首页|车路协同环境下数据驱动的混合交通流速度调控方法

车路协同环境下数据驱动的混合交通流速度调控方法

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
针对车联网和自动驾驶背景下的混合车流调控问题,提出通过路侧设备调节车流中受控自动网联车速度以实现宏观交通流速度调控的方法.首先将单个受控自动网联车对车流状态的调节过程描述为马尔可夫决策过程,设计衡量受控车辆对整体车流影响程度的动态网格,以受控车辆的实时期望速度为动作,并综合考虑受控车对网格车流的影响程度、对期望速度的响应快慢以及车流的安全性构造报酬函数,采用深度确定性策略梯度算法进行策略寻优.考虑到受控自动车周边交通环境的多样性,提出基于关键参数量的控制策略模型簇.其次,面向车流速度的实时调控,根据路段车辆簇的划分,提出多个路侧设备控制同一车辆簇的控制权时序交接策略.路测设备以簇为单位对信号覆盖范围内的车辆进行标记与跟踪,实时计算受控车辆的期望控制信号并下发.最后,将所提出的方法在多种场景下进行训练验证,结果表明所提出的方法在适用场景中能够实现车流的平稳调速.
Data-driven speed control method for mixed traffic flow in vehicle-road cooperative environment
Aiming at the problem of mixed traffic flow control in the context of development of cooperative vehicle infrastructure systems and autonomous driving technology,a speed control method of controlled connected autonomous vehicle based on road side unit(RSUs)to regulate the speed of macroscopic traffic flow is proposed.Firstly,this paper describes the adjustment process of a single controlled connected autonomous vehicle to the traffic state as a Markov decision process.A dynamic grid is designed to measure the influence of the controlled vehicle on the overall traffic flow.Considering the influence of the controlled vehicle on the traffic flow in the dynamic grid,the response speed to the expected target and the safety factor of the overall traffic flow,the reward function is constructed.The deep deterministic policy gradient(DDPG)algorithm is adopted for policy optimization.Considering the diversity of the traffic environment around the controlled connected autonomous vehicle,a control strategy model cluster based on key parameters is proposed for RSUs to select in real time according to traffic flow states.Secondly,for the real-time adjustment of traffic average speed,the division method of vehicle clusters in road sections and the timing handover strategy of control rights for multiple roadside units to control the particular vehicle cluster are proposed.The RSUs track vehicles within the signal coverage area in a vehicle cluster,and calculate the expected control signal in real time,then send it to the controlled vehicles.Finally,the proposed method is verified in different scenarios,which can smoothly and efficiently regulate the speed of traffic in multiple scenarios.

traffic flow controldata-drivencooperative vehicle infrastructure systemdeep reinforcement learningconnected autonomous vehiclemixed traffic flow

张辰、徐云雯、李德伟

展开 >

上海交通大学自动化系,上海 200240

车流控制 数据驱动 车路协同 深度强化学习 网联自动车 混合交通流

国家重点研发计划项目国家自然科学基金项目国家自然科学基金项目

2020YFB16004006200321062333015

2024

10.13195/j.kzyjc.2023.0700

控制与决策
东北大学

控制与决策

CSTPCD北大核心
影响因子：1.227
ISSN：1001-0920
年,卷(期)：2024.39(9)
  • 3