基于深度强化学习的单线路公交动态驻站控制策略研究

Single-line Bus Operations Dynamic Holding Control Strategy Based on Deep Reinforcement Learning

公交运行中,车辆车头时距波动过大会导致公交系统出现串车等运行不稳定现象,针对该问题,本文提出一种基于深度强化学习的动态驻站控制策略,实现公交系统的稳定运行,以及避免出现串车问题。首先,构造线形公交系统,并确定车辆运行和乘客行为规则;然后,介绍基于深度强化学习建立动态控制方法,定义强化学习框架的各要素,并开发事件驱动的模拟器环境,训练和测试智能体;最后,利用仿真模拟对所提方法与基准方法进行大量的数值实验,选取不同评价指标进行对比分析,并实施敏感性分析。实验结果发现,本文方法实现了最稳定的车辆运行轨迹和最小的载客分散度;在车头时距变动上,比无控制策略、基于时刻表控制策略和基于车头时距控制策略分别降低61.90%、60.98%和37.98%;在平均等待时间上,分别降低28.36%、26.53%和23.61%。此外,所提方法在不同行驶时间变异性和车头时距情景下,具有很强的鲁棒性。

Large headway and fluctuations in bus operations can lead to instability of the bus operation system,such as the bus bunching phenomena.This paper proposes a dynamic holding control strategy based on deep reinforcement learning to improve the stability of bus system operations and avoid bus bunching.A linear bus system is established,and the operating rules for vehicles and passenger behavior are defined.Then,a dynamic control method is introduced based on deep reinforcement learning,the elements of the reinforcement learning framework are defined,and an event driven simulator environment is developed to train and test the agents.Extensive simulation experiments are conducted to compare the proposed method with traditional methods.Various evaluation metrics are selected for comparative analysis,and the sensitivity analysis is also performed.The experimental results show that the proposed method achieves the most stable vehicle trajectories and the smallest passenger occupancy dispersion.The headway variation was reduced respectively by 61.90%,60.98%,and 37.98%compared to the no control strategy,the schedule-based control strategy,and the headway-based control strategy.The average waiting time was reduced by 28.36%,26.53%,and 23.61%compared to the aforementioned strategies.The proposed method also demonstrates strong robustness under varying travel time variability and headway conditions.