基于深度强化学习的多应急车辆信号优先控制  

Multi-emergency Vehicle Signal Priority Control Based on Deep Reinforcement Learning

作　　者：刘翔 李艾 成卫[1] LIU Xiang;LI Ai;CHENG Wei(School of Traffic Engineering,Kunming,Kunming University of Science and Technology,Kunming 650504,China;Traffic Police Detachment of Yuxi Public Security Bureau,Yuxi 653100,China)

机构地区：[1]昆明理工大学交通工程学院,昆明650504 [2]玉溪市公安局交通警察支队,玉溪653100

出　　处：《武汉理工大学学报（交通科学与工程版）》2021年第6期1056-1061,共6页Journal of Wuhan University of Technology（Transportation Science & Engineering）

摘　　要：针对突发事件下交叉口可能出现不同方向上多辆应急车辆竞争绿灯相位的情况,文中提出了一种基于深度强化学习的应急车辆信号优先控制方法,通过构造应急车辆状态表示与设计相应的奖励函数,不需要对交叉口进行复杂建模,直接将实时的交通情况作为输入,根据不同的交通状况下交叉口信号控制反馈的结果,不停迭代,动态调整交通信号,从而满足应急车辆的优先通行并尽量保证其他社会车辆因此增加的延误能足够小.基于SUMO的仿真实验表明,在多辆应急车可能同时到达交叉口的情况下,该方法与感应优先控制相比能够降低60%左右的应急车辆延误,并能够在保证应急车辆通行的同时,减少应急车辆通行后对社会车辆的影响.In view of the situation that several emergency vehicles in different directions may compete for the green light phase at the intersection under emergency,a signal priority control method of emergency vehicles based on deep reinforcement learning was proposed in this paper.By constructing the reward function corresponding to the emergency vehicle state representation and design,the real-time traffic situation was directly taken as input without complex modeling of intersections.According to the feedback results of intersection signal control under different traffic conditions,the traffic signals were dynamically adjusted through iteration,so as to meet the priority of emergency vehicles and try to ensure that the delay caused by other social vehicles is small enough.The simulation experiment based on SUMO shows that when multiple emergency vehicles may arrive at the intersection at the same time,this method can reduce the delay of emergency vehicles by about 60%compared with inductive priority control,and can ensure the emergency vehicles to pass,and at the same time reduce the influence of emergency vehicles on social vehicles.

关 键 词：深度强化学习 应急车辆 信号优先控制 智能交通 

分 类 号：U491[交通运输工程—交通运输规划与管理]