基于深度学习的交通信号灯实时决策研究

Research on Real-time Decision Making for Traffic Lights Based on Deep Learning

为有效缓解城市交通压力,提升交通信号控制的智慧化水平,提出一种基于图像识别并融合深度学习的多交叉口交通信号灯实时决策模型方法。采用图像识别的方法判别拥堵状态,搭建区域多交叉口交通信号灯实时决策模型,以总调度期内通行评分最高为目标函数构建深度学习网络;采用机器学习思想优化决策方案,通过预训练增加决策方案容量并缩短现场决策时间,达到实时决策的目的;将研究模型应用于湖北省武汉市武昌区中山路路段,并进行模型论证和结果分析。研究结果表明,所提出的模型方法能有效解决区域多交叉口交通信号灯联合调度问题,可为交通管理者提供更合理的决策方案。

In order to effectively alleviate the pressure of urban traffic and improve the intelligent level of traffic signal control,a real-time decision-making model method for multi-intersection traffic lights based on image recognition fusion deep learning is proposed.The image recognition method is used to identify the congestion status,a real-time decision-making model for regional multi-intersection traffic lights is built,a deep learning network is constructed with the highest traffic score in the total scheduling period as the objective function,and the machine learning idea is used to optimize the decisionmaking scheme.The capacity of the decision-making scheme is increased through pre-training,and the on-site decision-making time is shortened to achieve the purpose of real-time decision-making.The research model is applied to the Zhongshan Road Section of Wuchang District,Wuhan City,Hubei Province for model demonstration and result analysis.And the analysis results show that the model method established in the research can effectively solve the problem of joint scheduling of traffic lights at multiple intersections,and provide more reasonable decision-making schemes for traffic managers.