The invention discloses a deep reinforcement learning-based double-pantograph cooperative control method for a reconnection-motor train unit. The method specifically comprises the following steps: acquiring pantograph state information and train operation information by an information acquisition unit; establishing a controller control action and contact network interaction sample data set; constraining a continuous and stable behavior strategy by using a comprehensive reward; in order to quantify the influence of catenary fluctuation propagation caused by a front bow control action on a rear bow, a double-bow interaction mechanism is established by using a reward propagation channel; learning an optimal behavior strategy by using a multi-agent deep reinforcement learning network based on the established interaction sample data set; the optimal behavior strategy serves as a controller, and the compensation action of the controller is output to a precise pressure regulating valve on a pantograph air valve plate, so that the pressure intensity of an air bag is controlled. By means of the method, accurate and advanced control over the front bow and the rear bow of the reconnection-motor train unit can be effectively achieved, good contact between the pantograph and a contact net is guaranteed, the current collection quality of the train is improved, and the offline rate of the rear bow is reduced; abrasion of contact parts is reduced, and the service life is prolonged.
本发明公开了一种基于深度强化学习的重联‑动车组双弓协同控制方法,具体为:信息采集单元获取受电弓状态信息、列车运行信息;建立控制器控制动作与接触网交互样本数据集;使用综合奖励约束持续平稳的行为策略;为量化前弓控制动作引发的接触网波动传播对后弓的影响,使用奖励传播通道建立双弓交互机制;基于所建立的交互样本数据集,利用多智能体深度强化学习网络学习最优行为策略;根据最优行为策略作为控制器,将控制器补偿动作输出受电弓气阀板上的精密调压阀从而控制气囊压强。本发明能有效对重联‑动车组前后弓的精准、提前控制,保证受电弓和接触网的良好接触,提升列车的受流质量,降低后弓的离线率;降低接触部件的磨损、提升服役寿命。
Deep reinforcement learning-based double-bow cooperative control method for reconnection-motor train unit
一种基于深度强化学习的重联-动车组双弓协同控制方法
2023-04-28
Patent
Electronic Resource
Chinese
Logic control unit applied to train reconnection and train reconnection method
European Patent Office | 2024
|Pantograph reconnection control circuit and method of reconnection train and reconnection train
European Patent Office | 2025
|European Patent Office | 2024
|Deep reinforcement learning cooperative control method for EMS type maglev train suspension frame
European Patent Office | 2024
|Panograph rising control circuit and method of reconnection train and reconnection train
European Patent Office | 2024
|