基于深度强化学习的综合干线协调控制方法

针对社会车辆和专用道公交干线运行特性差异大、协调控制效果差的问题，提出一种集成社会车辆干线协调控制和公交干线优先控制的综合干线协调控制方法。首先，基于两者路段行程时间的分布差异，结合公交车辆上下游路口不停车通行概率分析，确定干线协调的关联交通状态和对应信号调整策略；然后，结合信号调整策略对车辆延误损失和公交优先收益的量化分析确定奖惩机制；最后，提出一种深度强化学习框架用于最佳信号调整策略的实时求解。仿真实验分析发现：本文方法在人均延误上比社会车辆干线协调、公交干线协调控制分别提升 38.63%和 27.43%，在公交停车次数上比社会车辆干线协调提升52.17%，证明该方法能够有效提高公交车辆和社会车辆的通行效率。

Priority of Dedicated Bus Arterial Control Based on Deep Reinforcement Learning

Due to the differences in operating characteristics of the social vehicles and dedicated buses and the poor performance of the coordination control, this paper proposes a comprehensive arterial line coordination control method that integrates social vehicle arterial coordination control and public transportation arterial priority control. With the analysis of the dedicated bus non-stop probability of the upstream and downstream, the associated state spaces and the corresponding action decisions were determinded by analysing the difference in the travel time distribution of the social vehicle and dedicated bus. We combined with the influence of signal adjustment strategy on vehicle delay loss and bus priority gains to determine the reward and punishment mechanism. And a deep reinforcement learning framework is proposed to solve the best signal adjustment strategy in real time. Finally, the simulation experiment indicates that the proposed method can reduce the per-capita delay by 38.63% and 27.43%, and the stop times of bus at intersections can decreased by 52.17% compared with the social vehicle arterial coordination, which proves that the method can effectively increase the efficiency of buses and social vehicles.