Deep Deterministic Policy Gradient Virtual Coupling control for the coordination and manoeuvring of heterogeneous uncertain nonlinear High-Speed Trains

The new railway mobility paradigm Virtual Coupling, recognized as one of the most promising solutions to face the ever-increasing demand in railway transportation and the saturation of network capacity, enables two or more trains to virtually couple in a single convoy, hence reducing the headway among them. Within this context, the design of an effective and robust Virtual Coupling control strategy for uncertain nonlinear heterogeneous train convoys, able to simultaneously cope with uncertain nonlinearities and unexpected/unpredictable external factors, is an open challenge in the railway field. This is very crucial for High-Speed Trains, where, due to the high-speed operating ranges, uncertain factors have a stronger impact on Virtual Coupling performance. To deal with this issue, this work exploits the ability of Deep Reinforcement Learning based strategies in defining an optimal control policy via an iterative exploration of the surrounding unknown environment and without detailed knowledge of the plant dynamics. Specifically, we propose a novel Deep Deterministic Policy Gradient based control strategy to coordinate and manage the High-Speed Trains convoy such that this latter can autonomously adapt its behaviour to all the encountered driving scenarios. The effectiveness of the proposed approach is evaluated via simulation analyses, carried-out via an ad-hoc implemented Virtual Coupling Train System simulation platform. After verifying the efficiency of the training process in ensuring the fulfilment of the Virtual Coupling control objectives, extensive non-trivial simulations, also involving cooperative manoeuvres, are performed for the validation phase. Results confirm how the proposed model-free Deep Deterministic Policy Gradient approach guarantees the Virtual Coupling for nonlinear heterogeneous High-Speed Train convoys despite the co-presence of uncertainties and unknown external factors. Finally, the advantages and the benefits of the proposed data-driven control are disclosed via a comparison analysis against model-based control strategies.