Research On Short-term Network-wide Traffic Prediction Based On Deep Learning

With the improvement of urbanization level in China and the increase of national travel demand,traffic congestion has become a general problem in most major cities.Intelligent Transportation System is an effective way to alleviate urban traffic congestion through real-time and efficient traffic induction and control.Based on massive traffic data resources,traffic flow prediction can realize traffic conditions prediction of road network in the future periods precisely,which provides theoretical reference for realizing fine active control and travel guidance,and sets foundation for the development of Intelligent Transportation System.Based on the spatio-temporal correlation characteristics of traffic flow,this article builds a short-term traffic flow prediction model of expressway based on the deep learning methods.Furthermore,the research object is extended to the range of road network to achieve short-term traffic flow prediction of urban road network considering the spatial topology structure and the spatio-temporal characteristics of traffic flow.First,this paper respectively takes traffic flow of expressway section and urban road network as the research object,introduces the traffic flow data collection technology and then carries out the data preprocessing work.This paper explores distribution characteristics of three traffic flow parameters and spatio-temporal correlation characteristics,and then analyzes the spitio-temporal correlation of the traffic flow in road network.Then,this paper instroduces the deep learning methods of short-term traffic flow prediction,providing data and technical support for establishing short-term traffic flow prediction of road network.Second,in order to fully excavate the spatio-temporal characteristics of traffic flow,this paper puts forward a short-term expressway traffic flow prediction model named BSTCN based on Convolution Neural Network.This model uses Convolutional Neural Network to mine the spatial characteristics of traffic flow,and applies Bidirectional Long-Short Term Memory Network to obtain the temporal characteristics.Then Self-Attention mechanism is used to establish the long-term relationship of the sequence and acquire effective feature expression.The prediction performance of this model is proved to be better than that of other comparative models by using real expressway coil speed data.Finally,the research scope is extended to the urban road network level.Considering the complex spatial topology structure of the urban road network and the spatio-temporal characteristics of traffic flow,this paper establishs two short-term traffic flow prediction framework named BSTGN and BSTGAN of road network based on Graph Convolutional Neural Network.In the spatial dimension,BSTGN uses Graph Convolutional Neural Network to model the spatial connectivity and global correlation in road network.BSTGAN applies Graph Attention Network to model the spatial characteristics of road sections and effectively aggregate spatial structure characteristics.The Bidirectional Long-Short Term Memory is still used to obtain the temporal characteristics,and Self-Attention mechanism is used to obtain expression important features.The applicability and effectiveness of the two models are verified in the real urban road network dataset.And BSTGAN is superior to BSTGN in both prediction accuracy and efficiency,which reflects the importance of the spatial information for short-term traffic flow prediction of road network.There are 47 figures,24 tables and 72 references in this paper.