干线动态协调控制的深度Q网络方法  

A deep Q network method for dynamic arterial coordinated control

作　　者：郭瑝清 陈锋[1,2] Guo Huangqing;Chen Feng(School of Information Science and Technology,University of Science and Technology of China,Hefei 230027,China;Anhui LoongSec Science and Technology Ltd.,Hefei 230088,China)

机构地区：[1]中国科学技术大学信息科学技术学院,安徽合肥230027 [2]安徽中科龙安科技股份有限公司,安徽合肥230088

出　　处：《信息技术与网络安全》2020年第6期1-6,共6页Information Technology and Network Security

基　　金：安徽省对外科技合作项目(1804b06020376)。

摘　　要：为有效降低城市交通干线的车均延误与停车次数,将深度Q网络引入干线协调控制,给出了一种干线动态协调控制的DDDQN(Dueling Double Deep Q Network)方法。该方法结合双重深度Q网络与基于竞争架构深度Q网络,并将干线作为整体处理,通过深度神经网络挖掘干线各交叉口协调控制的相关性,基于Q学习进行交通信号控制决策。通过仿真实验,在近饱和流量和干线存在初始排队的情况下,将DDDQN方法与现有绿波方法,以及经典深度Q网络、双重深度Q网络、基于竞争架构深度Q网络的干线协调控制算法进行对比,实验结果表明基于DDDQN的干线动态协调控制算法性能优于其他四种方法。In order to effectively reduce the average delay and number of stops for urban traffic trunk roads, a deep Q network is introduced to arterial coordinated control, and a DDDQN (Dueling Double Deep Q Network) method is present-ed in this paper. This method combines the double deep Q network and the dueling deep Q network, and views the trunk road as a whole. The deep neural network is applied to find the correlation of the coordinated control for each in-tersection in the trunk road, and Q learning makes these decisions for traffic signal control. Through simulation platform,in the condition of near saturation and initial queue, the method proposed in this paper is compared with the existing green wave method, the arterial coordinated methods respectively based on deep Q network, double deep Q network, and dueling deep Q network. The experimental results show that the performance of DDDQN algorithm is better than the oth-er four methods.

关 键 词：城市交通 干线协调控制 深度Q网络 双重深度Q网络 基于竞争架构深度Q网络 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]