Reinforcement Learning-Based Security Enhancement for Controlled Optimization of Phases in Intelligent Traffic Signal System

• Published in: IEEE transactions on industrial cyber-physical systems 2024, Vol.2, p.575-587
• Main Authors: Qiao, Ziyan, Xiang, Yingxiao, Baker, Thar, Li, Gang, Wu, Yalun, Tong, Endong, Peng, Shuanghe, Zhu, Ye, Xu, Dongwei, Niu, Wenjia
• Format: Article
• Language: English
• Subjects: congestion attack; connected vehicle; Connected vehicles; Cyber-physical systems; Delays; Industrial cyber-physical systems; intelligent traffic signal systems; Long short term memory; Optimization; Real-time systems; Roads; Security; traffic signal control; Transportation; Turning
• ISSN: 2832-7004; 2832-7004
• DOI: 10.1109/TICPS.2024.3476455

With the rise of intelligent devices within Industrial Cyber-Physical Systems (ICPS), encompassing applications in traffic signal control, the vulnerability of devices such as On-Board Units and Roadside Units to attacks of data spoofing becomes a critical issue. Congestion attacks pose a significant threat to traffic security, capable of causing traffic jams by manipulating traffic signal control methods on the Internet of Vehicles. One such method is the Controlled Optimization of Phases algorithm, which is susceptible to congestion attacks. In this paper, we propose a security enhancement approach integrating advantage actor-critic reinforcement learning with a long short-term memory network. Our work extends the application of security enhancement methodologies to the context of intelligent traffic systems, that synchronously detects and disposes of attacks, ensuring the physical safety and reliable operation of ICPS. The experimental results and analysis exhibit the efficiency of our approach in terms of processing time and effectiveness.