Deep Reinforcement Learning-based Resilient Control Method for CBTC Systems through Train-to-Train Communications under Adversarial Attacks

In recent years, intelligent transportation systems and train autonomous driving technologies have been developed rapidly. In the communication-based train control (CBTC) system through the train-to-train communication, the train-to-train communication system is an important subsystem of the train control system, which uses wireless communication channels to transmit control commands. When subjected to malicious attacks, for example the DoS attack, the traditional control methods cannot adapt well. Therefore, the resilient control method based on deep reinforcement learning is proposed in this paper under Denial of Service (DoS) attacks. By training the controller in the case of the DoS attack, the resilient control method makes the train continue to operate with the acceptable performance decline, instead of emergency braking under traditional control methods. The experimental results show that the proposed method can effectively control the train operation under the DoS attack.