随着城市化进程的加快和人口的增长，城市轨道交通客流量持续攀升，这对轨道交通的安全、舒适和稳定运营构成了极大的挑战。为解决城市轨道交通高峰小时区间满载率过高的问题，本文提出一种基于深度强化学习的城市轨道交通协同限流控制方法。该方法利用历史客流数据建立线网层面的限流仿真环境和智能体模型，以区间满载率为状态，以限流策略为动作，以客流体验为奖励，通过多轮强化学习训练产生最优的限流方案。随后利用北京地铁线网数据构建仿真实验并验证了该方法的有效性。仿真结果表明，协同限流方法可以有效降低断面客流量，缓解高峰小时区间拥挤程度，提高乘客出行舒适度。

The rapid urbanization and population increase have led to a continuous rise in passenger flow in urban rail transit, which presents significant challenges to the safety, comfort, and stability of rail transit operation. In order to solve the problem of excessive load rate of urban rail transit during peak hours, we propose a cooperative passenger flow control method for urban rail transit, which is based on deep reinforcement learning. This method uses the full load rate between intervals as its state, the flow restriction strategy as its action, and the passenger flow experience as its reward. Then, it generates the optimal flow restriction scheme through multi-round reinforcement learning training. Subsequently, we validate the effectiveness of this method by constructing simulation experiments using the Beijing subway network data. The simulation results show that the cooperative passenger flow control method can effectively reduce the passenger flow in the section, relieve the congestion in the peak hour section, and improve the passenger travel comfort.

U231

1.国家重点研发计划资助(2020YFB1600702) 2.北京市科技新星计划项目(Z211100002121098)