Coordinated Operation Strategy Design for Virtually Coupled Train Set: A Multiagent Reinforcement Learning Approach

A virtually coupled train set (VCTS) is regarded as a complete train. To better serve passengers, train units (TUs) in a VCTS require synchronous stopping and departure with small tracking distances during the station operation stage. However, under the existing predecessor–follower framework, the TUs operate independently. The latter TU is limited by vehicle performance and safety constraints, making it unable to operate synchronously with the front TU. Moreover, the characteristics of a few midway evaluation indicators and complex constraints also make it difficult to solve the operation strategy of the VCTS. This article realizes the synchronous operation of a VCTS through coordination and designs coordinated operation guidance. On this basis, an optimization model for the synchronous operation of a VCTS is constructed, and it is solved using a multiagent reinforcement learning algorithm to obtain a coordinated driving curve. First, guidance curves are generated based on the guidance of coordinated operation to guide the initial search of the algorithm. Second, the algorithm takes multimirror scene synchronization training to quickly accumulate data and design a prediction-based action management module to adjust violation decisions. Finally, the data supervision and management module ensures the effective learning of valuable data. The coordinated driving curve generated based on the Hebei Jingche test line has good synchronization indicators, and the realization of the field vehicle tracking curve proves the engineering usability.