基于异步优势强化学习的交通信号控制策略

Traffic signal control strategy based on asynchronous advantage actorcritic algorithm

为解决交通信号控制中的信号灯配时调度不合理、路口拥堵等问题,提出一种基于行动者-评论家算法的城市智能交通控制算法。该算法是一种基于异步优势的算法,可对交通状态特征进行抽象表征,并以多线程并行实现对交通状态的精确感知。该算法还参考了强化学习算法,能在最短时间内不断迭代优化其内部参数,得到交通信号控制的最优方案。为验证该算法的有效性,采用交通仿真软件SUMO,对该算法和其他3种典型的交通信号控制算法进行模拟仿真,并对仿真结果进行比较和分析。研究结果表明:与这3类典型算法中效果最好的Qlearning算法相比,该算法的交叉口车辆平均延误时间减少了14.1%,平均队列长度缩短了13.1%,平均等待时间减少了13.5%。该交通信号控制算法能有效地改善城市道路拥堵,提高道路交叉口的通行效率。

and represent traffic state features,enabling accurate perception of traffic conditions through parallel multithreading.Drawing inspiration from reinforcement learning techniques,the algorithm iteratively optimizes its internal parameters to obtain the optimal solution for traffic signal control within the shortest possible timeframe.To assess the algorithm's effectiveness,we conducted simulated experiments using the traffic simulation software SUMO,comparing its performance with three other commonly used traffic signal control algorithms.The simulation results reveal that compared to the Q-learning algorithm,this algorithm reduces the average delay time of vehicles at intersections by 14.1%,decreases the average queue length by 13.1%,and lowers the average waiting time by 13.5%.This traffic signal control algorithm can effectively alleviate urban road congestion and improve the traffic efficiency of road intersections.