深度强化学习在城市交通信号控制方法中的应用综述  

A survey of application of deep reinforcement learning in urban traffic signal control methods

作　　者：周斌[1,2] 吴晓东 马东方 邱红桐[3] ZHOU Bin;WU Xiaodong;MA Dongfang;QIU Hongtong(Ocean College,Zhejiang University,Zhoushan 316021,China;Pengcheng Laboratory,Shenzhen 518000,China;Traffic Management Research Institute of the Ministry of Public Security,Wuxi 214151,China)

机构地区：[1]浙江大学海洋学院,浙江舟山316021 [2]鹏城实验室,广东深圳518000 [3]公安部交通管理科学技术研究所,江苏无锡214151

出　　处：《现代交通与冶金材料》2022年第3期84-93,共10页Modern Transportation and Metallurgical Materials

基　　金：山东省重大科技创新工程项目(2019TSLH0203);国家自然科学基金资助项目(52172334);浙江省自然科学基金资助项目(LR19F030002)。

摘　　要：信号控制是智能交通系统的关键组成部分,通过优化冲突交通流的时空资源分配保障车辆通行安全和最小化车辆通行时间。传统的信号控制主要依靠规则和模型两类手段优化信号参数,存在诸多理想假设,实际控制效果欠佳。近年来,随着交通数据的丰富、算力的增强和人工智能算法的成熟,以数据驱动为导向的交通信号控制方法成为新的研究方向,其中结合深度强化学习的优化方法是最主要的研究热点。本文综述基于深度强化学习的前沿信号控制技术,系统介绍深度强化学习在城市交通信号控制中的应用,从基本概念、主要影响因素和实验设计三个维度进行分类阐述。最后,讨论深度强化学习在智能交通领域应用中的主要挑战和亟待解决的技术难题。In the intelligent transportation system,traffic signal control is one of the key links,because it can achieve the goal of minimizing vehicle travel time by coordinating vehicles at intersections.Howerer the current traffic signal control is highly dependent on manual rules and model.The development of technology,the enrichment of data,the enhancement of computing ability and other factors have brought new research directions for the data-driven traffic signal control,among which the intelligent transportation combined with deep reinforcement learning is one of the main research hotspots.This survey reviews the latest research on traffic signal control based on deep reinforcement learning.It broadly summarizes the application of deep reinforcement learning in urban traffic signal control and classifies them from basic concepts,main influencing factors and experimental design.Finally,we discuss the challenges and problems to be solved in traffic applications based on deep reinforcement learning.

关 键 词：交通信号控制 智能交通系统 深度强化学习 数据驱动 

分 类 号：U491.51[交通运输工程—交通运输规划与管理]