大规模智慧交通信号控制中的强化学习和深度强化学习方法综述  

Review of reinforcement learning and deep reinforcement learning methods inlarge-scale intelligent traffic signal control

作　　者：翟子洋 郝茹茹[1] 董世浩 Zhai Ziyang;Hao Ruru;Dong Shihao(School of Information Engineering,Chang’an University,Xi’an 710064,China)

机构地区：[1]长安大学信息工程学院,西安710064

出　　处：《计算机应用研究》2024年第6期1618-1627,共10页Application Research of Computers

基　　金：国家重点研发计划资助项目(2021YFA1000300,2021YFA1000303);国家青年基金资助项目(202006565013)。

摘　　要：当前在交通信号控制系统中引入智能化检测和控制已是大势所趋,特别是强化学习和深度强化学习方法在可扩展性、稳定性和可推广性等方面展现出巨大的技术优势,已成为该领域的研究热点。针对基于强化学习的交通信号控制任务进行了研究,在广泛调研交通信号控制方法研究成果的基础上,系统地梳理了强化学习和深度强化学习在智慧交通信号控制领域的分类及应用;并归纳了使用多智能体合作的方法解决大规模交通信号控制问题的可行方案,对大规模交通信号控制的交通场景影响因素进行了分类概述;从提高交通信号控制器性能的角度提出了本领域当前所面临的挑战和未来可能极具潜力的研究方向。At present,it is a general trend to introduce intelligent detection and control into traffic signal control system,especially reinforcement learning and deep reinforcement learning methods show great technical advantages in scalability,stability and extensibility,and have become a research hotspot in this field.This paper studied traffic signal control tasks based on reinforcement learning,systematically sorted out the classification and application of reinforcement learning and deep reinforcement learning in the field of intelligent traffic signal control on the basis of extensive research results on traffic signal control methods,and summarized feasible solutions to large-scale traffic signal control problems by using multi-agent cooperation.This paper classified and summarized the factors affecting the traffic scene of large-scale traffic signal control,put forward the current challenges and potential research directions in this field from the perspective of improving the performance of traffic signal controllers.

关 键 词：智能交通 交通信号控制 强化学习 交通信号灯 多智能体 大规模交通网络 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]