DDPG-based Energy-Efficient Train Speed Trajectory Optimization under Virtual Coupling

Train operation control under virtual coupling has become an efficient method to increase the capacity of the railway, especially in the scenario of emergencies represented by the temporary speed restriction. To realize safe, efficient, and energy-saving operations for the following train under virtual coupling, this paper proposes a deep reinforcement learning method for follower speed trajectory optimization based on the given speed trajectory of the leader. The agent outputs the continuous values in [ 1] to control the follower to brake or accelerate. A collision-avoidance protection mechanism is conducted in the learning process to ensure feasible action. Numerical experiments are carried out based on real railway data. The results indicate that the proposed method can achieve a more energy-efficient process of tight and safe following between the adjacent trains in less computation time.