively find local optima. Still, they may require information on the derivative of the problem and are prone to get stuck in local

S.-C. Chu · Z.-Y. Shao · X. Zhang (B) College of Computer Science and Engineering, Shandong University of Science and Technology, Qingdao 266590, China e-mail: [email protected] C.-S. Shieh Department of Electronic Engineering, National Kaohsiung University of Science and Technology, Kaohsiung 807618, Taiwan e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_27

353

354

S.-C. Chu et al.

optima. However, meta-heuristic algorithms can effectively avoid these problems and show more powerful optimization capabilities than deterministic algorithms. A Genetic Algorithm (GA) is a classical evolutionary algorithm proposed according to the law of biological evolution [6]. Particle swarm optimization (PSO) is proposed based on the social behavior of birds. It finds the optimal solution according to the optimal historical position and optimal global position of particles [7]. Ant Colony Optimization (ACO) simulates the social behavior of ants foraging and seeks the optimal solution by adjusting the concentration of pheromones on the path [8]. Cuckoo Search (CS) improves the ability to find optimal solutions by simulating cuckoos’ reproduction and flight behavior [9]. However, not all problems can be optimally solved using one algorithm. According to the No Free Lunch Theorem [10], different problems may require different algorithms to solve. Therefore, different new algorithms are proposed every year, for example, SCA [11], EO [12], AO [13], PPE [14], HHO [15], GOA [16], and AHA [17]. A common feature of most meta-heuristics is the exploration and exploitation process to find optimal solutions [18]. Each algorithm can’t balance exploration and exploitation with the best of both worlds. Therefore, we can improve the algorithm to solve practical problems better. There are various improvement methods, and multiple algorithms have been enhanced in the literature [19–26]. The most common method is the parallel method. The idea of parallelism is to divide the population into multiple groups, and each group handles the problem independently. Particles in each group communicate with each other to jump out of the optimal local solution. Another standard method is the compact scheme, which reduces storage capacity and speeds up computation by mapping the population to a distribution function. These two methods are used in [27–33]. This paper presents two methods for improving the artificial hummingbird algorithm, organized as follows. Section 27.2 presents the compact scheme and the original AHA algorithm. In Sect. 27.3, we offer the enhanced AHA algorithm. Section 27.4 uses the cec2013 test set to test the improved algorithm and compare the enhanced AHA algorithm with other improved ones. Finally, Sect. 27.5 gives the conclusion.

27.2 Related Work 27.2.1 Compact Scheme The Estimation of Distribution Algorithm (EDA) maps the population into a probability model and operates on the population through the probability model [34]. The compact scheme is one of its specifics. The compact scheme can reduce memory usage, and operating the entire population through a probabilistic model can reduce the amount of computation and shorten the algorithm’s running time. The compact

27 Artificial Hummingbird Algorithm with Parallel Compact Strategy

355

scheme replaces the whole population with a virtual population, encoding the population into a data structure, the perturbation vector, denoted by PV. The PV vector is continuously updated as the algorithm runs. Its representation is: where it is used to represent the mean of the perturbation vector, the standard deviation of the perturbation vector, and is used to describe the current iteration number. Each pair of mean and standard deviation in the PV vector corresponds to a probability density function (PDF), the PDF is truncated within [−1, 1], and the PDF normalizes the area of the amplitude to 1 [35]. The solution of the population can be generated by the inverse cumulative distribution function (CDF). After using the PV vector to develop two solutions, compare the fitness values of the two solutions. The winner is the one with good fitness value and the loser with poor fitness value. The generated two solutions update the PV. The updated rules are as follows: μt+1 = μt +  δ

t+1

=

1 (winner − loser) Np

2  1 (winner 2 − loser 2 ) (δ t )2 + (μt )2 − μt+1 − Np

(27.1)

(27.2)

27.2.2 AHA Algorithm The AHA algorithm is a meta-heuristic algorithm based on hummingbirds’ flight skills, memory ability, and foraging strategy. It mainly consists of three parts: Food sources, Hummingbirds, and Visit table. In AHA, each hummingbird is permanently assigned to a specific food source, and the hummingbird has the exact location of the food source. Hummingbirds can remember the place and rate of nectar replenishment for this particular food source and share this information with other hummingbirds in the population. In addition, each hummingbird can also remember how long each food source has not been visited by itself. The information on hummingbird visits is recorded in the visit table, and when the number of visits is the same, hummingbirds tend to fly to food sources with a high rate of honey filling. The AHA algorithm finds the optimal solution by simulating the three foraging behaviors of hummingbirds. Mathematical models of three foraging behaviors of hummingbirds: guided foraging, territorial foraging, and migration foraging. Guided foraging Hummingbirds share information to find the best food source. It always tends to look for those food sources that have not been visited for a long time. When the access levels of these food sources are the same, it will choose the food source with the highest nectar rate. After finding a suitable food source, hummingbirds will fly toward it. There are three flight modes: axial flight, diagonal flight, and omnidirectional flight. The formula is as follows:

356

S.-C. Chu et al.

D D(i) =



(i)

 =

1 if i = r and i([1, d ]) i = 1, . . . d else 0

(27.3)

1 if i = p(j), j = [1, k], p = r and perm(k), k[2, r1 ∗ (d − 2) + 1] else 0 i = 1, . . . , d

(27.4)

D(i) = 1 i = 1, . . . , d

(27.5)

Randomly select the 1D dimension of a particle and set its dimension value to 1, indicating axial flight. Randomly select some of the 1D dimensions and set the dimension value to 1 to indicate diagonal flight, and set the value of all dimensions to 1 to indicate omnidirectional flight. The mathematical modeling of the guided foraging strategy is expressed as follows: vi (t + 1) = xi,tar (t) + a ∗ D ∗ (xi (t) − xi,tar (t)) a ∼ N (0, 1)  xi (t + 1) =

xi (t) f (xi (t) ≤ f vi (t + 1)) vi (t + 1) f (xi (t) > f vi (t + 1))

(27.6)

(27.7)

xi (t) is the position of the ith food source at time t, and xi,tar (t) is the target food source position that the ith hummingbird intends to visit. a is the guiding factor, which obeys the normal distribution with mean = 0 and standard deviation = 1. Equation (27.7) indicates that if the nectar refill rate of the candidate food source is better than that of the current food source, the hummingbird will give up the current food source and stay on the candidate food source obtained by Eq. (27.6) and continue to eat. Territorial foraging After foraging in one food source, hummingbirds fly to the next food source to forage. However, some foods around it tend to have more honey filling rates, so it will fly to the next distant food source or stay in the local search for selection, with a probability of 0.5. The following is a territorial search formula: vi (t + 1) = xi (t) + b ∗ D ∗ xi (t) b ∼ N (0, 1)

(27.8)

where b is a territorial factor, which obeys the normal distribution with mean = 0 and standard deviation = 1. Migration foraging When hummingbirds frequently forage in areas where they often forage, the rate of feeding the food source decreases, at which point hummingbirds migrate to more distant locations to provide. A migration coefficient is defined in the AHA algorithm, and hummingbirds will migrate after a certain number of iterations. The mathematical formula is as follows:

27 Artificial Hummingbird Algorithm with Parallel Compact Strategy

xwor (t + 1) = low + rand ∗ (up − low)

357

(27.9)

The xwor (t + 1) was the food source with the worst rate of nectar replenishment in the population. rand is a random number between [0–1]. up and low represent the maximum and minimum values of the boundary, respectively.

27.3 Improved Artificial Hummingbird Algorithm 27.3.1 Compact Artificial Hummingbird Algorithm In the process of using the compact scheme to improve meta-heuristics, most of them use the PV vector to generate one solution. They then use the meta-heuristic to create another solution. The advantage of this is that it can effectively save memory capacity and improve the computing speed of the computer. However, this method also has obvious shortcomings. That is, it uses a probabilistic model to represent the population. Two particles are extracted from the probabilistic model for iteration, and the diversity of the population is reduced. As the PV vector converges to a specific region, the particles removed from it will not be able to jump out of the local optimum. Considering this situation, combine the three flight modes of hummingbirds in the AHA algorithm and the attributes of the visit table. In this paper, three particles are extracted from the PV vector, and the AHA algorithm generates three particles, thereby increasing the diversity of the population, which can effectively jump out of the local optimum and find the global optimum solution. We use the PV vector to generate three particles, each of which chooses a way to fly. The AHA algorithm updates the particle position, generating three new particles. The compact scheme compares the particles with the best and worst fitness values as winners and losers. Update the PV vector with these two particles. Another improvement of the compact artificial hummingbird algorithm is the improvement of the migration foraging strategy and the territorial foraging strategy. On the one hand, the compact artificial hummingbird algorithm uses PV vectors to generate fewer particles, and the initialization has little impact on the entire probability model. On the other hand, the overall optimization ability decreases when a poor group is initialized. Therefore, this paper does not use the migration foraging strategy. In formula (27.8), the moving direction of each particle is based on its position, so the moving range is minimal, and there is no clear direction. So, we move the particle toward the optimal particle so that it can find the food source faster. Change the formula as follows: vi (t + 1) = xi (t) + b ∗ D ∗ BestX BestX represents the global optimal particle.

(27.10)

358

S.-C. Chu et al.

27.3.2 Parallel Strategy The parallel strategy in the meta-heuristic Algorithm is to divide the population into different groups and exchange information between the groups through different communication strategies so that the population can find the optimal solution faster. In this section, we divide the population into three groups. Each group has its PV vector and iteratively updates its PV vector. First, initialize the PV vector, we let mean = 0 and standard deviation = 10. Then operate 3.1 for each group. Finally, we get the optimal global solutions for each group and exchange information. The communication strategy is as follows: we let each group communicate with other groups. First, we compare the fitness values of the optimal particles of the two groups and replace the particles with poor fitness values with another particle. Then perturb the particle with a good fitness value, and replace it if the obtained particle fitness value is better. The pseudocode is shown in Algorithm. Algorithm: PCAHA Require: Virtual population size, Max iteration, Fitness function, Dimension, Upper and Lower bounds Ensure: Global best value Global_BestF 1: Set the number of groups g 2: Initialize the G(g).PV , G(g).BestX and G(g).BestF and VisitTable of each group 3: Global_BestX=G(1).BestX; Global_BestF=G(1).BestF ; count=1 4: for It=2:iter_max do 5: count=count+1; 6: for g=1:G do 7: Generate three particles using PV vector and calculate their fitness value 8: Each particle selects a flight mode according to Eq.(3) and Eq.(4) and Eq.(5) 9: for i=1:3 do 10: if rand 0}. d is the camera center o to pixel unit direction vector. The cumulative color C(r ) along this ray is the pixel color, which is calculated by the following formula. ∞ T (t)σ (r (t))c(r (t), n(r (t)), g(r (t)), d)dt

C(r ) =

(30.4)

0

⎛ T (t) = exp ⎝−

t

⎞ σ (r (s))ds ⎠

(30.5)

0

⎧

⎨ α exp SDF(x) 2 β

σα,β (x) = ⎩α − α exp − SDF(x) 2 β

if SDF(x) ≤ 0 if SDF(x) > 0

(30.6)

where T (t) is cumulative transmission probability along ray r (t). σ (r (ti )) denotes the probability that ray r (t) terminates at ti , where α, β are learnable parameters. c(r, n, g, d) represents light field, also known as color field, which is the amount of light emitted in the direction of each spatial point, and can also be considered as RGB color values. The light field is constrained by the normal n of surface and the spatial point feature g, both n (r (ti )) = ∇x f θ,sd f (r (ti )) and g (r (ti )) = f θ,g (r (ti )) ⊕ f I (r (ti )). Spatial point feature g is generated by concatenating pixel feature f I (r (ti )) and global shape feature f θ,g (r (ti )). This fusing method of ours considers not only how the bidirectional reflectance distribution function (BRDF) surface normals of common materials are encoded in volume rendering, but also the effects of different wavelengths of incident light and global shape features. Equation (30.4) is approximately solved by numerical integration. N sampling ˆ ) along the ray points xi are sampled along a ray r (t) , and the cumulative color C(r r (t):

390

X. Huang et al. N 

ˆ )= C(r

i=1

Ti (1 − exp(−σi (xi ))δi ))ci (xi , n(xi ), g(xi ), d)  wher e Ti = exp(− σ j (x j )δ j )

(30.7)

jρi

j

34.3.3 Assignment Strategy Combining Shared Nearest Neighbor and Natural Nearest Neighbor Weighted Similarity In the sample assignment strategy, the DPC algorithm only contemplated the distance-density relationship between samples, which makes the density peak clustering algorithm perform better on some simple shaped datasets. However, in the face of manifold dataset, the DPC algorithm is easy to assign the samples originally belonging to a certain cluster to other clusters in the clustering process incorrectly, and then transfer the assigned error to the next sample, which leads to continuous error transmission and makes the clustering accuracy not high. In this paper, we inaugurate the shared nearest neighbors and natural nearest neighbors of samples, and redefine the inter-point similarity of samples. The new definition of sample similarity makes the similarity of samples more dependent on the local environment in which the samples are located, which is more consistent with the sample distribution than the traditional allocation strategy, in order to avoid the cascading effect generated by sample misallocation. Definition 1 (Shared nearest neighbors (SNN)) The intersection of two samples i and the K-nearest neighbor samples of sample j is called the shared nearest neighbor of sample i and sample j, which is given by SNN(xi , x j ) = KNN(xi ) ∩ KNN(x j )

(34.11)

Definition 2 (Natural nearest neighbors (NNN)) A sample i is said to be a natural nearest neighbor of sample j if sample j is incorporated in the K-nearest neighbors of sample i and sample i is included in the K-nearest neighbors of sample j. If the samples are not natural nearest neighbors of each other, then the value is 0. The formula is described as

34 Density Peaks Clustering Algorithm for Manifold Data Based …

 NNN(i, j) =

  1, x j ∈ KNN(xi ), xi ∈ KNN x j 0, otherwise

445

(34.12)

Definition 3 (Sample similarity) For samples i and j in the dataset, the sample similarity is defined as follows: 

A(xi , x j ) = xv

1 1 + dv j ∈[KNN(x ),x ] i

(34.13)

i

  Sim(xi , x j ) = (SNN(xi , x j ) + NNN(xi , x j )) · (A(xi , x j ) + A(x j , xi )) (34.14) A(xi , x j ) denotes the total of the resemblance of sample i and its K-nearest neighbors to sample j. It can be seen that the lower distance of sample points, the greater the  similarity between sample points. SNN(xi , x j ) denotes the number of constituents of the shared nearest neighbor ensemble of samples i and j. The sum of similarity of samples is weighted by the number of constituents of shared nearest neighbors and natural nearest neighbors, and similarity exists only when there are shared nearest neighbors or mutual natural nearest neighbors between samples. The new assignment strategy can fully consider the terms in which the samples are positioned and avoid the domino continuity error during the assignment of samples. To verify the advantages of the allocation strategy of this algorithm, experiments are conducted in this paper using the cth3 dataset. Under the premise of uniformly using Gaussian kernel as the local density calculation formula, Fig. 34.1 shows the clustering results acquired by using the allotment strategy of DPC algorithm, and Fig. 34.2 shows the clustering results acquired by using the allocation strategy of weighted similarity. From Fig. 34.1, we can see that the DPC algorithm can only accurately distinguish the middle three clusters and find the exact density peak points, but for the outer manifold clusters, the outer clusters are assigned to the middle three clusters by angle because the geometry of the samples is ignored, while Fig. 34.2 uses the weighted similarity allotment approach to significantly improve the clustering results on the manifold cluster, which proves the effectiveness of the weighted similarity assignment strategy.

34.3.4 Algorithm Steps The clustering steps of the DPC-GWNN algorithm are as follows: Input: Dataset S = {X 1 , X 2 , ..., X n }, number of samples nearest neighbors. Output: Clustering result. Step 1: Data standardization. Step 2: Constructing the connectivity graph according to the number of nearest neighbors K, and then using the Dijkstra algorithm to compute the shortest distance

446 Fig. 34.1 Clustering results obtained by the assignment strategy of the DPC algorithm

X.-Y. Hu et al. 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0.7

0.8

0.9

1

1

Fig. 34.2 Clustering results obtained by weighted similarity assignment strategy

0.9 0.8 0.7

y

0.6 0.5 0.4 0.3 0.2 0.1 0

0

0.1

0.2

0.3

0.4

0.5

x

0.6

between two points to generate the geodesic distance matrix, and calculating the distance between samples according to steps (7)–(8). Step 3: Computed the sample local density according to steps (9)–(10). Step 4: Computed the decision value of the sample using the geodesic distance of the sample and the local density calculated in Step 3 and draw a decision diagram to select the cluster center. Step 5: Computed the sample similarity according to steps (13)–(14). Step 6: Loop operation for all assigned samples to find the highest similarity and not yet assigned samples, the point can be allocated to the cluster where the assigned samples are located, until the end of clustering.

34 Density Peaks Clustering Algorithm for Manifold Data Based … Table 34.1 Basic characteristics of manifold dataset

Dataset

No. of records

No. of attributes

447 No. of cluster

Jain

373

2

2

Pathbased

300

2

3

Spiral

312

2

3

Lineblobs

266

2

3

Cth

1016

2

4

Db

630

2

4

34.4 Experimental Results and Analysis 34.4.1 Experimental Setting To verify the performance of the DPC-GWNN algorithm, six manifold datasets are used to experiment the DPC-GWNN algorithm in this paper, and the datasets are shown in Table 34.1. The experimental results of the DPC-GWNN algorithm are compared with the DPC [26], DPC-GD [35], and FKNN-DPC [32] algorithms. In this paper, the parameters of the comparison algorithms are adjusted, and the parameters of the DPC algorithm are selected in the range of 0.1–5%; the K-nearest neighbor parameters of the DPC-GWNN and FKNN-DPC algorithms are selected in the range of 1–50. The K-nearest neighbor parameter of DPC-GD takes values in the field [2–4, 6, 10] and the truncation distance parameter takes values in the range [0.1, 0.2, 0.5, 1, 2, and 6%]. The experimental environment adopted for all clustering algorithms in this paper were Intel(R) Core(TM) i5-1035G4 CPU @ 1.10 GHz, 8G RAM, Windows 10 64bit operating system, and MATLAB 2021a programming environment. In this paper, the evaluation metrics for clustering effectiveness are based on Adjusted Mutual Information (AMI) [36], Adjusted Rand Index, (ARI) [36], and Fowlkes-Mallows index (FMI) [37]. The closer the value of each index is to 1 the better the algorithm’s results on the dataset.

34.4.2 Experimental Results and Analysis of the Manifold Dataset The basic characterization of the experimentally selected manifold dataset is shown in Tables 34.1 and 34.2 indicates three evaluation metrics of all the 4 clustering algorithms on the results of 6 manifold datasets, and the bold font in the table indicates the optimal clustering outcomes on this dataset. A comprehensive comparison of the clustering results of all algorithms on the six manifold datasets shows that the DPCGWNN algorithm achieves the foremost clustering results on the Jain and Pathbased

448

X.-Y. Hu et al.

Table 34.2 Performance of 4 cluster algorithms on 6 manifold datasets Algorithms

AMI

ARI

FMI

Jain

AMI

ARI

FMI

Pathbased

DPC-GWNN

1

1

1

0.9482

0.9695

0.9796

DPC

0.6183

0.7146

0.8819

0.5573

0.5082

0.6848

DPC-GD

0.6183

0.7146

0.8819

0.5294

0.4797

0.6703

FKNN-DPC

0.7092

0.8224

0.9359

0.9305

0.9499

0.9665

1

1

1 0.8148

Spiral DPC-GWNN

1

Lineblobs 1

1

DPC

1

1

1

0.8375

0.7179

DPC-GD

1

1

1

1

1

1

FKNN-DPC

1

1

1

1

1

1

DPC-GWNN

1

1

1

1

1

1

DPC

0.6866

0.5135

0.6473

0.5185

0.2794

0.5853

Cth

Db

DPC-GD

1

1

1

1

1

1

FKNN-DPC

1

1

1

0.5107

0.2718

0.5793

datasets compared to the comparison algorithms. The clustering results of DPCGWNN algorithm on Spiral, Lineblobs, Cth, and Db datasets are better than or equal to those of other comparative algorithms, indicating that DPC-GWNN algorithm can handle manifold clusters with complex shapes and has excellent clustering results on manifold datasets. Due to the limitation of space, two typical manifold datasets are selected in this paper. Figures 34.3 and 34.4 indicate the clustering results of DPC-GWNN, DPC, DPC-GD, and FKNN-DPC algorithms on Jain and Cth datasets. The different colors in the results indicate distinct clusters, and the star dots spot the cluster centers. The Jain dataset consists of 373 samples and contains two arc-shaped manifold cluster. Figure 34.3 shows that the DPC algorithm and the DPC-GD algorithm misidentified the upper part of the cluster center within the lower part of the cluster, and FKNN-DPC found the correct cluster center, but all three algorithms are due to the fact that the two FKNN-DPC found the correct clustering center, but all three algorithms are inaccurate due to the misallocation of a point where the clusters are close to each other and lead to a continuum error, resulting in inaccurate clustering results. Only the DPC-GWNN algorithm can find the clustering center accurately and at the same time, the clusters are completely distinguished. The Db dataset consists of 630 samples and contains four manifold clusters with curvilinear patterns of large differences in size. In Fig. 34.4, the DPC algorithm incorrectly elites two cluster centers on the longest manifold cluster. resulting in an extensive error in sample assignment. The FKNN-DPC, on the other hand, identifies three centers on the largest manifold cluster identifies three centroids, and only the

34 Density Peaks Clustering Algorithm for Manifold Data Based …

449

Fig. 34.3 Clustering results of 4 algorithms on Jain dataset

outermost cluster correctly identifies the centroids, leading to errors in sample assignment and thus reducing the clustering quality. Only the DPC-GWNN and DPC-GD algorithms correctly recognized the cluster centers, and although the two algorithms identified slightly different centroids, they eventually correctly distinguished the four clusters.

34.5 Conclusion The DPC-GWNN algorithm firstly uses geodesic distance in place of Euclidean distance to reckon the sample spacing, which makes the distance between samples on the same manifold cluster closer; combines weighted K-nearest neighbors to redefine the local density, draws a new decision map, and accurately finds the cluster center of manifold cluster, which improves the clustering effect of manifold dataset; uses shared nearest neighbors and natural nearest neighbors to weight the similarity. The new inter-sample similarity is defined to fully integrate the local information

450

X.-Y. Hu et al.

Fig. 34.4 Clustering results of 4 algorithms on Db dataset

of samples, while avoiding the cascading effect of errors, and then achieve good clustering effect. The experimental results demonstrate that the DPC-GWNN algorithm can precisely recognize the manifold cluster and then find the cluster center, avoiding the problem that samples are easily misallocated and bring cascading errors, and finally achieve accurate clustering in the manifold dataset. Acknowledgements This research was clamped by the National Natural Science Foundation of China under Grant (No. 52069014), Jiangxi Provincial Key Research and Development Program under Grant (Nos. 20192BBE50076, 20203BBGL73225).

34 Density Peaks Clustering Algorithm for Manifold Data Based …

451

References 1. Xu, J., Lou, H.: X-strip Y-band location clustering algorithm for network topology inference. J. Inf. Hiding Multimed. Signal Process. 12(4), 175–185 (2021) 2. Pan, J.S., Hu, P.: Novel parallel heterogeneous meta-heuristic and its communication strategies for the prediction of wind power. Processes 7(11), 845 (2019) 3. Zhang, S., Su, X., Jiang, X.: A traffic prediction method of bicycle-sharing based on long and short term memory network. J. Netw. Intell. 4(2), 17–29 (2019) 4. Wang, J., Cui, J., Chen, C.: A prediction method of consumer buying behavior based on attention mechanism and CNN-BiLSTM. J. Netw. Intell. 7(2), 375–385 (2022) 5. Ma, S., Cui, J.: An improved Bi-LSTM EEG emotion recognition algorithm. J. Netw. Intell. 7(3), 623–639 (2022) 6. Wu, T., Lin, J.: An efficient algorithm for fuzzy frequent, itemset mining. J. Intell. Fuzzy Syst. 38(5), 5787–5797 (2020) 7. Wu, T., Lin, C.: A grid-based swarm intelligence algorithm for privacy-preserving data mining. Appl. Sci. 9(4), 774 (2019) 8. Chen, C., Chen, L.: Discovering high utility-occupancy patterns from uncertain data. Inf. Sci. 546, 1208–1229 (2021) 9. Zhang, F., Wu, T.: Human motion recognition based on SVM in VR art media interaction environment. HCIS 9(1), 1–15 (2019) 10. Wu, H., Xiao, R.: Flexible wolf pack algorithm for dynamic multidimensional knapsack problems. Research 2020 (2020) 11. Wu, H., Xue, J., Xiao, R.: Uncertain bilevel knapsack problem based on improved binary wolf pack algorithm. Front. Inf. Technol. Electron. Eng. 21(9), 1356–1368 (2020) 12. Kumar, S., Damaraju, A.: LSTM network for transportation mode detection. J. Internet Technol. 22(4), 891–902 (2021) 13. Karim, M.R., Beyan, O.: Deep learning-based clustering approaches for bioinformatics. Brief. Bioinform. 22(1), 393–415 (2021) 14. Liang, T.M., Chiu, M.S.: Applying image processing technology to face recognition. J. Inf. Hiding Multimed. Signal Process. 13(2), 106–112 (2022) 15. Kang, L., Chen, R.S.: Selecting hyper-parameters of Gaussian process regression based on non-inertial particle swarm optimization in internet of things. IEEE Access 7, 59504–59513 (2019) 16. Pan, J., Chai, Q.W.: 3-D terrain node coverage of wireless sensor network using enhanced black hole algorithm. Sensors 20(8), 2411 (2020) 17. Ng, R.T., Han, J.: CLARANS: a method for clustering objects for spatial data mining. IEEE Trans. Knowl. Data Eng. 14(5), 1003–1016 (2002) 18. Khan, I., Luo, Z.: Nonnegative matrix factorization based consensus for clusterings with a variable number of clusters. IEEE Access 6, 73158–73169 (2018) 19. Likas, A., Vlassis, N.: The global k-means clustering algorithm. Pattern Recogn. 36(2), 451– 461 (2003) 20. Cheng, D., Zhu, Q.: A novel cluster validity index based on local cores. IEEE Trans. Neural Netw. Learn. Syst. 30(4), 985–999 (2018) 21. Karypis, G., Han, E.H.: Chameleon: hierarchical clustering using dynamic modeling. Computer 32(8), 68–75 (1999) 22. Duan, L., Xu, L.: A local-density based spatial clustering algorithm with noise. Inf. Syst. 32(7), 978–986 (2007) 23. Sharma, A., Sharma, A.: KNN-DBSCAN: Using k-nearest neighbor information for parameterfree density-based clustering. In: 2017 International Conference on Intelligent Computing, Instrumentation and Control Technologies (ICICICT), pp.787–792. IEEE, Kerala India (2017) 24. Khan, I., Huang, J.Z.: Incremental density-based ensemble clustering over evolving data streams. Neurocomputing 191, 34–43 (2016) 25. Patel, K.A.: An efficient and scalable density-based clustering algorithm for normalize data. Procedia Comput. Sci. 92, 136–141 (2016)

452

X.-Y. Hu et al.

26. Mehmood, R., Bie, R.: Adaptive cutoff distance: clustering by fast search and find of density peaks. J. Intell. Fuzzy Syst. 31(5), 2619–2628 (2016) 27. Zhao, J., Tang, J.: Density peaks clustering based on circular partition and grid similarity. Concurr. Comput. Pract. Exp. 32(7), e5567 (2020) 28. Fan, T., Yao, Z.: Density peaks clustering based on k-nearest neighbors sharing. Concurr. Comput. Pract. Exp. 33(5), e5993 (2021) 29. Li, C., Zhang Y.: Density peak clustering based on relative density optimization. Math. Probl. Eng. 2020 (2020) 30. Zhao, J., Tang, J.: Improved density peaks clustering based on firefly algorithm. Int. J. BioInspir. Comput. 15(1), 24–42 (2020) 31. Zhao, J., Chen, D.: Multi-strategy ensemble firefly algorithm with equilibrium of convergence and diversity. Appl. Soft Comput. 123, 108938 (2022) 32. Xie, J., Gao, H.: Robust clustering by detecting density peaks and assigning points based on fuzzy weighted K-nearest neighbors. Inf. Sci. 354, 19–40 (2016) 33. Lotfi, A., Moradi, P.: Density peaks clustering based on density backbone and fuzzy neighborhood. Pattern Recogn. 107, 107449 (2020) 34. Yu, D., Liu, G.: Density peaks clustering based on weighted local density sequence and nearest neighbor assignment. IEEE Access 7, 34301–34317 (2019) 35. Du, M., Ding, S.: Density peaks clustering using geodesic distances. Int. J. Mach. Learn. Cybern. 9(8), 1335–1349 (2018) 36. Vinh, N.X., Epps, J.: Information theoretic measures for clusterings comparison: is a correction for chance necessary? In: Proceedings of the 26th annual international conference on machine learning, pp.1073–1080. Association for Computing Machinery, Montreal Quebec Canada (2009). 37. Fowlkes, E.B., Mallows, C.L.: A method for comparing two hierarchical clusterings. J. Am. Stat. Assoc. 78(383), 553–569 (1983)

Chapter 35

Optimizing the Layout of Nucleic Acid Test Sites for COVID-19 Based on Gannet Optimization Algorithm Ruo-Bin Wang , Rui-Bin Hu , Fang-Dong Geng , and Lin Xu

Abstract Optimizing the layout of Nucleic Acid Test Sites (NATS) is a primary issue for the blocking-up of COVID-19. In this paper, we propose an optimization model to address this issue with Gannet Optimization Algorithm (GOA) under multiple constraints. The experimental results show that the GOA-based model performed well in searching optima. It is also revealed that optional solutions with different amounts of NATS can be found, which will provide multiple references for decisionmakers in planning the layout of NATS.

35.1 Introduction COVID-19 is still threatening human life in today’s world. As an efficient measure for epidemic prevention, Nucleic Acid Test is widely employed for the detection of infections before further spreading. Therefore, the layout of Nucleic Acid Test Sites (NATS) is the key issue for quick and convenient detection, especially in areas with big populations and complex shifting. However, it is challenging to complete this task. From the perspective of resource allocation, the layout planning of NATS is an issue of optimization under multiple constraints. Therefore, it is feasible to propose a mathematical model to search the optima for this issue. In recent years, a growing number of complex and realistic optimization problems are solved with meta-heuristic algorithms. On the one hand, improved meta-heuristic algorithms are R.-B. Wang (B) · R.-B. Hu · F.-D. Geng School of Information Science and Technology, North China University of Technology, Beijing 100144, China e-mail: [email protected] R.-B. Wang Beijing Urban Governance Research Center, North China University of Technology, Beijing 100044, China L. Xu (B) University of South Australia, Adelaide 5095, Australia e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_36

453

454

R.-B. Wang et al.

extensively applied for optimization in many engineering problems, such as data mining [1], wireless network node localization [2], and distribution network reactive power optimization [3]. On the other hand, meta-heuristic algorithms are employed for the optimization of social resource allocation issues, such as the optimization of senior citizen centers’ layout of location [4] and the optimization of city traffic network [5]. In this paper, we propose a model for the optimization of NATS’s layout planning. Gannet Optimization Algorithm [6], which is a novel meta-heuristic algorithm proposed recently, is employed to address the issue of NATS’s layout planning. The main contributions of this research are summarized as follows: • A mathematical model is constructed under multiple constraints for the planning of NATS’s layout. • Gannet Optimization Algorithm is employed in the mathematical model to obtain multiple solutions with different amounts of NATS. The organization of this article is as follows: Sect. 35.2 discusses the related works. Section 35.3 introduces the model of NATS. Section 35.4 describes the GOA. Section 35.5 presents the results of the experiment and gives an analysis and discussion. Finally, Sect. 35.6 summarizes the research results.

35.2 Related Work In the past, the solution of optimization problems generally relied on traditional optimization techniques [7, 8]. With the emergence of heuristic algorithms, many optimization problems have employed novel algorithms for improved solutions. Inspired by random phenomena in nature, meta-heuristic algorithms combine stochastic algorithms with local algorithms, which have a certain probability of escaping the local optimum and are more likely to obtain the global optimal solution [9]. Meta-heuristic algorithms are broadly classified into four categories: swarm intelligence algorithms [10], biological evolution-based algorithms, chemistry-based algorithms, and human intelligence-based method heuristics. The Particle Swarm Optimization (PSO) algorithm [11] is one of the most classical swarm intelligence algorithms inspired by the migration and clustering behavior of bird populations during foraging. Although PSO has the advantages of few parameters, fast convergence and easy implementation, it has poor local search ability and low search accuracy. Similarly, the Ant Colony Optimization (ACO) algorithm [12] was inspired by the foraging behavior of ants. The ant colony algorithm was first used to solve the TSP problem and has shown great advantages because of its distributed nature, robustness, and easy integration with other algorithms, but it also has the disadvantages of slow convergence and the tendency to trap in local optimality. Gannet optimization algorithm is a swarm intelligence algorithm inspired by the predatory behavior of Gannets. The imitation of Gannet’s U-shaped and V-shaped diving methods makes it fast and effective in exploring space, and the mechanism of sudden rotation and random

35 Optimizing the Layout of Nucleic Acid Test Sites for COVID-19 Based …

455

wandering greatly enhances exploitation capabilities. GOA performs well on global search with a good capability of escaping local optimization. Therefore, we choose GOA as the primary algorithm to solve the issue of NATS’s layout optimization.

35.3 Model of NATS’s Optimization Taking an urban region as an example, we need to collect the information, such as amounts of residential communities and their corresponding locations and populations in a certain area. For the convenience of people’s routine life, a 15-min walking distance for adults in a given area is defined as the acceptable distance. For NATS’s layout issues, the constraints are as follows: 1. Each residential community is served by at least one NATS. 2. Guarantee that each NATS in the solution cannot exceed its service capacity C. 3. The distance between NATS and the residential community being served must be equal to or less than the acceptable distance. 4. The solution should minimize the sum of the distances from all communities to the corresponding NATS. Based on the above definitions, the objective function is as follows: min F =



ωi di j Z i j

(35.1)

i∈N j∈Mi

The constraint is expressed as follows: ⎧  Z i j = 1, i ∈ N ⎪ ⎪ ⎪ j∈Mi ⎪ ⎪ ⎪ ⎪ Z i j ≤ h j , i ∈ N , j ∈ Mi ⎪ ⎪ ⎪ ⎪  hj = p ⎪ ⎪ ⎪ ⎪ j∈Mi ⎨ Z i j , h j ∈ {0, 1}, i ∈ N , j ∈ Mi ⎪ ⎪ d ij ≤ s ⎪ ⎪ ⎪ ⎪ s = vt ⎪ ⎪  ⎪ ⎪ ⎪ ωi Z i j , j ∈ Mi C j = ⎪ ⎪ ⎪ j∈Mi ⎪ ⎩ Ct ≤ C

(35.2)

where N = {1, 2, 3 . . . , n} is the collection of residential communities, Mi is the set of NATS whose distance to the residential community is less than s, ωi denotes the population of the corresponding community, and di j denotes the distance from the community i to the nearest NATS. Z i j is a 0–1 variable indicating the residential community’s service allocation relationship with NATS. When it is 1, it means that the residential community i is served by NATS j, otherwise it is 0. For hj , it is a

456

R.-B. Wang et al.

0–1 variable, where 1 indicates that point j is selected for Nucleic Acid Testing, otherwise it is 0. Where s is the comfort distance mentioned above, defined as the result of multiplying the walking speed of an average adult with 15 min C j denotes the population served by NATS j and C denotes the maximum population served by a single NATS. The objective function and constraints form a NATS layout optimization model that uses the shortest distance principle to plan the location of NATS within a certain region.

35.4 Gannet Optimization Algorithm The four predation strategies on GOA are separated into two phases: exploration and exploitation. Where the exploration phase consists of U-shaped dive and Vshaped dive. Likewise, the exploitation phase contains sudden rotation and random wandering. Initialization. A random set of solutions is generated starting from this as in (35.3), where N denotes the number of populations and Dim denotes the dimensionality of the solution, and x i is the position of the ith individual. Each x i,j is updated from (35.4). ⎡

xi, j

x1,1 . . . x1, j . . . x1, Dim - 1 ⎢ x ⎢ 2,1 . . . x2, j . . . x2, Dim - 1 ⎢ . .. .. .. .. ⎢ .. . . . . ⎢ ⎢ X = ⎢ . . . . . . xi, j . . . ... ⎢ . .. .. .. .. ⎢ . ⎢ . . . . . ⎢ ⎣ x N −1,1 . . . x N −1, j . . . x N −1, Dim - 1 x N ,1 . . . x N , j . . . x N , Dim - 1   = r1 × U B j − L B j + L B, i = 1, 2, . . . , N , j

x1, Dim x2, Dim .. . ... .. .

x N −1, Dim x N , Dim

⎤ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦

= 1, 2, . . . , Dim

(35.3)

(35.4)

where UBj is the upper bound and LBj is the lower bound on the given problem in dimension j, and r 1 is a random number between 0 and 1. Exploration. Gannets usually look for prey from the air and determine how far the prey is under the water, using the u-dive mode if the prey is deeper in the water, otherwise the v-dive mode is used, as indicated by the following equation: t =1−

It Tmax_ iter

(35.5)

35 Optimizing the Layout of Nucleic Acid Test Sites for COVID-19 Based …

 V (x) =

− πx + 1, x ∈ (0, π ) x − 1, x ∈ (π, 2π ) π

457

(35.6)

where Tmax_iter denotes the maximum number of iterations for a population of individuals, I t indicates the number of iterations of individuals in the current population, and r 2 and r 3 are both random numbers between 0 and 1. The location of individuals in the population is updated as follows:  M X i (t + 1) =

X i (t) + u1 + u2, q ≥ 0.5 X i (t) + v1 + v2, q < 0.5

(35.7)

u2 = (2 ∗ r4 − 1) ∗ 2 ∗ cos(2 ∗ π ∗ r2 ) ∗ t ∗ (X i (t) − X r (t))

(35.8)

v2 = (2 ∗ r5 − 1) ∗ 2 ∗ V(2 ∗ π ∗ r3 ) ∗ t ∗ (X i (t) − X m (t))

(35.9)

where r 4 and r 5 are two different random numbers in the interval [0, 1], u 1 and v1 also correspond to two different random numbers in the interval [−a, a] and [−b, b], X m (t) is the mean value of the location of individuals, X i (t) is the ith individual, and X r (t) is a random individual in the current population. X m (t) =

N 1  X i (t) N i=1

(35.10)

Exploitation. When the Gannet enters the water to hunt, the prey will usually turn suddenly and try to escape, and the Gannet will use a lot of energy to turn suddenly and pursue. Over time, the Gannet’s own energy decreases and when it reaches a certain point it cannot continue to hunt and turns away in search of its next target. The whole process is shown in the following equation: Capturability = t2 = 1 + R=  M X i (t + 1) =

1 R ∗ t2

It Tmax_ iter

M ∗ vel2 0.2 + (2 − 0.2) ∗ r6

(35.11) (35.12)

(35.13)

t × delta × (X i (t) − X Best (t)) + X i (t), Capturability ≥ c X Best (t) − (X i (t) − X Best (t)) × P × t, Capturability < c (35.14)

delta = Capturability × |X i (t) − X Best (t)|

(35.15)

458

R.-B. Wang et al.

P = Levy(Dim) = 0.01 × ⎛



(1 + β) × sin ⎜ σ =⎝   ×β ×2  1+β 2

μ×σ 1

|v| β

(35.16)

1  ⎞β

πβ 2 ⎟  ⎠ β−1 2

(35.17)

where M and vel are the weight of the gannet and the velocity of its movement in water and r 6 is a random number in the interval [−1, 1]. The remaining letters represent some fixed coefficients in the algorithm, the exact values of which are determined experimentally.

35.5 Experiments 35.5.1 Experimental Data Pre-processing The experiment was set within a square area of three square kilometers, on which a fixed number of 10 residential communities were randomly generated. Meanwhile, the population of residents in each community is a random number in the interval [500, 1500]. The coordinates and numbers of people in each residential community are shown in Table 35.1. The residential community distribution map is shown in Fig. 35.1. As mentioned in (35.2), the computational model constructed is a multiconstrained non-linear minimum value optimization model, setting the population size of the GOA to 30 and the number of iterations to 300. NATS is set within the Table 35.1 Number of people and coordinates of each residential community

Residential area number

Number of people (unit: person)

Coordinates (unit: m)

1

1213

(321.7, 381.9)

2

814

(736.4, 1981.0)

3

1119

(1943.2, 1232.5)

4

809

(952.2, 2858.9)

5

782

(2194.6, 1704.6)

6

1445

(1850.7, 1145.5)

7

1193

(2636.7, 1444.4)

8

554

(1489.4, 421.2)

9

998

(2561.0, 2695.4)

10

1496

(778.0, 1159.4)

35 Optimizing the Layout of Nucleic Acid Test Sites for COVID-19 Based …

459

Fig. 35.1 Map of residential communities

square area described above and the service range is an acceptable distance. An acceptable distance of 900 m is calculated with an adult walking speed of 1 m/s. Each NATS has the same set-up cost and can serve a maximum of 3000 people. The pseudocode for GOA-based optimization of nucleic acid test site layout is shown in Table 35.2.

35.5.2 Simulation Experiments According to 5.1, four sets of experiments were implemented with the amount of NATS as the variable, and the optimization schemes are shown in Table 35.3 and the corresponding NATS’s layout diagram is shown in Fig. 35.2. The experimental results show that the actual demand is not met when the amount of NATS is less than 5, so no solution is found. At least five NATS are required to meet the nucleic acid testing needs of the population in a given area. Since the objective function measures the sum of the distances of all residents to the corresponding NATS, a smaller value represents a more convenient nucleic acid testing for the residents. As the amount of NATS increases, the value of the objective function decreases, which indicates improved convenience. However, more NATS mean more investments. Decision-makers should balance convenience and financial budget to choose the appropriate solution according to the actual situation.

460

R.-B. Wang et al.

Table 35.2 Pseudocode for nucleic acid test site layout optimization

Pseudocode for GOA-based optimization of nucleic acid test site layout Input: N: Population size is 30; : 300; Dim: problem dimension is double the number of NATS; LB: Set the lower bound of the NATS coordinate to 0; UB: Set upper bound of NATS coordinates to 3000. Output: the location of NATSs and its fitness value; 1: Generate the initial population from (1) and calculate the fitness. 2: 3: 4:

for It 0.5 then for MXi do

5:

if rand > 0.5 then

6:

update position from (7a);

7:

else

8:

update position from (7b);

9: 10: 11: 12:

end if end for Else for MXi do if capturability > c then

13: 14:

update position from (14a);

15:

else

16:

update position from (14b);

17:

end if

18:

end for

19:

for MXi do

20:

calculating the fitness based on the current position;

21:

If fitness is the current minimum, record it and its position.

22: 23:

end for end for

35.6 Conclusions Through an in-depth investigation and analysis of the NATS’s layout problem, this paper proposes an optimization model and then uses GOA to solve this issue. The experimental results show that feasible solutions to the NATS’s layout problem with multiple constraints can be obtained with the GOA-based optimization model

35 Optimizing the Layout of Nucleic Acid Test Sites for COVID-19 Based …

461

Table 35.3 Experimental results Number

Number of NATS

Optimal coordinate

Fitness value

1

4

No solution

Inf

2

5

(931.72, 2766.57)

3.19e+06

(2442.56, 1803.23) (867.16, 1071.44) (321.70, 381.90) (1913.47, 1215.03) 3

6

(778.00, 1159.50)

2.07e+06

(952.20, 2858.90) (592.64, 391.60) (2561.00, 2695.40) (2636.70, 1444.40) (1918.27, 1209.32) 4

7

(2636.70, 1444.40)

1.21e+06

(1850.70, 1145.50) (1489.40, 421.20) (778.00, 1159.50) (2561.00, 2695.40) (952.20, 2858.90) (321.70, 381.90)

Fig. 35.2 NATS layout diagram

described above. Meanwhile, the corresponding feasible solutions can be obtained separately according to the amount of NATS, which makes it easier for public decision-makers to choose the appropriate scheme to perform nucleic acid test. In the future, we will consider more complex and realistic constraints to improve the NATS’s layout optimization model. The improvements on GOA are also considered for a better performance of the model.

462

R.-B. Wang et al.

References 1. Wu, T.Y., Lin, J.C.W., Zhang, Y., Chen, C.H.: A grid-based swarm intelligence algorithm for privacy-preserving data mining. Appl. Sci. 9(4), 774 (2019) 2. Wang, R.B., Wang, W.F., Xu, L., Pan, J.S., Chu, S.C.: Improved DV-Hop based on parallel and compact whale optimization algorithm for localization in wireless sensor networks. Wirel. Netw., 1–18 (2022) 3. Kong3d, L., Pan, J.S.: Parallel Seagull Optimization Algorithm for Application in Distribution Network Reactive power optimization. J. Netw. Intell. 7(2), 466–479 (2022) 4. Wang, W.F., Wang, R.B., Yin, S., An, Z.W., Xu, L.: Location optimization of service centers for seniors based on an improved particle swarm optimization algorithm. In: Advances in Smart Vehicular Technology, Transportation, Communication and Applications, pp. 249–256. Springer, Singapore (2022) 5. Zhang, F., Wu, T.Y., Wang, Y., Xiong, R., Ding, G., Mei, P., Liu, L.: Application of quantum genetic optimization of LVQ neural network in smart city traffic network prediction. IEEE Access 8, 104555–104564 (2020) 6. Pan, J.S., Zhang, L.G., Wang, R.B., Snášel, V., Chu, S.C.: Gannet optimization algorithm: a new metaheuristic algorithm for solving engineering optimization problems. Math. Comput. Simul. 202, 343–373 (2022) 7. Smith, S.T.: Optimization techniques on Riemannian manifolds. Fields Inst. Commun. 3(3), 113–135 (1994) 8. Kolda, T.G., Lewis, R.M., Torczon, V.: Optimization by direct search: new perspectives on some classical and modern methods. SIAM Rev. 45(3), 385–482 (2003) 9. Abdel-Basset, M., Abdel-Fatah, L., Sangaiah, A.K.: Metaheuristic algorithms: a comprehensive review. Comput. Intell. Multimed. Big Data Cloud Eng. Appl., 185–231 (2018) 10. Beheshti, Z., Shamsuddin, S.M.H.: A review of population-based meta-heuristic algorithms. Int. J. Adv. Soft Comput. Appl. 5(1), 1–35 (2013) 11. Kennedy, J., Eberhart, R.: Particle swarm optimization. In: Proceedings of ICNN’95International Conference on Neural Networks, vol. 4, pp. 1942–1948. IEEE (1995) 12. Dorigo, M., Maniezzo, V., Colorni, A.: Ant system: optimization by a colony of cooperating agents. IEEE Trans. Syst. Man Cybern. Part B (Cybernetics) 26(1), 29–41 (1996)

Chapter 36

Two Factors that Influence Our Selection of Digital Avatars: Gender Performativity and Historical Culture Shutang Liu, Yongyu Li, Minggui Li, Yin Guan, Hang Jiang, and Lei Jiang

Abstract In this study, 74 college student volunteers were recruited as experimental participants, and 108 game 3D images of the Arena of Valor game were used as the selected samples. The digital avatar is presented to the subject through AR technology. Then, the participants were asked to choose five digital avatar that they would probably go on to use in the future metaverse. Finally, after excluding the effects of game familiarity and other socio-statistical variables, it was found that the gender of the selected Digital avatar was positively correlated with the actual participants’ gender, and it was further found that females were more gender specific and more aggregated in terms of Gender Expression than males, while males were less gender specific and more dispersed. This study found that although there was certain influence of historical culture on the selection aspect of the digital avatar, it was

S. Liu · L. Jiang (B) Faculty of Humanities and Arts, Macau University of Science and Technology, 999078 Macau, China e-mail: [email protected] S. Liu e-mail: [email protected] S. Liu Academic Affairs Office, Minjiang University, Fuzhou, China Y. Li · Y. Guan College of Computer and Control Engineering, Minjiang University, Fuzhou, China e-mail: [email protected] Y. Guan e-mail: [email protected] M. Li College of Mathematics and Data Science, Minjiang University, Fuzhou, China e-mail: [email protected] H. Jiang Culture and Creation Department, Fuzhou Polytechnic, Fuzhou, China e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_37

463

464

S. Liu et al.

not significant. Local game characters and Japanese game characters have equally important influence.

36.1 Introduction In January 2022, Microsoft Corporation announced the acquisition of Activision Blizzard, Inc, in which Microsoft Corporation emphasized the concept of a metaverse. The executive chairman and CEO of Microsoft -Satya Nadella said: “Today, it’s the largest and fastest-growing form of entertainment, and as the digital and physical worlds come together, it will play a critical role in the development of metaverse platforms.” [1] The main reason is that Activision Blizzard, as the most successful game company in the American PC era, has a large number of 3D raw modeling of characters and scenes. As we all know, the uncanny valley effect [2] does not exist only when humans face robots, but also have the same effect when they face the human digital body in VR. Mainstream game companies in the PC era and the mobile Internet era have accumulated a great amount of digital games and popular 3D game characters. In fact, this provides a rich source of choices for people in the metaverse to choose their own digital avatars [3], which provides a predetermined exploration space for us to explore how humans will choose their digital avatars in the metaverse in the future. In the Chinese MOBA game “Arena of Valor”, it also has a large number of character 3D modeling. Based on this hypothesis, this study recruited college student as volunteers to conduct research on how to choose their own digital avatars through AR technology.

36.2 Literature Review 36.2.1 The Use of 5G and the Formation of Metaverse Concept The emergence of 5G has greatly promoted the development of the metaverse. With the technical features of ultra-high speed, ultra-low delay and super-large connection, 5G greatly improves users’ perception experience. At the same time, 5G transmits high-definition data, making command response faster. Not only that, but it solves the major problem of secure transmission [4]. Speed and security are essential in the metaverse, and as we build our digital avatars, we certainly want to be able to trust them to keep our digital possessions from being stolen or leaked. The security of transmission is not easy to be attacked by ill-intentioned people, and the improvement of 5G network environment proves the security and effectiveness of the protocol [5]. A mature key management scheme based on 5G network proves that 5G technology

36 Two Factors that Influence Our Selection of Digital Avatars: Gender …

465

has excellent security performance and we can fully rely on it [6]. With excellent sensory experience and super-strong security, 5G lays a foundation for the metaverse.

36.2.2 The Formation of the Concept of Digital Avatar Under the Concept of Metaverse The study of avatars originated from the study of games. In 1998, Michael Mateas first proposed the concept of subjective avatars, arguing that these subjective avatars, in fact, are virtual characters that bring users into the game world, exploring and experiencing stories from an autonomous other perspective [7]. Many people think that the development of interactive computer games such as Super Mario can be considered the beginning of a metaverse philosophy. But when the concept of the metaverse was proposed, people rethought the concept of the digital body, and found that the convergence of new technologies such as VR, AR, and 5G, and the growth of digital capabilities changed interactive games, so that the user’s perception could be closer to the real elements of the physical world [8]. However, in special fields such as medicine, aerospace, military, etc., the digital avatar of a person that resembles one’s real appearance is more valuable [9], in other areas are usually not like this, such as social, gaming, entertainment, and experience. Ukrainian scholar Kostenko. O. V points out that today, digital avatars rarely replicate typical appearance or human behavior, and do not reproduce the appearance of the true owner of the avatar or its users. In most cases, electronic avatars are fictional generalized or idealized images of an impersonal person or fictional fantasy character [8].

36.2.3 Select of Digital Avatars The select of digital avatars is first and foremost based on functional purposes. In the old video game era, the choice of digital avatar was strictly limited, and it was all contextual. You only use this digital avatar when you enter the game. According to the imagination of the mobile Internet media era, the digital avatar needs of people based on digital space social scenes, work scenes, leisure and entertainment scenes in the future meta-universe era are long-term. Some customized digital avatars even become digital possessions, as solely for their own use [10]. This gives users a certain right to choose their own digital avatars. This article is based on the select of human digital avatars from a personal perspective, not a functional perspective. Digital avatars and the performative nature of gender For natural persons, gender is the first attribute of natural persons. It is the first consideration to be relate to in the study of human select in the digital body. The famous Queer theorist Judith Butler gender theory provides us with an important

466

S. Liu et al.

research frame. In her view, the subject’s gender identity is not established and fixed, but uncertain and unstable, that is, performative [11]. That is to say, people have the instinct of gender performance. In recent years, the most researched in this field has been in the field of cosplay for young people. In recent years, cosplay has become popular among young people, and people can play as their favorite characters, which come from fictional characters such as comics, anime, and games. Cosplay not only has role playing but also gender play, and gender play in cosplay is like a work of art. In the Cosplay circle, In the cosplay circle, there are more successful cases of playing non-own gender roles to gain fans. Transgender performances are more likely to achieve artistic effects and attract the attention of the audience due to their novelty. Of course, in cosplay, more information is obtained from makeup and digital beautification of human face features and clothing whether it is a male or female. In the metaverse, without the influence of the natural human body, human beings can freely choose digital images that are not of their natural sex, or even genderless for perform exercises. More importantly, in the metaverse, due to the processability of sound information and the anonymity, this kind of performance is more stealthy and less detectable. Therefore, many researchers believe that gender equality can be achieved in the metaverse [12]. Based on this we propose the following assumptions: H1: People ‘s choosing of digital avatar influenced by their natural gender. Men are more inclined to choose the digital avatar with male characteristics, and women are more inclined to choose the digital avatar with female gender characteristics. H2: Due to the existence of gender performance, there will be transgender selection of digital avatars when multiple avatars can be selected. Digital avatars and the influence of social history and culture In the past game research, more and more research prove that there is a relationship between games and players’ historical cognition. The book “Playing with the Past: Digital Games and Historical Simulations, edited by Kapell and Elliott, includes more than 20 papers that combine multiple game cases to explore topics such as “history as a process”, "history written by the West", “user-generated history”, “the politics of reproduction” and “historical perspectives from the end of the world” [13]. Adam Chapman’s monograph further explores “digital games as history”, distinguishes between games as historical representations, historical action systems, and historical forms of existence, and proposes five ways of analyzing historical games: analog form and epistemology, time, space, narrative, and affordance, thus making the foundation of interdisciplinary research in games and history more solid [14]. Liu Mengfei’s case study of the revival of the British Governor’s Idydrianism from “discovery” to “being invented” to “being played” convincingly shows that digital games can not only reproduce history, but even participate in the shaping of history; And “the initiative and practice of the player is perhaps the most advantageous, yet the most undervalued, tool for shaping our current historical understanding and cultural identity.” In the metaverse, the select of digital avatar will certainly be influenced by history and culture. However, unlike the game world, we estimate that the select of digitized avatar in the future metaverse may be largely due to the personality and image

36 Two Factors that Influence Our Selection of Digital Avatars: Gender …

467

characteristics of a certain historical figure, which are representative and consistent with the characteristics of the selector himself, or the desired characteristics. Based on this we make the following assumptions. H3: Certain historical figures with distinctive characteristics will be more popular. H4: Newly produced persona digital figures, due to lack of understanding, is less likely to be chosen.

36.3 Method VR is a virtual scene, which has a high accuracy in the recognition of human movements [15]. AR, on the other hand, is a real scene that can enhance reality [16]. VR is more suitable for the scene in the meta-universe, but due to limited experimental resources, it does not need a high sense of experience, so this study chose AR instead of VR. Either the questionnaire survey method based on human as the research object or the experimental method based on stimulus response are all set on the real-world space [17], The digital avatar itself exists in the digital virtual space, hence on the one hand, this research allows research participants to watch digital avatar images through AR experimental equipment. Then, according to their own ideas, they can choose the digital avatar.

36.3.1 Selection of Digital Images As mentioned in the introduction part of the article, “Arena of Valor”, the most popular game in mainland China, has the most digital 3D images. This experiment takes the 108 digital characters from the game where players can choose as the selected objects.

36.3.2 Selection of Study Participants In this study, the research participants were collected from college students in a certain university, and the principle of voluntary registration was adopted. We put the experiment equipment in the registration area. Participants who are interested in registration can try it out within 5 min. The process of this experiment itself is interesting enough, there were no extra gifts offered to participants. After a week, we recruited 74 study participants, among whom, 48 were female and 26 were male. The specific sample distribution will be further described in the subsequent part of the article.

468

S. Liu et al.

36.4 Experiment Before conducting the digital image experiment, we first conducted a pre-test to determine the specific personal information of study participants, to facilitate subsequent analysis. Both general social statistics and information related to this experiment were included. The participants then viewed digital avatar images with AR devices and selected the digital avatars they preferred.

36.4.1 Pre-test Data of the Experimenter’s Situation We publish the basic information of the relevant participants as follows: The minimum age of the participants is 18-year-old, the oldest is 32-year-old, and the mean age of all is 21-year-old. The mode is age 20. There are 16 students at the age of twenty, 68 participants at the age from 18- to 23-year-old, 3 students at the age of 24, 1 student at the age of, 1 student at the age of 28, and 1 student at the age of 32. The data has showed that the sample is mainly undergraduate students, with a small number of graduate students participating. Considering that familiarity with the game may affect their selection of digital avatars, the research team conducted a pre-test on the participants’ familiarity with “Arena of Valor”. The test results are as follows: Among them, 27 students are very familiar with the game and playing regularly, accounting for 36.5%. Followed by those who know how to play, but play occasionally due to busy work and study, a total of 25 students, accounting for 33.8%. In addition, 16.2% of participants, which means 12 students, are not players of this game but have been invited by others to play a few times. At last, about 13.5% of participants, 10 students know about the game but never try before.

36.4.2 Experiment Procedure We divided 74 participants randomly, into 9 groups. Group 1 to Group 8, with nine participants in each group and two participants in Group 9. Each subject was given 20 min to perform the viewing of the 108 digital images AR presented and to complete the selection of digital images within the time limit. Since we only had one set of experimental equipment, we needed four and a half days to complete the experiment. The specific experimental groups and time schedule are listed below (Table 36.1). In addition, we also provide here a brief description of the technical implementation of the experiment. The first thing that took place was the presentation of the introductory words. We set the lead words as follows: “In 2022, Microsoft Corporation acquisition of Activision Blizzard, Inc. was commented by the industry as an important move for Microsoft Corporation to deploy the metaverse. The main reason

36 Two Factors that Influence Our Selection of Digital Avatars: Gender …

469

is that Activision Blizzard, as the most successful game company in the American PC era, has a large number of 3D original modeling of characters and scenes. In contrast, the MOBA game “Arena of Valor” in my country also has a large number of 3D modeling of characters. Based on this hypothesis, this study investigates the tendency of humans to choose their own digital avatar in the metaverse.” Let the subject understand the meaning of what he or she is about to do. Since the “Arena of Valor” game comes with the AR Camera function, it can realize the presentation of three-dimensional digital images. But phone or tablet screen resolution is limited, so we don’t think this is enough to generate enough experimental stimuli. Therefore, we use the tablet projection method to project the screen on the high-resolution screen to present the AR digital image. The subject’s manipulation is then facilitated by a remote sensing mouse device. We asked participants to finally choose five digital avatars of their own preference, and in the future, these digital avatars will be used as their own digital avatars in the metaverse, and we recorded the selection process of all participants in detail. The experiment roadmap is shown in Fig. 36.1.

36.5 Result 36.5.1 Elimination of Interfering Factors For academic background and familiarity with the game, we tested the correlation between the results of these two options and the select of the final digital image. The results show that there is no correlation between the two variables and the final digital avatar choosing. Therefore, the functions of these two items can be excluded, and will not be described in detail here.

36.5.2 Calculation and Influence of Gender Factors We tested the gender factor on the aspect of digital image selection, we first created a dummy variable. The variable is named “Gender expression value of the selected digital image”. Based on Risman’s character role theory [18], we believe that most people will choose gender expression that conforms to social norms, that is, they will choose a digital avatar 3D image that is consistent with their actual gender for Gender expression. We coded the biological sex of the study participants as: +1 for males and −1 for females. The gender roles of the selected digital images are coded as: + 1 for males, 0 for animals or robots whose gender cannot be determined, and –1 for females. The coding diagram samples are shown in Table 36.2. Finally, we get the following results by comparing birth gender with gender expression: Both gender and gender expression passed the 2-tailed Test, Pearson

470

S. Liu et al.

Table 36.1 Experimental grouping and time arrangement First day a.m

Second day Third day

Fourth day

Fifth day

Total

9 (group 1) 9 (group 3) 9 (group 5) 9 (group 7) 1 (group 9) 37

p.m

9 (group 2) 9 (group 4) 9 (group 6) 9 (group 8) 1 (group 9) 37

Time(minute)

360

Number of participants 18

360

360

360

40

1480

18

18

18

2

74

Table 36.2 Digital image gender role coding sample Sample

Gender Figure 1

Sample1

+1 male

Figure 2

Figure 3

Figure 4

Figure 5

Gender expression 5

(+1)

(+1)

(+1)

(+1)

(+1)

Sample2 – 1 female

– 4

(−1)

(−1)

(−1)

(−1)

(0)

Fig. 36.1 Experiment roadmap

Correlations is 0.615, Correlation is significant at the 0.01 level (2-tailed). The SPSS results are as follows Table 36.3. Since the Pearson coefficient is positive, we can conclude that H1 is validated. To further clarify H2, we also conducted a comparative analysis of the selection of digital images of men and women. It was found that 5 out of 48 female participants chose to have a Gender Expression via digital avatar as male, or approximately 10.4% of the total, and 7 out of 26 males chose to have a female Gender Expression via digital

36 Two Factors that Influence Our Selection of Digital Avatars: Gender …

471

Fig. 36.2 Histogram of digital avatars chosen by male (at list twice)

Fig. 36.3 Histogram of digital avatars chosen by female (at list twice) Table 36.3 Result of Pearson correlation coefficient between gender and gender expression

Gender

Pearson correlation

Gender

Gender expression

1

0.615**

Sig. (2-tailed) Gender expression

0.000

N

74

74

Pearson Correlation

0.615**

1

Sig. (2-tailed)

0.000

N

74

**Correlation is significant at the 0.01 level (2-tailed)

74

472

S. Liu et al.

Table 36.4 Descriptive Statistics comparation between male and female N

Minimum

Maximum

Mean

Std. Deviation

Gender expression (female)

48

−5.00

3.00

−2.5000

2.04211

Gender expression (male)

26

−5.00

5.00

1.5769

3.25175

avatar, accounting for 26.9% of the total. Not only did this prove the H2 hypothesis, but the research team immediately realized that men’s transgender gender expression was different from women’s in deviation degree. So, the research team went further. Statistical comparative analysis was carried out on the selection of digital avatar 3D images of men and women, as shown in Table 36.4. Compare the mean and standard deviation of the Gender Expression for the digital images chosen by men and women. We can preliminarily draw a conclusion that it seems that the numerical measurement results of Gender Expression in women can see that the absolute value of Gender Expression of women is 2.5, which is bigger than the mean absolute value of men 1.5769. And the standard deviation is also smaller than that of men. In terms of the Gender Expression of digital avatars, women reflect more of their own gender temperament and are more aggregated than men, while men are the opposite.

36.5.3 The Historical Impact of Digital Avatar Selection To investigate this aspect, the team initially studied at just how often these avatars were chosen. It was found that among the digital images that were widely selected, female images were the majority. As H1 demonstrated, we realized that this was due to the gender imbalance in the sample itself. So, we separated the digital images chosen by male and female. The first thing we need to explain here is the translation of character names. Due to the cross-cultural reasons, the international version of the “Arena of Valor” has made relevant changes to the characters’ names. As a result, some names that originated from Chinese historical figures have been changed into names that are easily accepted by the West. For such characters, the article still retains their names in Chinese Pinying for the sake of scientific explanation, and also marks that the characters come from real history. The histogram of the frequency of digital avatars of characters which was selected more than twice, as the Figs. 36.2, 36.3 show. Tables 36.5 and 36.6 for attribute analysis. From the above analysis, it is not difficult to see that although historical figures such as Lu Bu, Xi Shi, Da Qiao, Ma Chao, and Cao Cao are popular, but the above figures are not chosen more frequently than the characters from mythology, characters from local Chinese games, and characters from Japanese games. Characters in local games and characters in Japanese games are also popular. So H3 is not supported. The popularity of the characters in local games and the characters in Japanese games shows that H4 is rejected.

36 Two Factors that Influence Our Selection of Digital Avatars: Gender …

473

Table 36.5 Attributes and frequency distribution of male selected digital avatars Chinese name

English name

Attribute

Selected frequency

Valid percentage (%)

澜

Lan

Image

Game character figure (AOV Co.)

9

6.92

西施

Xi Shi

Historical figure

8

6.15

吕布

Lv Bu

Historical figure

8

6.15

沈梦溪

Shen Mengxi

Game character figure (AOV Co.)

7

5.38

马超

Ma Chao

Historical figure

7

5.38

大乔

Da Qiao

Historical figure

6

4.62

橘右京

Ukyo Tachibana

Game character figure (SNK Co.)

6

4.62

不知火舞

Mai Shiranui

Game character figure (SNK Co.)

5

3.85

曹操

Cao Cao

Historical figure

4

3.07

(continued)

474

S. Liu et al.

Table 36.5 (continued) Chinese name

English name

艾琳

Irene

Image

Attribute

Selected frequency

Valid percentage (%)

Game character figure (AOV Co.)

3

2.31

Table 36.6 Attributes and frequency distribution of women’s selected digital avatars Chinese name

English name

Attribute

Selected frequency

Valid percentage (%)

西施

Xi Shi

Image

Historical figure

27

11.25

嫦娥

Chang’e

Mythical character figure

22

9.17

大乔

Da Qiao

Historical figure

17

7.08

艾琳

Irene

Game character figure (AOV Co.)

15

6.25

梦奇

Munchee

Game character figure (AOV Co.)

11

4.58

不知火舞

Mai Shiranui

Game character figure (SNK Co.)

9

3.75

公孙离

Gongsun Li

Game character figure (AOV Co.)

9

3.75

(continued)

36 Two Factors that Influence Our Selection of Digital Avatars: Gender …

475

Table 36.6 (continued) Chinese name

English name

Attribute

Selected frequency

Valid percentage (%)

沈梦溪

Shen Mengxi

Image

Game character figure (AOV Co.)

8

3.33

安琪拉

Angela

Game character figure (AOV Co.)

8

3.33

澜

Lan

Game character figure (AOV Co.)

7

2.92

36.6 Conclusion and Future Work This study found that female is more gendered and more aggregated than men in terms of Gender Expression of digital avatars, while men are less gendered and more dispersed, which contributes to the question of gender performativity in the metaverse. As a sample of undergraduates, the select of digital avatars is limited by history, but local games and Japanese games have a relatively large impact, which is different from previous studies. This research is insufficient in terms of historical impact, and more granular and more scientifically rigorous research should be added in the next step. This research explores how people choose their own digital avatars in the future metaverse, and the research results have significance for some enterprise technology research and development of digital avatars in the metaverse. Acknowledgements The paper is supported by the foundation of the Chinese Central Government Guided Local Science and Technology Development Project—“Fujian Mental Health HumanComputer Interaction Technology Research Center” (No.2020L3024), and the Minjiang University Scientific Research Project in 2020— “Research on related ethical issues in higher education reform in the era of weak artificial intelligence—Based on the perspective of Technology Intermediary Theory” (No. MYS20007).

476

S. Liu et al.

References 1. CNBC Homepage.: https://www.cnbc.com/2022/05/07/in-microsofts-activision-deal-a-futureworld-is-at-stake.html. Last Accessed 07 May 2022 2. Mori, M., MacDorman, K.F., Kageki, N.: The uncanny valley. IEEE Robot. Autom. Mag. 19(2), 98–100 (2012) 3. Willumsen, E.C.: Is my avatar my avatar? character autonomy and automated avatar actions in digital games. In: Fassone, R., Bittant, M. (eds.) The digital games research association conference 2018, DiGRA, vol. 1, pp. 1–13. ETC Press, Pittsburgh (2018) 4. Wu, T., Guo, X., Chen, Y., Kumari, S., Chen, C.: Amassing the security: An enhanced authentication protocol for drone communications over 5G networks. Drones 6(1), 10 (2021) 5. Wu, T.Y., Lee, Z., Obaidat, M.S., Kumari, S., Kumar, S., Chen, C.M.: An authenticated key exchange protocol for multi-server architecture in 5G networks. IEEE Access 8, 28096–28108 (2020) 6. Yang, L., Chen, Y. C., Wu, T. Y.: Provably secure client-server key management scheme in 5g networks. Wirel. Commun. Mob. Comput. 2021(Article ID 4083199), (2021) 7. Mateas, M.: Subjective avatars (poster). In: Proceedings of the second international conference on autonomous agents, pp. 461–462. Association for computing machinery, New York (1998) 8. Kostenko, O.V.: Electronic jurisdiction, metaverse, artificial intelligence, digital personality, digital avatar, neural networks: theory, practice, perspective. World Science 1(73), 1–13 (2022) 9. Maniadi, E., Kondylakis, H., Spanakis, E. G., Spanakis, M., Tsiknakis, M., Marias, K., Dong, F.: Designing a digital patient avatar in the context of the My Health Avatar project initiative. In: Nikita, K.S., Fotiadis, D.I. (eds.) 13th IEEE international conference on BioInformatics and BioEngineering, BIBE, vol. 9999, pp. 1–4. IEEE, Chania (2013) 10. Banta, N.M.: Property interests in digital assets: The rise of digital feudalism. Cardozo Law Rev. 38(3), 1099–1156 (2017) 11. Wenjing, L.: Judith Butler’s theory of gender performance. Guangxi Normal University doctoral dissertation, Guangxi (2015) 12. Yang, Y.: The art worlds of gender performance: cosplay, embodiment, and the collective accomplishment of gender. J. Chin. Sociol. 9(1), 1–23 (2022) 13. Kapell, M.W., Elliott, A.B.: Playing with the past: Digital games and the simulation of history. Bloomsbury Publishing, USA (2013) 14. Chapman, A.: Digital games as history: how videogames represent the past and offer access to historical practice. Routledge, New York (2016) 15. Zhang, F., Wu, T.Y., Pan, J.S., Ding, G., Li, Z.: Human motion recognition based on SVM in VR art media interaction environment. HCIS 9(1), 1–15 (2019) 16. Chen, C., Peng, S.L.: Multi Barcode Scanning and Decoding Technology Based on AR Smart Glasses, J. Inf. Hiding Multimed. Signal Process. 12(4), 226–238, (2021) 17. Mengfei, L.: The game into a history: a study of the image of druid, the natural priest. Tsinghua University doctoral dissertation, Beijing (2019) 18. Risman, B.J., Davis, G.: Property interests in digital assets: The rise of digital feudalism. Curr. Sociol. 61(56), 733–755 (2013)

Part IV

Cybersecurity Threats and Innovative Solutions

Chapter 37

Path Planning Method of UAV Cluster Against Forgery Attack Under Differential Boundary Constraint Jianchen Wang, Yanlong Li, Yabin Zhang, Jianjun Wu, Wei Sun, and Xuyang Zhou Abstract With the development of UAV ad hoc network and networking, various attacks in network environment pose the more and more increasing threats to UAV cluster system. To solve this problem, this paper selects a forgery attack as the research object and proposes an ant colony algorithm based on octree to carry out the path planning task of a UAV cluster in a complex environment. Firstly, the attack detection algorithm based on differential constraint and the processing method of converting the attacked UAV into static obstacles are proposed, so that the UAV cluster can detect and process forgery attacks in real time; secondly, on this basis, the UAV cluster defense model is established, and the algorithm based on octree is proposed to improve the efficiency of solving the model. The simulation results show that the Oc-ACO algorithm can solve the problem of UAV cluster encountering forgery attack safely and quickly.

J. Wang · Y. Li · J. Wu Northwestern Polytechnical University, 365Th Research Institute, Xi’An 710072, China e-mail: [email protected] Y. Li e-mail: [email protected] J. Wu e-mail: [email protected] Y. Zhang · W. Sun (B) · X. Zhou School of Aerospace Science and Technology, Xidian University, Xi’An 710071, China e-mail: [email protected] Y. Zhang e-mail: [email protected] X. Zhou e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_38

479

480

J. Wang et al.

37.1 Introduction In the modern battlefield of high technology, using UAV cluster formation [1–3] to perform the task has become one of the hot research directions. With the development of the UAV ad hoc network, various attacks in the network environment pose new threats to the UAV cluster system. Therefore, in the case of forgery attack in complex environment, it is worth further research on how UAV cluster can safely accomplish tasks. At present, the results achieved in the research of protection in complex network environments are mainly in the protection of routing, determination of node information, and management of keys [4–6]. However, these protection means provide better means only at the network layer and are not applicable for UAV clusters, which are mobile bodies performing missions. To address the problem of forgery attacks on UAV clusters in the network environment, this paper first establishes the constraint function of the UAV formation model in the obstacle environment [7, 8], and then optimally solves the UAV formation avoidance and defense model [9, 10] in the network environment by the improved ant colony algorithm [11, 12]. First, the differential model boundary constraint is added so that the UAV cluster detects forgery attacks in real time during the flight [13, 14]. If a forgery attack is detected, the forged nodes and the attacked nodes are converted to static obstacles for processing; if no forgery attack is detected, normal cluster trajectory planning is performed. Secondly, for the formation flight process, the map is too large leading to extremely complicated iterative calculation of ant colony, the octree method [15, 16] is proposed to update the map without traversing the environmental information, and data update for a single node is sufficient, which simplifies the data model and improves the solution rate. Finally, B-sample polynomials [17] are used for path optimization. In summary, this paper proposes an algorithm (Oc-ACO), which inherits the characteristics of the classical ACO such as fast search speed, but also increases the differential model boundary constraints, adds equivalent static obstacles, cluster replanning, and other processing, which can effectively solve the problem of forgery attacks on UAV clusters in complex network environments.

37.2 Establishment and Analysis of Forgery Attack Model 37.2.1 Implementation of Forgery Attack Forgery attack, that is, by issuing false routing to other nodes through malicious nodes, forgery RREP as target nodes, resulting in the target nodes being isolated from the network; and it can publish the wrong data package, which leads to the wrong decision-making of the decision system. The process of the forgery node is shown in Fig. 37.1.

37 Path Planning Method of UAV Cluster Against Forgery Attack Under …

481

Fig. 37.1 The process of forgery node

The specific attack steps of forgery node are as follows: (1) Attack nodes join ad hoc network to receive RREQ information. (2) Any node within the maximum distance allowed by the communication network is selected as the target node to issue false routing information to other nodes in the network. (3) The attack node intercepts the packets requested by other nodes in the network instead of the target node and sends false packets. (4) Other nodes receive the wrong data packet and send it back to the decision system. The decision system continues with the original planning.

37.2.2 Analysis of Formation Optimization Model for Forgery Attack 1. Formation aircraft collision constraints. Namely,   xi − x j  ≥ dsa f e

i, j = 1, 2, ν, N F i = j

(37.1)

In which, xi , x j are the coordinate vectors of U AVi , U AV j , dsa f e is the safe distance between UAVs. 2. Communication constraints between formation aircraft. Namely,   xi − x j  ≤ dmax i, j = 1, 2, . . . , N F i = j

(37.2)

In which, xi , x j are the coordinate vectors of U AVi , U AV j , dmax is the max distance between UAVs. 3. Forgery attack constraints. As shown in Fig. 37.2, the attacked nodes are transformed into equivalent static obstacle model. The equivalent static obstacle model is established as follows:

482

J. Wang et al.

Fig. 37.2 Equivalent state obstacle model

⎧ ⎪ ⎨ x − xi  = Ri x − xi+1  = Ri ⎪ ⎩ Ri = max{dsa f e , ρi }

(37.3)

In which, x, xi , xi+1 are the coordinates of the surface position of the obstacle, the attacked UAV and the forged node, Ri is the radius of the static obstacle, dsa f e is the safe distance between the UAVs, ρi is the step size of the ant colony algorithm.

37.3 Construction of Forgery Attack Model for UAV Formation 37.3.1 Detection and Processing of Forgery Attack Considering that the main information transmitted between nodes is location information, for forgery attack, each node carries out differential constraint on the dynamic information of each node to determine whether there is forgery attack. As shown in Eq. (37.4), if the conditions in equation are satisfied, the UAV cluster is considered not to be attacked by forgery, and transforms the attacked node into a static obstacle; if no, the UAV cluster is considered to be attacked by forgery.   k−1  x − xik  < ρi , i = 1, 2, 3, 4, 5, 1 ≤ k ≤ num i   k x − x k  < Dmax , i, j = 1, 2, 3, 4, 5, i = j i

j

(37.4)

37 Path Planning Method of UAV Cluster Against Forgery Attack Under …

483

In which, xik is the real-time position of the i-th UAV at k time, ρi is the step size of ant colony algorithm at k time, num is the number of iterations of ant colony algorithm, Dmax is the maximum distance between UAV clusters.

37.3.2 The Model of UAV Formation Forgery Attack Defense The goal of this study is that when there is a forgery attack in the network environment, the UAV formation can complete the task safely and reach the command area quickly. Therefore, based on the analysis of the above formation optimization model, the UAV formation flight model is established as follows: min J (x(k), u(k)) =

NF 

Ji (xi (k), u i (k))

i=1

=

NF  i=1

(||xi (k + 1) − xi (k)||2P +

NF 

||xj (k) − xi (k)||2Q + ||U ||2R )

j=1

⎧ ⎪ U = [vi , rs ] ⎪   ⎪ ⎪   ⎪ ⎪ ⎪  xi − xj ≤ dmax i, j = 1, 2, 3 . . . , N F i = j ⎪ ⎨ xi − x j  ≥ dsa f e i, j = 1, 2, 3 . . . , N F i = j s.t. ⎪  k−1  ||xi − rs || ≥ dsa f e ⎪ ⎪ k  ⎪ x − x ⎪ i i  < ρi i = 1, 2, 3, 4, 5, 1 ≤ k ≤ num ⎪ ⎪  ⎪ k k ⎩ xi − x j   < Dmax , i, j = 1, 2, 3, 4, 5, i = j

(37.5)

In which, J (·, ·) is the cost function of the overall formation, Ji (·, ·) is the formation safety cost function of the i-th UAV, rs is the center point coordinate of the obstacle, P, Q, R are the symmetric positive definite matrix, N F is the number of UAVs in the UAV cluster, U is the velocity vector of UAV and the vector set between UAV and obstacle center.

37.4 Formation Route Planning Based on Oc-ACO Algorithm Ant-colony algorithm has been widely concerned because of its strong robustness and fast search speed in path planning. The traditional ant colony algorithm needs to traverse and store all the node information in map conversion, which leads to huge information, large amount of calculation, and slow map update speed. To solve this problem, this paper uses Octree for map conversion. When updating the map, only update the data of nodes in the forgery attack area, without traversing

484

J. Wang et al.

the map information, greatly speeding up the speed of the algorithm. According to the environment information and UAV cluster security constraints, size constraint on leaf node is set, namely, dsa f e ≤ s ≤ 2 ∗ dsa f e

(37.6)

Aiming at the unsmooth node path, uniform B-spline function is used to optimize the path node. B-spline curve of order k–1 can be expressed by the following formula: p(t) =

n 

pi Bi,k (t)

(37.7)

i=0

In which, pi is the control node of time ti , Bi,k (t) is the k–1 B-spline basis function, which can be obtained by De Boor-Cox recursive formula 7 . The improved Oc-ACO algorithm introduces boundary constraints, increases the detection of forgery attack and the processing method of converting the uninjured into equivalent static obstacles, combines the Octree method, and uses B-spline interpolation function to optimize the path. The main process is as follows: Step 1: Set the initialization parameters and the maximum number of iterations; Step 2: The environment model is established, and the Octree is used for map conversion to generate grid block map; Step 3: When the number of iterations is less than the maximum number of iterations, whether there is a forgery attack is detected. If there is a forgery attack, Step 4 is executed. If there is no forgery attack, Step 7 is executed; Step 4: The coordinates of the attacked UAV and the forged nodes are taken as the core to update the equivalent static obstacle model and generate new obstacle constraints; Step 5: Roulette method is used to select the local best point for the next step; Step 6: Calls the real-time ant colony algorithm for path planning; Step 7: Judges whether the best point obtained in Step 5 is the target node. If it is, stop planning and optimize the spline interpolation to get the ideal path. If it is not, go back to Step 3.

37.5 Simulation Results and Analysis In order to verify the effectiveness of the algorithm proposed in this paper, the MATLAB is used for simulation verification. The classic ACO algorithm and the improved Oc-ACO algorithm are selected, respectively, to test the route planning of UAV formation in the case of forgery attack.

37 Path Planning Method of UAV Cluster Against Forgery Attack Under …

485

Table 37.1 Initial parameters of the algorithm in simulation experiment 1 Algorithm name

Heuristic factor

Maximum number of iterations

Pheromone evaporation coefficient …

ACO algorithm

5

2000

0.7

Oc-ACO algorithm

5

2000

0.7

Assuming that the UAV is equivalent to a particle in the air, and the wind speed and other factors that interfere with the UAV speed are ignored. Simulation Experiment 1: In order to verify the effectiveness of the algorithm, the parameters of UAV algorithm in ideal obstacle free scene are shown in Table 37.1. The simulation results are shown in Fig. 37.3. Five UAVs are selected to form a formation, from the starting position to the ending position. Figure 37.3a shows the path planning of ACO algorithm without attack detection in the forgery attack environment. Figure 37.3b shows the track planning of the improved Oc-ACO algorithm in the forgery attack environment. The simulation results show that the improved Oc-ACO algorithm can complete the detection of forgery attack and complete the original planning task, which proves the effectiveness of the algorithm. Through the data comparison in Table 37.2, it can be seen that compared with the traditional algorithm, the time consumption of map updating by Octree is reduced by at least 57%. Thus, the improved algorithm has obvious advantages over the traditional algorithm. Simulation Experiment 2: In order to further prove the effectiveness of the proposed method in complex environment, UAV cluster formation is tested in multi-obstacle scenes. The initial parameters of the algorithm are shown in Table 37.3. The simulation results are shown in Fig. 37.4. Figure 37.4b adds forgery attack on the basis of Fig. 37.4a. The real-time detection is used to determine whether the forgery attack occurs in the network. The UAV cluster formation is used to deal with the forgery attack model, and the formation control is constrained. From the simulation results, it can be seen that UAV cluster can plan obstacle avoidance in the case of forgery attack, and can maintain a relatively safe distance, which verifies the effectiveness of this method.

37.5.1 Conclusion In view of the forgery attack in the network environment, this paper proposes a multi-UAV cluster formation obstacle avoidance method based on Oc-ACO algorithm. Compared with ACO algorithm, Oc-ACO algorithm increases the boundary constraint of the difference model and judges forgery attack in real time, which

486

J. Wang et al.

Fig. 37.3 Comparison of planned tracks before and after improvement of ACO algorithm in obstacle free environment

(a)ACO algorithm route planning

(b)Oc-ACO algorithm route planning Table 37.2 Time comparison table of ant colony algorithm before and after improvement Table 37.3 Initial parameters of the algorithm in simulation experiment 2

Algorithm

ACO algorithm

Oc-ACO algorithm

Time

270.43 s

113.96 s

Heuristic factor

5

Maximum number of iterations

2000

Pheromone evaporation coefficient

0.7

37 Path Planning Method of UAV Cluster Against Forgery Attack Under … Fig. 37.4 Path planning of improved ACO algorithm in multi-obstacle environment

(a) No forgery attack

(b) Encounter forgery attacks

487

488

J. Wang et al.

provides a powerful criterion for path replanning. And the strategy of UAV formation to deal with forgery attack is proposed. By using the method of equivalent static obstacles and the method of cluster replanning, the path planning of UAV cluster is carried out to avoid the collision of UAV cluster. Acknowledgements This paper was supported by the National Natural Science Foundation of China (61671356); Shaanxi Provincial Key R&D Plan (2020GY-136); Shaanxi Key R&D Plan Key Industry Innovation Chain Project (2019ZDLGY14-02-03, 2022ZDLGY03-01); China College Innovation Fund of Production, Education and Research (2021ZYA08004); Xi’an Science and Technology Plan Project (2022JH-RGZN-0039); Graduate Innovation Fund in Xidian University (YJS2217).

References 1. D’Amato, E., Mattei, M., Notaro, I.: Bi-level flight path planning of uav formations with collision avoidance. J. Intell. Robot. Syst., 93(1), (2018) 2. Seo, J., Kim, Y., Kim, S., et al.: Collision avoidance strategies for unmanned aerial vehicles in formation flight. IEEE Trans. Aerosp. Electron. Syst. 53(6), 2718–2734 (2017) 3. Yao, P., Wang, H., Su, Z.: Real-time path planning of unmanned aerial vehicle for target tracking and obstacle avoidance in complex dynamic environment. Aerosp. Sci. Technol. 47, 269–279 (2015) 4. Wu, T.Y., Guo, X.,Chen, Y.C., Kumari, S., Chen, C.M.: Amassing the security: an enhanced authentication protocol for drone communications over 5g networks. Drones, 6(1), 10 (2022) 5. Khan, M.A., Ullah, I., Alsharif, M.H.,Alghtani, A.H., Aly, A.A., Chen, C.H.: An efficient certificate-based aggregate signature scheme for internet of drones. Secur. Commun. Netw. 2022, 9718580 (2022) 6. Liu, S., Chen, C.-M.: Comments on a secure and lightweight drones-access protocol for smart city surveillance. IEEE Trans. Intell. Transp. Syst. (2022). https://doi.org/10.1109/TITS.2022. 3198045 7. Mellinger, D., Kumar, V.: Minimum snap trajectory generation and control for quadrotors//In: IEEE International conference on robotics & automation. IEEE, (2011) 8. Achtelik, M.W., Lynen, S., Weiss, S., et al.: Motion—and Uncertainty—aware path planning for micro aerial vehicles. J. Field Robot., 2014 9. Lin,Y., Saripalli, S.: Sampling-based path planning for UAV collision avoidance. IEEE Trans. Intell. Transp. Syst., 18(11), 3179–3192 (2017) 10. Esfahani, N.R., Khorasani, K.A.: distributed model predictive control (MPC) fault recon— figuration strategy for formation flying satellites. Int. J. Control. 89(5), 1–31 (2015) 11. Lee, M.-G., Yu, K.-M.: dynamic path planning based on an improved ant colony optimization with genetic algorithm. IEEE Asia-Pacific Conference on Antennas and Propagation (APCAP) 2018, 1–2 (2018) 12. Zong, C., Yao, X., Fu, X.: Path planning of mobile robot based on improved ant colony algorithm. In: 2022 IEEE 10th Joint International Information Technology and Artificial Intelligence Conference (ITAIC), pp. 1106–1110 (2022) 13. Nardi, L., Stachniss, C.: Uncertainty-aware path planning for navigation on road networks using augmented MDPs. International Conference on Robotics and Automation (ICRA) 2019, 5780–5786 (2019) 14. Banfi, J., Woo, L., Campbell, M.: Is it worth to reason about uncertainty in occupancy grid maps during path planning? International Conference on Robotics and Automation (ICRA) 2022, 11102–11108 (2022)

37 Path Planning Method of UAV Cluster Against Forgery Attack Under …

489

15. Kothari, M., Postlethwaite, I., Gu, D.W.: A Suboptimal path planning algorithm using rapidlyexploring random trees. Int. J. Aerosp. Innov. 2(1), 93–104 (2010) 16. Jia, F., Lei, Y., et al.: A neural network constructed by deep learning technique and its application to intelligent fault diagnosis of machines. Neurocomputing, (2018) 17. Yang, K., Moon, S., Yoo, S., et al.: Spline-based RRT path planner for non-holonomic robots. J. Intell. & Robot. Syst. 73(1–4), 763–782 (2014)

Chapter 38

To Analyze Security Requirements of Two AKA Protocols in WBAN and VANET Haozhi Wu, Saru Kumari, and Tsu-Yang Wu

Abstract The emergence of the Internet of Things (IoT) is gradually changing people’s lives and is widely used in various fields. Wireless body area networks (WBAN) and vehicular ad-hoc networks (VANET) are typical examples of IoT applications. In both environments, an attacker can tamper, intercept, and eavesdrop on data transmitted in the public channel. Therefore, it is necessary to design authentication and key agreement (AKA) protocols to protect the communication of entities. Recently, Wu et al. proposed a novel AKA protocol in WBAN. Jagriti et al. proposed an anonymous AKA protocol in VANET. In this paper, we analyse of Wu et al.’s protocol and Jagriti et al.’s protocol, respectively. Unfortunately, we find that Wu et al.’s protocol cannot resist sensor node capture attacks, and Jagriti et al.’s protocol cannot resist session key disclosure attacks, on board unit (OBU) capture attacks, and man-in-the-middle attacks. Finally, we propose some suggestions for the two protocols.

38.1 Introduction With the development of wireless communication, sensor and low-power integrated circuit technology, the Internet of Things (IoT) [10, 20, 24] came into being. The application areas of IoT are very wide, such as smart home [26], agriculture [25], healthcare [5, 9, 16], VANET [4, 11], etc. Wireless body area network (WBAN) [1, 6, 22] can easily monitor physiological signals to avoid interrupting patients’ H. Wu · T.-Y. Wu (B) College of Computer Science and Engineering, Shandong University of Science and Technology, Qingdao 266590, China e-mail: [email protected] H. Wu e-mail: [email protected] S. Kumari Department of Mathematics, Chaudhary Charan Singh University, Meerut 250004, Uttar Pradesh, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_39

491

492

H. Wu et al.

everyday life. Vehicular ad-hoc network (VANET) [7, 18, 21] is formed using wireless communication technology with vehicles and transportation facilities as nodes. Using VANET to plan vehicle movement rationally and realize smart transportation can improve the convenience of people’s lives. Because data is transmitted through public channels, attackers can intercept this data to launch several attacks, for example, node capture attacks, known session-specific temporary information attacks, impersonation attacks, insider attacks, lack of user anonymity, untraceability, etc. Therefore, it is crucial to ensure the transmitted data’s privacy and confidentiality. In recent years, WBAN has played an important role in the medical system. In 2017, Li et al. [17] proposed an anonymous mutual authentication and key agreement protocol based on WBAN, and claimed that their protocol can ensure higher security with lower consumption. However, Gupta et al. [13] found that their protocol did not resist to sensor node impersonation attacks and hub node impersonation attacks, and they proposed secure authentication protocol. VANET has emerged as a promising area of smart transportation. In 2019, Jia et al. [15] proposed a three-party AKA protocol based on bilinear pairings. However, Zhang et al. [27] point out that the protocol suffered from known session-specific temporary information attacks (or called ephemeral secret leakage attacks). In the same year, Ma et al. [19] proposed an efficient and provably secure AKA protocol. However, Chen et al. [3] point out that their protocol is vulnerable to known session-specific temporary information attacks. In 2020, Cui et al. [8] proposed an authentication scheme for VANET in a multi-cloud environment. Unfortunately, Zhang et al. [27] point out that the scheme is vulnerable to impersonation attacks and man-in-the-middle attacks. Recently, Wu et al. [23] proposed a new AKA protocol in WBAN. Jagriti et al. [14] proposed an efficient and anonymous AKA protocol in VANET. In this paper, we find that Wu et al.’s protocol can not resist the node capture attacks. In addition, we find that several security problems in Jagriti et al.’s protocol, including session key disclosure attacks, on board unit (OBU) capture attacks, man-in-the-middle attacks. Finally, we suggest improvements to their protocols.

38.2 Review of Protocols In this section, we review Wu et al.’s protocol [23] and Jagriti et al.’s protocol [14]. The notations used in this paper are shown in Table 38.1.

38.2.1 Review Wu et al.’s Protocol Their protocol can be divided into three phases, including “Initialization”, “Registration”, and “Login and Authentication” phases. The initialization and registration phases can be referred to [23]. The login and authentication process is shown in Fig. 38.1, and the specific steps describe as follows.

38 To Analyze Security Requirements of Two AKA Protocols … Table 38.1 Notations

493

Symbol

Definition

I Nj idi id I N tid S N kH N R SU RI D

Intermediate node Identity of S Ni Identity of I N j Temporary identity of S Ni Master key of H N Road side unit Pseudo identity of Vi

SNi

INj

Picks random number r1 Generate timestamp T1 Compute xN = r1 ⊕ aN Compute tidSN = h(idi ⊕ r1 , T1 ) {xN , tidSN , T1 } −−−−−−−−−−−→

HN

{idIN , xN , tidSN , T1 } −−−−−−−−−−−−−−−−→

Check that idIN exisits Check validity of T1 Compute a∗ N = kHN ⊕ kN r1∗ = xN ⊕ a∗ N ∗ idi = bN ⊕ h(kHN , kN ) ∗ ∗ tid∗ = H(id ⊕ r SN i N , T1 ) ?

Compute r ∗ = W1 ⊕ aN H ∗ = h(r ∗ , idi , W1 , W2 , W3 )

{W1 , W2 , W3 , H} ←−−−−−−−−−−−−

Checks tid∗ SN = tidSN Pick two random numbers R and K Compute W1 = R ⊕ a∗ N V1∗ = K ⊕ kH N V2+ = id∗ i ⊕ h(K, k( HN )) W2 = V1∗ ⊕ h(r ∗ , id∗ i) W3 = b∗ N ⊕R H = h(R, id∗ , M , M , 1 2 V3 ) i ∗ Ks = h(r ∗ , id∗ i , b N , xN , r 1 ) {idI N, W1 , W2 , W3 , H} ←−−−−−−−−−−−−−−−−−

?

CheckH ∗ = H Compute V1∗ = W2 ⊕ h(r ∗ , idi )V2+ = r ∗ ⊕ W3 Ks = h(r ∗ , idi , V2+ , xN , r1 ) Replace(V1∗ , V2+ ) with (V1 , V2 ) Store session key Ks

Fig. 38.1 Login and key agreement phase of Wu et al.’s protocol

Step 1: First, the sensor node (S Ni ) picks a random number r1 , and generates a timestamp T1 . Then S Ni calculates x N = r1 ⊕ a N , tid S N = h(idi ⊕ r1 , T1 ). The S Ni sends the message {X n , tid S N , T1 } to I N j . Step 2: I N j adds its identity id I N to the message and sends it to the hub node (H N ). Step 3: First, H N searches its database to find entries that match id I N . If id I N exists, H N continues checking for timestamp T1 . H N calculates a ∗N = k H N ⊕ k N , r1∗ = x N ⊕ a ∗N , idi∗ = b N ⊕ h(k H N , k N ), tid S∗N = h(idi∗ ⊕ r1∗ , T1 ). The authenticity of the message is judged by comparing the calculated tid S∗N with the transmitted tid S∗N . If the message is true, H N picks two random numbers R and K . H N

494

H. Wu et al.

calculates W1 = R ⊕ a ∗N , V1∗ = K ⊕ k H N , V2+ = idi∗ ⊕ h(K , k H N ), W2 = V1∗ ⊕ h(r ∗ , idi∗ ), W3 = b∗N ⊕ R, H = h(R, idi∗ , W1 , W2 , W3 ), K s = h(r ∗ , idi∗ , b∗N , x N , r1 ). The H N sends the message {id I N , W1 , W2 , W3 , H } to I N j . Step 4: After I N j receives the message from H N . I N j sends the message {W1 , W2 , W3 , H } to S Ni . Step 5: When S Ni receives the message, it computes r ∗ = W1 ⊕ a N , H ∗ = h(r ∗ , idi , W1 , W2 , W3 ). The authenticity of the message is determined by checking whether H ∗ is equal to H . Then S Ni computes V1∗ = W2 ⊕ h(r ∗ , idi ), V2+ = r ∗ ⊕ W3 , K s = h(r ∗ , idi , V2+ , x N , r1 ). S Ni is going to replace (V1∗ , b∗N ) with (a, b). Finally, the session key K s is stored.

38.2.2 Review Jagriti et al.’s Protocol Their protocol can be divided into three main phases: “Initialization”, “Registration”, and “Login and authentication”. We mainly introduce the login and authentication phase. After the user successfully login, the O BU and RSU achieve mutual authentication and establish the session key. Step 1: O BU generates a timestamp t1 and a random number R1 . Then, O BU computes V1 = h(Bi ) ⊕ R1 , P I Di = h(R1 ) ⊕ R I D, and V2 = h(R1 ||P I Di ||Di || Ti ||t xi ). The O BU sends the message {P I Di , Bi , V1 , V2 , Di , t xi , Ti } to the RSU over the public channel. Step 2: RSU selects timestamp Ti and expiration time t xi . RSU uses P S K to calculates Ai = Di ⊕ P S K , N1 = V1 ⊕ h(Bi ), R I D = h(R1 ) ⊕ P I Di , M2 = ?

h(N1 ||P I Di ||Di ). RSU checks whether M2 = V2 . If the equation does not hold, the RSU rejects the request. Otherwise, RSU generates a random number R2 , and calculates session key S K i j = h(N1 ||R2 ), V3 = R2 ⊕ h(N1 ), V4 = h(S K i j ||R2 ). The RSU sends the message {V3 , V4 } to the O BU . Step 3: O BU receives the message. O BU calculates R2 = V3 ⊕ h(R1 ), S K i j = h(R1 ||R2 ). O BU checks that the calculated H (S K i j ||R2 ) is equal to the transmitted V4 . If the validation equation does not hold then the session is aborted. Otherwise, O BU calculates the V5 = S K i j ⊕ h(R2 ). The O BU sends message V5 to the RSU . Step 4: After receiving the message, the RSU computes H (N2 ) = S K i j ⊕ V5 . RSU checks that h(N2 ) is equal to H (R2 ). In this way, RSU determines whether S K i j generated by O BU is equal. The authentication process is shown in Fig. 38.2.

38 To Analyze Security Requirements of Two AKA Protocols … V ehicle(Vi )/OBU

495 RSU (RSUi )

Generate timestamp Ti Create nonce Ni V1 = h(Bi ) ⊕ Ni P IDi = h(Ni ) ⊕ RID V2 = h(R1 ||P IDi ||Di ||Ti ||txi )

{P IDi , Bi , V1 , V2 , Di , txi , Ti } −−−−−−−−−−−−−−−−−−−−−→

Check timestamp Ti Check expiration time txi Ai = Di ⊕ P SK RID = h(R1 ) ⊕ P IDi N1 = V1 ⊕ h(Bi ) M2 = h(N1 ||P IDi ||Di ||Ti ||txi ) ?

R2 = V3 ⊕ h(R1 ) SKij = h(R1 ||R2 )

{V3 , V4 } ←−−−−−

Check M2 = V2 Generate R2 SKij = h(N1 ||R2 ) V3 = R2 ⊕ h(N1 ) V4 = h(SKij ||R2 )

?

Check if h(SKij ||R2 ) = V4 V5 = SKij ⊕ h(R2 )

{V5 } −−→

h(N2 ) = SKij ⊕ V5 Check if h(N2 ) = h(R2 )

Fig. 38.2 Authentication phase of Jagriti et al.’s protocol

38.3 Cryptanalysis of Protocols 38.3.1 Attacker Model We use the well-known Dolev-Yao (DY) [12] and Canetti-Krawczyk (CK) [2] threat models to assume that the attacker (A) has the following capabilities. 1. A can obtain or intercept the messages transmitted on the public channel. 2. A can obtain the entity’s long-term key or random number during the establishment of each session. 3. A can obtain sensitive information stored on smart cards or physical devices. 4. A can register and login in as a legitimate user.

38.3.2 Cryptanalysis of Wu et al.’s Protocol In this section, we analyze Wu et al.’s protocol [23], and found that their protocol cannot resist node capture attacks.

496

H. Wu et al. SNi

INj

Pick rN Generate timestamp T1 Compute xN = r1 ⊕ aN Compute tidSN = h(idi ⊕ r1 , T1 ) {xN , tidSN , T1 } −−−−−−−−−−−→

HN

{idIN , xN , tidSN , T1 } −−−−−−−−−−−−−−−−→ Check that idIN exisits Check validity of T1 Computes a∗ N = kHN ⊕ kN r1∗ = xN ⊕ a∗ N id∗ i=

bN ⊕ h(kHN , kN )

∗ SN = H(id∗ i ⊕ r 1 , T1 ) ?

Check tid∗ SN = tidSN Pick two random numbers R and K Compute

W1 = R ⊕ a∗ N

V1∗ = K ⊕ kH N

V2+ = id∗ i ⊕ h(K, k( HN )) W2 = V1∗ ⊕ h(r ∗ , id∗ i) W 3 = b∗ N ⊕R H = h(R, id∗ , V , V , 1 2 V3 ) i

∗ Ks = h(r ∗ , id∗ i , b N , xN , r 1 )

Compute r ∗ = W1 ⊕ aN H ∗ = h(r ∗ , idi , W1 , W2 , W3 )

{W1 , W2 , W3 , H} ←−−−−−−−−−−−−

{idI N, W1 , W2 , W3 , H} ←−−−−−−−−−−−−−−−−−

?

CheckH ∗ = H Compute V1∗ = W2 ⊕ h(r ∗ , idi )V2+ = r ∗ ⊕ W3 Ks = h(r ∗ , idi , V2+ , xN , r1 ) Replace(V1∗ , V2+ ) with (a, b) Store session key Ks

Fig. 38.3 Node capture attacks in [23]

38.3.2.1

Node Capture Attacks

Suppose that A can capture the hub node and has access to the {id I N , k N , b N , k H N } stored in the database. The node capture attacks as shown in Fig. 38.3 and the specific attack steps are described as follows. Step 1: First, A can calculating a ∗N = k H N ⊕ k N , r1∗ = x N ⊕ a ∗N , idi∗ = b N ⊕ h(k H N , k N ). Then, A calculates = W1 ⊕ a ∗N , V1∗ = W2 ⊕ h(R, idi∗ ), K = V1∗ ⊕ k H N , V2+ = idi∗ ⊕ h(K , k H N ). Step 2: After obtaining the above parameters, A calculates the session key K s = h(R, idi∗ , V2+ , x N , r1 ). Therefore, Wu et al.’s protocol cannot resist node capture attacks.

38 To Analyze Security Requirements of Two AKA Protocols …

497

38.3.3 Cryptanalysis of Jagriti et al.’s Protocol In this section, we analyze Jagriti et al.’s protocol [14], and find their protocols cannot resist session key disclosure attacks, OBU capture attacks, and man-in-the-middle attacks. These three attacks as shown in Fig. 38.4.

38.3.3.1

Session Key Disclosure Attacks

Suppose that A can intercept {V1 , Bi , V3 } on the public channel. A can calculate the session key. The specific steps are as follows. Step 1: A calculates N1 = V1 ⊕ h(Bi ), R2 = V3 ⊕ h(N1 ). Step 2: After A calculates the random numbers generated by O BU and RSU . A can calculate the session key S K i j = h(N1 ||R2 ). Thus, Jagriti et al.’s protocol cannot resist session key disclosure attacks.

38.3.3.2

OBU Capture Attacks

Suppose that A can capture Bi stored in OBU. The attacker can compute the session key. The specific steps are as follows. A needs to intercept the message V1 , V3 on the public channel. A can figure out R1 = V1 ⊕ h(Bi ), R2 = V3 ⊕ h(R1 ). Finally, A can calculate the session key S K i j = h(R1 ||R2 ). Therefore, Jagriti et al.’s protocol cannot resist OBU capture attacks.

38.3.3.3

Man-in-the-Middle Attacks

In this subsection, we present the man-in-the-middle attacks. The specific steps are described as follows. Step 1: A can intercept messages {P I Di , Bi , V1 , V2 , Di , t xi , Ti }, {V3 , V4 } and {V5 } on the public channel. Then, A calculates R1 = V1 ⊕ h(Bi ), R I D = P I Di ⊕ h(R1 ). Next, A generates random number R1∗ , and calculates V1∗ = h(Bi ) ⊕ R1∗ , P I Di∗ = h(R1∗ ) ⊕ R I D, V2∗ = h(R1∗  P I Di∗  Di  Ti∗  t xi∗ ). Finally, A sends the messages {P I Di∗ , Bi , V1∗ , V2∗ , Di , t xi∗ , Ti∗ } to RSU . Step 2: Upon receiving the messages, RSU first checks the t xi∗ and Ti∗ . Then, RSU compute Ai = Di ⊕ P S K , R I D = h(R1∗ ) ⊕ P I Di∗ , R1∗ = V1∗ ⊕ h(Bi ), V2∗ = h(R1∗  P I Di∗  Di  Ti∗  t xi∗ ), and check the validity of V2∗ . Next, RSU generates R2 , and computes S K i∗j = h(R1∗  R2 ), V3 = R2 ⊕ h(R1∗ ), V4 = h(S K i∗j  R2 ). Finally, RSU sends {V3 , V4 } to O BU . Step 3: A intercepts the messages {V3 , V4 }. A computes R2 = V3 ⊕ h(R1∗ ), ∗ ∗ S K i∗j = h(R1∗  R2 ). Then, A selects R2∗ , and calculates S K i∗∗ j = h(R1  R2 ), V3 =

498

H. Wu et al. V ehicle(Vi )/OBU

RSU (RSUi )

Generate timestamp Ti Create nonce R1 V1 = h(Bi ) ⊕ R1 P IDi = h(R1 ) ⊕ RID V2 = h(R1 ||P IDi ||Di ||Ti ||txi ) {P IDi , Bi , V1 , V2 , Di , txi , Ti } −−−−−−−−−−−−−−−−−−−−−→

Check timestamp Ti Check expiration time txi Ai = Di ⊕ P SK RID = h(R1 ) ⊕ P IDi N1 = V1 ⊕ h(Bi ) M2 = h(N1 ||P IDi ||Di ||Ti ||txi ) ?

Check M2 = V2 Generate R2 SKij = h(N1 ||R2 ) V3 = R2 ⊕ h(N1 ) V4 = h(SKij ||R2 ) {V3 , V4 } ←−−−−−

R2 = V3 ⊕ h(R1 ) SKij = h(R1 ||R2 ) ?

Check if h(SKij ||R2 ) = V4 V5 = SKij ⊕ h(R2 )

{V5 } −−→

h(N2 ) = SKij ⊕ V5 Check if h(N2 ) = h(R2 )

Fig. 38.4 Session key disclosure attacks, OBU capture attacks, and man-in-the-middle attacks in [14]

∗ ∗ ∗ R2∗ ⊕ h(R1 ), V4∗ = h(S K i∗∗ j  R2 ). Then, A transmits the messages {V3 , V4 } to O BU . Step 4: On receiving the {V3∗ , V4∗ }, O BU calculates R2∗ = V3∗ ⊕ h(R1 ), S K i∗∗ j = ∗ ∗ h(R1  R2∗ ), V4∗ = h(S K i∗∗ j  R2 ), and checks the validity of V4 . Then, O BU com∗ putes V5 = S K i∗∗ j ⊕ h(R2 ), and sends the {V5 } to RSU . Step 5: A intercepts the {V5 }, and computes V5∗ = S K i∗j ⊕ h(R2 ). Then, RSU sends {V5∗ } to RSU . Step 6: Upon receiving the {V5∗ }, RSU computes h(R2 ) = S K i∗j ⊕ V5∗ , and checks ∗ the validity of h(R2 ). Finally, A establishes S K i∗∗ j with O BU and S K i j with RU S, respectively. Therefore, Jagriti et al.’s protocol cannot resist the man-in-the-middle attacks.

38 To Analyze Security Requirements of Two AKA Protocols …

499

38.4 Conclusion and Suggestion In this paper, we review Wu et al.’s protocol and Jagriti et al.’s protocol. We found node capture attacks in Wu et al.’s protocol. We also found session key disclosure attacks, OBU capture attacks, and man-in-the-middle attacks in Jagriti et al.’s protocol. Moreover, we have provided several suggestions for both protocols. In Wu et al.’s protocol, the master key kn of S Ni should not be stored in H N . Furthermore, the master key k H N of H N should be separated from {id I N , k N , b N } in the database. In this way, it will be difficult to obtain the session key when A captures the H N . In Jagriti et al.’s protocol, the user’s biometric information is transmitted through public channels, which will lead to the leakage of user information. Therefore, it is recommended to cancel the transmission of Bi on the public channel, or use an encryption algorithm to encrypt Bi during transmission. Acknowledgements This research was partially supported by Natural Science Foundation of Shandong Province, China (Grant no. ZR202111230202).

References 1. Alizadeh, J., Safkhani, M., Allahdadi, A.: Isaka: improved secure authentication and key agreement protocol for wban. Wirel. Pers. Commun. 126(4), 2911–2935 (2022) 2. Canetti, R., Krawczyk, H.: Universally composable notions of key exchange and secure channels. In: International Conference on the Theory and Applications of Cryptographic Techniques, pp. 337–351. Springer (2002) 3. Chen, C.M., Huang, Y., Wang, K.H., Kumari, S., Wu, M.E.: A secure authenticated and key exchange scheme for fog computing. Enterp. Inf. Syst. 15(9), 1200–1215 (2021) 4. Chen, J.N., Huang, Z.J., Zhou, Y.P., Zou, F.M., Chen, C.M., Wu, J.M.T., Wu, T.Y.: Efficient certificate-based aggregate signature scheme for vehicular ad hoc networks. IET Netw. 9(6), 290–297 (2020) 5. Chen, J.N., Zhou, Y.P., Huang, Z.J., Wu, T.Y., Zou, F.M., Tso, R.: An efficient aggregate signature scheme for healthcare wireless sensor networks. J. Netw. Intell. 6(1), 1–15 (2021) 6. Cheng, Q., Li, Y., Shi, W., Li, X.: A certificateless authentication and key agreement scheme for secure cloud-assisted wireless body area network. Mob. Netw. Appl. 27(1), 346–356 (2022) 7. Cui, J., Zhang, X., Zhong, H., Zhang, J., Liu, L.: Extensible conditional privacy protection authentication scheme for secure vehicular networks in a multi-cloud environment. IEEE Trans. Inf. Forensics Secur. 15, 1654–1667 (2019) 8. Cui, J., Zhang, X., Zhong, H., Zhang, J., Liu, L.: Extensible conditional privacy protection authentication scheme for secure vehicular networks in a multi-cloud environment. IEEE Trans. Inf. Forensics Secur. 15, 1654–1667 (2020) 9. Deebak, B.D., Al-Turjman, F., Aloqaily, M., Alfandi, O.: An authentic-based privacy preservation protocol for smart e-healthcare systems in iot. IEEE Access 7, 135632–135649 (2019) 10. Deebak, B.D., Fadi, A.T.: Lightweight authentication for iot/cloud-based forensics in intelligent data computing. Futur. Gener. Comput. Syst. 116, 406–425 (2021) 11. Deng, L., Shao, J., Hu, Z.: Identity based two-party authenticated key agreement scheme for vehicular ad hoc networks. Peer-to-Peer Netw. Appl. 14(4), 2236–2247 (2021) 12. Dolev, D., Yao, A.: On the security of public key protocols. IEEE Trans. Inf. Theory 29(2), 198–208 (1983)

500

H. Wu et al.

13. Gupta, A., Tripathi, M., Sharma, A.: A provably secure and efficient anonymous mutual authentication and key agreement protocol for wearable devices in wban. Comput. Commun. 160, 311–325 (2020) 14. Jagriti, J., Lobiyal, D.: An efficient and anonymous authentication key agreement protocol for smart transportation system. In: 2021 International Conference on Computational Performance Evaluation (ComPE), pp. 190–194. IEEE (2021). https://doi.org/10.1109/ComPE53109.2021. 9751950 15. Jia, X., He, D., Kumar, N., Choo, K.K.R.: Authenticated key agreement scheme for fog-driven iot healthcare system. Wirel. Netw. 25(8), 4737–4750 (2019) 16. Kim, Y.J., An, D.J., Sin, K.S., Kim, S.G.: Security analysis of a recent pairing-based certificateless authenticated key agreement protocol for blockchain-based wbans. Cryptology ePrint Archive (2022). https://eprint.iacr.org/2022/844 17. Li, X., Ibrahim, M.H., Kumari, S., Sangaiah, A.K., Gupta, V., Choo, K.K.R.: Anonymous mutual authentication and key agreement scheme for wearable sensors in wireless body area networks. Comput. Netw. 129, 429–443 (2017) 18. Ma, M., He, D., Wang, H., Kumar, N., Choo, K.K.R.: An efficient and provably secure authenticated key agreement protocol for fog-based vehicular ad-hoc networks. IEEE Internet Things J. 6(5), 8065–8075 (2019) 19. Ma, M., He, D., Wang, H., Kumar, N., Choo, K.K.R.: An efficient and provably secure authenticated key agreement protocol for fog-based vehicular ad-hoc networks. IEEE Internet Things J. 6(5), 8065–8075 (2019) 20. Tso, R., Huang, K., Chen, Y.C., Rahman, S.M.M., Wu, T.Y.: Generic construction of dual-server public key encryption with keyword search on cloud computing. IEEE Access 8, 152551– 152564 (2020) 21. Wu, T.Y., Lee, Z., Yang, L., Chen, C.M.: A provably secure authentication and key exchange protocol in vehicular ad hoc networks. Secur. Commun, Netw (2021) 22. Wu, T.Y., Yang, L., Meng, Q., Guo, X., Chen, C.M.: Fog-driven secure authentication and key exchange scheme for wearable health monitoring system. Secur. Commun, Netw (2021) 23. Wu, X., Xu, J., Huang, W., Jian, W.: A new mutual authentication and key agreement protocol in wireless body area network. In: 2020 IEEE International Conference on Smart Cloud (SmartCloud), pp. 199–203. IEEE (2020) 24. Yang, L., Chen, Y.C., Wu, T.Y.: Provably secure client-server key management scheme in 5g networks. Wirel. Commun. Mob. Comput. 2021, 4083199 (2021) 25. Yang, X., Shu, L., Chen, J., Ferrag, M.A., Wu, J., Nurellari, E., Huang, K.: A survey on smart agriculture: development modes, technologies, and security and privacy challenges. IEEE/CAA J. Autom. Sin. 8(2), 273–302 (2021) 26. Yu, S., Jho, N., Park, Y.: Lightweight three-factor-based privacy-preserving authentication scheme for iot-enabled smart homes. IEEE Access 9, 126186–126197 (2021) 27. Zhang, J., Zhong, H., Cui, J., Xu, Y., Liu, L.: Smaka: secure many-to-many authentication and key agreement scheme for vehicular networks. IEEE Trans. Inf. Forensics Secur. 16, 1810–1824 (2020)

Chapter 39

A Method of Expressway Congestion Identification Based on the Electronic Toll Collection Data Ziyang Lin, Fumin Zou, Feng Guo, Xiang Yu, Nan Li, and Chenxi Xia

Abstract With the rapid development of domestic expressway ETC system, the trend of intelligent and digital for expressway management began to mature, which provides a solid foundation for the fully connected vehicle–road-cloud intelligent perception and collaborative decision-making system. At present, as one of the largest Internet of Vehicles (IOV) in the world, ETC system can provides multidimensional data for expressway and covers the vehicle information of each road sections. Among them, the traffic congestion identification plays an important role for vehicle collaborative decision-making and it has tremendous research value. In order to improve the accuracy and stability of traffic congestion identification, this paper proposes a Fuzzy Comprehensive Evaluation Adaptive Matching Algorithm (FACM) by deeply mining the dimensional information of expressway ETC transaction data. This method introduces the Section Portrait into dimensional analysis, and uses Analytic Hierarchy Process (AHP) to weighted average the dimensions of each section, combined with Fuzzy Comprehensive Evaluation (FCE), the level division of section congestion is carried out. The experimental results show that the average congestion recognition accuracy of FCAM is 98.03%, which is 4.98% and 3.07% Z. Lin (B) · F. Zou · X. Yu · N. Li · C. Xia Fujian Key Lab for Automotive Electronics and Electric Drive, Fujian University of Technology, Fuzhou 350118, China e-mail: [email protected] F. Zou e-mail: [email protected] X. Yu e-mail: [email protected] N. Li e-mail: [email protected] C. Xia e-mail: [email protected] F. Guo College of Computer and Data Science, Fujian University, Fuzhou 350118, China e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_40

501

502

Z. Lin et al.

higher than FCE method and K-means method respectively. The proposed method has high stability and high recognition rate.

39.1 Introduction IOV is an open network environment [1], it realizes the transmission of information between objects. With the rapid development of domestic Internet of Things, the trend of intelligent and digital for expressway management began to mature, and we can get the overall information of road conditions in real time [2]. The technology of IOV have led to the rapid development of global informatization, and promoted the development of digital informatization [3]. The on-board system includes roadside units and on-board units, which has become one of the key technologies of intelligent transportation [4]. ETC as an IOV system that can meet most of the vehicle collaborative decision-making needs, that hide a lot of information that can be mined [5], it plays a great role in promoting the development of the IOV [6]. At present, expressway is the implementation scenario of ETC system and the important infrastructure supporting China’s economic operation, that plays a great supporting role in improving the IOV technology. By the end of 2020, the mileage of expressway in China has exceeded 160,000 km, ranking first in the world. However, according to the data released by the Ministry of transport of China, the economic losses caused by traffic congestion account for 20% of the disposable income of the urban population, equivalent to 5–8% of China’s annual Gross Domestic Product (GDP). In addition to economic losses, traffic congestion is also considered to be an important factor in carbon dioxide emissions and global warming [7]. Therefore, how to effectively solve this problem has become one of the great challenges for expressway management departments. In the past few decades, people have done a lot of researches on solving the problem of expressway traffic congestion. At present, there are three mainstream methods: physical-based congestion mitigations, policy-based congestion managements and algorithm-based congestion identifications. Physical-based congestion mitigations mainly relieves congestion by changing the expressway network structure, such as expand lanes, adjust ramp and add new routes. Gregory fields et al. [8] proposed the appropriate scheme to increase transport capacity and reduce traffic congestion by building ARC remote model; policy-based congestion managements refers to government regulations to avoid traffic congestion, and these policies contain various types of systems to help alleviate congestion, including differential tolls on some sections to guide vehicle diversion through toll price leverage to alleviate congestion in some expressway sections; however, the above methods are not the appropriate solutions to alleviate expressway traffic congestion, an algorithm-based approach to identifying congestion and dynamically adjusting vehicle travel, thus improve expressway congestion is considered the most effective method to improve traffic congestion.

39 A Method of Expressway Congestion Identification Based …

503

The current algorithms for judging traffic congestion can be divided into three categories: (1) Direct detection method, such as directly judging the congestion of expressway section through video Chetouan et al. [9] developed four research schemes based on surveillance video to detect congestion. The implementation effect of congestion is proportional to the requirements of hardware, and the dimensions of identification are scarce, so it is difficult to achieve the desired congestion identification effect; (2) Indirect detection method mainly detects the existence of events according to the impact of traffic indicators on traffic flow, such as analyzing the current road congestion through common traffic indicators of speed and flow. These methods have low cost and simple operation, but the detection rate is low and the false alarm rate is high, Atikom et al. [10] used speed to classify the congestion level, however, the test effect of this method fluctuates significantly, and the traffic state does not fluctuate every day like the traffic volume, resulting in a low detection rate of the identification method through indicators. Just as Lin [11] once said, “The effect of detecting the traffic state by detecting the traffic volume is not as significant as that of directly detecting the traffic state.” Therefore, if one wants to identify the traffic congestion of the expressway, one must directly identify the traffic state. Therefore, most people began to judge the traffic state through traffic indicators. Tu et al. [12] generated the congestion index by mining the free flow speed and flow; Gao et al. [13] explored the time pattern of multiple congestion points and the spatial pattern of frequently congested section, using multiple indicators for analysis and good results have been achieved; however, the above methods ignore the correlation between traffic indicators, which often leads to congestion miscalculation when identifying congestion, to unavoidable congestion events, and the identification effect is not ideal; (3) Based on theoretical model: Traditional algorithm to judge traffic indicators, which includes some mature theoretical models such as grey system theory, cluster analysis, Fuzzy Comprehensive Evaluation. Most people use SVM [14] or LSTM [15] to predict traffic conditions, Liao et al. use Bi-direction LSTM to identify Missing POI [16], Zhang et al. [17] made traffic prediction for single vehicle based on the theory of long and short term memory network. Wu et al. [18] introduce a new wavelet kernel to save more data details. Yu et al. [19] used a weighted FCM algorithm to detect expressway congestion; Kalinic et al. [20] used the fuzzy reasoning model to solve the problem of correlation between variables, and took the flow and density as the input of the model to evaluate the traffic status of Ulaanbaatar; Liu et al. [21] proposed a multi metric Fuzzy Comprehensive Evaluation TCSA scheme in the environment based on 5 g LOV. This method uses three traffic congestion indicators and divides the congestion evaluation into five levels, achieving good results. Based on the above content, this paper proposes a scheme based on improved fuzzy evaluation method to accurately identify congestion, which can greatly avoid a false identification of congestion caused by characteristic differences between sections on expressway. The ETC transaction data can provide dimensional support for this method. The ETC system generates a large amount of ETC data, with up to 103 data types, wide coverage, high reliability, convenient analysis and processing, rich results.

504

Z. Lin et al.

Therefore, for the expressway section congestion identification, this paper proposes a FCAM method to accurately identify the congestion of expressway section, which can avoid the problem of a false identification of congestion caused by differences of section on the expressway. Firstly, extract and construct the feature vector matrix of the expressway section, mine the characteristic dimension of the section every 15 min, and use AHP to weight the dimension data. Using FCE for adaptive matching to find the appropriate congestion identification scheme, then the congestion classification is carried out. The test result shows that the method can identify the congestion degree of the section, making avoid a false identification of the expressway congestion and missed detection of congestion. The structure of this paper is as follows. The first section introduces the research methods of expressway traffic congestion identification. The second section defines the relevant concepts of this study. The third section introduces the overall framework of FCAM. The fourth part presents the experimental results and analysis. The fifth paragraph draws conclusions and future work.

39.2 Relevant Definitions Definition 1 Transaction Node: The expressway network is connected by plenty of node ND. The nodes are mainly composed of Gantry FND, Toll Station SND and Junction Point YND, of which FND and SND are collectively referred to as Boundary Point DND.

ND =

⎧ ⎨ ⎩

 DND =

FND SND

(39.1)

YND

Definition 2 Section: The adjacent DND on the expressway are combined into Section QD.

QD = {DND1 , DND2 }

(39.2)

Definition 3 Expressway Road Network: The expressway network is composed of plenty of QD, including all the time–space information of QD, which is called LW. The relationship between each DND pair is called Topo. Definition 4 Uncertainty Factor: There are various types of section on the expressway, and their influencing factors account for different proportions. Among them, the uncertainty factor determines the type of section on the expressway, which is expressed by UF, including SND, YND and Service Area VND.

39 A Method of Expressway Congestion Identification Based …

505

Definition 5 Vehicle Driving Section: Every vehicle will drive through the QD on the expressway, and the section in which the vehicle passes can be divided into different Vehicle Driving Section due to UF, this Vehicle Driving Section is called VDS. According to the UF, the section type can be further determined.

V DS = {DND1 , UF, DND2 }

(39.3)

When UF does not exist, it is normal section FD FD = {DND1 , DND2 }

(39.4)

When UF is SND, QD is Toll Section SD SD = {DND1 , SND, DND2 }

(39.5)

When UF is VND, QD is Service Area Section VD V D = {DND1 , V ND, DND2 }

(39.6)

When UF is YND, QD is Junction Section YD YD = {DND1 , YND, DND2 }

(39.7)

Definition 6 Vehicle Trajectory: Each vehicle has its own driving trajectory on the expressway, which is composed of continuous DND by which the vehicle passed, Called Traj (n ≥ 2).

Traj = {DND1 , DND2 , ..., DNDn }

(39.8)

Definition 7 Vehicle Trajectory Set: The above definition shows that the trajectory of a single vehicle is Traj. Since most of the time, there are multiple vehicles on the expressway. The trajectory of plenty of vehicles is defined as a set, which is called vehicle trajectory set TrajS.

TrajS = {Traj1 , Traj2 , ..., Trajn }

(39.9)

506

Z. Lin et al.

39.3 Methodology 39.3.1 Data Pre-processing ETC transaction data is composed of continuous discrete points obtained by sampling. Obviously, the workload of mining congestion information directly from a large amount of transaction data is large and inefficient. And after constructing the VDS data set, the congestion information obtained by mining is more realistic, effective and intuitive. After the initial cleaning of the transaction data collected by the ETC gantry system, the processed data constructs the vehicle trajectory dataset TrajS in chronological order. In detail, the vehicle trajectory dataset is a collection of ETC gantries of multiple vehicles in the process of expressway driving. We aggregate the dimensional information of the transaction data based on a single trip code and obtain the TrajS after deduplication of part of the data. TrajS contains congested section and uncongested section, of which congested section account for only a minority of the total. When the number of congested sections are small, there must be insufficient congestion information. In order to mine the congestion information of the section more fully, we may cut TrajS, and then selects congested sections by eliminating TrajS. We can quickly identify section congestion information in massive ETC trajectories, and then realize deep mining of congested section.

39.3.2 Evaluation of Congestion Information for Section on Expressway In this subsection, this paper realizes the congestion identification of the section by evaluating the congestion information of the expressway section. On the basis of FCE, the proposal uses AHP to adaptively match the evaluated section, which makes up for the shortcoming that FCE is too subjective to the index weight vector, which not only strengthens the connection between traffic indicators but also distinguishes the characteristics between different section to achieve accurate identification, as shown in Algorithm 1.

39 A Method of Expressway Congestion Identification Based …

507

Algorithm 1: FCAM Input: The set of VDS = { |1 ≤ i ≤ n}; LW; Topo; Output: , |1 ≤ j ≤ n}; The VDS Score = { Algorithm Steps: 1: Type = []; Score = []; Level = [] ; // Initialize set 2: |1 ≤ i ≤ n}, 0 = Topo; 0= { // Data pre-processing 3: X = Data_Preprocessing( 0 , 0 ); 4: T = Extractfeature(X); // Extract data dimension features 5: S = Normalization( ); // Normalization processing 6: W = AHP( ); // AHP dimension weighting 7: Foreach( in Type): 8: ℎ = Match(W, ); // Match the weight 9: End for 10: Foreach( in Score): = FCE( // FCE congestion level assessment , ℎ ); 11: ); // congestion scores = CongestionClassification( 12: 13: End for

In order to better evaluate the congestion information of expressway, this paper constructs three parameters commonly used for traffic congestion according to the objective rules of driving on expressway. The three parameters are section speed, section traffic flow and section delay time. This paper adopts AHP to analysis and weight these three dimensions to ensure the scientific rationality and universality of the framework of expressway congestion identification. Firstly, this paper analysis the relationship between the various factors in the system. In a single section of an expressway, each factor is interrelated and interacts with each other. The average speed of a section plays a positive or negative role on both traffic volume and average delay. Then, we need to make a two-by-two comparison of the importance of an element in the same level about an element in the previous level to obtain a two-by-two comparison matrix. Finally, we construct the judgment matrix X of all dimensions, the xij is the comparison value element. ⎛

x11 . . . ⎜ .. . . X =⎝ . . xm1 . . .

⎞ x1n .. ⎟ . ⎠

(39.10)

xmn

We fill in the comparison value of the importance degree of the two elements in X, and the importance degree is shown in Table 39.1. Traffic indicators of expressway have different criteria for judging congestion information. After referring to a large number of papers [22, 23] and experimental validation, we construct the judgment matrix X based on three dimensions and calculate the characteristic values and feature vector of X. Because the scores of the judgment matrix may be conflicting, the experiment needs to pass a test of the consistency

508

Z. Lin et al.

Table 39.1 Comparison of the importance of the elements

Scale

Implication

1

Both factors are equally important

3

One factor is slightly more important than the other

5

One factor is significantly more important than the other

7

One factor is more strongly important than the other

9

One factor is extremely important than the other

2, 4, 6, 8

The median of the above two adjacent judgments

Table 39.2 Index RI of random consistency n

1

2

3

4

5

6

7

8

9

10

11

RI

0

0

0.58

0.90

1.12

1.24

1.32

1.41

1.45

1.49

1.51

of X before the final weights can be obtained. We test the characteristic values and feature vector of the matrix, and judge any element of the matrix with positive numbers. Referring to Table 39.2, if the calculated result of CR is less than 0.1, the consistency judgment is passed, and the weight set A is obtained by normalizing the feature vector at this time. We define the traffic state of the section as fuzzy sets μA → [0, 1], and use vector notation to represent degrees of membership and use the assignment method to determine the membership function. We selected a partial-large membership function to assign the average speed of the section and a partial-small membership function to assign the delay time of the section according to the objective rules and general needs of the three parameters of the zone: average speed, traffic flow and delay time in the section on the expressway. Partial-small membership function:

μA =

μA =

⎧ ⎨ ⎩ ⎧ ⎨ ⎩

1, x ≤ a ≤x≤b 0, x ≥ b

(39.11)

1, x ≤ a ≤x≤b 0, xνb

(39.12)

b−x ,a b−a

x−a ,a b−a

We construct the fuzzy matrix R by the relationship between traffic indicators and congestion evaluation levels. ⎤ ⎡ ⎤ r11 r12 r13 r14 r15 r16 R1 R = ⎣ R2 ⎦ = ⎣ r21 r22 r23 r24 r25 r26 ⎦ R3 r31 r32 r33 r34 r35 r36 ⎡

(39.13)

39 A Method of Expressway Congestion Identification Based …

509

Table 39.3 Definition of congestion status Speed threshold (km/h)

Congestion scores

Congestion levels

Speed ≥ 60

C ≥ 0.6

Smooth

50 ≤ Speed ≤ 60

0.5 ≤ C ≤ 0.6

Stable

35 ≤ Speed ≤ 50

0.35 ≤ C ≤ 0.5

Bascially stable

20 ≤ Speed ≤ 35

0.20 ≤ C ≤ 0.35

Slight congestion

10 ≤ Speed ≤ 20

0.10 ≤ C ≤ 0.20

Moderate congestion

0 ≤ Speed ≤ 10

0 ≤ C ≤ 0.10

Congestion

We combine the weights A which obtained by AHP with the fuzzy matrix R for a comprehensive evaluation. B=A∗R

(39.14)

According to the principle of maximum membership, we select the maximum score in B as the congestion score of the section in this period. We define the congestion score as C. In this study, the congestion score C of expressway section through FCE, we define the output of C as a continuous value [0, 1]. According to the technical specification for highway travel information service, the section traffic speed is regarded as the main index for evaluating expressway traffic congestion, and C is classified according to the threshold value of speed, with the section traffic flow and the section delay time as auxiliary indexes, so as to judge the congestion status. If C ≥ 0.35, we think there is no congestion in this section, or we think that congestion has occurred in this section, as shown in Table 39.3.

39.4 Results and Discussion 39.4.1 Pre-introduction The large-scale deployment of the ETC system records the traffic status of most vehicles on the expressway, which provides a solid data foundation for studying the traffic congestion identification of key section of the expressway. This article takes the distribution of expressway ETC systems in 8 prefecture-level cities and 13 county-level cities in Fujian Province as the study area, and the spatial positions of its gantry and Toll Station are shown in the Fig. 39.1. There are three categories of experimental data. One is the intelligent transaction data generated by ETC gantry of Fujian Provincial expressway for 5 days from May 1st, 2021 to May 5th, 2021, as shown in Table 39.4, there are 54,474,098 pieces of transaction data, of which 4,611,924 pieces of trajectory data and 20,231,317

510

Z. Lin et al.

Fig. 39.1 Fujian Provincial Highway Road Network Map

pieces of section data are fitted, and the other data is the topological relationship map between each gantry of the expressway and the distance between them. The third type of data is the latitude and longitude information of the gantry and toll stations. All these data come from Fujian Expressway Information Technology Co., Ltd. Table 39.4 ETC transaction data Name

Types

Examples

PassID

String

01350119382305xxxxxxxx0501163019

Obuplate

String

xxxx9742

Entime

DateTime

2021-05-01 xx:xx:xx

Enstation

FixedString(4)

35xx

Flagid

String

34xx01

Tradetime

DateTime

2021-05-01 xx:xx:xx

39 A Method of Expressway Congestion Identification Based …

511

Table 39.5 Random samples Samples

Speed(km/h)

Flow

Delay(s)

Results

3

107.7841

43

169.1628

Smooth

16

22.6235

199

656.5075

Congestion

22

53.47994

198

363.4533

Slight congestion

56

86.57878

125

146

65.35857

86

564

95.45739

45

40.42222

Smooth

19

25.94737

Smooth

788

107.0518

65.71987 391.6512

Stable Basically stable

39.4.2 Result Analysis This article uses FCAM method to identify 20,231,317 section data of expressway. First, we randomly extracted some data from the section data to test the congestion. The test results are shown in Table 39.5. Then, based on different weighting schemes for different section type after the consistency test of AHP, the congestion identification data set is constructed with a time interval of 15 min, and 2501 possible congested section are selected from 2668 section, among which the number of 6 section with obvious congestion characteristics are selected, congestion identification based on days as shown in Fig. 39.2. In order to test the recognition accuracy of section congestion, we compare the proposed congestion identification method with the FCE method and the K-means algorithm. This paper tests the section with obvious congestion characteristics on May 1, 2021, and the test results are as follows in Table 39.6. In this article, if the congestion level is above Moderate Congestion, it is regarded as congestion, and the ETC transaction data from May 1, 2021 to May 5, 2021 provided by Fujian Expressway Information Technology Co., Ltd. are used as the validation set for verification, and the section congestion identification is accurate as verified by video data Fig. 39.3.

39.5 Results and Discussion This paper proposes an expressway congestion identification method FCAM based on ETC transaction data. Firstly, the section data set is constructed based on ETC trajectory data. After analysing the traffic characteristics of expressway, three data dimensions are constructed, and the AHP is used to weight the data. Then, the corresponding weighting scheme is selected according to the adaptive matching of the section weight. Finally, the Fuzzy Comprehensive Evaluation are combined to evaluate the score, which ensures the integrity and accuracy of the congested section.

512

Z. Lin et al.

Fig. 39.2 Experimental diagram of FCAM Table 39.6 Comparison of identification accuracy

Section

FCE (%)

K-means (%)

FCAM (%)

340D19-340D1B

95.83

96.53

97.92

340D1B-350133

91.67

90.97

98.61

340D19-340,401

88.89

95.83

98.61

340,603-340,607

93.06

97.22

97.92

350D15-350D14

94.44

97.91

98.61

340,261-340,263

90.28

90.97

95.83

35,016,701-350D15

98.61

96.52

99.31

340,263-340,265

90.97

91.67

94.44

341B0B-340507

85.41

88.89

97.92

341F05-341F07

96.53

97.91

98.61

350A07-350A05

92.36

96.52

99.31

350A17-350A15

98.61

98.61

99.31

39 A Method of Expressway Congestion Identification Based …

513

Fig. 39.3 Video data

The experimental results show that the combination of Analytic Hierarchy Process and section portrait improves the result of FCE, so that the FCAM can make a reasonable evaluation of the fuzzy situation in the section, effectively improve the identification of the situation in the congested section and reduce the occurrence of congestion false identification, and can evaluate the congestion information in combination with the characteristics of different section. However, there are still some unsolved problems in the work: 1. This work will be affected by some special conditions of the expressway, such as weather factors, road maintenance factors, traffic accidents and so on. More data features can be considered to carry out the analysis, extract and model them to improve the recognition effect. 2. Due to the differences in the speed limit of different expressway section, there are plenty of differences in the average speed between each section, which has a certain impact on the experimental results.

References 1. Wu, T.Y., Guo, X., Yang, L., Meng, Q., Chen,C.M.: A lightweight authenticated key agreement protocol using fog nodes in Social Internet of vehicles. Mob. Inf. Syst., 2021, 3277113 (2021) 2. Kumar, V., Kumar, R., Kumar, V., A. Kumari., Kumari, S.: RAVCC: Robust authentication protocol for rfid based vehicular cloud computing. J. Netw. Intell. 7, (3), 526–543, (2022) 3. Wu T.Y., Guo, X., Chen, Y.C., Kumari, S., Chen, C.M.: SGXAP: SGX-based authentication protocol in iov-enabled fog computing. Symmetry, 14(7), 1393 (2022) 4. Chen, J.N., Huang, Z.J., Zhou, Y.P., Zou, F.M., Chen, C.M., Wu, J.M.T., Wu, T.Y.: Efficient certificate-based aggregate signature scheme for vehicular ad hoc networks. IET Networks, 9(6), 290–297, (2020) 5. Chen, C-M., Chen, L., Gan, W., Qiu, L., Ding, W.: Discovering high utility-occupancy patterns from uncertain data. Inf. Sci. 546, 1208-1229 (2021) 6. Chen, R-F., Luo, H., Huang, K-C., Nguyen, T-T., Pan, J-S.: An improved honey badger algorithm for electric vehicle charge orderly planning. J. Netw. Intell. 7(2), 332–346, (2022) 7. Aung, N., Zhang, W., Dhelim, S., Ai, Y.: T-Coin: Dynamic traffic congestion pricing system for the internet of vehicles in smart cities. Inf. 11(3), (2020)

514

Z. Lin et al.

8. Fields, G., Hartgen, D., Moore, A., Poole, R.W., Jr.: Relieving congestion by adding road capacity and tolling. Int. J. Sustain. Transp. 3(5-6), 360–372, (2009) 9. Chetouane, A., Mabrouk, S., Jemili, I., Mosbah, M.: Vision-based vehicle detection for road traffic congestion classification. Concurr. Comput. Pract. Experi-Ence: Distrib. Comput. Smart Netw. 34(7), (2020) 10. Pattara-Atikom, W., Pongpaibool, P., Thajchayapong, S.: Estimating road traffic congestion using vehicle velocity. In: 2006 6th International conference on ITS telecommunications proceedings. 11. Lin, W.H., Lu, Q., Dahlgren, J.: dynamic procedure for short-term prediction of traffic conditions. Transp. Res. Rec. 1783(1), (2002) 12. Tu, Y., Lin, S., Qiao, J., Liu, B.: Deep traffic congestion prediction model based on road segment grouping. Appl. Intell. 51, 8519–8514, (2021) 13. Gao, H., Yan, Z., Hu, X., Yu, Z., Luo, W., Yuan, L., Zhang, J.: A method for exploring and analyzing spatiotemporal patterns of traffic congestion in expressway networks based on origin– destination data. Int. J. GeoInformation. 10(5), (2021) 14. Zhang, F., Wu, T.Y., Pan, J.S., Ding, G., Li, Z.: Human motion recognition based on SVM in VR art media interaction environment. Hum-centric Comput. Inf. Sci. 9, 40 (2019) 15. Kumar, S., Damaraju, A., Kumar, A., Kumari, S., Chen, C.M.: LSTM network for transportation mode detection. J. Internet Technol. 22 (4), 891–902, (2021) 16. Liao, L., Lin, J., Zhu, Y., Bi, S., Lin, Y.: A Bi-direction LSTM attention fusion model for the missing POI identification. J. Netw. Intell. 7(1), 161–174, (2022) 17. Zhang, S., Su, X., Jiang, X., Chen, M., Wu, T.Y.: A traffic prediction method of bicycle-sharing based on long and short term memory network. J. Netw. Intell. 4(2), 17–29, (2019) 18. Wu, Q., Zang, B., Zhang, Y., Qi, Z.: Wavelet kernel twin support vector machine. J. Inf. Hiding Multimed. Signal Process. 12(2), 93–101, (2021) 19. Yu, X., Zou, F.M., Guo, F., Chen, Z.Y., Luo,Y.Y., Chen, H.: An expressway congestion detection algorithm based on etc data mining. In: Advances in intelligent information hiding and multimedia signal processing, part of the smart innovation, systems and technologies book series, vol. 277, pp. 269–279, (2022) 20. Kalinic, M., Krisp, J.M.: Fuzzy inference approach in traffic congestion detection. Ann. GIS. 25(4), (2019) 21. Liu, L., Lian, M., Lu, C., Zhang, S., Liu, R., Xiong, N.N.: TCSA: A traffic congestion situation assessment scheme based on multi-index fuzzy comprehensive evaluation in 5G-IoV. Electron. 11(7), (2022) 22. Yun, H.: Traffic jam detection based on fuzzy analytic hierarchy process. Dalian University of Technology, (2021) 23. Xin, Y., Henan, B., Jianmin, N., Wenjuan, Y., Honggen, Z., Xingyu, J., Pengfei, Y.: Coating matching recommendation based on improved fuzzy comprehensive evaluation and collaborative filtering algorithm. Sci. Rep. Article no:14035, 2021

Chapter 40

Privileged Insider Attacks on Two Authentication Schemes Yiru Hao, Saru Kumari, Kuruva Lakshmanna, and Chien-Ming Chen

Abstract Due to the epidemic, it is difficult for people to go out. As a result, telemedicine systems are widely used, making it possible for people to see a doctor without leaving home. However, one of the essential issues facing online medical diagnosis is whether the patient’s privacy can be protected. Seno et al. proposed that mutual authentication between the patient and the medical server is required when using a telemedicine system to address this issue. Nevertheless, Seno et al. did not consider comprehensively in the authentication phase when designing this protocol, which led to the security protocol still being flawed. Another critical aspect of the telemedicine system is that the real-time nature of the network must be guaranteed. Due to the increase of various modern wireless applications, it is necessary to manage a large amount of traffic on the server. The operation speed may be inefficient if only a single server is used for data processing. Hence, Kumar et al. propose a protocol for wireless applications in the multi-server environment. In this paper, we analyze the security of the protocols proposed by Seno et al. and Kumar et al. We found that both protocols have some security vulnerabilities, including privileged insider attacks.

Y. Hao · C.-M. Chen (B) College of Computer Science and Engineering, Shandong University of Science and Technology, Qingdao 266590, China e-mail: [email protected] Y. Hao e-mail: [email protected] S. Kumari Department of Mathematics, Ch. Charan Singh University, Meerut, India K. Lakshmanna School of Information Technology and Engineering, Vellore Institute of Technology, Vellore, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_41

515

516

Y. Hao et al.

40.1 Introduction Today, Internet of Things (IoT) application is snowballing, and wireless applications are everywhere in our daily life [1, 2]. For example, telemedicine systems [3, 4], multi-server environments [5, 6], Internet of Vehicles [7, 8], online banking, smart home, smart cities [9, 10], etc. While the popularity of IoT brings excellent convenience to people, it also involves network security issues [3, 11–14]. Attackers can easily access the information transmitted by legitimate users during the communication process and even tamper with the information obtained. If a user’s information is leaked or tampered with, it would have severe consequences for this user. To solve these security problems, we mainly consider two aspects. First, the identities of both sides of the communication need to be authenticated mutually. Second, we need to encrypt sensitive messages transmitted through a public channel. It means that we need a session key shared between both entities. Researcher has proposed various authentication schemes [15, 16] to achieve these two aspects. A Telemedicine system is one of the most important applications of IoT, especially under the influence of COVID-19. Telemedicine systems can break time and space constraints, and doctors can share information about patients’ conditions and make diagnoses over the Internet. All sensitive patient information is transmitted online; thus, patients’ privacy would be compromised. Besides, a large amount of patient treatment information is stored on a server. If this server is hacked by an attacker, this attacker can obtain all these confidential data [17–19]. In 2012, Wu et al. [20] proposed a secure authentication scheme for telemedicine information systems with smart cards. However, this scheme cannot resist privileged insider attacks and imitation attacks. Subsequently, He et al. [21] introduced a two-way authentication scheme and claimed that it could meet various security requirements and resist all attacks. Unfortunately, Seno et al. [22] found that the protocol [21] could not provide perfect forward secrecy [23] and was not resistant to denial of service attacks. To solve these weaknesses, Seno et al. [22] explored an improved authentication and key agreement scheme for patients’ privacy. Regrettably, we further demonstrate that Seno et al.’s scheme [22] is still vulnerable to privileged insider attacks. On the other hand, the rapid increase of various wireless applications makes a single-server environment not meet the current needs of people using it. Besides, in a single-server environment, a failure of that server can bring down the entire communication system. Therefore, the emergence of a multi-server environment (MSE) is necessary [24–27]. In 2020, Haq et al.[28] proposed a lightweight 5G network authentication scheme for multi-server environments. Two years later, Kumar et al. [29] revealed some vulnerabilities in Haq et al.’s scheme [28], such as the inability to resist imitation attacks, replay attacks, etc. They then described an improved scheme for multi-server environments based on elliptic curve cryptosystems. Although the authors claimed that this scheme [29] is secure, sadly, we found that this scheme is still not resistant to privileged insider attacks in this paper. The organization of this paper is as follows. In Sect. 40.2, we give a brief review of Seno et al.’s scheme [22]. In Sect. 40.3, we show that Seno et al.’s scheme is

40 Privileged Insider Attacks on Two Authentication Schemes Table 40.1 Notations in this section Notations Ui Sj I Di ,P Wi I Dj a, b T1 , T2 y SK SC h(·) · ⊕  A ?

=

517

Descriptions i-th user j-th sensor Identity and password of Ui Identity of S j Random integers Current timestamps Secret keys of S j Session key User’s smart card Hash function Scalar multiplication of elliptic curves Bitwise XoR functions Concatenation Attacker/adversary Equality checker

vulnerable to privileged insider attacks. We also review Kumar et al.’s scheme [29] in Sect. 40.4 and then cryptanalysis of this scheme in Sect. 40.5. Finally, Sect. 40.6 concludes.

40.2 Review of Seno et al.’s Protocol Here we briefly review Seno et al.’s scheme [22]. The protocol is composed of two main phases: the user(patient) registration phase, and the login and authentication phase. Readers can refer to their paper for detail. Table 40.1 describes the symbols used in this scheme.

40.2.1 Registration Phase (1) Ui chooses their own username I Di , password P Wi , a random number a and computes V1 = h(I Di  P Wi  a). Then, Ui transmits registration request {I Di , V1 } to Sever S j via a secure channel. (2) the sever S j picks two random numbers b, d. Next, S j calculates P I D = h(b  I D i ), R1 = h(V1  P I D), V2 = h(R1  I Di  V1 ), V3 = h(P I D  y), V3 . Apart from that, S j stores {d, V3 } in tamper proof memory and G 1 = R1

518

Y. Hao et al.

{b, d, V2 , G 1 , h(·)} into a smart card. In the end, S j sends the smart card back to Ui .  V1 and adds (3) After getting the smart card from S j , Ui calculates P A = G 1 a, P A in this smart card. At this stage, this smart card stored {a, b, d, V2 , G 1 , P A, h(·)}.

40.2.2 Login and Mutual Authentication Phase Figure 40.1 illustrates this phase of Seno et al.’s scheme [22]. The detailed steps are as follows: 1. Ui inserts SC into a device and inputs his I Di and P Wi . Immediately after, SC calculates V1 = h(I Di  P Wi  a), P I D = h(b  I Di ) and R1 = h(V1  ? P I D). Then, SC checks V2 = h(R1  I Di  V1 ). If the equation holds, Ui logs in successfully. Next,SC captures its current timestamp T1 and computes G 1 =  G 1 , M = h(V3  G 1  P I D  R1  T1 ). Ultimately, Ui P A V1 , V3 = R1 connects with S j and sends a message including the values {R1 , P I D, M, T1 }. 2. S j checks the freshnessof T1 . If the verification passes, S j calculates V3 = V3 . After that, S j authenticates the user’s legality by h(P I D  y), G 1 = R1

Fig. 40.1 Login And Mutual Authentication Phase

40 Privileged Insider Attacks on Two Authentication Schemes

519

?

verifying that the equation holds M = h(V3  G 1  P I D  R1  T1 ). If the verification holds, S j computes session key S K = h(V3  d  G 1 ) and validation message V E R = h(S K  G 1  V3  T2 ), where T2 is current timestamp. In the end, S j transmits authentication message V E R and timestamp T2 to Ui . 3. As soon as Ui receives the sever’s message, Ui checks the freshness of T2 , which prevents replay attack. If time freshness verification passes, Ui calculates S K = h(V3  d  G 1 ). Finally, Ui verifies the reliability of the shared session key by ? verifying V E R = h(S K  G 1  V3  T2 ) that the equation holds. Assuming that the equation holds, then Ui and S j will communicate with messages using S K .

40.3 Cryptanalysis of Seno et al.’s Protocol In this section, we demonstrate that Seno et al.’s scheme suffers from the problem of privileged insider attack. In privileged insider attacks, an attacker A is able to obtain the data stored in the server’s database that stores users’ registration information. With the information captured from the server and the messages transmitted online, A can further calculate S K . The explicit steps of our attacks are as follows. (1) An attacker A who is a privileged insider obtains information {d, V3 } stored in S j ’s database. (2) A steals information R1 , which is transmitted via a public channel. (3) With d, V3 , R1 , A calculates:  G 1 = R1 V3 (4) Now A can further computes S K where S K = h(V3  d  G 1 ) Obviously, Seno et al.’s scheme is actually vulnerable to privileged insider attacks.

40.4 Review of Kumar et al.’s Protocol In this section we focus on reviewing the scheme designed by Kumar et al. [29]. This scheme contains the initialization phase, the user and server registration phase, and the login and authentication phase. Notations used in this scheme are defined in Table 40.2.

520 Table 40.2 Notations listed in this section

Y. Hao et al. Notations

Descriptions

Ui Sj RC I Di , P Wi SI Dj x, y

i-th user j-th sensor Registration center identity and password Ui S j ’s identity Secret keys of Ui , S j respectively RC’s private keys Public key of Ui Public key of S j Public key of RC Session key Elliptic curve base point Allowed transmission delay Allowed transmission delay Hash function Scalar multiplication of elliptic curves Bitwise XOR operation Elliptic curve

s RC PUi PU j PU RC SK P T1 , T2 , T3 T h(·) · ⊕ E

40.4.1 Initialization Phase Kumar et al.’s scheme [29] embraces three roles: Ui , S j and RC. In this phase, RC selects a generator point P ∈ E p . In the same way, RC calculates its own public key PU RC = s RC · P, where s RC is a secret key of RC. Similarly, Ui selects x as his/her private key and calculates its own public key PUi = x · P. Eventually, RC discloses the parameters{E,P,PU RC ,h(·)} and Ui publicly announces a parameter PUi .

40.4.2 User Registration Phase (1) Ui picks up a real identity I Di for himself and a a0 ∈ Z q∗ . Subsequently, Ui calculates R E G = h(I Di  a0 ) and posts R E G to RC. (2) RC calculates B I D = h(R E G  s RC ). Then RC deposits B I D into SC and sends it to Ui .

40 Privileged Insider Attacks on Two Authentication Schemes

521

 (3) Ui sets a password P Wi and calculates Z 1 = h(I Di  P Wi ) a0 , P I D = h(I Di  B I D  a0 ). Lastly, Ui saves {Z 1 , P I D} into a SC. SC now has parameters {Z 1 , P I D, B I D}.

40.4.3 Sensor Registration Phase (1) S j chooses S I D j and y as its secret key. Then S j calculates PU j = y · P , PUsu = y · PUi and transmits parameters {S I D j , PUsu } to RC. (2) On receiving the parameters of S j , RC calculates S I F = h(S I D j  PUsu ) · s RC . Ultimately, RC transmits S I F to S j . (3) Upon getting the message, S j stores S I F in its own database. Moreover, S I D j and PU j are publicly announced.

40.4.4 Login and Mutual Authentication Phase Figure 40.2 illustrates this phase. The detailed steps are as follows: (1) Ui inserts his/her SC into a mobile device and inputs his/her {I Di , P Wi }  ? pair. Mobile device computes a0 = Z 1 h(I Di  P Wi ) and checks P I D = h(I Di  B I D  a0 ). If the equation holds, Ui chooses a random number b1 and timestamp T1 . Mobile device calculates S I F1 = h(S I D j  x · PU j ) · PU RC , S I F1 ,G 1 = h(I Di  b1  T1 ) and Z 3 = h(S I F1  G 1 ). After that Z 2 = b1 Ui sends the message {Z 2 , Z 3 , G 1 , T1 } through public channel to S j . authentica(2) After receiving the message, S j chooses the reception timestamp of ? tion request T2 and checks the received timestamp as (T2 − T1 ) ≤ T , where S I F · P. T is a valid time interval. If it holds, S j computes b1 = Z 2 Then, S j confirms the reliability of the received information by verifying ?

Z 3 = h(S I F · P  G 1 ). If verification is successful, S j selects a random num S I F · P, S K = h(b1  b2 ), R = ber b2 . Futhermore, S j computes G 2 = b2 h(S K  b1  b2  T3 ), where T3 is current timestamp. Eventually, {G 2 , R, T3 } is sent to Ui . ?

(3) Ui checks the freshness of T3 by verifying (T4 − T3 ) ≤ T , where T4 indicates the message from S j . After successful verifithe time when the Ui receives S I F1 , S K = h(b1  b2 ). Finally, Ui verifies cation, SC calculates b2 = G 2 ? R = h(S K  b1  b2  T3 ). If the condition is true, Ui considers S j to be an authenticated entity and S K is key for further use.

522

Y. Hao et al.

Fig. 40.2 Login And Mutual Authentication Phase

40.5 Cryptanalysis of Kumar et al.’s Protocol Here we show that Kumar et al.’s scheme is also vulnerable to privileged insider attack. Assuming that attacker A is a privileged insider. It means that A get access to the database information of S j . The following steps will show that A can successfully calculate the session key. (1) A steals the information S I F in the server’s database. (2) A intercepts messages {Z 2 , G 2 } over public channels. (3) With the messages A obtains, he computes  b1 = Z 2  S I F · P, SI F · P b2 = G 2 (4) Once the A solves the values of b1 , b2 , A can successfully compute S K between Ui and S j . S K = h(b1  b2 ).

40 Privileged Insider Attacks on Two Authentication Schemes

523

40.6 Conclusion In this paper, we review the protocols proposed by Kumar et al. and Seno et al. We also analyze the authentication phase of these two protocols and find that both are not resistant to privileged insider attacks. An attacker can use the information stored in the server and the public communication channel to compute the session key during communication efficiently. We hope that the research will provide guidance for the development of more secure protocols.

References 1. Chen, C.-M., Deng, X., Gan, W., Chen, J., Islam, S.K.: A secure blockchain-based group key agreement protocol for IoT. J. Supercomput. 77(8), 9046–9068 (2021) 2. Yavari, M., Safkhani, M., Kumari, S., Kumar, S., Chen, C.-M.: An improved blockchain-based authentication protocol for IoT network management. Secur. Commun. Netw. 2020 (2020) 3. Chen, C.-M., Li, C.-T., Liu, S., Tsu-Yang, W., Pan, J.-S.: A provable secure private data delegation scheme for mountaineering events in emergency system. Ieee Access 5, 3410–3422 (2017) 4. Kumari, A., Kumar, V., Yahya Abbasi, M., Kumari, S., Chaudhary, P., Chen, C.-M.: Csef: cloudbased secure and efficient framework for smart medical system using ECC. IEEE Access 8, 107838–107852 (2020) 5. Akram, M.A., Ghaffar, Z., Mahmood, K., Kumari, S., Agarwal, K., Chen, C.-M.: An anonymous authenticated key-agreement scheme for multi-server infrastructure. Hum.-Centric Comput. Inf. Sci. 10(1), 1–18 (2020) 6. Amintoosi, H., Nikooghadam, M., Kumari, S., Kumar, S., Chen, C.-M.: Tama: three-factor authentication for multi-server architecture. Hum.-Centric Comput. Inf. Sci. 11 (2021) 7. Kumar, V., Kumar, R., Kumar, V., Kumari, A., Kumari, S.: RAVCC: robust authentication protocol for RFID based vehicular cloud computing. J. Netw. Intell. 7 (2022) 8. Chaudhry, S.A.: Combating identity de-synchronization: an improved lightweight symmetric key based authentication scheme for IOV. J. Netw. Intell. 6, 12 (2021) 9. Wang, E.K., Wang, F., Kumari, S., Yeh, J.-H., Chen, C.-M.: Intelligent monitor for typhoon in IoT system of smart city. J. Supercomput. 77(3), 3024–3043 (2021) 10. Chen, C.-M., Chen, L., Huang, Y., Kumar, S., Wu, J.M.-T.: Lightweight authentication protocol in edge-based smart grid environment. EURASIP J. Wirel. Commun. Netw. 2021(1), 1–18 (2021) 11. Triantafyllou, A., Sarigiannidis, P., Lagkas, T.D.: Network protocols, schemes, and mechanisms for internet of things (IoT): features, open challenges, and trends. Wirel. Commun. Mob. Comput. 2018 (2018) 12. Ravanbakhsh, N., Nazari, M.: An efficient improvement remote user mutual authentication and session key agreement scheme for e-health care systems. Multimed. Tools Appl. 77(1), 55–88 (2018) 13. Abbasinezhad-Mood, D., Nikooghadam, M.: Efficient design of a novel ECC-based public key scheme for medical data protection by utilization of NanoPi fire. IEEE Trans. Reliab. 67(3), 1328–1339 (2018) 14. Kim, M., Moon, J., Won, D., Park, N.: Revisit of password-authenticated key exchange protocol for healthcare support wireless communication. Electronics 9(5), 733 (2020) 15. Chen, C.-M., Fang, W., Wang, K.-H., Tsu-Yang, W.: Comments on “an improved secure and efficient password and chaos-based two-party key agreement protocol”. Nonlinear Dyn. 87(3), 2073–2075 (2017)

524

Y. Hao et al.

16. Liu, S., Chen, C.-M.: Comments on “a secure and lightweight drones-access protocol for smart city surveillance. IEEE Trans. Intell. Transp. Syst. (2022) 17. Ostad-Sharif, A., Abbasinezhad-Mood, D., Nikooghadam, M.: An enhanced anonymous and unlinkable user authentication and key agreement protocol for TMIS by utilization of ECC. Int. J. Commun. Syst. 32(5), e3913 (2019) 18. Radhakrishnan, N., Karuppiah, M.: An efficient and secure remote user mutual authentication scheme using smart cards for telecare medical information systems. Inf. Med. Unlocked 16, 100092 (2019) 19. Radhakrishnan, N., Muniyandi, A.P.: Dependable and provable secure two-factor mutual authentication scheme using ECC for IoT-based telecare medical information system. J. Healthc. Eng. 2022 (2022) 20. Zhen-Yu, W., Lee, Y.-C., Lai, F., Lee, H.-C., Chung, Y.: A secure authentication scheme for telecare medicine information systems. J. Med. Syst. 36(3), 1529–1535 (2012) 21. He, D., Zeadally, S., Kumar, N., Lee, J.-H.: Anonymous authentication for wireless body area networks with provable security. IEEE Syst. J. 11(4), 2590–2601 (2016) 22. Hosseini Seno, S.A., Budiarto, R.: An efficient lightweight authentication and key agreement protocol for patient privacy. Comput. Mater. Continua 69 (2021) 23. Ge, M., Kumari, S., Chen, C.-M.: AuthPFS: a method to verify perfect forward secrecy in authentication protocols. J. Netw. Intell. (2022) 24. He, D., Zeadally, S., Kumar, N., Wei, W.: Efficient and anonymous mobile user authentication protocol using self-certified public key cryptography for multi-server architectures. IEEE Trans. Inf. Forensics Secur. 11(9), 2052–2064 (2016) 25. Braeken, A., Kumar, P., Liyanage, M., Hue, T.T.K.: An efficient anonymous authentication protocol in multiple server communication networks (EAAM). J. Supercomput. 74(4), 1695– 1714 (2018) 26. Ying, B., Nayak, A.: Lightweight remote user authentication protocol for multi-server 5g networks using self-certified public key cryptography. J. Netw. Comput. Appl. 131, 66–74 (2019) 27. Tomar, A., Dhar, J.: An ECC based secure authentication and key exchange scheme in multiserver environment. Wirel. Pers. Commun. 107(1), 351–372 (2019) 28. Wang, J., Zhu, Y., et al.: Secure two-factor lightweight authentication protocol using selfcertified public key cryptography for multi-server 5g networks. J. Netw. Comput. Appl. 161, 102660 (2020) 29. Kumar, P., Om, H.: A secure and efficient authentication protocol for wireless applications in multi-server environment. Peer-to-Peer Netw. Appl., 1–14 (2022)

Chapter 41

Secure Communication in Digital Twin-enabled Smart Grid Platform with a Lightweight Authentication Scheme Jiaxiang Ou, Mi Zhou, Houpeng Hu, Fan Zhang, Hangfeng Li, Fusheng Li, and Pengcheng Li Abstract The concepts of Digital Twin let the evolution of new energy services and more decentralized business models where people and energy industries are becoming essential participants in donating to smart grid sustainability dreams. In order to provide secure communication in a digital twin-enabled smart grid platform, we propose a lightweight authentication scheme. In our work, each smart meter and its related digital twin entity will calculate a shared session key for further use. The proposed scheme is indeed secure and efficient because we delivered a detailed security analysis and performance evaluation. Compared with other works, the proposed scheme is more suitable for creating the future smart grid industry.

J. Ou · H. Hu · H. Li · P. Li Guizhou Power Grid Co., Ltd., Guizhou, China e-mail: [email protected] H. Hu e-mail: [email protected] H. Li e-mail: [email protected] P. Li e-mail: [email protected] M. Zhou (B) · F. Zhang · F. Li Institute of Metrology Technology, Electric Power Research Institute, CSG, Guangzhou, China e-mail: [email protected] F. Zhang e-mail: [email protected] F. Li e-mail: [email protected] Guangdong Provincial Key Laboratory of Intelligent Measurement and Advanced Metering of Power Grid, Guangzhou, China © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_42

525

526

J. Ou et al.

41.1 Introduction With the continuous evolution of network technologies, Internet of Things (IoT) [5, 9, 33], and information engineering, Smart Grid [12, 13, 29] appears in people’s vision. A smart grid (SG) is an electrical grid to improve automation, communication, and connectivity among energy networks. The emergence of SG makes it possible for smart electricity meters to realize automatic meter reading, transfer, and payment. Obviously, a SG can solve the constraints of traditional power grids, such as centralized generation, manual monitoring, and one-way communication. In a SG infrastructure, security is considered one of the significant challenges because of its long-range communication on public networks. Various malicious behaviors and attacks have been described in [32, 34] recently. A malicious attacker can impersonate a legal user or device and eavesdrop on the transmissions among these devices. For these reasons, providing secure communication in a smart grid environment is essential [3, 23, 31]. On the other hand, a Digital Twin (DT ) [15, 20, 25, 28] is a real-time digital representation of a physical entity in the real world. More specifically, DT connecting the physical and virtual worlds lets data analysis and systems monitoring avoid problems before they occur or predict the future. It can reflect the life cycle of the corresponding physical equipment. Many researchers have recently tried integrating DT into a smart grid infrastructure [10, 22, 24]. A DT-enabled smart grid infrastructure can help electric companies process the user’s electricity meter information more efficiently. Besides, DT can simulate the household situation of real users and analyze the habits of different users using electricity from the model to improve power supply reliability. In conclusion, integrating DT with SG digitalizes various smart grid management processes such as energy production and consumption monitoring, load prediction, etc. In this paper, we establish secure communication in a digital twin-enabled smart grid platform. A secure platform should provide authenticity of all participants and confidentiality of messages transmitted online [17, 21]. Thus, we propose a lightweight authentication scheme for this kind of platform. In order to show that the proposed scheme is secure, we utilize a well-known Real-or-Random (ROR) model to demonstrate our work is provably secure. Compared with related schemes, the proposed scheme has great advantages in terms of communication and computation costs. The rest of the article is arranged as follows. Section 41.2 introduces related work. In Sect. 41.3, we define the system model used in this paper. Section 41.4 described the proposed scheme. Section 41.5 and Sect. 41.6 analyze the security and performance of our design. Finally Sect. 41.7, we summarize this paper.

41 Secure Communication in Digital Twin-enabled Smart Grid Platform …

527

41.2 Related Work With the development of IoT, security issues have received extensive attention. Recently, various schemes for protecting data security have been proposed [1, 2, 16, 19, 27, 30]. Amin et al. [1] proposed an anonymous monitoring system for wireless sensor networks (WSN). Wu et al. [30] proposed a lightweight two-factor authentication method for personalized systems based on WSN, which can keep users away from user tracking and resist simulated attacks. In 2017, Liu et al. [19] gave another user authentication method for WSN. The authors claimed that their approach could protect the data’s confidentiality while measuring body temperature, heart rate, and blood pressure and can resist existing attacks. However, Challa et al. [2] pointed out that Liu et al.’s scheme has security vulnerabilities and cannot resist privileged insider attacks, stolen smart card attacks, offline password guessing attacks, and user simulation attacks. They further use Elliptic Curve Cryptography (ECC) [6–8] to design an effective three-factor user authentication method. Unfortunately, Soni et al. [26] found that their method could not resist sensor capture attacks. In addition, their scheme has loopholes in the user registration phase. Therefore, Soni et al. proposed an improved authentication scheme. However, we think Soni et al.’s method consumed higher communication and computation costs.

41.3 System Model The system model is illustrated in Fig. 41.1. A digital twin server contains various digital twin entities. Each electric energy meter has a corresponding digital twin entity. Each digital twin entity stores user’s household electricity meter data through its corresponding energy meter. After collecting these data from its related to the energy meter, it can further establish a power model for the energy meter to make a timely analysis of users’ use habits. For security considerations, we must ensure electric energy meters and digital twin entities are legitimate. More specifically, all these participants should register with a trust registration authority server. This registration authority server not only provides the authenticity of meters and digital twin entities but also establishes secure communication between them. It means that this registration authority server helps them generate a session key for further use. Communication between each energy meter and its related digital twin entity will be protected by their session key.

41.4 Proposed Scheme In this paper, the proposed scheme includes four phases, the energy meter registration phase, the digital twin entity registration phase, the session key establishment phase,

528

J. Ou et al.

Fig. 41.1 System model Table 41.1 Notations used in this paper Notations RC E Mi DTi H I Di H P Wi D I Di D P Wi ri , r j , r g KR SK h(.) ⊕ A 

Descriptions Registration authority server The i-th energy meter, i = 1, 2, 3 ... The i-th digital twin entity E Mi ’s identity E Mi ’s password DTi ’s identity DTi ’s password Random numbers Master key of RC Session key Hash function XOR operation The attacker Concatenation operation

and the data transmission phase. All communication in the energy meter registration phase and digital twin entity registration phase are through a secure channel. On the other hand, communications in the session key establishment phase and data trasmission phase are via a public channel. The symbols used in this scheme are defined in Table 41.1.

41 Secure Communication in Digital Twin-enabled Smart Grid Platform … EMi Input HIDi , HP Wi compute SHP Wi = h(HIDi  HP Wi )

RC HID ,HP W

i −−−−i−−−−−→

H

Store {Hi , SHP Wi } into device

529

Compute Hi = h(HIDi  SHP Wi  KR ) store {HIDi , Hi } into memory

i ←−−

Fig. 41.2 Energy meter registration phase DTi Input DIDi , DP Wi compute SDP Wi = h(DIDi  HDP Wi )

RC DID ,SDP W

i −−−−i−−−−−−→

Hj

Store {Hj , SDP Wi } into device

Compute Hj = h(DIDi  SDP Wi  KR ) store {DIDi , Hj } into memory

←−−

Fig. 41.3 Digital twin entity registration phase

41.4.1 Energy Meter Registration Phase Before E Mi communicates with his DTi , E Mi must registeres with RC. E Mi first selects its own identity H I Di and password H P Wi , and then uses a hash function to obtain the pseudo-identity S H P Wi = h(H I Di  H P Wi ). Now E Mi transmits its identity H I Di and S H P Wi to RC. After receiving {H I Di , S H P Wi }, RC calculates Hi with its master key K R along with H I Di and S H P Wi . Then, RC stores H I Di and Hi in its memory. Finally, RC transmits Hi back to E Mi through a secure channel. Now, E Mi stores Hi and S H P Wi in its memory. This detailed process of this phase is also shown in Fig. 41.2.

41.4.2 Digital Twin Entity Registration Phase Similarly, all digital twin entities should register to RC. Assume that DTi desires to register to RC, DTi selects a unique identity D I Di and password D P Wi and utilizes a hash function to compute S D P Wi = h(D I Di  D P Wi ). After that, DTi transmits {D I Di , S D P Wi } to RC. After receiving {D I Di , S D P Wi }, RC uses its master key K R to compute H j = h(D I Di  S D P Wi  K R ) and then further stores {D I Di , H j } in its memory. Now RC sends H j back to DTi . At this time, DTi stores the received H j and S D P Wi in its memory. The procedure is shown in Fig. 41.3.

530

J. Ou et al.

EMi select a random number ri W1 = ri ⊕ Hi

RC

D Ti

{HID ,W1 }

i − −−−−− −−− →

select a random number rj M1 = rj ⊕ Hj {DID ,M1 }

ri = W1 ⊕ Hi rj = M1 ⊕ Hj select a random number rg SK = h(ri  rj  rg ) W2 = SK ⊕ Hi M2 = SK ⊕ Hj

i ← −−−−− −−− −

{M }

2 −−−− → {W }

SK = W2 ⊕ Hi

←−−2−− SK = M2 ⊕ Hj

Fig. 41.4 Session key establishment phase

41.4.3 Session Key Establishment Phase Assume that E Mi needs to communicate with its corresponding DTi , these steps are executed. 1. E Mi selects a random number ri and calculatess W1 = ri ⊕ Hi . Then, E Mi transmits {H I Di , W1 } to RC. 2. DTi selects another random number r j and computes M1 = r j ⊕ H j . Then, DTi transmits {D I Di , M1 } to RC. 3. After RC receives the message from E Mi and DTi , RC performs the XOR operation on W1 and M1 to obtain ri and r j where ri = W1 ⊕ Hi , r j = M1 ⊕ H j . Then, RC selects another random number r g to further compute S K = h(ri  r j  r g ). Now RC calculates W2 and M2 where W2 = S K ⊕ Hi and M2 = S K ⊕ H j . Finally, RC sends W2 and M2 to E Mi and DTi respectively. 4. E Mi can compute this session key S K where S K = W2 ⊕ Hi . Similarly, DT can obtain the session key S K by computing S K = M2 ⊕ H j . By now, the session key S K between E Mi and DTi has been established. Detailed steps are shown in Fig. 41.4.

41.4.4 Data Transmission Phase After E Mi and DTi have established the session key, communication between E Mi and DTi can be protected by the session key. E Mi can encrypt the sensitive data with this key, and DTi can obtain the sensitive data by decrypting the ciphertext with the same session key.

41 Secure Communication in Digital Twin-enabled Smart Grid Platform …

531

41.5 Formal Security Analysis Recently, security analysis based on Real-Or-Random (ROR) model has been widely used in many authentication schemes [4, 14, 18]. In this section, we also use the ROR model to prove the security of the scheme proposed in this paper. The proposed scheme defines three entities: digital twin entity DT , energy meter E M, and regy z x , HE M and H RC istration authority server RC. In the proof process, we define H DT y z x are the x-th DT , the y-th E M, and the z-th RC, D = {H DT , HE M , H RC }. A is an attacker who can fully access the public channel in polynomial time. Here we assume that A has the following functions. E xecute(D): This operation is a passive attack simulated by A, through which A can intercept messages transmitted by D through a public channel. Reveal(D): This operation simulates an active attack. With this operation, A can destroy the session key negotiated between various entities. Send(D, msg): By performing this operation, A can receive and send messages to various entities. Corr upt Device(E M): This operation simulates a power device stolen attack. Where A can get the information stored in the energy meter memory. T est (D): At the beginning of the game, a coin c with uniform texture is thrown, and the output result is only confidential to A. This result determines whether the T est query results are consistent. If c = 1, A executes the query and the session key is fresh, the session key is provided. If c = 0, a null value is returned. Theorem 1 We specify that A is the attacker who destroys the security of scheme P in polynomial time. We assume that S is the password dictionary and t is the length of the password dictionary. Then A has the advantage Adv PA of destroying the session key between E M and DT where Adv PA ≤ qh2 /|H ash| + qs /2t−1 |S|. Note that qs and qh represent the number of hash operations and Send query operations, and |H ash| and |S| represent the space size of S and size of hash operations. Proof In order to prove the above theorem, we define four games G i (i = 0, 1, 2, 3). In addition, Pr [SU CCi ] (i = 0, 1, 2, 3) represents A’s probability of winning in the game, and A’s advantage of winning in G i can be expressed as Adv PA ≤ qh2 /|H ash| + qs /2t−1 |S|. The specific process of proof is as follows. G 0 : The first game simulates the real attack of A choosing a random number c, and can get the results Adv PA =|2Pr [Succ0 ] − 1|. G 1 : This game introduces eavesdropping attacks on the basis of the first game. A can execute E xecute query and T est query. Then A checks whether the session key is a random number or a real session key. The session key S K = h(ri  r j  r g ) between E M and DT allows A to intercept information {H I Di , W1 }, {D I Di , M1 }, {M2 } and {W2 } through a public channel, but the intercepted information is not helpful for cracking the session key. Session key cracking also requires random numbers ri , r j and r g . This means that A’s probability of cracking the session key has not increased, so A’s probability of winning in G 1 and G 0 is the same. On this basis, the results can be obtained Pr [SU CC1 ] = Pr [SU CC1 ].

532

J. Ou et al.

G 2 : On the basis of G 1 , this operation introduces Send query and Hash operation. A can modify the received messages {H I Di , W1 }, {D I Di , M1 }, {M2 } and {W2 }. However, the long-term key K R of RC is protected by hash operation. In addition, the temporary random number set in the scheme makes the messages sent each time different. Therefore, the birthday paradox can be used to get the results |Pr [SU CC1 ] − Pr [SU CC2 ]| ≤ qh2 /2|H ash|. G 3 : A introduces the Corr upt Device operation on the basis of G 2 . By executing this operation, A can obtain the information {S D P Wi } stored in the memory of the E M. In addition, with the help of the data dictionary, A can try to guess the user’s password H P Wi , and use the obtained information S D P Wi to verify the password. However, the number of times A guesses the user’s password is limited, so the probability that A can guess successfully is 1/2t . In addition, H j is required to obtain the random number, but H j is stored in the RC memory, so A cannot successfully obtain the session key between DT and E M. So we can get the result |Pr [SU CC2 ] − Pr [SU CC3 ]| ≤ qs /2t |S|. In addition, since the session key between DT and E M is randomly generated, A does not know the bit of c, so we can get the result |Pr [SU CC3 ] = 1/2. According to the above conclusions, we can draw a conclusion 1/2 Adv PA = |Pr [SU CC0 ] − 1/2| = |Pr [SU CC0 ] − Pr [SU CC3 ]| = |Pr [SU CC1 ] − Pr [SU CC3 ]| ≤ qh2 /2|H ash| + qs /2t |S|. Finally, we can successfully prove the correctness of Theorem 1 Adv PA ≤ qh2 / |H ash| + qs /2t−1 |S|.

41.6 Performance Analysis This section analyzes the performance of the proposed scheme in terms of computation and communication costs. Here we refer to the experiment results from [26]. The time Th consumed by a single hash operation is 0.005 s, the time Tecm needed for a elliptic curve point multiplication operation is 0.063075 s, the time bp for computing a bilinear pairing operation is 0.189225 s, and the time consumed by fuzzy extractor operation f e is 0.063075 s. We also assume that the length of an identity, a password, a random number, and a timestamp are all 64 bits, the length of a hash function’s output is 160 bits, and the length of elliptic curve point multiplication is 320 bits. We analyze and compare the proposed scheme with five other related methods. Results are listed in Table 41.2. Das et al.’s scheme [11] uses hash function operations, elliptic curve point multiplication operations, and fuzzy extractor operations. It’s scheme totally consumes 0.330875 s. Soni et al.’s scheme [26] is another method that uses hash operation, elliptic curve point multiplication operation, and fuzzy extractor operation. This scheme totally consumes 0.4570 s. The computation costs of Wu et al.’s scheme, Liu et al.’s scheme, and Challa et al.’s scheme are 0.16 s, 0.2563 s, 0.2618 s, respectively. In the proposed scheme, we only use the hash func-

41 Secure Communication in Digital Twin-enabled Smart Grid Platform …

533

Table 41.2 Computational and communicational costs comparison Scheme Computation total cost (s) Communication cost (bits) Das et al. [11] Wu et al. [30] Liu et al. [19] Challa et al. [2] Soni et al. [26] Our

31Th + 4Tecm + f e ≈ 0.330875 32Th ≈ 0.16 8Th + Tecm + 3bp ≈ 0.2563 19Th + 3Tecm + f e ≈ 0.2618 31Th + 6Tecm + f e ≈ 0.4570 Th ≈ 0.005

2944 1731 1408 1536 2464 1216

Fig. 41.5 Computation time

tion operation, so the time cost consumed is less than 0.005 s. We can observe from Fig. 41.5 that our design’s computation cost is far lower than other schemes. Here we consider the communication costs in the session key establishment phase. It can be seen from Table 41.2 that the communication cost of the proposed work is only 1216 bits. Compared with other schemes, the communication cost of Das et al.’s scheme [11], Wu et al.’s scheme [30], Liu et al.’s scheme, Challa et al.’s scheme [2], and Soni et al.’s scheme [26] are 2944, 1731, 1408, 1536, and 2464 bits respectively. We can also observe from Fig. 41.6 that the communication cost of the proposed scheme is far lower than other schemes. To sum up, our scheme has great advantages both in terms of time cost and communication cost.

534

J. Ou et al.

Fig. 41.6 Communication costs

41.7 Conclusion This paper proposed a way to establish secure communication in a digital twinenabled smart grid platform with a lightweight authentication scheme. In our work, a lightweight authentication scheme is described. A session key is generated between a smart meter and its associate digital twin entity. To prove the security of our work, we use a well-known ROR model. With this model, we can claim that our design is provably secure. Experimental results also show that the proposed scheme performs better in communication and computation. Therefore, this scheme is more suitable for developing the future smart grid industry. Acknowledgements This research was partially supported by National Key Research and Development Program of China (Grant Nos. 2019YFE0118700), Science and Technology Project of China Southern Power Grid Corporation (Grant No. 066600KK52200016).

References 1. Amin, R., Islam, S.H., Biswas, G., Khan, M.K., Kumar, N.: A robust and anonymous patient monitoring system using wireless medical sensor networks. Futur. Gener. Comput. Syst. 80, 483–495 (2018) 2. Challa, S., Das, A.K., Odelu, V., Kumar, N., Kumari, S., Khan, M.K., Vasilakos, A.V.: An efficient ECC-based provably secure three-factor user authentication and key agreement protocol

41 Secure Communication in Digital Twin-enabled Smart Grid Platform …

535

for wireless healthcare sensor networks. Comput. Electr. Eng. 69, 534–554 (2018) 3. Chen, C.M., Chen, L., Huang, Y., Kumar, S., Wu, J.M.T.: Lightweight authentication protocol in edge-based smart grid environment. EURASIP J. Wirel. Commun. Netw. 2021(1), 1–18 (2021) 4. Chen, C.M., Chen, Z., Kumari, S., Lin, M.C.: Lap-IOHT: a lightweight authentication protocol for the internet of health things. Sensors 22(14), 5401 (2022) 5. Chen, C.M., Deng, X., Gan, W., Chen, J., Islam, S.: A secure blockchain-based group key agreement protocol for IoT. J. Supercomput. 77(8), 9046–9068 (2021) 6. Chen, J.N., Huang, Z.J., Zhou, Y.P., Zou, F.M., Chen, C.M., Wu, J.M.T., Wu, T.Y.: Efficient certificate-based aggregate signature scheme for vehicular ad hoc networks. IET Netw. 9(6), 290–297 (2020) 7. Chen, J.N., Zhou, Y.P., Huang, Z.J., Wu, T.Y., Zou, F.M., Tso, R.: An efficient aggregate signature scheme for healthcare wireless sensor networks. J. Netw. Intell. 6(1), 1–15 (2021) 8. Chen, J., Zou, F., Wu, T.Y., Zhou, Y.P.: A new certificate-based aggregate signature scheme for wireless sensor networks. J. Inf. Hiding Multimed. Signal Process. 9(5), 1264–1280 (2018) 9. Chen, Y.Q., Zhou, B., Zhang, M., Chen, C.M.: Using IoT technology for computer-integrated manufacturing systems in the semiconductor industry. Appl. Soft Comput. 89, 106065 (2020) 10. Danilczyk, W., Sun, Y.L., He, H.: Smart grid anomaly detection using a deep learning digital twin. In: 2020 52nd North American Power Symposium (NAPS), pp. 1–6. IEEE (2021) 11. Das, A.K., Kumari, S., Odelu, V., Li, X., Wu, F., Huang, X.: Provably secure user authentication and key agreement scheme for wireless sensor networks. Secur. Commun. Netw. 9(16), 3670– 3687 (2016) 12. Ekanayake, J.B., Jenkins, N., Liyanage, K., Wu, J., Yokoyama, A.: Smart Grid: Technology and Applications. Wiley (2012) 13. Fang, X., Misra, S., Xue, G., Yang, D.: Smart grid-the new and improved power grid: a survey. IEEE Commun. Surv. Tutor. 14(4), 944–980 (2011) 14. Ge, M., Kumari, S., Chen, C.M.: AUTHPFS: a method to verify perfect forward secrecy in authentication protocols. J. Netw. Intell. 7(3), 734–750 (2022) 15. Jones, D., Snider, C., Nassehi, A., Yon, J., Hicks, B.: Characterising the digital twin: a systematic literature review. CIRP J. Manuf. Sci. Technol. 29, 36–52 (2020) 16. Li, W., Li, B., Zhao, Y., Wang, P., Wei, F.: Cryptanalysis and security enhancement of three authentication schemes in wireless sensor networks. Wirel. Commun. Mob. Comput. 2018 (2018) 17. Li, X., Liu, S., Kumari, S., Chen, C.M.: PSAP-WSN: a provably secure authentication protocol for 5g-based wireless sensor networks. CMES-Comput. Model. Eng. Sci. 135(1), 711–732 (2023) 18. Li, Z., Miao, Q., Chaudhry, S.A., Chen, C.M.: A provably secure and lightweight mutual authentication protocol in fog-enabled social internet of vehicles. Int. J. Distrib. Sens. Netw. 18(6), 15501329221104332 (2022) 19. Liu, C.H., Chung, Y.F.: Secure user authentication scheme for wireless healthcare sensor networks. Comput. Electr. Eng. 59, 250–261 (2017) 20. Liu, M., Fang, S., Dong, H., Xu, C.: Review of digital twin about concepts, technologies, and industrial applications. J. Manuf. Syst. 58, 346–361 (2021) 21. Liu, S., Chen, C.M.: Comments on “a secure and lightweight drones-access protocol for smart city surveillance”. IEEE Trans. Intell. Transp. Syst. (2022) 22. Lopez, J., Rubio, J.E., Alcaraz, C.: Digital twins for intelligent authorization in the b5g-enabled smart grid. IEEE Wirel. Commun. 28(2), 48–55 (2021) 23. Luo, Y., Zheng, W., Chen, Y.C.: An anonymous authentication and key exchange protocol in smart grid. J. Netw. Intell. 6(2), 206–215 (2021) 24. Melillo, A., Linder, E., Barahona, B., Schuetz, P.: Statistical analysis of 200 digital twins for thermal load of swiss buildings created from smart grid monitoring data. In: Journal of Physics: Conference Series, vol. 2042, p. 012009. IOP Publishing (2021) 25. Semeraro, C., Lezoche, M., Panetto, H., Dassisti, M.: Digital twin paradigm: a systematic literature review. Comput. Ind. 130, 103469 (2021)

536

J. Ou et al.

26. Soni, P., Pal, A.K., Islam, S.H.: An improved three-factor authentication scheme for patient monitoring using WSN in remote health-care system. Comput. Methods Progr. Biomed. 182, 105054 (2019) 27. Srinivas, J., Mishra, D., Mukhopadhyay, S.: A mutual authentication framework for wireless medical sensor networks. J. Med. Syst. 41(5), 1–19 (2017) 28. Tao, F., Zhang, H., Liu, A., Nee, A.Y.: Digital twin in industry: state-of-the-art. IEEE Trans. Ind. Inf. 15(4), 2405–2415 (2018) 29. Tuballa, M.L., Abundo, M.L.: A review of the development of smart grid technologies. Renew. Sustain. Energy Rev. 59, 710–725 (2016) 30. Wu, F., Li, X., Sangaiah, A.K., Xu, L., Kumari, S., Wu, L., Shen, J.: A lightweight and robust two-factor authentication scheme for personalized healthcare systems using wireless medical sensor networks. Futur. Gener. Comput. Syst. 82, 727–737 (2018) 31. Wu, T.Y., Lee, Y.Q., Chen, C.M., Tian, Y., Al-Nabhan, N.A.: An enhanced pairing-based authentication scheme for smart grid communications. J. Ambient Intell. Hum. Comput. (2021). https://doi.org/10.1007/s12652-020-02740-2 32. Wu, T.Y., Meng, Q., Yang, L., Kumari, S., Nia, M.P.: Amassing the security: an enhanced authentication and key agreement protocol for remote surgery in healthcare environment. Comput. Model. Eng. Sci. 134(1), 317–341 (2023) 33. Wu, T.Y., Wang, L., Guo, X., Chen, Y.C., Chu, S.C.: SAKAP: SGX-based authentication key agreement protocol in IoT-enabled cloud computing. Sustainability 14(17), 11054 (2022) 34. Yang, L., Chen, Y.C., Wu, T.Y.: Provably secure client-server key management scheme in 5g networks. Wirel. Commun. Mob. Computin. 2021, 4083199 (2021)

Chapter 42

A Secure Authentication Scheme for Smart Home Based on Trusted Execution Environment Houpeng Hu, Jiaxiang Ou, Bin Qian, Yi Luo, Yanhong Xiao, and Zerui Chen

Abstract With the rapid growth of the smart home, remote control of devices within a smart home environment has become a primary and essential function. However, there are significant security vulnerabilities in the process of remote control. This paper proposes a secure authentication scheme based on a trusted execution environment for smart home remote control. The proposed scheme generates a session key between smart home devices and users. This session key ensures that a user can securely control devices remotely and resist various well-known attacks. We also utilize the Real-or-Random model to demonstrate our scheme is provably secure. Besides, our work has lower computation and communication costs than other related methods.

H. Hu · J. Ou · Y. Xiao · Z. Chen Guizhou Power Grid Co., Ltd., Guizhou, China e-mail: [email protected] J. Ou e-mail: [email protected] Y. Xiao e-mail: [email protected] Z. Chen e-mail: [email protected] B. Qian · Y. Luo (B) Institute of Metrology Technology, Electric Power Research Institute, CSG, Guangzhou, China e-mail: [email protected] B. Qian e-mail: [email protected] Guangdong Provincial Key Laboratory of Intelligent Measurement and Advanced Metering of Power Grid, Guangzhou, China © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_43

537

538

H. Hu et al.

42.1 Introduction With the constant advance of the financial level and the continuous development of information technology, people’s requirements for quality of life are also growing [9]. In particular, it puts forward higher needs for the home environment, closely related to our life. The emergence of wireless communications and the Internet of Things (IoT) [4, 6, 26] meets people’s needs and makes the traditional home more intelligent [10, 13, 18]. The smart home can form a comprehensive system of home appliances through the network, which can be managed systematically and make our lifestyle more convenient and efficient. Smart home, also known as home automation, connects all kinds of equipment in the house, such as lighting, sound, air conditioning, ventilator, alarm, electric curtains, sensors, and various other household appliances, through a dedicated network. All these devices can be triggered by receiving sensing signals wirelessly [21]. However, all these conveniences come at a price. Smart home devices are prone to many security vulnerabilities, such as identity theft, password utilization, location tracking, burglary, equipment or property damage, and data manipulation [3, 19, 25]. An authentication scheme is the core strategy to ensure information and network security [5, 15, 17]. A standard authentication scheme provides two functionalities [1]. First, it provides mutual authentication. All parties in the network should be authenticated. Second, it calculates a session key. All participants can use this session key to encrypt every message submitted via public channels. A secure authentication scheme should satisfy the following security needs: 1. Resist offline password guessing attacks: An attacker cannot calculate a user’s password even if he obtains all messages transmitted publicly. 2. Resist sensor capture attacks: Even if an attacker compromises a sensor deployed in a smart home, he still cannot get the session key. 3. Perfect forward security (PFS): PFS indicates that the leakage of a long-term key does not result in the leakage of a previous short-term key. 4. Resist replay attacks: An attacker continuously sends instructions to the sensor or user according to the intercepted public channel information, resulting in a waste of system resources. 5. Resist user simulation attacks: An attacker can impersonate a legitimate user and establish communication with a device according to the information intercepted in the public channel or retrieved from a captured device, then carry out illegal acts. Recently, various authentication schemes for smart home environments have been proposed. However, some of these schemes have been proven insecure. In this paper, we offer a novel Trusted Execution Environment (TEE)-based authentication scheme. With the TEE, the proposed scheme can provide a higher security level. Besides, this solution has better time and communication efficiency than related works.

42 A Secure Authentication Scheme for Smart Home …

539

The rest of the work is organized as follows. Section 42.2 provides a summary of previous work. In Sect. 42.3, we define the system model and adversary mode, and in Sect. 42.4, we describe the proposed scheme. Sections 42.5 and 42.6 evaluate the performance of our design. Finally, Sect. 42.8 concludes.

42.2 Related Work A smart home lets us control and receive information remotely, bringing great convenience. However, a smart home environment would have some security problems with the remote control of smart homes. Because of such defects, researchers have proposed various authentication schemes to solve these security problems. In 2012, Jeong et al. [11] suggested a user authentication scheme. Their proposed scheme can authenticate smart home users and allow users to deliver real-time permission control in a secure home network. However, this scheme is sent in clear text while transmitting user identity. Besides, this scheme cannot provide mutual authentication. Vaidya et al. [24] proposed an efficient authentication scheme based on strong passwords, which can provide users with the function of remote control of a smart home. However, Kim et al. [12] discovered their scheme [5] is not secure and does not offer PFS and user anonymity. Therefore, Kim et al. proposed an enhanced scheme. This scheme can deploy new applications to the system at runtime. In 2013, Li et al. [16] proposed a lightweight key agreement scheme for a smart home energy management system. However, their scheme is inconvenient for managing key certificates and cannot provide mutual authentication. In 2015, Santoso et al. [22] proposed a method combining strong security when deploying the IoT for smart home systems. Unfortunately, the proposed scheme cannot resist privileged internal attacks and cannot provide user anonymity. Also, in 2015, Kumar et al. [14] described a key agreement scheme. Unfortunately, their scheme cannot give the users untraceability, and anonymity attributes [20]. Naoui et al. [20] also proposed a new smart home authentication scheme. Fakroon et al. [8] described an anonymous remote control smart home scheme, which avoids the problem of clock synchronization. In addition, Dey et al. [7] also described an identity authentication scheme in the smart home environment.

42.3 System Model 42.3.1 TrustZone Model This scheme uses the Trusted Execution Environment (TEE) in the TrustZone architecture to design a secure authentication scheme for the remote control of smart homes. TrustZone is a security procedure based on hardware. Adjusting the initial

540

H. Hu et al.

Fig. 42.1 TrustZone architecture

hardware architecture introduces two protection domains with separate permissions at the processor level—a general world and a secure world. The secure world always uses a secure TEE kernel to provide a trusted environment where confidential data can be stored and accessed. In this way, even if the operating system in the general world is damaged or invaded (for example, iOS has been jailbroken or Android has been rooted), hackers still cannot obtain confidential data stored in TEE. Figure 42.1 shows the TrustZone architecture adopted on cortex-A and cortex-M.

42.3.2 Remote Control Smart Home Model Figure 42.2 is a schematic architecture of the proposed scheme. Suppose a user wants to use his smartphone to remotely control a smart device deployed in a smart home, such as electric lights, video cameras, and air conditioners. In that case, this user needs to establish a session key in advance with this device and then conduct a secure communication. In addition, it should be noted that the TrustZone architecture adopted on cortex-A is embedded in these devices and smartphones used by users. TrustZone provides a TEE system through a secure world isolated by hardware. By storing users’ confidential data in TEE built based on TrustZone, even if hackers fully control the operating system in Rich Execution Environment (REE), they cannot obtain these confidential data. It means that devices and users’ smartphones can store secret information in TEE. In our design, before a user establishes a session key with the device, the user must first register with this device. After finishing the registration phase, this device and smartphone store sensitive information in their TEE.

42 A Secure Authentication Scheme for Smart Home …

541

Fig. 42.2 Remote control smart home model

Table 42.1 Notations

Notations

Descriptions

Ui Dj I Di , P Wi Ks Ti SK h(·) ⊕ 

i-th user j-th device Ui ’s identity, password Master key of D j Timestamps Session key Hash function Bitwise XOR functions Concatenation

42.3.3 Attacker Model In this paper, we suppose that an attacker would have the following abilities: 1. Attackers can intercept, modify, replay, delete, and insert information transmitted through the public channel. 2. Attackers can obtain the information stored in the user’s intelligent device. 3. Attackers can capture devices deployed in smart homes and extract information stored in these devices. 4. Attackers cannot compromise information stored in TEE.

42.4 Proposed Scheme Our scheme includes a system initialization phase, a user registration phase, a session key establishment phase, and a user remote control phase. The specific steps of each phase are shown below. Notations used in this paper are defined in Table 42.1.

542

H. Hu et al.

Fig. 42.3 User registration phase

42.4.1 System Initialization Phase All smart devices should be initialized before being deployed in a smart home environment. Besides, users’ smartphones also need to be initialized. In this phase, a TrustZone architecture adopted on cortex-A is embedded into smartphones and devices. As we mentioned above, TEE in a TrustZone architecture can be used to store private and sensitive information.

42.4.2 User Registration Phase Suppose that Ui wishes to register with a device D j . This phase is performed. Messages submitted in this phase are via a secure channel. Step 1: Ui generates an identity I Di and password P Wi . Ui uses his smartphone to send a registration request along with I Di and P Wi to D j . Step 2: After receiving the registration request from Ui , D j first computes U I Di = h(I Di ||K s ) with his master key K s , generates a random number R I Di , and then computes V = h(I Di ||P Wi ||U I Di ). After that, D j stores R I Di and U I Di in the TEE system, and transmits R I Di , V , and U I Di to Ui . Step 3: After receiving the information from D j , Ui stores the R I Di , V , and U I Di in the TEE system embedded in his smartphone. The detailed steps are also shown in Fig. 42.3.

42.4.3 Session Key Establishment Phase This phase is performed if Ui and D j want to establish a session key. Detailed steps are illustrated in Fig. 42.4. Messages in this phase are via a public channel. Step 1: Ui first inputs I Di and P Wi to the smartphone, and then the smartphone calculates V  = h(I Di ||P Wi ||U I Di ) according to the identity, password, and U I Di stored in the TEE system. The smartphone compares the calculated V  with V stored in the TEE system to check whether it is consistent. If it is consistent, it indicates that it is a legal user; otherwise, the operation will be terminated. If Ui is a legitimate

42 A Secure Authentication Scheme for Smart Home …

543

Fig. 42.4 Session key establishment phase

user, the smartphone generates a timestamp T1 , calculates A1 = h(U I Di ||T1 ), and then transmits R I Di , A1 , and timestamp T1 to D j . Step 2: After receiving the message from Ui , D j first verifies whether the value of the timestamp Tc − T1 is within the legal time threshold. If it is within the set threshold range, it indicates that it has not suffered replay attacks. After verifying the validity of the timestamp, D j finds the matching U I Di in the TEE system according to the received R I Di , and then generates A1 = h(U I Di ||T1 ) according to the received timestamp T1 and U I Di . D j compares the generated A1 with the received A1 to check whether it is consistent. Now D j sends a successful request to Ui to remind Ui that the shared session key S K is U I Di . Step 3: Ui set U I Di as a session key which is shared with D j .

42.4.4 User Remote Control Smart Home Phase Once Ui establishes a shared session key with D j , they can use this session key to transmit the messages securely. Ui can encrypt the operation with the session key and then transmit it to D j . D j uses the session key to decrypt and get the operation that Ui wants to perform.

42.5 Security Analysis Here, we use the Real-Or-Random (ROR) model to demonstrate the security of our y design. We use IUx i and I D j to represent the x-th user and the y-th device, respectively, y where T = {IUx i , I D j }. The following are the capabilities that an attacker A has:

544

H. Hu et al.

1. E xecute(T ): With this query, A can intercept, delete, alter, and insert messages transmitted via a public channel. 2. Send(T, m): With this query, A can receive messages from Ui and D j , and send messages to Ui or D j . 3. Corr upt Smar t phone(IUx i ): With this query, A can obtain the information stored in Ui ’s smartphone. y 4. Corr upt Device(I D j ): With this query, A can get the sensitive information in the memory of D j . 5. T est (T ): With this query, A can destroy the security of the session key. During each session, A can only perform the T est operation once. After receiving the T est query request from A, Oracle will throw a coin c. If c = 1, it returns the session key S K between Ui and D j ; otherwise, it returns a random number between 0 and 1. Theorem 1 If Adv AP ≤ , the advantage of A in breaking a scheme P can be ignored, then we conclude that P is secure. In the ROR model, if A can break the semantic security of the scheme P, then the advantage of A breaking P is q2 qs Adv AP ≤ |Hh | + 2m−1 where C represents Ui ’s password dictionary, q represents |C| the number of Ui ’s passwords, qh , qs , |H |, and |C| respectively represent the number of hash queries, the number of send queries, the space size of hash queries, and the space size of Ui ’s password dictionary C. Proof We specify Game0 to Game3 to simulate the attack process of A. In the proof process, SU CCi represents A’s winning in each game, and Pr [SU CCi ] represents A’s probability of winning in each game. Adv AP represents A to undermine the advantages of our proposed scheme. The detailed simulation process starts from Game0 . Game0 : In this first game, A only performs the operation of selecting byte b, so Adv AP = |Pr [SU CC0 ] − 1| Game1 : Based on the first game, A can perform E xecute(T ). If A wants to obtain S K between Ui and D j , A must first perform the T est (T ) operation. In our proposed scheme, S K = U I Di , but A can only intercept the information R I Di , A1 , and T1 transmitted through a public channel, so A cannot really obtain S K between Ui and D j . Therefore, compared with Game0 , Game1 does not increase the possibility of A winning the game, so |Pr [SU CC1 ]| = |Pr [SU CC0 ]| Game2 : On the basis of the previous game, we introduced hash query and send query operations. A can tamper with the information intercepted through a public channel and retransmit the tampered information to each entity. Tampering with the forgery A1 requires a hash operation. However, the hash operation is unidirectional and noncolliding, so it is very difficult to forge A1 using the hash operation. Therefore, based

42 A Secure Authentication Scheme for Smart Home …

545

on the birthday paradox, we can conclude that |Pr [SU CC1 ] − Pr [SU CC2 ]| ≤

qh2 2 H| y

Game3 : In this game, Corr upt Smar t Phone(IUx i ) and Corr upt Device(I D j ) will be executed by A. However, the Ui ’s smartphone and device do not store any useful information, so A cannot obtain the information related to S K . Besides, with the help of Ui ’s password dictionary, A can guess the Ui ’s login password. The probability that A can successfully guess Ui ’s password is 1/2m . If A cannot obtain Ui ’s password, then A cannot disguise as a legitimate user to establish a key with D j . If the system can only allow A to try a limited number of password input times, then we can get |Pr [SU CC2 ] − Pr [SU CC3 ]| ≤

qs . m 2 |C|

Because A does not know whether the guessed byte b is generated by Ui or D j as the session key, we have 1 |Pr [SU CC3 ]| = 2 According to the formula mentioned above, we can draw a conclusion: 1 Adv AP = |Pr [SU CC0 ] − 1/2| 2 = Pr [SU CC1 ] − Pr [SU CC3 ]| ≤ |Pr [SU CC1 ] − Pr [SU CC2 ]| + |Pr [SU CC2 ] − Pr [SU CC3 ]| =

qh2 qs + 2 H | 2m |C|

Therefore, we prove that theorem 1 Adv AP ≤ posed scheme is secure.

qh2 |H |

+

qs 2m−1 |C|

is correct, so our pro-

42.6 Performance Evaluation Now we compare the communication cost and running time of the proposed scheme with other related works [2, 7, 8, 20, 23, 27]. To ensure the preciseness of the experiment, we refer to Soumya et al.’s paper [2] for some data required for the experiment. In the communication process, Elliptic Curve Cryptography (ECC) point multiplication requires 320 bits, the single hash operation requires 160 bits, and the random number and identity require 128 bits. In the process of comparing the running time, the time required for bilinear pairing Tbp is 32.713 ms, the time required for

546

H. Hu et al.

Fig. 42.5 Communication costs

dot multiplication Tm is 13.405 ms, and the time required for biological information fuzzy extraction Tb is 13.405 ms, and the time required for single hash operation Th is 0.056 ms.

42.6.1 Communication Cost In our proposed scheme, each entity exchanges two sets of messages during Ui and D j login authentication phase. Ui sends R I Di , A1 , and T1 to D j , and D j transmits a key establishment SU CC E SS message to Ui . The communication cost required by the pseudo-identity R I Di of Ui is 128 bits. We compare the proposed scheme with other relevant schemes in Table 42.2. Shuai et al. [23] exchanged 4 groups of messages with a communication cost of 296 bits, Yu et al. [27] exchanged 8 groups of messages with a communication cost of 1072 bits, Naoui et al. [20] exchanged 3 groups of messages with a communication cost of 212 bits, Fakroon et al. [8] exchanged 3 groups of messages with a communication cost of 288 bits, and Dey et al. [7] exchanged 5 groups of messages. The total communication cost is 420 bits. Banerjee et al. [2] exchanged 4 messages, and the required communication cost is 236 bits. In conclusion, we can see that the communication cost of other relevant schemes is higher than that of our proposed scheme. Therefore, our proposed scheme has a great advantage in communication cost. Figure 42.5 more intuitively shows the communication comparison process between various schemes.

42 A Secure Authentication Scheme for Smart Home … Table 42.2 Communication cost comparison Scheme No. of bytes Shuai et al.’s Yu et al.’s Naoui et al.’s Fakroon et al.’s Dey et al.’s Banerjee et al.’s Proposed

296 1072 212 288 420 236 128

547

No. of messages 4 8 3 4 5 4 2

Fig. 42.6 Computation costs

42.7 Computation Cost In our proposed scheme, Ui only uses two hash operations in the login and authentication phase, and D j only uses two hash operations. Therefore, the running time of our proposed scheme is 0.618 ms. Comparisons with other relevant schemes are shown in Table 42.3. The running time required for Shuai et al.’s scheme [23] is 27.604 ms, the running time for Yu et al.’s scheme [27] is 762.343 ms, the running time for Naoui et al. [20] is 68.332 ms, the running time for Fakroon et al. [8] is 1.848 ms, the running time for Dey et al.’s scheme [7] is 63.954 ms, and the running time for Banerjee et al.’s scheme [2] is 14.749 ms. In conclusion, it can be seen that the computation of our work is lower than other relevant schemes. Therefore, our proposed scheme has great advantages in running time. Figure 42.6 more intuitively shows the running time comparison process between various schemes.

548

H. Hu et al.

Table 42.3 Computation cost comparison Schemes User (ms) Gateway (ms) [23] [27] [20] [8] [7] [2] Ours

6Th + 1Tm 7Th + 14Tm 12Th + 2Tm + 3Tsym 4Th 4Th + 2Tm + 3Tsym 10Th + 1Tb 2Th

7Th + 1Tm 12Th + 19Tm + 4Tbp 13Th + 2Tm + 4Tsym 5Th ≈0 10Th ≈0

Device (ms)

Total cost (ms)

3Th 7Th + 14Tm 1Th + 1Tsym 24Th 3Th + 2Tm + 3Tsym 4Th 1Th

≈27.604 ≈762.343 ≈68.332 ≈1.848 ≈63.954 ≈14.749 ≈0.168

42.8 Conclusion This paper presents a lightweight security scheme for remote control of smart homes based on Trusted Execution Environment (TEE). This scheme is designed in combination with the TEE system. Compared with other related schemes, the scheme proposed in this paper is more efficient and secure. It has great advantages in communication time and running time. In addition, the resource cost is reduced in the user registration phase and the key establishment phase. In general, our proposed scheme is superior to other relevant schemes and more suitable for resource-constrained smart home devices. Acknowledgements This research was partially supported by National Key Research and Development Program of China (Grant No. 2019YFE0118700), and Science and Technology Project of China Southern Power Grid Corporation (Grant No. 066600KK52200016).

References 1. Amintoosi, H., Nikooghadam, M., Kumari, S., Kumar, S., Chen, C.M.: TAMA: three-factor authentication for multi-server architecture. Hum.-Centric Comput. Inf. Sci. 11 (2021) 2. Banerjee, S., Odelu, V., Das, A.K., Chattopadhyay, S., Park, Y.: An efficient, anonymous and robust authentication scheme for smart home environments. Sensors 20(4), 1215 (2020) 3. Bao, Z., Shi, W., Kumari, S., Kong, Z.Y., Chen, C.M.: Lockmix: a secure and privacy-preserving mix service for bitcoin anonymity. Int. J. Inf. Secur. 19(3), 311–321 (2020) 4. Chen, C.M., Deng, X., Gan, W., Chen, J., Islam, S.: A secure blockchain-based group key agreement protocol for IoT. J. Supercomput. 77(8), 9046–9068 (2021) 5. Chen, C.M., Xiang, B., Wang, K.H., Yeh, K.H., Wu, T.Y.: A robust mutual authentication with a key agreement scheme for session initiation protocol. Appl. Sci. 8(10), 1789 (2018) 6. Chen, Y.Q., Zhou, B., Zhang, M., Chen, C.M.: Using IoT technology for computer-integrated manufacturing systems in the semiconductor industry. Appl. Soft Comput. 89, 106065 (2020) 7. Dey, S., Hossain, A.: Session-key establishment and authentication in a smart home network using public key cryptography. IEEE Sens. Lett. 3(4), 1–4 (2019) 8. Fakroon, M., Alshahrani, M., Gebali, F., Traore, I.: Secure remote anonymous user authentication scheme for smart home environment. Internet Things 9, 100158 (2020)

42 A Secure Authentication Scheme for Smart Home …

549

9. Gomez, C., Paradells, J.: Wireless home automation networks: a survey of architectures and technologies. IEEE Commun. Mag. 48(6), 92–101 (2010) 10. Gubbi, J., Buyya, R., Marusic, S., Palaniswami, M.: Internet of things (IoT): a vision, architectural elements, and future directions. Futur. Gener. Comput. Syst. 29(7), 1645–1660 (2013) 11. Jeong, J., Chung, M.Y., Choo, H.: Integrated OTP-based user authentication scheme using smart cards in home networks. In: Proceedings of the 41st Annual Hawaii International Conference on System Sciences (HICSS 2008), pp. 294–294. IEEE (2008) 12. Kim, H.J., Kim, H.S.: AUTHHOTP -HOTP based authentication scheme over home network environment. In: International Conference on Computational Science and Its Applications, pp. 622–637. Springer (2011) 13. Kim, J.E., Boulos, G., Yackovich, J., Barth, T., Beckel, C., Mosse, D.: Seamless integration of heterogeneous devices and access control in smart homes. In: 2012 Eighth International Conference on Intelligent Environments, pp. 206–213. IEEE (2012) 14. Kumar, P., Gurtov, A., Iinatti, J., Ylianttila, M., Sain, M.: Lightweight and secure session-key establishment scheme in smart home environments. IEEE Sens. J. 16(1), 254–264 (2015) 15. Kumari, A., Kumar, V., Abbasi, M.Y., Kumari, S., Chaudhary, P., Chen, C.M.: CSEF: cloudbased secure and efficient framework for smart medical system using ECC. IEEE Access 8, 107838–107852 (2020) 16. Li, Y.: Design of a key establishment protocol for smart home energy management system. In: 2013 Fifth International Conference on Computational Intelligence, Communication Systems and Networks, pp. 88–93. IEEE (2013) 17. Li, Z., Miao, Q., Chaudhry, S.A., Chen, C.M.: A provably secure and lightweight mutual authentication protocol in fog-enabled social internet of vehicles. Int. J. Distrib. Sens. Netw. 18(6), 15501329221104332 (2022) 18. Liu, S., Chen, C.M.: Comments on “a secure and lightweight drones-access protocol for smart city surveillance”. IEEE Trans. Intell. Transp. Syst. (2022) 19. Mei, Q., Xiong, H., Chen, Y.C., Chen, C.M.: Blockchain-enabled privacy-preserving authentication mechanism for transportation cps with cloud-edge computing. IEEE Trans. Eng. Manag. (2022) 20. Naoui, S., Elhdhili, M.H., Saidane, L.A.: Novel smart home authentication protocol LRP-shap. In: 2019 IEEE Wireless Communications and Networking Conference (WCNC), pp. 1–6. IEEE (2019) 21. Renuka, K., Kumar, S., Kumari, S., Chen, C.M.: Cryptanalysis and improvement of a privacypreserving three-factor authentication protocol for wireless sensor networks. Sensors 19(21), 4625 (2019) 22. Santoso, F.K., Vun, N.C.: Securing iot for smart home system. In: 2015 International Symposium on Consumer Electronics (ISCE), pp. 1–2. IEEE (2015) 23. Shuai, M., Yu, N., Wang, H., Xiong, L.: Anonymous authentication scheme for smart home environment with provable security. Comput. Secur. 86, 132–146 (2019) 24. Vaidya, B., Park, J.H., Yeo, S.S., Rodrigues, J.J.: Robust one-time password authentication scheme using smart card for home network environment. Comput. Commun. 34(3), 326–336 (2011) 25. Xiong, H., Hou, Y., Huang, X., Zhao, Y., Chen, C.M.: Heterogeneous signcryption scheme from IBC to PKI with equality test for WBANs. IEEE Syst. J. (2021) 26. Yavari, M., Safkhani, M., Kumari, S., Kumar, S., Chen, C.M.: An improved blockchain-based authentication protocol for IoT network management. Secur. Commun. Netw. 2020 (2020) 27. Yu, B., Li, H.: Anonymous authentication key agreement scheme with pairing-based cryptography for home-based multi-sensor internet of things. Int. J. Distrib. Sens. Netw. 15(9), 1550147719879379 (2019)

Chapter 43

Comments on “Two Authentication and Key Agreement Protocols in WSN Environments” Fangfang Kong, Saru Kumari, and Tsu-Yang Wu

Abstract Wireless sensor network (WSN) is a self-organizing network composed of distributed sensor nodes. Its appearance has dramatically changed people’s lives and is now widely used in various fields. Due to the openness of wireless channel, malicious attackers can eavesdrop, intercept or tamper with data in the channel. In the communication process, the user’s privacy is easy to leak, which leads to the user’s inability to communicate securely with the sensor node. Hence, it is necessary to design secure authentication and key agreement (AKA) protocols to enhance the communication security of users in the WSN environment. Recently, Jawad et al. proposed an anonymous three-factor authentication protocol based on symmetric encryption in WSN. Polai et al. proposed an authentication protocol using lightweight primitives in wireless body area networks. In this paper, we analyze Jawad et al.’s and Polai et al.’s protocol and find that both protocols have security vulnerabilities. We prove their protocols cannot resist sensor node capture attacks, known temporary information disclosure attacks, and hub node stolen database attacks. Finally, we put forward suggestions for the improvement of these two protocols.

43.1 Introduction With the development of the Internet and wireless communication technology [28], wireless sensor networks (WSN) came into being. WSN [6, 15, 31] is a kind of wireless self-organizing network system composed of some small wireless sensors, F. Kong · T.-Y. Wu (B) College of Computer Science and Engineering, Shandong University of Science and Technology, Qingdao 266590, China e-mail: [email protected] F. Kong e-mail: [email protected] S. Kumari Department of Mathematics, Chaudhary Charan Singh University, Meerut 250004, Uttar Pradesh, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_44

551

552

F. Kong et al.

which have the ability to perceive, collect and transmit data. WSN and cloud computing technology [24, 26] combination has dramatically improved the data processing capability. Because of its low cost and energy consumption [29], WSN is widely used in smart city [9, 25], agriculture [1, 25], healthcare [7, 11, 27], vehicle [5, 16, 22], smart home [20, 22], and other fields. The wide application of WSN has brought great convenience to people’s life [3, 4, 12]. However, in this environment, when data is transmitted through public channels, malicious attackers can eavesdrop, intercept or tamper with it [18, 30], resulting in users suffering from tracking and privacy data leakage. Therefore, protecting communication security in WSNs is an important issue. Authentication and key agreement (AKA) protocol can enhance communication security in WSN environments. In recent years, many scholars have proposed a variety of AKA protocols in WSN. In 2020, Moghadam et al. [19] proposed an efficient authentication protocol under the elliptic curve Diffie–Hellman hard problem. They claimed that their protocol could resist various attacks. Unfortunately, Kwon et al. [13] proved that their protocol could not resist insider attacks and violates perfect forward secrecy. They designed a lightweight authentication protocol to improve the security and computing cost of Moghadam et al.’s protocol [19]. Liu et al. [17] proposed an anonymous one-round authentication protocol, but Li et al. [14] found that their protocol had security vulnerabilities and proposed an improved protocol. Unfortunately, Sowjanya et al. [23] found that Li et al.’s protocol [14] could not guarantee the security of the protocol. Then, Sowjanya et al. [23] proposed an end-to-end authentication protocol to enhance security. Recently, Jawad et al. [10] proposed an anonymous three-factor authentication protocol using symmetric encryption. Polai et al. [21] proposed a lightweight authentication protocol. In this paper, we first analyze both protocols [10, 21]. Then, we point out that Jawad et al.’s protocol [10] cannot resist sensor node capture attacks and known temporary information disclosure attacks. Meanwhile, Polai et al.’s protocol [21] also does not resist hub node stolen database attacks and known temporary information disclosure attacks. Finally, we put forward suggestions for the improvement of these two protocols.

43.2 Review of Protocols In this section, we mainly review the login and authentication phase in Jawad et al.’s protocol [10] and Polai et al.’s protocols [21]. The main symbols used in this paper are shown in Table 43.1.

43 Comments on “Two Authentication and Key Agreement Protocols … Table 43.1 Notations

553

Notations

Description

Ui SNj HN GW N I DUi I DS j I Dhn P Wi B I Oi x y

ith user jth sensor node Hub node Gateway node Identity of Ui Identity of Ui Identity of H N Password of Ui Biometric information of Ui Private key of GW N Shared key between Ui and GW N Session key Cryptographic function Decryption function Concatenation operation XOR operation

SK Eec(·) Dec(·) || ⊕

43.2.1 Review of Jawad et al.’s Protocol Their protocol consists of three entities: user Ui , sensor node S N j , and gateway GW N . Their protocol mainly includes three phases: “Ui registration”, “S N j registration”, and “login and authentication” phases. For Ui registration and S N j registration phases, please refer to [10]. The login and authentication phase of the protocol is shown in Fig. 43.1. The detailed steps are described as follows: (1) First, Ui inputs biometric B I Oi , and then calculates ai∗ = γ ⊕ B I Oi∗ , checks ?

h(ai∗ ) = γ . If they are not equal, Ui is denied login; otherwise, Ui is allowed to enter I DUi and P Wi , calculates R P Wi = h(P Wi  bi ), Bi∗ = h(I DUi  ?

R P Wi  ai∗ ), checks Bi∗ = Bi . If equal, Ui logs in successfully. Ui generates a random number Ni and s, and then calculates E 1 = Ai ⊕ h(h(P Wi  bi )  ai∗ ), E 2 = s · P, E 3 = s · X , E 4 = Enc y (I DUi  Ni ), E 5 = E 1 ⊕ Ni , E 6 = h(I DUi  Ni ) ⊕ I DS j , and VU G = h(E 1  I DS j  E 3  Ni ). Finally, Ui sends M1 = {E 2 , E 4 , E 5 , E 6 , VU G } to GW N . (2) After receiving M1 , GW N first decrypts E 4 using the shared key y to get (I DUi∗  Ni∗ ) = Dec y (E 4 ), and verifies that Ui ’s identity is in the database. Then, GW N calculates E 3∗ = x · E 2 , E 1∗ = h(I DUi∗  K G S ), Ni∗ = E 5 ⊕ E 1∗ , I DS ∗j = E 6 ⊕ h(I DUi∗  Ni∗ ), VU∗ G = h(E 1∗  I DS ∗j  E 3∗  ?

Ni∗ ), and checks VU∗ G = VU G . GW N generates N g and calculates K G∗ S =

554

F. Kong et al.

h(I DS ∗j  x), E 7 = I DUi∗ ⊕ K G∗ S , E 8 = N g ⊕ h(I DUi∗  K G∗ S ), E 9 = N g ⊕ Ni∗ , VG S = h(I DUi∗  I DS ∗j  K G∗ S  Ni∗  N g ). Finally, GW N sends M2 = {E 7 , E 8 , E 9 , VG S } to S N j . (3) After receiving M2 , S N j first calculates I DUi∗∗ = E 7 ⊕ K G S , N g∗ = h(I DUi∗∗  K G S ) ⊕ E 8 , Ni∗∗ = N g∗ ⊕ E 9 , VG∗ S = h(I DUi∗∗  I DS j  K G S  ?

Ni∗∗  N g∗ ), and checks VG∗ S = VG S . If true, S N j generates N j and calculates E 10 = N j ⊕ K G S , S K j = h(I DUi∗∗  I DS j  Ni∗∗  N g∗  N j ), VSG = h(K G S  S K j  r j ). Finally, S N j sends M3 = {E 10 , VSG } to GW N . (4) After receiving M3 , GW N calculates N ∗j = E 10 ⊕ K G∗ S , S K GW N = h(I DUi∗  ?

∗ ∗ = h(K G∗ S  S K GW N  N ∗j ), and checks VSG = I DS ∗j  Ni∗  N g  N ∗j ), VSG ∗ ∗ ∗ VSG . If equal, GW N calculates E 11 = E 1 ⊕ N g , E 12 = Ni ⊕ N j , VGU = h(I DUi∗  S K GW N  N g  N ∗j ). Finally, GW N sends M4 = {E 10 , VSG } to Ui . (5) After receiving M4 , Ui calculates N g∗∗ = E 11 ⊕ E 1 , N ∗∗ j = E 12 ⊕ Ni , S K i = ∗ ∗∗ ∗∗ ), V = h(I DU h(I DUi  I DS j  Ni  N g∗∗  N ∗∗ i  S K i  N g  N j ), j GU ?

∗ = VGU . If equal, both entities have achieved mutual authenand checks VGU tication and established the session key S K . Ui and S N j store S K for secure communication in the future.

43.2.2 Review of Polai et al.’s Protocol Their protocol mainly includes three phases: “Initialization”, “S N j registration”, and “Authentication”. In the authentication phase, it realizes mutual authentication and key agreement between S N j and hub nodes H N . The detailed steps are described as follows. Note that, for the initialization and S N j registration phases please refer to Ref. [21] (see Fig. 43.2). (1) S N j inputs I D j , calculates R I D ∗j = h(I D j  a j ), Di∗ = h(I D j  h( p)  ?

R I D ∗j ), and checks Di∗ = Di . If true, S N j generates R j , and calculates h( p  q) = Ai  h(I D j  R I D ∗j ), W1 = h(I D j  P I D j  Bi  Ci  ri ), X N i = h(I Dhn  h( p  q)  h( p)), W2 = X N i ⊕ R j , V jn = h(W1  W2  R j  X N i ). Finally, S N j sends M1 = {W1 , W2 , V jn } to H N . ∗ = h(W1  W2  (2) After receiving M1 , H N first calculates R ∗j = X N i ⊕ W1 , V jn ?

∗ = V jn . Then, H N generates Rhn and calculates W3 = R ∗j  X N i ), and checks V jn h(I Dhn  X N i  Rhn ), W4 = W3 ⊕ R ∗j , Vn j = h(I Dhn  h( p)  W3  R ∗j ). Finally, H N sends M2 = {W4 , Vn j , I Dhn } to S N j . (3) After receiving M2 , S N j calculates W5∗ = h(W3  W4  R ∗j  I Dhn ), and checks ?

W5∗ = W5 . If equal, S j and H N will store S K = h(W3  I Dhn  X N i  R j ).

43 Comments on “Two Authentication and Key Agreement Protocols … User Ui {Ai , Bi , β, γ, Gen(·), bi }

GW N {IDUi , y}

555

Sensor Node SNj {IDSj , KGS }

BIOi∗

Inputs ∗ a∗ i = γ ⊕ BIOi ?

Checks h(a∗ i) = γ Inputs IDUi , P Wi RP Wi = h(P Wi  bi ) Bi∗ = h(IDUi  RP Wi  a∗ i) ?

Checks Bi∗ = Bi Generates Ni and s E1 = Ai ⊕ h(h(P Wi  bi )  a∗ i) E2 = s · P E3 = s · X E4 = Ency (IDUi  Ni ) E5 = E1 ⊕ Ni E6 = h(IDUi  Ni ) ⊕ IDSj VU G = h(E1  IDSj  E3  Ni ) {E2 , E4 , E5 , E6 , VU G } −−−−−−−−−−−−−−−−→

(IDUi∗  Ni∗ ) = Decy (M4 ) Checks if IDUi∗ in its database E3∗ = xE2 E1∗ = h(IDUi∗  KGS ) ∗ Ni = E5 ⊕ E1∗ IDSj∗ = E6 ⊕ h(IDUi∗  Ni∗ ) ∗ VU G = h(E1∗  IDSj∗  E3∗  Ni∗ ) ?

VGS

Checks VU∗G = VU G Generates Ng ∗ KGS = h(IDSj∗  x) ∗ E7 = IDUi∗ ⊕ KGS ∗ E8 = Ng ⊕ h(IDUi∗  KGS ) E9 = Ng ⊕ Ni∗ ∗ = h(IDUi∗  IDSj∗  KGS  Ni∗  Ng ) {E7 , E8 , E9 , VGS } −−−−−−−−−−−−−→

∗ VGS

∗ Nj∗ = E10 ⊕ KGS SKGW N = h(IDUi∗  IDSj∗  Ni∗  Ng  Nj∗ ) ∗ ∗ VSG = h(KGS  SKGW N  Nj∗ )

IDUi∗∗ = E7 ⊕ KGS Ng∗ = h(IDUi∗∗  KGS ) ⊕ E8 Ni∗∗ = Ng∗ ⊕ E9 = h(IDUi∗∗  IDSj  KGS  Ni∗∗  Ng∗ ) ?

∗ Checks VGS = VGS Generates Nj E10 = Nj ⊕ KGS SKj = h(IDUi∗∗  IDSj  Ni∗∗  Ng∗  Nj ) VSG = h(KGS  SKj  rj ) {E10 , VSG } ←−−−−−−−−

?

VGU Ng∗∗ Nj∗∗

= E11 ⊕ E1 = E12 ⊕ Ni SKi = h(IDUi  IDSj  Ni  Ng∗∗  Nj∗∗ ) ∗ VGU = h(IDUi  SKi  Ng∗∗  Nj∗∗ )

∗ Checks VSG = VSG E11 = E1∗ ⊕ Ng E12 = Ni∗ ⊕ Nj∗ = h(IDUi∗  SKGW N  Ng  Nj∗ ) {E11 , E12 , VGU } ←−−−−−−−−−−−−

?

∗ Checks VGU = VGU

Fig. 43.1 Login and authentication phase of Jawad et al.’s protocol

43.3 Cryptanalysis of Protocols 43.3.1 Adversary Model We adopted the widely used Dolev-Yao (DY ) [8] and Canetti and Krawczyk (C K ) [2] threat models. The capabilities of attacker ( A) are described as follows: (1) A can eavesdrop, intercept, and delete information transmitted in a public channel. (2) A can steal the information stored in the memory of the sensor node and the information in the hub node database. (3) A can obtain the random number generated by any entity in the session.

556

F. Kong et al. Sensor Node SNj {Aj , Bj , Cj , Dj , h(p), h(·), aj }

Hub Node HN {XN j , IDhn , h(p)}

Inputs IDj Computes RIDj∗ = h(IDj  aj ) Dj∗ = h(IDj  h(p)  RIDj∗ ) ?

Checks Dj∗ = Dj h(p  q) = Aj ⊕ h(IDj  RIDj∗ ) Generates a random number Rj W1 = h(IDj  RIDj∗  Bj  Cj  Rj ) XN i = h(IDhn  h(p  q)  h(p)) W2 = XN j ⊕ Rj Vjn = h(W1  W2  Rj  XN j ) {W1 , W2 , Vjn } −−−−−−−−−−→

Rj∗ = XN j ⊕ W2 ∗ Vjn = h(W1  W2  Rj∗  XN j ) ?

W3∗ = W4 ⊕ Rj ∗ Vnj = h(IDhn  h(p)  W3∗  Rj∗ )

∗ Checks Vjn = Vjn Generates a random number Rhn W3 = h(IDhn  XN j  Rhn ) W4 = W3 ⊕ Rj∗ Vnj = h(IDhn  h(p)  W3  Rj∗ ) {W4 , Vnj , IDhn } ←−−−−−−−−−−−−

?

∗ Checks Vnj = Vnj ∗ W5 = h(W3  W4  Rj  IDhn ) {W5 } −−−→

W5∗ = h(W3  W4  Rj∗  IDhn ) ?

Checks W5∗ = W5 Stores session key SK = h(W3∗  IDhn  XN j  Rj )

Stores session key SK = h(W3  IDhn  XN j  Rj∗ )

Fig. 43.2 Authentication phase of Polai et al.’s protocol

43.3.2 Cryptanalysis of Jawad et al.’s Protocol In this section, we analyze Jawad et al.’s protocol [10], and find that their protocol cannot resist sensor node capture attacks and known temporary information disclosure attacks. The attack process is shown in Fig. 43.3.

43.3.2.1

Sensor Node Capture Attacks

We assume that A can capture sensor node S N j and get the information{I DS j , K G S }. The specific steps to calculate S K are as follows. (1) A first intercepts messages {E 7 , E 8 , E 9 , VG S } and {E 10 , VSG } transmitted in the public channel.

43 Comments on “Two Authentication and Key Agreement Protocols … User Ui {Ai , Bi , β, γ, Gen(·), bi }

GW N {IDUi , y}

557

Sensor Node SNj {IDSj , KGS }

BIOi∗

Inputs ∗ a∗ i = γ ⊕ BIOi ?

Checks h(a∗ i) = γ Inputs IDUi , P Wi RP Wi = h(P Wi  bi ) Bi∗ = h(IDUi  RP Wi  a∗ i) ?

Checks Bi∗ = Bi Generates Ni and s E1 = Ai ⊕ h(h(P Wi  bi )  a∗ i) E2 = s · P E3 = s · X E4 = Ency (IDUi  Ni ) E5 = E1 ⊕ Ni E6 = h(IDUi  Ni ) ⊕ IDSj VU G = h(E1  IDSj  E3  Ni ) {E2 , E4 , E5 , E6 , VU G } −−−−−−−−−−−−−−−−→

(IDUi∗  Ni∗ ) = Decy (M4 ) Checks if IDUi∗ in its database E3∗ = xE2 E1∗ = h(IDUi∗  KGS ) Ni∗ = E5 ⊕ E1∗ IDSj∗ =E6 ⊕h(IDUi∗  Ni∗ )

VU∗G = h(E1∗  IDSj∗  E3∗  Ni∗ ) ?

VGS

Checks VU∗G = VU G Generates Ng ∗ KGS = h(IDSj∗  x) ∗ E7 = IDUi∗ ⊕ KGS ∗ E8 = Ng ⊕ h(IDUi∗  KGS ) ∗ E9 = Ng ⊕ Ni ∗ = h(IDUi∗  IDSj∗  KGS  Ni∗  Ng ) {E7 , E8 , E9 , VGS } −−−−−−−−−−−−−→

IDUi∗∗ =E7 ⊕KGS Ng∗ =h(IDUi∗∗ KGS )⊕E8 Ni∗∗ =Ng∗ ⊕E9 ∗ VGS = h(IDUi∗∗  IDSj  KGS  Ni∗∗  Ng∗ ) ?

∗ Checks VGS = VGS Generates Nj

E10 =Nj ⊕KGS SKj = h(IDUi∗∗  IDSj  Ni∗∗  Ng∗  Nj )

Nj∗

∗ = E10 ⊕ KGS SKGW N = h(IDUi∗  IDSj∗  Ni∗  Ng  Nj∗ ) ∗ ∗ VSG = h(KGS  SKGW N  Nj∗ ) ? ∗ Checks VSG = VSG E11 = E1∗ ⊕ Ng E12 = Ni∗ ⊕ Nj∗ VGU = h(IDUi∗  SKGW N  Ng  Nj∗ )

Ng∗∗ = E11 ⊕ E1 Nj∗∗ = E12 ⊕ Ni SKi = h(IDUi  IDSj  Ni  Ng∗∗  Nj∗∗ ) ∗ VGU = h(IDUi  SKi  Ng∗∗  Nj∗∗ )

VSG = h(KGS  SKj  rj ) {E10 , VSG } ←−−−−−−−−

{E11 , E12 , VGU } ←−−−−−−−−−−−−

?

∗ Checks VGU = VGU

Fig. 43.3 Sensor node capture attacks and known temporary information disclosure attacks in Jawad et al.’s protocol

(2) Secondly, A obtains the information {I DS j , K G S } stored in the memory of S N j , and then calculates S K composition parameters {I DUi∗ , N g∗ , Ni∗∗ , N j }, where I DUi∗∗ = E 7 ⊕ K G S , N g∗ = h(I DUi∗∗  K G S ) ⊕ E 8 , Ni∗∗ = N g∗ ⊕ E 9 . (3) Finally, A can successfully calculate S K j = h(I DUi∗∗  I DS j  Ni∗∗  N g∗  N j ). Therefore, Jawad et al.’s protocol cannot resist sensor node capture attacks.

558

43.3.2.2

F. Kong et al.

Known Temporary Information Disclosure Attacks

We assume that A can obtain a random number N j generated by S N j . A can calculate S K and sensor node identity I DS j through the following steps: (1) A first intercepts messages {E 2 , E 4 , E 5 , E 6 , VU G }, {E 7 , E 8 , E 9 , VG S }, and {E 10 , VSG } transmitted in the public channel. (2) Secondly, A obtains random number N j generated by S N j , and then calculates S K composition parameters {I DUi∗ , I DS j , N g∗ , Ni∗∗ }, where I DUi∗∗ = E 7 ⊕ K G S , N g∗ = h(I DUi∗∗  K G S ) ⊕ E 8 , Ni∗∗ = N g∗ ⊕ E 9 , N ∗j = Ni∗ ⊕ E 12 , I DS ∗j = E 6 ⊕ h(I DUi∗  Ni∗ ). (3) Finally, A can successfully calculate S K j = h(I DUi∗∗  I DS j  Ni∗∗  N g∗  N j ), also can calculate S N j identity I DS ∗j = E 6 ⊕ h(I DUi∗  Ni∗ ). Therefore, Jawad et al.’s protocol cannot resist known temporary information disclosure attacks.

43.3.3 Cryptanalysis of Polai et al.’s Protocol In this section, we analyze Polai et al.’s protocol [21], and find that their protocol cannot resist hub node stolen database attacks and known temporary information disclosure attacks. The attack process is shown in Fig. 43.4.

43.3.3.1

Hub Node Stolen Database Attacks

We assume that A can capture hub node H N and get the information {X N i , I Dhn , h( p)}. The specific steps to calculate S K are as follows: (1) A first intercepts messages {W1 , W2 , V jn } and {W4 , Vn j , I Dhn } transmitted in the public channel. (2) Secondly, A obtains the information {X N i , I Dhn , h( p)} stored in the memory of the S N j , and then calculates S K composition parameters {R ∗j , W3 }, where R ∗j = X N i ⊕ W1 , W3∗ = W4 ⊕ R j . (3) Finally, A can successfully calculate S K = h(W3  I Dhn  X N i  R ∗j ). Therefore, Polai et al.’s protocol cannot resist hub node stolen database attacks.

43.3.3.2

Known Temporary Information Disclosure Attacks

We assume that A can obtain random number R j generated by S N j . The specific steps to calculate S K are as follows:

43 Comments on “Two Authentication and Key Agreement Protocols … Sensor Node SNj {Aj , Bj , Cj , Dj , h(p), h(·), aj }

559

Hub Node HN {XN j , IDhn , h(p)}

Inputs IDj Computes RIDj∗ = h(IDj  aj ) Dj∗ = h(IDj  h(p)  RIDj∗ ) ?

Checks Dj∗ = Di h(p  q) = Ai ⊕ h(IDj  RIDj∗ ) Generates a random number Rj W1 = h(IDj  RIDj∗  Bj  Cj  Rj ) XN j = h(IDhn  h(p  q)  h(p)) W2 =XN j ⊕Rj Vjn = h(W1  W2  Rj  XN j ) {W1 ,W2 , Vjn } −−−−−−−−−−→

Rj∗ =XN j ⊕W2 ∗ Vjn

= h(W1  W2  Rj∗  XN j ) ?

∗ Checks Vjn = Vjn Generates a random number Rhn W3 = h(IDhn  XN j  Rhn )

W4 =W3 ⊕Rj∗

∗ Vnj

W3∗ =W4 ⊕Rj

Vnj = h(IDhn  h(p)  W3  Rj∗ ) {W4 , Vnj ,IDhn } ←−−−−−−−−−−−−

= h(IDhn  h(p)  W3∗  Rj∗ ) ?

∗ Checks Vnj = Vnj W5 = h(W3∗  W4  Rj  IDhn ) {W5 } −−−→

W5∗ = h(W3  W4  Rj∗  IDhn )

Stores session key

Stores session key

SK = h(W3∗  IDhn XN j  Rj )

SK = h(W3  IDhn  XN j Rj∗ )

Fig. 43.4 Hub node stolen database attacks and known temporary information disclosure attacks in Polai et al.’s protocol

(1) A first intercepts messages {W1 , W2 , V jn } and {W4 , Vn j , I Dhn } transmitted in the public channel. (2) Secondly, A obtains random numbers R j generated by S N j , and then calculates S K composition parameters {X N i , W3∗ }, where W2 = X N i ⊕ R j , W3∗ = W4 ⊕ Rj. (3) Finally, A can successfully calculate S K = h(W3∗  I Dhn  X N i  R j ). Therefore, Polai et al.’s protocol cannot resist known temporary information disclosure attacks.

560

F. Kong et al.

43.4 Conclusion and Discussion In this paper, we first describe the WSN environment and its applications and review the relevant authentication protocols in this environment. After that, we first analyze Jawad et al.’s protocol and Polai et al.’s protocol and then prove that the two protocols have security vulnerabilities, such as cannot resist sensor node capture attacks, known temporary value disclosure attacks, and hub node stolen database attacks. Moreover, we have put forward some suggestions for improving both protocols. In Jawad et al.’s protocol, we suggest sensor nodes can perform XOR or encryption operations for the values I DS j to resist sensor node capture attacks. Meanwhile, we can use the perudo-identity of the user and sensor node, for example, using hash operation to real identity, to ensure the anonymity of users and sensor nodes. Similarly, we suggest to replace the transmitted identity I Dhn with a pseudo-identity to resist known temporary information disclosure attacks in Polai et al.’s protocol. Acknowledgements This research was partially supported by Natural Science Foundation of Shandong Province, China (Grant no. ZR202111230202).

References 1. Bothe, A., Bauer, J., Aschenbruck, N.: RFID-assisted continuous user authentication for IoTbased smart farming. In: 2019 IEEE International Conference on RFID Technology and Applications (RFID-TA), pp. 505–510. IEEE (2019) 2. Canetti, R., Krawczyk, H.: Analysis of key-exchange protocols and their use for building secure channels. In: International Conference on the Theory and Applications of Cryptographic Techniques, pp. 453–474. Springer (2001) 3. Chen, C.M., Li, C.T., Liu, S., Wu, T.Y., Pan, J.S.: A provable secure private data delegation scheme for mountaineering events in emergency system. IEEE Access 5, 3410–3422 (2017) 4. Chen, C.M., Xiang, B., Wang, K.H., Yeh, K.H., Wu, T.Y.: A robust mutual authentication with a key agreement scheme for session initiation protocol. Appl. Sci. 8(10), 1789 (2018) 5. Chen, J.N., Huang, Z.J., Zhou, Y.P., Zou, F.M., Chen, C.M., Wu, J.M.T., Wu, T.Y.: Efficient certificate-based aggregate signature scheme for vehicular ad hoc networks. IET Netw. 9(6), 290–297 (2020) 6. Chen, J.N., Zhou, Y.P., Huang, Z.J., Wu, T.Y., Zou, F.M., Tso, R.: An efficient aggregate signature scheme for healthcare wireless sensor networks. J. Netw. Intell. 6(1), 1–15 (2021) 7. Deebak, B.D., Al-Turjman, F.: Smart mutual authentication protocol for cloud based medical healthcare systems using internet of medical things. IEEE J. Sel. Areas Commun. 39(2), 346– 360 (2020) 8. Dolev, D., Yao, A.: On the security of public key protocols. IEEE Trans. Inf. Theory 29(2), 198–208 (1983) 9. Gupta, M., Kumar, R., Shekhar, S., Sharma, B., Patel, R.B., Jain, S., Dhaou, I.B., Iwendi, C.: Game theory-based authentication framework to secure internet of vehicles with blockchain. Sensors 22(14), 5119 (2022) 10. Jawad, K., Mansoor, K., Baig, A.F., Ghani, A., Naseem, A.: An improved three-factor anonymous authentication protocol for WSNs based IoT system using symmetric cryptography. In: 2019 International Conference on Communication Technologies (ComTech), pp. 53–59. IEEE (2019). https://doi.org/10.1109/COMTECH.2019.8737799

43 Comments on “Two Authentication and Key Agreement Protocols …

561

11. Kumar, P., Lee, S.G., Lee, H.J.: E-sap: Efficient-strong authentication protocol for healthcare applications using wireless medical sensor networks. Sensors 12(2), 1625–1647 (2012) 12. Kumari, A., Kumar, V., Abbasi, M.Y., Kumari, S., Chaudhary, P., Chen, C.M.: CSEF: cloudbased secure and efficient framework for smart medical system using ECC. IEEE Access 8, 107838–107852 (2020) 13. Kwon, D.K., Yu, S.J., Lee, J.Y., Son, S.H., Park, Y.H.: WSN-slap: Secure and lightweight mutual authentication protocol for wireless sensor networks. Sensors 21(3), 936 (2021) 14. Li, X., Peng, J., Kumari, S., Wu, F., Karuppiah, M., Choo, K.K.R.: An enhanced 1-round authentication protocol for wireless body area networks with user anonymity. Comput. Electr. Eng. 61, 238–249 (2017) 15. Li, X., Liu, S., Kumari, S., Chen, C.M.: PSAP-WSN: a provably secure authentication protocol for 5g-based wireless sensor networks. Comput. Model. Eng. Sci. 135(1), 711–732 (2023) 16. Li, Z., Miao, Q., Chaudhry, S.A., Chen, C.M.: A provably secure and lightweight mutual authentication protocol in fog-enabled social internet of vehicles. Int. J. Distrib. Sens. Netw. 18(6), 15501329221104332 (2022) 17. Liu, J., Zhang, L., Sun, R.: 1-RAAP: an efficient 1-round anonymous authentication protocol for wireless body area networks. Sensors 16(5), 728 (2016) 18. Liu, S., Chen, C.M.: Comments on “a secure and lightweight drones-access protocol for smart city surveillance”. IEEE Trans. Intell. Transp. Syst. (2022). https://doi.org/10.1109/TITS.2022. 3198045 19. Moghadam, M.F., Nikooghadam, M., Al Jabban, M.A.B., Alishahi, M., Mortazavi, L., Mohajerzadeh, A.: An efficient authentication and key agreement scheme based on ECDH for wireless sensor network. IEEE Access 8, 73182–73192 (2020) 20. Naoui, S., Elhdhili, M.E., Saidane, L.A.: Lightweight and secure password based smart home authentication protocol: LSP-SHAP. J. Netw. Syst. Manag. 27(4), 1020–1042 (2019) 21. Polai, M., Mohanty, S., Sahoo, S.S.: A lightweight mutual authentication protocol for wireless body area network. In: 2019 6th International Conference on Signal Processing and Integrated Networks (SPIN), pp. 760–765. IEEE (2019). https://doi.org/10.1109/SPIN.2019.8711643 22. Sadri, M.J., Rajabzadeh Asaar, M.: A lightweight anonymous two-factor authentication protocol for wireless sensor networks in internet of vehicles. Int. J. Commun. Syst. 33(14), e4511 (2020) 23. Sowjanya, K., Dasgupta, M., Ray, S.: An elliptic curve cryptography based enhanced anonymous authentication protocol for wearable health monitoring systems. Int. J. Inf. Secur. 19(1), 129–146 (2020) 24. Tso, R., Huang, K., Chen, Y.C., Rahman, S.M.M., Wu, T.Y.: Generic construction of dual-server public key encryption with keyword search on cloud computing. IEEE Access 8, 152551– 152564 (2020) 25. Vasudev, H., Deshpande, V., Das, D., Das, S.K.: A lightweight mutual authentication protocol for v2v communication in internet of vehicles. IEEE Trans. Veh. Technol. 69(6), 6709–6717 (2020) 26. Wu, T.Y., Wang, L., Guo, X., Chen, Y.C., Chu, S.C.: SAKAP: SGX-based authentication key agreement protocol in IoT-enabled cloud computing. Sustainability 14(17), 11054 (2022) 27. Wu, T.Y., Yang, L., Meng, Q., Guo, X., Chen, C.M.: Fog-driven secure authentication and key exchange scheme for wearable health monitoring system. Secur. Commun. Netw. 2021, 8368646 (2021) 28. Yang, L., Chen, Y.C., Wu, T.Y.: Provably secure client-server key management scheme in 5g networks. Wirel. Commun. Mob. Comput. 2021, 4083199 (2021) 29. Yick, J., Mukherjee, B., Ghosal, D.: Computer networks. Wirel. Sens. Netw. Surv. 52(12), 2292–2330 (2008) 30. Zhang, D., Qian, Y., Wan, J., Zhao, S.: An efficient RFID search protocol based on clouds. Mob. Netw. Appl. 20(3), 356–362 (2015) 31. Zhang, Z., Mehmood, A., Shu, L., Huo, Z., Zhang, Y., Mukherjee, M.: A survey on fault diagnosis in wireless sensor networks. IEEE Access 6, 11349–11364 (2018)

Chapter 44

Security Analysis of Two Authentication and Key Agreement Protocols Based on Wireless Sensor Networks Liyang Wang, Saru Kumari, and Tsu-Yang Wu

Abstract In wireless sensor network (WSN), various data is collected via sensors and transmitted over a public channel. However, during the transmission, an attacker can intercept and eavesdrop on data and use it to launch some attacks. Therefore, it is necessary to design secure authentication and key agreement (AKA) protocols in WSN environments, which ensures secure communication between entities. Recently, Yu et al. designed an authentication protocol in WSN and claimed their proposed protocol can resist well-known attacks. In addition, Wang et al. proposed another three-party mutual authentication protocol in IoT-enabled wireless sensor networks. In this paper, we find some security vulnerabilities in both protocols, including sensor node capture attacks, known session-specific temporary information attacks, and violating perfect forward secrecy. Finally, we introduce several suggestions for the improvement of both protocols.

44.1 Introduction Wireless sensor network (WSN) [10, 13, 24], an application of the Internet of Things (IoT) [5, 21, 28, 29], is a network composed of sensor nodes, which can monitor, perceive and collect various data in the node deployment area. With the rapid development of wireless communication, WSN can be applied to many fields, such as intelligent transportation [6, 14, 22], smart grid [11, 16, 25], health care [1, 7, 15], L. Wang · T.-Y. Wu (B) College of Computer Science and Engineering, Shandong University of Science and Technology, Qingdao 266590, China e-mail: [email protected] L. Wang e-mail: [email protected] S. Kumari Department of Mathematics, Chaudhary Charan Singh University, Meerut 250004, Uttar Pradesh, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_45

563

564

L. Wang et al.

and smart city [2, 27]. However, due to the openness of the wireless network, attackers can capture data transmitted on the public channel to launch various attacks, such as sensor node capture attacks, known session-specific temporary information attacks, etc. Therefore, it is necessary to design authentication and key agreement (AKA) protocols to ensure secure communication between entities. In 2015, Chang et al. [4] designed an AKA protocol for dynamic identity in WSN. However, Park et al. [19] proved that their protocol was susceptible to off-line password guessing attacks and did not ensure perfect forward secrecy. Shin and Kwon [20] designed a lightweight AKA protocol based on WSN for smart homes. However, Mo et al. [17] demonstrated that their protocol suffered user impersonation attacks, desynchronization attacks, and sensor node capture attacks. Wu et al. [24] proposed a three-factor AKA protocol in WSN. However, Wu et al. [26] found that their protocol was susceptible to known session-specific temporary information attacks, key compromise impersonation attacks, and violating perfect forward secrecy. In 2020, Moghadam et al. [18] proposed an authentication protocol using elliptic curves based on WSN. However, Kwon et al. [12] discovered that their protocol was not resistant to privileged insider attacks, known session-specific temporary information attacks, and also violating perfect forward secrecy [9]. Recently, Yu et al. [30] proposed a user authentication protocol in a WSN environment. Wang et al. [23] proposed a three-party AKA protocol in IoT-enabled wireless sensor networks. In this paper, we demonstrate that Yu et al.’s protocol is susceptible to sensor node capture attacks and known session-specific temporary information attacks. Meanwhile, we discover that Wang et al.’s protocol cannot resist sensor node capture attacks and violate perfect forward secrecy. A very interest method [9] is adopted to verify perfect forward secrecy. Finally, we provide several suggestions for both protocols to enhance the security.

44.2 Review of Protocols In this section, we review Yu et al.’s protocol [30] and Wang et al.’s protocol [23]. Table 44.1 lists the notations used in this paper.

44.2.1 Review of Yu et al.’s Protocol Their protocol has three phases including “sensor node registration,” “user registration,” and “authentication and key agreement (AKA)” phases. Here, we mainly review the AKA phase, while the registration phase refers to Yu et al.’s protocol [30]. When user Ui wants to communicate with sensor node S N j , they need to establish a session key S K with the help of the gateway GW N to ensure communication security. The authentication and key agreement process are shown in Fig. 44.1, and the steps are shown below.

44 Security Analysis of Two Authentication and Key Agreement Protocols … Table 44.1 Notations

565

Notations

Description

Ui SNj GW N CS I Di SI Dj P Wi B I Oi Ks s, Pubcs SK Ti Gen(.)/Rep(.)

i-th user j-th sensor node Gateway node Cloud server Identity of Ui Identity of S N j Password of Ui Biometric information of Ui Private key of GW N Private-public key pair of C S Session key Timestamp Fuzzy generator/reproduction function

(1) Ui chooses I Di , P Wi , B I Oi . Next, Ui calculates U Q i = B H (B I Oi ), R I Di = U Ci ⊕ h(I Di  P Wi  U Q i ), R P Wi = h(P Wi  U Q i ), U X i = U Ai ⊕ ? h(R I Di  R P Wi ), U Bi∗ = h(U X i  R P Wi ), and check U Bi∗ = U Bi . Then, Ui selects ri , and computes W1 = U X i ⊕ ri , W2 = I Di ⊕ h(U X i  ri ), V1 = h(I Di  ri  U X i ). Finally, Ui transmits {W1 , W2 , V1 , R I Di } to S N j . (2) When S N j receives the messages from Ui , it computes W3 = P j ⊕ r j , V2 = h(GW N PS j  W1  r j ). Then, Ui transmit {W1 , W2 , V1 , R I Di , S I D j , W3 , V2 } to GW N . (3) When GW N receives the messages from S N j , it retrieves {Si } in database using R I Di , and computes U X i = h(R I Di  K s  Si ), ri = W1 ⊕ U X i , I Di = ? W2 ⊕ h(U X i  ri ), V1∗ = h(I Di  ri  U X i ), and check V1∗ = V1 . Then, GW N ?

calculates r ∗j = W3 ⊕ P j , V2∗ = h(GW N PS j  W1  r ∗j ), and check V2∗ = V2 . Next, GW N computes W4 = r j ⊕ h(I Di  ri ), W5 = ri ⊕ h(S I D j  r j ), V3 = h(ri  r j  GW N PS j ). Finally, GW N sends {W4 , V5 , V3 } to S N j . (4) Upon receiving {W4 , V5 , V3 }, S N j calculates ri = W5 ⊕ h(S I D j  r j ), V3∗ = ?

h(ri  r j  GW N PS j ), and check V3∗ = V3 . Next, GW N computes S K = h(ri  r j ), V4 = h(ri  r j  S K ). Then, S N j transmits {W5 , V4 } to Ui . (5) When Ui receives the messages from S N j , it calculates r j = W4 ⊕ h(I Di  ri ), ?

S K = h(ri  r j ), V4∗ = h(ri  r j  S K ), and check V4∗ = V4 . If it holds, which means the entire authentication process is achieved.

566

L. Wang et al.

Ui {U Ai , U Bi , U Ci } Input IDi , P Wi , BIOi Compute U Qi = BH(BIOi ) RIDi = U Ci ⊕ h(IDi  P Wi  U Qi ) RP Wi = h(P Wi  U Qi ) U Xi = U Ai ⊕ h(RIDi  RP Wi ) U Bi∗ = h(U Xi  RP Wi )

SNj {Pj }

GW N {Si .RIDi }, {Pj }

?

Check U Bi∗ = U Bi Generate ri W 1 = U Xi ⊕ r i W2 = IDi ⊕ h(U Xi  ri ) V1 = h(IDi  ri  U Xi ) {W1 , W2 , V1 , RIDi } − −−−−−−−−−−−−−− →

W3 = P j ⊕ rj V2 = h(GW N PSj  W1  rj ) {W1 , W2 , V1 , RIDi , SIDj , W3 , V2 } −−−−−−−−−−−−−−−−−−−−−−−−−−→

Retrieve {Si } in database using RIDi Compute U Xi = h(RIDi  Ks  Si ) r i = W1 ⊕ U X i IDi = W2 ⊕ h(U Xi  ri ) V1∗ = h(IDi  ri  U Xi ) ?

V2∗

ri = W5 ⊕ h(SIDj  rj ) V3∗ = h(ri  rj  GW N PSj )

Check V1∗ = V1 Compute rj∗ = W3 ⊕ Pj = h(GW N PSj  W1  rj∗ ) ?

Check V2∗ = V2 W4 = rj ⊕ h(IDi  ri ) W5 = ri ⊕ h(SIDj  rj ) V3 = h(ri  rj  GW N PSj ) {W4 , W5 , V3 } ←−−−−−−−−−

?

rj = W4 ⊕ h(IDi  ri ) SK = h(ri  rj ) V4∗ = h(ri  rj  SK)

Check V3∗ = V3 SK = h(ri  rj ) V4 = h(ri  rj  SK) {W5 , V4 } ←−−−−−−

?

Check V4∗ = V4

Fig. 44.1 Authentication and key agreement phase of Yu et al.’s protocol

44.2.2 Review of Wang et al.’s Protocol Their protocol has three phases: “sensor node registration,” “user registration,” and “login and authentication.” Here, we only review the login and authentication phase, while the details of the registration phases refer to Wang et al.’s protocol [23]. Here, we describe the process of the S N j and Ui establishing the S K with the cloud server C S. The login and authentication process is shown in Fig. 44.2, and the detailed steps are shown below. (1) Initially, Ui selects I Di , P Wi and imprints B I Oi . Next, Ui computes σi = Rep(B I Oi , τi ), U Ai = h(I Di  P Wi  σi ), and check if U Ai is correct. Then, Ui generates a timestamp Ti and a random number a, and computes U Ci = U Bi ⊕ h(I Di  σi  P Wi ), K = a · Pubcs , P I Di = I Di ⊕ K x . Finally, Ui sends {P I Di , T1 } to S N j . (2) When S N j receives the {P I Di , T1 }, it computes T I D = S I D j ⊕ h(SC j  T2  T1 ). Next, the S N j sends {T I D, R j , T2 } to Ui .

44 Security Analysis of Two Authentication and Key Agreement Protocols …

567

(3) Upon receiving {T I D, R j , T2 }, Ui calculates C = a · P, V1 = h(I Di  T I D  T Mmt  U Ci  K y ). Then, Ui transmit {P I D, T I D, R j , T1 , T2 , T Mmt , C, V1 } to C S. (4) When C S receives the messages from Ui , it first verifies freshness of T1 , and T2 . Next, C S computes SC j = h(s · R j ), S I D j = T I D ⊕ h(SC j  T2  T1 ), K = s · C, I Di = P I D ⊕ K x , V1 = h(I Di  T I D  T Mmt  U Ci  K y ), and check if V1 is correct. Then, C S generates T3 and calculates S K = h(P I D  S I D j  K  T3  SC j ), M SC K = h(I Di  U Ci  K y  K x  T Mmt ), W1 = S K ⊕ h(MC S K  T3 ), V2 = h(K y  T2  T3  S K ), W2 = S K ⊕ h(SC j  T2  T3  S I D j ), P T I D = P I D  T I D. Finally, C S sends {P T I D, W1 , V2 , W2 , T3 } to Ui . (5) Upon receiving the messages from C S, Ui computes T I D = P T I D ⊕ P I D, and check if T I D is correct. Next, Ui calculates M SC K = h(I Di  U Ci  K y  K x  T Mmt ), S K = W1 ⊕ h(MC S K  T3 ), V2 = h(K y  T2  T3  S K ), V3 = h(P I D  T3  S K ), and checks V2 . Finally, Ui sends {P T I D, W2 , V3 , T3 } to SNj. (6) When S N j receives the messages, S N j verifies freshness of T3 , and calculates P I D = P T I D ⊕ T I D, S K = W2 ⊕ h(SC j  T2  T3  S I D j ), V3 = h(P I D  T3  S K ). If the result of verification V3 is correct, which means that the entire authentication phase is over, and the Ui , S N j , and C S successfully establish the S K .

44.3 Cryptanalysis of Protocols In this section, we make cryptanalysis of Yu et al.’s protocol [30] and Wang et al.’s protocol [23], respectively, and point out the security vulnerabilities in both of their protocols.

44.3.1 Threat Model We use the Dolev-Yao [8], and the Canetti-Krawczyk [3] models to define that an attacker (A) has the following capabilities. (1) A can eavesdrop, replay, intercept, and tamper with messages on the public channel. (2) A can be an insider of an entity to gain access to the data. (3) A can capture sensor nodes and extract the data in memory through power analysis. (4) A can get access to the entity’s long-term keys and random numbers.

568

L. Wang et al. SNj {SIDj , SCj , Rj }

Ui {τi , U Ai , U Bi } Input IDi , P Wi Imprint BIOi Compute σi = Rep(BIOi , τi )

CS {IDi , T Ci }

?

T ID = SIDj ⊕ h(SCj  T2  T1 ) {T ID, Rj , T2 } −−−−−−−−−−→

Check U Ai = h(IDi  P Wi  σi ) Generate T1 , a U Ci = U Bi ⊕ h(IDi  σi  P Wi ) K = a · P ubcs P IDi = IDi ⊕ Kx {P IDi , T1 } ←−−−−−−−− C =a·P V1 = h(IDi  T ID  T Mmt  U Ci  Ky ) {P ID, T ID, Rj , T1 , T2 , T Mmt , C, V1 } − −−−−−−−−−−−−−−−−−−−−−−−−−−−− →

Check |Tc − T1 | ≤ ΔT Check |Tc − T2 | ≤ ΔT SCj = h(s · Rj ) SIDj = T ID ⊕ h(SCj  T2  T1 ) K =s·C IDi = P ID ⊕ Kx

?

?

T ID = P T ID ⊕ P ID M CSK = h(IDi  U Ci  Ky  Kx  T Mmt ) SK = W1 ⊕ h(M CSK  T3 )

V1 = h(IDi  T ID  T Mmt  T Ci  By ) SK = h(P ID  SIDj  K  T3  SCj ) M CSK = h(IDi  U Ci  Ky  Kx  T Mmt ) W1 = SK ⊕ h(M CSK  T3 ) V2 = h(Ky  T3  SK) W2 = SK ⊕ h(SCj  T2  T3  SIDj ) P T ID = P ID ⊕ T ID P T ID, W1 , V2 , W2 , T3 ←−−−−−−−−−−−−−−−−

?

Check |Tc − T3 | ≤ ΔT P ID = P T ID ⊕ T ID SK = W2 ⊕ h(SCj  T2  T3  SIDj )

V2 = h(Ky  T3  SK) V3 = h(P ID  T3  SK) {P T ID, W2 , V3 , T3 } ← −−−−−−−−−−−−−− −

?

V3 = h(P ID  T3  SK)

Fig. 44.2 Login and authentication phase of Wang et al.’s protocol

44.3.2 Cryptanalysis of Yu et al.’s Protocol We perform cryptanalysis Yu et al.’s protocol [30] and prove that the protocol suffered from sensor node capture and known session-specific temporary information attacks. The process of the attack method has been marked in Fig. 44.3.

44.3.2.1

Sensor Node Capture Attacks

Assume that A can obtain the information {P j } in the S N j and calculate the S K through the following steps. (1) A can eavesdrop on messages {W1 , W2 , V1 , R I Di , S I D j , W3 , V2 } and {W4 , W5 , V3 } on the public channel. (2) A can calculate r j = W3 ⊕ P j , and ri = W5 ⊕ h(S I D j  r j ). (3) Finally, A can successfully calculate the S K , where S K = h(ri  r j ). Thus, Yu et al.’s protocol cannot resist sensor node capture attacks.

44 Security Analysis of Two Authentication and Key Agreement Protocols … Ui {U Ai , U Bi , U Ci } Input IDi , P Wi , BIOi Compute U Qi = BH(BIOi ) RIDi = U Ci ⊕ h(IDi  P Wi  U Qi ) RP Wi = h(P Wi  U Qi ) U Xi = U Ai ⊕ h(RIDi  RP Wi ) Check U Bi∗ = h(U Xi  RP Wi )

SNj {Pj }

569

GW N {Si .RIDi }, {Pj }

?

U Bi∗ = U Bi Generate ri

W 1 = U Xi ⊕ r i W2 = IDi ⊕ h(U Xi  ri ) V1 = h(IDi  ri  U Xi ) {W1 , W2 , V1 , RIDi } − −−−−−−−−−−−−−− →

W3 = Pj ⊕ rj V2 = h(GW N PSj  W1  rj ) {W1 , W2 , V1 , RIDi , SIDj , W3 , V2 } −−−−−−−−−−−−−−−−−−−−−−−−−−→

Retrieve {Si } in database using RIDi Compute U Xi = h(RIDi  Ks  Si ) ri = W1 ⊕ U X i IDi = W2 ⊕ h(U Xi  ri ) ∗ V1 = h(IDi  ri  U Xi ) ?

Check V1∗ = V1 Compute rj∗ = W3 ⊕ Pj V2∗ = h(GW N PSj  W1  rj∗ ) ?

ri = W5 ⊕ h(SIDj  rj )

Check V2∗ = V2 W4 = rj ⊕ h(IDi  ri ) W5 = ri ⊕ h(SIDj  rj ) V3 = h(ri  rj  GW N PSj ) {W4 , W5 , V3 } ←−−−−−−−−−

V3∗ = h(ri  rj  GW N PSj ) ?

Check V3∗ = V3 SK = h(ri  rj )

rj = W4 ⊕ h(IDi  ri )

V4 = h(ri  rj  SK) {W5 , V4 } ←−−−−−−

SK = h(ri  rj ) V4∗ = h(ri  rj  SK) ?

Check V4∗ = V4

Fig. 44.3 Sensor node capture attacks and known session-specific temporary information attacks in Yu et al.’s protocol

44.3.2.2

Known Session-specific Temporary Information Attacks

Suppose that A can obtain the random number ri generated by the Ui , and A can compute the S K by following the steps below. (1) A can eavesdrop on messages {W1 , W2 , V1 , R I Di }, {W5 , V4 } on the public channel. (2) Then, A using r j to compute U X i = W1 ⊕ ri , I Di = W2 ⊕ h(U X i  ri ), and r j = W4 ⊕ h(I Di  ri ). (3) Finally, A can obtain the S K through computing S K = h(ri  r j ). Thus, Yu et al.’s protocol cannot resist known session-specific temporary information attacks.

570

L. Wang et al.

44.3.3 Cryptanalysis of Wang et al.’s Protocol We analyzed Wang et al.’s protocol [23] and found some security vulnerabilities, including sensor node capture attacks and violating perfect forward secrecy. The process of the attack method has been marked in Fig. 44.4.

44.3.3.1

Sensor Node Capture Attacks

Suppose that A can obtain the data {S I D j , SC j , R j } in the sensor node and calculate the S K through the following steps. (1) A can eavesdrop on messages {T I D, R j , T2 } and {P T I D, W2 , V3 , T3 } on the public channel. (2) A can successfully calculate the S K , where S K = W2 ⊕ h(SC j  T2  T3  S I D j ). Therefore, Wang et al.’s protocol is subjected to sensor node capture attacks.

44.3.3.2

Perfect Forward Secrecy

In this section, we use two methods to prove that Wang et al.’s protocol cannot ensure perfect forward secrecy. 1. Assume that A has access to the long-term key s of the cloud server and can compute the S K by following the steps below. (a) A can eavesdrop on messages {T I D, R j , T2 }, {P I D, T I D, R j , T1 , T2 , T Mmt , C, V1 }, and {P T I D, W1 , V2 , W2 , T3 } on the public channel. (b) Then, A using s to compute SC j = h(s · R j ), S I D j = h(SC j  T2  T1 ), and K = s · C. (c) Finally, A can obtain the S K through calculating S K = h(P I D  S I D j  K  T3  SC j ). 2. We use the concept of Ge et al.’s method [9] to prove that their protocol violates perfect forward secrecy. (a) First, the S K computation consists of variables including {P I D, S I D j , K , T3 , SC j }, where S K = h(P I D  S I D j  K  T3  SC j ). New variables are added one by one according to the rules proposed by Ge et al. [9], and treat the newly added variable as a node with an arrow pointing to the result node. For example, the computation of K requires s and C, and the nodes C and s point to the K node. (b) Next, coloring all nodes that are long-term confidential or messages transmitted on public channels. These nodes are {P I D, T I D, T1 , T2 , C, s, T3 , R j }.

44 Security Analysis of Two Authentication and Key Agreement Protocols … SNj {SIDj , SCj , Rj }

Ui {τi , U Ai , U Bi } Input IDi , P Wi Imprint BIOi Compute σi = Rep(BIOi , τi )

571

CS {IDi , T Ci }

?

T ID = SIDj ⊕ h(SCj  T2  T1 ) {T ID, Rj , T2 } −−−−−−−−−−→

Check U Ai = h(IDi  P Wi  σi ) Generate T1 , a U Ci = U Bi ⊕ h(IDi  σi  P Wi ) K = a · P ubcs P IDi = IDi ⊕ Kx {P IDi , T1 } ←−−−−−−−− C =a·P V1 = h(IDi  T ID  T Mmt  U Ci  Ky ) {P ID, T ID, Rj , T1 , T2 , T Mmt , C, V1 } − −−−−−−−−−−−−−−−−−−−−−−−−−−−− →

Check |Tc − T1 | ≤ ΔT Check |Tc − T2 | ≤ ΔT SCj = h(s · Rj ) SIDj = T ID ⊕ h(SCj  T2  T1 )

?

K =s·C IDi = P ID ⊕ Kx

V1 = h(IDi  T ID  T Mmt  T Ci  By ) SK = h(P ID  SIDj  K  T3  SCj )

?

T ID = P T ID ⊕ P ID M CSK = h(IDi  U Ci  Ky  Kx  T Mmt ) SK = W1 ⊕ h(M CSK  T3 )

M CSK = h(IDi  U Ci  Ky  Kx  T Mmt ) W1 = SK ⊕ h(M CSK  T3 ) V2 = h(Ky  T3  SK) W2 = SK ⊕ h(SCj  T2  T3  SIDj ) P T ID = P ID ⊕ T ID P T ID, W1 , V2 , W2 , T3 ←−−−−−−−−−−−−−−−−

?

Check |Tc − T3 | ≤ ΔT P ID = P T ID ⊕ T ID

V2 = h(Ky  T3  SK) V3 = h(P ID  T3  SK) {P T ID, W2 , V3 , T3 } ← −−−−−−−−−−−−−− −

SK = W2 ⊕ h(SCj  T2  T3  SIDj ) ?

V3 = h(P ID  T3  SK)

Fig. 44.4 Sensor node capture attacks and violating perfect forward secrecy in Wang et al.’s protocol

(c) Finally, all the input edges of the colored nodes are removed, and the graph composed of the remaining nodes is used to determine whether the protocol satisfies the perfect forward secrecy. The final result is shown in Fig. 44.5. Since all leaf nodes in the graph are colored, the A can access these variables to calculate the S K . Therefore, the protocol does not satisfy perfect forward secrecy.

44.4 Conclusion and Suggestion In this paper, we first describe the applications of the WSN in people’s lives and review the AKA protocols based on WSN environments. Secondly, we review and make cryptanalysis of Yu et al.’s protocol and Wang et al.’s protocol. We point out security vulnerabilities in both protocols, including sensor node capture attacks, known session-specific temporary information attacks, and violating perfect forward secrecy.

572

L. Wang et al.

Fig. 44.5 The result of Wang et al.’s protocol.

Then, several suggestions for both protocols are provided. In Yu et al.’s protocol, we suggest W1 = U X i ⊕ I Di and W2 = ri ⊕ h(U X i  I Di ), so the protocol can be resistant to known session-specific temporary information attacks. Moreover, a pseudo-identity can be generated for the sensor node in the protocol to resist sensor node capture attacks. Then the real identity transmitted on the public channel can be replaced with a pseudo-identity. In Wang et al.’s protocol, the private value of the cloud server must be added to the operation with the private key to ensure perfect forward secrecy. Acknowledgements This research was partially supported by Natural Science Foundation of Shandong Province, China (Grant no. ZR202111230202)

References 1. Alladi, T., Chamola, V., et al.: HARCI: a two-way authentication protocol for three entity healthcare IoT networks. IEEE J. Sel. Areas Commun. 39(2), 361–369 (2020) 2. Bera, B., Das, A.K., Balzano, W., Medaglia, C.M.: On the design of biometric-based user authentication protocol in smart city environment. Pattern Recognit. Lett. 138, 439–446 (2020) 3. Canetti, R., Krawczyk, H.: Analysis of key-exchange protocols and their use for building secure channels. In: International Conference on the Theory and Applications of Cryptographic Techniques, vol. 2045, pp. 453–474. Springer (2001) 4. Chang, I.P., Lee, T.F., Lin, T.H., Liu, C.M.: Enhanced two-factor authentication and key agreement using dynamic identities in wireless sensor networks. Sensors 15(12), 29841–29854 (2015) 5. Chen, C.M., Deng, X., Gan, W., Chen, J., Islam, S.: A secure blockchain-based group key agreement protocol for IoT. J. Supercomput. 77(8), 9046–9068 (2021) 6. Chen, J.N., Huang, Z.J., Zhou, Y.P., Zou, F.M., Chen, C.M., Wu, J.M.T., Wu, T.Y.: Efficient certificate-based aggregate signature scheme for vehicular ad hoc networks. IET Netw. 9(6), 290–297 (2020) 7. Chen, J.N., Zhou, Y.P., Huang, Z.J., Wu, T.Y., Zou, F.M., Tso, R.: An efficient aggregate signature scheme for healthcare wireless sensor networks. J. Netw. Intell. 6(1), 1–15 (2021) 8. Dolev, D., Yao, A.: On the security of public key protocols. IEEE Trans. Inf. Theory 29(2), 198–208 (1983)

44 Security Analysis of Two Authentication and Key Agreement Protocols …

573

9. Ge, M., Kumari, S., Chen, C.M.: AuthPFS: a method to verify perfect forward secrecy in authentication protocols. J. Netw. Intell. 7, 734–750 (2022) 10. Ghani, A., Mansoor, K., Mehmood, S., Chaudhry, S.A., Rahman, A.U., Najmus Saqib, M.: Security and key management in IoT-based wireless sensor networks: an authentication protocol using symmetric key. Int. J. Commun. Syst. 32(16), e4139 (2019) 11. Kumar, N., Aujla, G.S., Das, A.K., Conti, M.: ECCAuth: a secure authentication protocol for demand response management in a smart grid system. IEEE Trans. Ind. Inf. 15(12), 6572–6582 (2019) 12. Kwon, D.K., Yu, S.J., Lee, J.Y., Son, S.H., Park, Y.H.: WSN-slap: Secure and lightweight mutual authentication protocol for wireless sensor networks. Sensors 21(3), 936 (2021) 13. Li, X., Liu, S., Kumari, S., Chen, C.M.: PSAP-WSN: a provably secure authentication protocol for 5g-based wireless sensor networks. Comput. Model. Eng. Sci. 135(1), 711–732 (2023) 14. Li, Z., Miao, Q., Chaudhry, S.A., Chen, C.M.: A provably secure and lightweight mutual authentication protocol in fog-enabled social internet of vehicles. Int. J. Distrib. Sens. Netw. 18(6), 15501329221104332 (2022) 15. Liu, S., Chen, C.M.: Comments on “a secure and lightweight drones-access protocol for smart city surveillance”. IEEE Trans. Intell. Transp. Syst. (2022). https://doi.org/10.1109/TITS.2022. 3198045 16. Luo, Y., Zheng, W., Chen, Y.C.: An anonymous authentication and key exchange protocol in smart grid. J. Netw. Intell. 6(2), 206–215 (2021) 17. Mo, J., Hu, Z., Shen, W.: A provably secure three-factor authentication protocol based on chebyshev chaotic mapping for wireless sensor network. IEEE Access 10(3), 12137–12152 (2022) 18. Moghadam, M.F., Nikooghadam, M., Al Jabban, M.A.B., Alishahi, M., Mortazavi, L., Mohajerzadeh, A.: An efficient authentication and key agreement scheme based on ECDH for wireless sensor network. IEEE Access 8, 73182–73192 (2020) 19. Park, Y., Park, Y.: Three-factor user authentication and key agreement using elliptic curve cryptosystem in wireless sensor networks. Sensors 16(12), 2123 (2016) 20. Shin, S., Kwon, T.: A lightweight three-factor authentication and key agreement scheme in wireless sensor networks for smart homes. Sensors 19(9), 2012 (2019) 21. Tso, R., Huang, K., Chen, Y.C., Rahman, S.M.M., Wu, T.Y.: Generic construction of dual-server public key encryption with keyword search on cloud computing. IEEE Access 8, 152551– 152564 (2020) 22. Vasudev, H., Deshpande, V., Das, D., Das, S.K.: A lightweight mutual authentication protocol for v2v communication in internet of vehicles. IEEE Trans. Veh. Technol. 69(6), 6709–6717 (2020) 23. Wang, Z., Sun, P., Luo, N., Guo, B.: A three-party mutual authentication protocol for wearable IoT health monitoring system. In: 2021 IEEE International Conference on Smart Internet of Things (SmartIoT), pp. 344–347. IEEE (2021). https://doi.org/10.1109/SmartIoT52359.2021. 00063 24. Wu, F., Li, X., Xu, L., Vijayakumar, P., Kumar, N.: A novel three-factor authentication protocol for wireless sensor networks with IoT notion. IEEE Syst. J. 15(1), 1120–1129 (2020) 25. Wu, T.Y., Lee, Y.Q., Chen, C.M., Tian, Y., Al-Nabhan, N.A.: An enhanced pairing-based authentication scheme for smart grid communications. J. Ambient Intell. Hum. Comput. (2021). https://doi.org/10.1007/s12652-020-02740-2 26. Wu, T.Y., Yang, L., Lee, Z., Chu, S.C., Kumari, S., Kumar, S.: A provably secure three-factor authentication protocol for wireless sensor networks. Wirel. Commun. Mob. Comput. 2021, 5537018 (2021) 27. Xie, Q., Li, K., Tan, X., Han, L., Tang, W., Hu, B.: A secure and privacy-preserving authentication protocol for wireless sensor networks in smart city. EURASIP J. Wirel. Commun. Netw. 2021(1), 1–17 (2021) 28. Yang, L., Chen, Y.C., Wu, T.Y.: Provably secure client-server key management scheme in 5g networks. Wirel. Commun. Mob. Comput. 2021, 4083199 (2021)

574

L. Wang et al.

29. Yavari, M., Safkhani, M., Kumari, S., Kumar, S., Chen, C.M.: An improved blockchain-based authentication protocol for IoT network management. Secur. Commun. Netw. 2020, 8836214 (2020) 30. Yu, S., YoungLee, J., Kim, M., Park, Y.: A secure biometric based user authentication protocol in wireless sensor networks. In: 2020 10th Annual Computing and Communication Workshop and Conference (CCWC), pp. 0830–0834. IEEE (2020). https://doi.org/10.1109/CCWC47524. 2020.9031136

Chapter 45

Face Mask Detection Based on YSK Neural Network for Smart Campus Li Yu

Abstract In this paper, a new YSK neural network is proposed to detect face mask wearing to prevent COVID-19 or other infectious diseases. And the aim is not only to detect the face mask wearing person, but also can detect, track and warn the notwearing face mask person. The technique we applied is referred to as object detection based on deep learning. Several experiments were made to test the performance of the proposed model and it showed that it has better performance than other common detection models and the result is excellent.

45.1 Introduction With the spread of COVID-19, which is a highly infectious respiratory disease to affect all over the world, it is necessary for the people to wear a face mask when being away from home for prevention of infection especially in the campus and other public places. What’s more, driven by the need for efficient face mask wearing detection, deep learning is of great importance. Therefore, it has become more and more essential to detect face mask wearing using deep learning method through the camera or other mobile devices to prevent COVID-19 or other infectious diseases. It is common that the detection method [1–5, 13] based on deep learning often falls into two major groups. One is one-stage detector [1, 2], and the other is twostage detector [3–5]. And the information of category and location can be accessed through the backbone network in one-stage detector while the RPN should be trained extra in two-stage detector. So one-stage detector is faster in speed but the accuracy is slightly lower than that of the two-stage detector. Although there is a variety of face mask wearing detection research thathas been gained around in recent years, how to improve and balance the face mask detection speed and accuracy is still the problem to be studied. What’s more, most of studies L. Yu (B) University of International Relations, Beijing 100094, China e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6_46

575

576

L. Yu

mainly studied face mask wearing detection but have neglected the warning of the not-wearing person. Therefore, to solve the problems mentioned in the previous paragraph, in our study, a new YSK network is proposed to detect face mask wearing in real time and track and warn the not-wearing face mask person. And it can be applied in public places especially in the campus.

45.2 Related Work In recent years, there are many literature based on deep learning applied in all kinds of fields [14–22]. In 2021, Mohamed et al. [19] proposed to detect image steganography using multimodal deep learning. In 2022, Ma et al. [14] detected lymph node based on deep learning. And Chen et al. [18] proposed to detect contraband using deep learning method in 2022. Therefore, it is popular that many studies on detecting face mask wearing use deep learning detection method. In 2021, Boulila et al. [6] presented to use MobileNetV2 to detect facemask in real time. And Pushparaj et al. [7] proposed the face mask detection using CNN method also in 2021. During the same period, Khan et al. [8] utilized the transfer learning method for face mask detection during COVID19. Rahman et al. [9] put forward a mask surveillance system using deep learning, Internet of Thing and Blockchain method in 2022. Saravanan et al. [10] proposed to use pretrained model VGG16 to detect face mask in the same year. In this section, the architecture of the network and the data used in the work will be introduced.

45.2.1 YSK Backbone Structure The backbone network referenced in the proposed YSK neural network can be shown in Fig. 45.1, and more module detail can be shown in Fig. 45.2. In Fig. 45.1, it can be seen that the input image can be resized to 640 × 640 and then to the backbone network. And through the head layer network, it outputs the feature maps of three layers with different sizes. Finally, the prediction results can be output through REP and convolution layers. And in Fig. 45.2, it is shown that there are 50 layers in the backbone network. Firstly, the input image will pass through four convolution layers namely CBS module, which is consisted of Conv, BN and SiLU. Then it will pass through ELAN module, which is composed of many CBS modules. Then it is followed by three MP adding ELAN module. The ELAN module is an efficient network architecture and more robust. It can control the shortest and longest gradient paths to learn more features. There are two branches in the ELAN module. And in the ELAN module, the input and output feature sizes remain the same while the number of channels will be changed in the

45 Face Mask Detection Based on YSK Neural Network for Smart Campus

577

Fig. 45.1 An illustration of backbone network (The figure is presented by zhangjianhu)

Fig. 45.2 Details of the structure of backbone network

first two CBS modules and the following input channels are consistent with the output channels. So the final required channel is output by the last CBS module. And there are also two branches in MP module to perform down-sample. The first branch goes through a maxpool to perform down-sample and then changes the number of channels by the 1 × 1 convolution. And the second branch first goes through a 1 × 1 convolution to change the number of the channel, then goes through a 3 × 3 convolution kernel with a step size of 2 convolution block, which is also used for downsampling. Finally, the results of the first branch and the second branch are added together to obtain the results of super downsampling.

578

L. Yu

The function of SPP module is to increase the receptive field so that the algorithm can adapt to different resolution images, in which the different receptive fields are obtained by maxpooling. And in CSP module, the feature is divided into two parts. One part is processed by conventional processing, and the other part is processed by SPP structure. Therefore, the SPPCSPC module in the backbone network can merge the two parts together to reduce the calculation amount by half, which makes the speed faster and the accuracy improved. The REP module is also divided into two parts, one is training and the other is deploying, which is reasoning. In reasoning module, a 3 × 3 convolution and stride 1 are contained. It is transformed by the reparameterization of the training module.

45.2.2 KCF Tracker In our work, KCF tracker is utilized in our proposed network to track the face mask person especially the not-wearing face mask person. The main idea of the Kernel Correlation Filter [11] is to expand the number of negative samples to enhance the performance of the tracker, and the way to expand the negative samples is to use the cyclic matrix construction method. In KCF, the gray level feature that can only be used with a single channel is improved to HOG feature or other features that can be used with multiple channels, and it performs better in the existing algorithm. The advantages of KCF are that the calculation amount is greatly reduced and the calculation speed is improved. Therefore, it can meet the real-time requirements. The circulant matrix from the 1D vector can be shown in Fig. 45.3 and the shift of 2D images after different number of cycles can be shown in Fig. 45.4.

Fig. 45.3 The circulant matrix from the 1D vector [11]

Fig. 45.4 The shift of 2D images after different number of cycles [11]

45 Face Mask Detection Based on YSK Neural Network for Smart Campus

579

Fig. 45.5 The structure of the SA-Net

45.2.3 SA-Net Attention Module In our work, SA-Net module shown in Fig. 45.5 is added to our neural network to improve the performance of the network without increasing the computational cost. It can be seen that in Fig. 45.5, there are five parts in the SA module, feature grouping, channel attention, spatial attention and aggregation. The feature grouping is mainly used to group input features. In channel attention, the gap, scale and sigmoid are fused to make it more light. And the spatial attention is the complement to the channel attention. Finally, it integrated the previous two attention calculations in the aggregation module.

45.2.4 Data In our work, the data augmentation method has been utilized. As shown in Fig. 45.6, the coco dataset is pretrained in our model. And COCO dataset is a large-scale dataset that can be used for image detection, semantic segmentation and image captioning. What’s more, it contains 1.5 million objects, 80 object categories and 91 stuff categories. Besides, each image contains a five-sentence description of an image, also together with 250,000 pedestrians with keypoints. What’s more, the face mask dataset is also trained as shown in Fig. 45.7 in our model to have better performance in detecting face mask. It has collected all kinds of persons wearing face mask. And data augmentation is used in our data to improve the generalization ability and robustness of the model. The color jittering, PCA jittering, random scale, random crop, horizontal and vertical flip, shift, rotation and reflection and noise are used in our data augmentation.

45.3 Experiments The experiments were tested on NVIDIA GeForce RTX 2060, which is 6 GB, and the system in our device is Win 10. Figure 45.8 shows the train loss, val loss, precision,

580

L. Yu

Fig. 45.6 The COCO dataset

Fig. 45.7 The face mask dataset

recall and mAP. It is found that the loss of our model is lower than common detection models, and the precision and mAP are higher than that. The face mask detection visualization can be shown in Fig. 45.9. And more detail of the model training metrics can be seen in Figs. 45.10, 45.11, 45.12. It shows the F1 metrics of the proposed model in Fig. 45.10, the precision of the proposed model in Fig. 45.11 and the recall of the proposed model in Fig. 45.12. It can be seen that the precision of the proposed model in detecting face mask wearing is 0.927, the recall is 0.708 and the mAP50 is 0.848, which is higher and has better performance than other common detection models.

45.4 Conclusion In conclusion, it can be seen that the proposed model has better performance than other common detection models through the testing experiments. And it can not only detect the face mask wearing person, but also the not-wearing face mask person can

45 Face Mask Detection Based on YSK Neural Network for Smart Campus

Fig. 45.8 The metrics of YSK model

Fig. 45.9 The visualization of the face mask detection

581

582

L. Yu

Fig. 45.10 The F1 of YSK model

Fig. 45.11 The precision of YSK model

be detected and warned. However, data amount remains the problem to be solved and more data needs to be collected and processed to improve the performance of the model.

45 Face Mask Detection Based on YSK Neural Network for Smart Campus

583

Fig. 45.12 The recall of YSK model

References 1. Liu, W., Anguelov, D., Erhan, D., et al.: SSD: Single shot multi box detector. (2015) 2. Joseph, R., Santosh, D., Ross, G., Ali, F.: You only look once: Unified, real-time object detection. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 779–788 (2016). 2 3. Ren, S., He, K., Girshick, R. et al.: Faster R-CNN: Towards real-time object detection with region proposal networks. (2015) 4. Lin, T.Y., Dollár, P., Girshick, R. et al.: Feature pyramid networks for object detection. arXiv e-prints, (2016) 5. He, K., Zhang, X., Ren, S., et al.: Spatial pyramid pooling in deep convolutional networks for visual recognition//European conference on computer vision. Springer, Cham, (2014) 6. Boulila, W., Alzahem, A., Almoudi, A., et al.: A deep learning-based approach for real-time facemask detection, (2021) 7. Pushparaj, A., Rajendran, R., Bj, S., et al.: Medical Face Mask Detection Using CNN. JETIR. (www.jetir.org), 2021 8. Khan, J.Y., Alamin, M.: A comparative analysis of machine learning approaches for automated face mask detection during Covid-19. (2021) 9. Rahman, W., Mudawi, N.A., Alazeb, A., et al.: IoT and blockchain-based mask surveillance system for covid-19 prevention using deep learning. 2022(007):000 10. Saravanan, T.M, Karthiha, K., Kavinkumar, R. et al.: A novel machine learning scheme for face mask detection using pretrained convolutional neural network. (2022) 11. Henriques, J.F., Rui, C., Martins, P., et al.: High-speed tracking with kernelized correlation filters. (2014) 12. Jie, H., Li, S., Samuel, A., Gang, S., Enhua, W., et al.: Squeeze-and-excitation networks, CVPR. (2019) 13. Zhang, F., Wu, T.Y., Zheng, G.: Video salient region detection model based on wavelet transform and feature comparison. EURASIP J. Image Video Process. 2019, 58 (2019) 14. Ma, Y., Peng, Y., Wu, T.Y.: Transfer learning model for false positive reduction in lymph node detection via sparse coding and deep learning. J. Intell. & Fuzzy Syst. 43 (2), 2121-2133 (2022)

584

L. Yu

15. Zhang, F.,Wu, T.Y., Wang, Y., Xiong, R., Ding, G., Mei, P., Liu, L.: Application of quantum genetic optimization of LVQ neural network in smart city traffic network prediction. IEEE Access. 8, 104555-104564 (2020) 16. Sun, Z., Pan, J.-S., Pan, T.-S., Chen, C.-H.: Deep learning-based probability model for traffic information estimation. Journal of Network Intelligence 7(3), 592–607 (2022) 17. Li, Y.: A suvey on edge intelligent video surveillance with deep reinforcement learning. Journal of Network Intelligence 7(1), 70–83 (2022) 18. Chen, H., Lu, Z.: Contraband detection based on deep learning. J. Inf. Hiding Multimed. Signal Process. 13(3), 165-177 (2022) 19. Elshafey, M.A., Amein, A.S., Badran, K.S.: Universal image steganography detection using multimodal deep learning framework. Journal of Information Hiding and Multimedia Signal Processing 12(3), 152–161 (2021) 20. Wu, M.E., Syu, J.H., Chen, C.M.: Kelly-based options trading strategies on settlement date via supervised learning algorithms. Comput. Econ. 59 (4), 1627-1644 (2022) 21. Kumar, S., Damaraju, A., Kumar, A., Kumari, S., Chen, C.-M.: LSTM network for transportation mode detection. Journal of Internet Technology 22(4), 891–902 (2021) 22. Chen, C.-M., Chen, L., Gan, W., Qiu, L., Ding, W.: Discovering high utility-occupancy patterns from uncertain data. Inf. Sci. 546, 1208–1229 (2021)

Author Index

B Bai, Li, 161

C Cai, Qiqin, 251, 291 Chen, Chien-Ming, 515 Chen, Ding, 3, 15 Chen, Dingjun, 67 Chengli, Zhu, 57 Chen, Mei-Feng, 229 Chen, Tao, 93 Chen, Xu, 161 Chen, Zerui, 537 Chiu, Yi-Jui, 203, 239 Chu, Shu-Chuan, 333, 343, 353

D Deng, Tian-Hang, 239 Dong, Ziqi, 133

F Fan, Yijie, 373 Fu, Zonglin, 321

G Gao, Fan, 161 Gao, Yiyuan, 27 Geng, Fang-Dong, 453 Guan, Yin, 463 Guo, Feng, 251, 263, 277, 291, 501 Guo, Xiuyun, 27, 189

Guo, Yundong, 321

H Han, Hui, 27 Hanji, Li, 363 Hao, Yiru, 515 He, Wenhui, 189 He, Youpeng, 425 Hou, Jia-Zheng, 439 Huang, Shibin, 263 Huang, Xin, 385 Hu, Bin, 3 Hu, Houpeng, 525, 537 Hu, Rui-Bin, 453 Hu, Xin-Yue, 439

I Islam SK Hafizul, 311

J Jiang, Hang, 463 Jiang, Jin, 219 Jiang, Lei, 463 Jiang, Pei-Zhou, 15 Jingjing, Huang, 363 Ju, Cheng-Xiao, 229

K Kang, Ping, 411 Kong, Fangfang, 551 Kumari, Saru, 491, 515, 551, 563

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 S. Ni et al. (eds.), Advances in Smart Vehicular Technology, Transportation, Communication and Applications, Smart Innovation, Systems and Technologies 347, https://doi.org/10.1007/978-981-99-0848-6

585

586 L Lakshmanna, Kuruva, 515 Liang, Lulu, 333 Liao, Lyuchao, 425 Liao, Qing, 385 Li, Ben, 425 Li, Fusheng, 525 Li, Hangfeng, 525 Li, Lexi, 67, 147 Li, Minggui, 463 Li, Nan, 501 Lin, Ziyang, 501 Li, Pengcheng, 525 Li, Tan, 79 Liu, Fei-Fei, 343 Liu, Kun, 175 Liu, Qiuqi, 189 Liu, Shutang, 463 Liu, Yang, 373, 385 Liu, Zhen, 111, 147 Li, Xueting, 111 Li, Yanlong, 479 Li, Yi, 93, 175 Li, Yongxin, 111 Li, Yongyu, 463 Lu, Hongxia, 133 Luo, Yi, 537 Luo, Yongyu, 251 Lv, Li, 411 Lv, Miaomiao, 175 M Mao, Qingyun, 219 Mondal, Shukla, 311 N Ni, Shaoquan, 57, 67, 147 O Ou, Jiaxiang, 525, 537 P Pal, Arup Kumar, 311 Pan, Jeng-Shyang, 321, 343 Pan, Jinshan, 43 Pan, Tien-Szu, 333, 343 Pan, Yan Peng, 397 Q Qian, Bin, 537

Author Index Qi, Haige, 219 Qingguo, Song, 57 Qi, Shuhan, 373, 385 R Ren, Qiang, 263 S Samanta, Debabrata, 311 Shao, Zhi-Yuan, 353 Shi, Fangyu, 43, 133 Shieh, Chin-Shiuh, 353 Shih, Yung-Hui, 203, 239 Snášel, Václav, 321 Song, Xiaohe, 161 Song, Zhi-bin, 411 Song, Zongying, 175 Sun, Wei, 479 Sun, Yunhao, 93 T Tang, Linlin, 373, 385 Tan, Haowen, 43 Tao, Jun, 219 Tian, Junshan, 291 W Wang, Bozhou, 67, 147 Wang, Haolin, 277 Wang, Jianchen, 479 Wang, Liyang, 563 Wang, Meng, 111, 133, 189 Wang, Pei, 219 Wang, Ruo-Bin, 453 Wang, Yongcheng, 93 Wenxin, Zhang, 57 Wu, Haozhi, 491 Wu, Jianjun, 479 Wu, Jie, 343 Wu, Run-Xiu, 439 Wu, Songyang, 251 Wu, Tsu-Yang, 333, 491, 551, 563 Wu, Yong, 239 Wu, Zhi, 43 X Xia, Chenxi, 501 Xiao, Hankun, 425 Xiao, Hui-Jun, 229 Xiao, Yanhong, 537

Author Index Xue, Chen Chen, 397 Xu, Gen, 291 Xu, Lin, 453 Y Yang, Li Jie, 397 Yang, Min, 43 Yang, Wen-Qi, 203 Ye, Xuze, 161 Yilin, Yang, 57 Yu, Li, 575 Yuan, Yu-Yang, 203 Yue, Mengyuan, 175 Yu, Xiang, 501 Z Zeng, Ren-xian, 411 Zeng, Xiaoxu, 161

587 Zhang, Bo, 219 Zhang, Chi, 111 Zhang, Fan, 525 Zhang, Fu Quan, 397 Zhang, Jiajia, 385 Zhang, Xiaoqing, 353 Zhang, Xue Ting, 397 Zhang, Yabin, 479 Zhang, Yong, 219 Zhang, Zhi-Cheng, 15 Zhao, Jia, 439 Zhao, Lifeng, 219 Zhao, Wenqing, 425 Zhao, Yong, 219 Zhou, Mi, 525 Zhou, Xuyang, 479 Zou, Fumin, 251, 263, 277, 291, 501 Zou, Xinqian, 27 Zou, Zhong-De, 3