基于团树传播的强化学习交通信号协调控制的研究

摘要

针对当前城市交通信号控制效率低下,致使车辆在道路交叉口等待时间较长,停车次数较多等问题,提出了一种新型的基于团树传播算法的强化学习控制方法来协调控制网络级交通.分别重点介绍强化学习算法与以联合树算法为代表的团树传播算法如何与交通控制相结合以及联合树算法是如何实现联合动作推理的.选取24个交叉口组成的路网为研究对象,在交通仿真软件VISSIM中进行仿真,软件可读取当前环境的状态,选取车辆的平均延误和平均停车次数作为性能指标,同时,分别与相邻路口简单协调的强化学习控制算法、无学习的LQF算法控制效果进行比较.%On account of the inefficiency of current traffic signal control system,resulting in long waits and more stops for most vehicles at the road intersections, this paper propose a novel reinforcement learning algorithm based on clique-tree propagation to optimize network wide traffic control.We focuses on how reinforcement learning and junction tree algorithm ,a typical Clique tree propagation algorithm,combined with traffic signal control and how the Junction tree algorithm achieve joint action reasoning. The algorithm is testd with a network containing 24 intersections,simulated in VISSIM, a traffic simulation software which can read the current state of the environment,choose the average vehicle delay and average number of stops as performance indicator.We also compare with simple reinforcement learning control algorithm which intersections coordinated with neighborhood and LQF algorithm.