Deep Deterministic Policy Gradient-based Virtual Coupling Control For High-Speed Train Convoys

This work addresses the problem of Virtual Coupling (VC) control for uncertain heterogeneous nonlinear autonomous trains convoys sharing information among each other with Radio Block Center (RBC) and via Train-2-Train (T2T) communication network. To solve the problem we propose a novel no-supervised actor-critic Deep Deterministic Policy Gradient-based (DDPG) controller which drives each train within the convoy to track the reference behaviour, as imposed by the RBC, while maintaining a desired inter-train distance w.r.t. the preceding train. The effectiveness of the proposed approach is evaluated via a numerical analysis which is carried out in Python environment. The first step of validation involves the efficiency of the training process and discloses how the agent has learned the correct behaviour to track the train ahead. Then, we numerically prove how the overall closed-loop trains convoy under the action of the DDPG controller reaches the VC formation despite the presence of external disturbances acting on the train dynamics.