基于深度强化学习的平滑车速控制系统  

Traffic Smoothing Control System Based on Deep Reinforcement Learning

作　　者：徐延军 陈建雄 XU Yanjun;CHEN Jianxiong(COSCO SHIPPING Technology Co.,Ltd.,Shanghai 200135,China;Shanghai Ship and Shipping Research Institute Co.,Ltd.,Shanghai 200135,China)

机构地区：[1]中远海运科技股份有限公司,上海200135 [2]上海船舶运输科学研究所有限公司,上海200135

出　　处：《上海船舶运输科学研究所学报》2023年第3期35-41,共7页Journal of Shanghai Ship and Shipping Research Institute

摘　　要：为调整不同路段的限速值,平滑交通流,从而提升高速公路车辆通行的安全性和效率,针对交通瓶颈区设计一种基于深度强化学习的平滑车速管控系统。该系统主要包含动态限速启动、限速值确定与更新和情报板动态发布等3个模块。将深度强化学习算法DDQN(Double Deep Q-Network)引入系统中,提出一种基于DDQN的平滑车速控制策略,从目标网络和经验回顾2个维度提升该算法的性能。基于元胞传输模型(Cellular Transmission Model,CTM)对宁夏高速公路某路段的交通流运行场景进行仿真,以车辆总通行时间和车流量为评价指标验证该系统的有效性,结果表明该系统能提高瓶颈区内拥堵路段车辆的通行效率。The in-depth reinforcement learning technology is introduced into the highway traffic management and a traffic smoothing control system is designed and physically developed.The system controls the speed of vehicles by adjusting the speed limit value of road sections according to the traffic situation in bottleneck area.The system consists of three functional modules:dynamic speed limit initiation,speed limit determination and update-information board dynamic releasing.The mainstream deep reinforcement learning algorithm DDQN(Double Deep Q-Network)is used in the system,and a traffic smoothing control strategy based on DDQN is developed to improve the performance of the algorithm by the destination network and experience replay technologies.Based on the CTM(Cellular Transmission Model),a traffic flow scenario at a road section of Ningxia expressway is simulated,and the control strategy is verified in such environment.The experiment shows that the system can improve the traffic efficiency in the bottleneck area in terms of volume of traffic and vehicle transit time.

关 键 词：平滑车速控制 交通瓶颈区 深度强化学习(DDQN)算法 元胞传输模型(CTM) 神经网络 

分 类 号：U491.54[交通运输工程—交通运输规划与管理]