APP下载

基于多路口车辆感知预测的协同信号配时技术

2021-07-25赵晋芳

科技创新导报 2021年6期

赵晋芳

摘  要:为了缓解大城市的交通拥堵现状,交叉口信号灯配时的研究越来越有必要。普遍的交通信号配时技术,是基于单路口、传统车辆检测方法而设计的,没有考虑影响实际交通状况的流量,存在配时不准确、不智能的局限性。通过基于强化学习的深度Q网络,提出基于多路口车辆感知预测的协同信号配时技术,将每个路口建模为一个代理,每个代理被训练从道路环境接受交通状态并采取最佳行动。实验表明,该方法不仅可以有效地进行交通流量预测,解决多路口协同的信号灯配时问题,还可以提高配时技术的智能性。

关键词:Q网络  强化学习  协同交通信号控制  交通流预测

中图分类号:U491.2          文献标识码:A                  文章编号:1674-098X(2021)02(c)-0113-04

Cooperative Traffic Signal Control with Vehicle Perception Prediction in Multi-Intersection

ZHAO Jinfang

(Xi'an Vocational University of Automobile, Xi'an, Shaanxi Province, 710600 China)

Abstract: In order to alleviate the current situation of traffic congestion in large cities, it is more and more necessary to study the signal timing at intersections. The universal traffic signal timing technology is designed based on single intersection and traditional vehicle detection method. It does not consider the flow affectsing the actual traffic situation, and has the limitations of inaccurate timing and intelligence. Based on the deep Q network of reinforcement learning, this paper proposes a cooperative signal timing technology based on multi intersection vehicle perception prediction. This technology will model each intersection as an agent, and each agent is trained to accept the traffic status from the road environment and take the best action. Experiments show that this method can not only effectively predict traffic flow, solve the problem of multi intersection coordinated signal timing, but also improve the intelligence of timing technology.

Key Words: Q network ; Reinforcement learning; Collaborative traffic signal control; Traffic flow prediction

最近,交通擁堵已成为大多城市的严重问题[1]。由于有限的道路通行能力,使得交通流量增加越来越快。目前,传统的静态信号灯长度,使得交通系统不能有效地应对交通拥堵现状[2]。为了应对这些局限性,自适应交通信号控制根据路口处观察到的交通量[3],实时切换交通信号相位。

各种传统的机器学习算法被用于研究自适应交通控制。其中,基于图像处理技术的自适应交通信号控制是主流方向[4],虽然有部分通过提取车辆位置和速度信息进行模拟的多路口协同交通信号控制研究,但是大多数快照效率不高,而且不能处理影响交通状况的现实变量,例如温度和天气等。

另外,强化学习被广泛应用于交通信号控制研究[5],以寻找最佳解决方案。……

登录APP查看全文