基于Deep Q Networks的交通指示灯控制方法

Road Signal Light Control Method Based on Deep Q Networks

交通指示灯的智能控制是当前智能交通研究中的热点问题;为更加及时有效地自适应动态交通,进一步提升街道路口车流效率,提出了一种基于Deep Q Networks的道路指示灯控制方法;该方法基于道路指示灯控制问题描述,以状态、行动和奖励三要素构建道路指示灯控制的强化学习模型,提出基于Deep Q Networks的道路指示控制方法流程;为检验方法的有效性,以浙江省台州市市府大道与东环大道交叉路口交通数据在SUMO中进行方法比对与仿真实验;实验结果表明,基于Deep Q Networks的交通指示灯控制方法在交通指示等的控制与调度中具有更高的效率和自主性,更有利于改善路口车流的吞吐量,对道路路口车流的驻留时延、队列长度和等待时间等方面的优化具有更好的性能。

The intelligent control of traffic lights is a hot issue in the research of intelligent traffic.In order to adapt to dynamic traffic in a more timely and effective manner and further improve traffic flow efficiency at street intersections,a road indicator light control method based on Deep Q Networks was proposed.This method is based on the description of the road indicator control problem,constructs the reinforcement learning model of the road indicator control with the three elements of state,action and reward,and proposes the road indicator control method flow based on Deep Q Networks.To test the effectiveness of the method,the traffic data at the intersection of Shifu Avenue and Donghuan Avenue in Taizhou City,Zhejiang Province were compared and simulated in SUMO(simulation of urban MObility).The experimental results show that the traffic light control method based on Deep Q Networks has higher efficiency and autonomy in the control and scheduling of traffic indications,is more conducive to improving the throughput of intersections traffic flow,and has better performance in optimizing the stay delay,queue length and waiting time of intersections traffic flow.