SAC-based collaborative tracking control of virtual coupling trains

Abstract To facilitate coordinated operation of virtual coupling trains, this study proposes a leader–follower cooperative control framework based on the soft actor-critic (SAC) algorithm. The proposed framework optimizes the leading train speed profile and trains the following train controller to achieve speed convergence and safe spacing. The research initially develops a leader-train dynamics model and examines train operation conditions. It formulates a reward function predicated on energy consumption and punctuality, subsequently utilizing a SAC algorithm to optimize the speed trajectory, so producing a recommended driving speed curve for the leading train. Secondly, a Markov decision process model is developed utilizing the speed curve of the leading train as input, while the speed error and position error between the leading and following trains function as rewards. A train operational environment is established, and the speed of the train unit is regulated using the SAC algorithm to attain formation coordination. Ultimately, simulation experiments are performed utilizing the Beijing Subway Yizhuang Line route and train data for validation purposes. The results indicate that, in comparison to the traditional traction braking curve, this algorithm realizes a 7.84% reduction in energy consumption while satisfying the punctuality standards for urban rail transit. It effectively attains the collaborative goals of speed convergence and consistent spacing with the follower train.