基于CNN-LSTM的城市道路平均车速预测方法研究  

Research on the prediction method of urban road average vehicle speed based on CNN-LSTM

作　　者：王雪梅 陈莹 WANG Xuemei;CHEN Ying(School of Transportation Engineering,Tongji Zhejiang College,Jiaxing 314000,China;School of Automotive Engineering,Changshu Institute of Technology,Changshu 215506,China;Operation Management Center,Suzhou Rail Transit,Suzhou 215004,China)

机构地区：[1]同济大学浙江学院交通运输工程学院,嘉兴314000 [2]常熟理工学院汽车工程学院,常熟215506 [3]苏州市轨道交通集团有限公司运营管理中心,苏州215004

出　　处：《青岛理工大学学报》2023年第1期117-126,140,共11页Journal of Qingdao University of Technology

基　　金：嘉兴市公益性研究计划项目(2020AY10032);浙江省软科学研究计划项目(2021C35030,2021C35004)。

摘　　要：城市道路平均车速预测不仅是智能交通信息服务系统必不可少的组成部分,也是智能交通控制和管理系统的重要支撑。以实际的城市区域路网为研究对象,构建路网空间权重矩阵,采用时空自相关函数(ST-ACF)分析实验路网平均车速的时空相关性。在此基础上,构建基于卷积神经网络-长短期记忆网络(CNN-LSTM)的短时交通速度预测模型,既能学习平均速度数据的动态变化特征来捕获时间依赖关系,又能学习复杂的拓扑结构来捕获空间依赖关系。通过对未来5 d内5 min时间粒度的平均车速值进行预测,并与ARIMA,BPNN,CNN3个基准模型的预测结果对比,验证了构建的CNN-LSTM模型能较好地适用于实际区域路网的平均车速预测。本研究有助于为交通管理者提供交通事故预判和交通拥堵缓解的决策依据,也是对城市道路交通治堵技术的一次有益补充。The prediction of the urban road average vehicle speed is not only an essential part of the intelligent traffic information service system, but also an important support for the intelligent traffic control and management system. Taking the actual urban regional road network as the research object, this study constructs the road network spatial weight matrix and uses the spatio-temporal autocorrelation function(ST-ACF) to analyze the spatio-temporal autocorrelation of the average vehicle speed of the experimental road network. On this basis, this study builds an average vehicle speed prediction model based on Convolutional Neural Network-Long Short-Term Memory(CNN-LSTM), which can not only learn the dynamically changing features of average vehicle speed data to capture temporal dependencies, but also learn the complex topological structures to capture spatial dependencies. The average vehicle speed values every 5 minutes in the next 5 days are predicted, and the proposed CNN-LSTM model is compared with ARIMA, BPNN and CNN. It is verified that the CNN-LSTM model constructed in this study can be better applied to the average vehicle speed prediction of the actual regional road network. This study helps to provide traffic managers with decision-making basis for predicting traffic accidents and alleviating traffic congestion. At the same time, it is also a useful supplement to the technology of urban road traffic congestion controlling.

关 键 词：城市道路 平均车速 时空相关性 预测方法 CNN-LSTM 

分 类 号：U412.1[交通运输工程—道路与铁道工程]