A Resilient Control Strategy for Train-to-Train Communications under Jamming Attacks

Communication-Based Train Control (CBTC) systems rely on wireless communications to enhance the efficiency of railway operations. Classical CBTC systems incorporate bidirectional train-to-wayside (T2W) communications through which trains send their status information to wayside units. T2T-CBTC systems represent a burgeoning direction in the future of CBTC. They adopt train-to-train (T2T) communications for adjacent trains to share status information. T2T communications simplify the architecture of traditional CBTC networks and reduce transmission delays. Wireless communications can introduce cybersecurity threats to inter-train communications. This work proposes two resilient control strategies for T2T-CBTC systems to mitigate the effects of jamming. Both strategies are based on multi-agent deep reinforcement learning and aim to control trains’ operations under jamming attacks, allowing them to continue operating safely instead of applying emergency braking. One strategy is based on the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm, and the other is based on the Multi-Agent Twin Delayed Deep Deterministic Policy Gradient (MATD3) algorithm. The strategies are implemented and compared with an existing strategy based on MADDPG. The experimental results indicate that the proposed MADDPG-based strategy shortens the convergence time by 32% to 42%, while the MATD3-based strategy achieves a reduction of 40% to 48%, compared to the baseline.