基于深度确定性策略梯度算法的交通协同智能控制技术分析  

Technology analysis of traffic cooperative intelligent control based on depth deterministic strategy gradient algorithm

作　　者：高兴媛[1] 和铁行[2] GAO Xingyuan;HE Tiexing(School of Computer and Information Technology,Zhejiang Changzheng Vocational and Technical College,Hangzhou 310023,China;School of Information Engineering,Hangzhou Medical University,Hangzhou 311300,China)

机构地区：[1]浙江长征职业技术学院计算机与信息技术学院,杭州310023 [2]杭州医学院信息工程学院,杭州311300

出　　处：《国外电子测量技术》2025年第1期54-61,共8页Foreign Electronic Measurement Technology

基　　金：浙江省高职教育“十四五”第二批教学改革项目(jg20240369)。

摘　　要：为提高城市交通系统的效率和稳定性,减少车辆等待时间,提高道路通行能力,采用深度确定性策略梯度(Deep Deterministic Policy Gradient,DDPG)算法作为核心控制策略。将城市路网建模为集中式控制系统,通过Agent控制路网中的多个交叉口,并提出多智能体系统(Multi-Agent System,MAS)结合异步优势行动者评论家(Asynchronous Advantage Actor-Critic,A3C),简称MA3C。结果表明,DDPG算法训练初期奖励值迅速上升,1000步后约稳定于150,表现优异。MA3C在高峰时奖励值为−5.94,延迟仅0.39 s,速度最高,其队列长度和等待时间显著低于其他算法。在不同车流密度下,所研究系统的车道平均占用率和平均速度均优于对比算法,高密度流量中车道平均占用率为0.9%,平均速度达14.89 m/s。低密度流量中车道平均占用率为0.4%,平均速度为17.68 m/s。所提方法不仅能够提高了交通系统的效率,还能增强交通控制的灵活性和适应性,推动了交通控制技术向智能化、自动化的方向发展。To improve the efficiency and stability of urban transportation system,reduce the waiting time of vehicles and improve the road capacity.DDPG algorithm was used as the core control strategy.The urban road network was modeled as a centralized control system,and multiple intersections in the road network were controlled by Agent.The MAS combined with A3C,referred to as MA3C was proposed.The results show that the reward value of DDPG algorithm increases rapidly at the initial stage of training,and stabilates at about 150 after 1000 steps,showing excellent performance.MA3C has a bonus value of-5.94 at peak,a latency of just 0.39 seconds,and the highest speed.Its queue length and waiting time are significantly lower than other algorithms.Under different traffic densities,the average lane occupancy rate and average speed of the research system are superior to the comparison algorithm.In high-density traffic,the average lane occupancy rate is 0.9%and the average speed reaches 14.89 m/s.The average lane occupancy rate in low-density traffic is 0.4,and the average speed is 17.68 m/s.The research method can not only improve the efficiency of traffic system,but also enhance the flexibility and adaptability of traffic control,and promote the development of traffic control technology to the direction of intelligence and automation.

关 键 词：交通系统 深度确定性策略梯度算法 路网 智能化 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]