| In recent years,the national intelligent transportation system has achieved a qualitative leap,with an increase in people’s travel frequency and a continuous increase in the number of motor vehicles.How to alleviate road traffic pressure and achieve real-time and efficient traffic information guidance and control has become an essential issue.Under the background of intelligent transportation system,traffic flow prediction tasks based on massive and unique traffic flow information data can timely discover the road traffic status at a certain time in the future,providing a strong theoretical reference for achieving refined road management and improving people’s travel experience.In view of the unique spatio-temporal correlation relationship of traffic road flow,this paper further analyzes and optimizes the traffic flow prediction model based on in-depth learning,deeply studies the traffic network from the perspective of graph,depicts the spatio-temporal characteristics of the traffic network graph,introduces the restart random walk algorithm(RWR: Random Walk with Restart)to reconstruct the road network topology,and realizes the road network traffic flow prediction research considering the spatial structure information of the road network and the traffic flow timing characteristics.First of all,this paper summarizes and analyzes the theoretical knowledge of traffic flow prediction,analyzes the time-space characteristics of expressway road network and urban traffic road network,introduces in detail the main acquisition methods and pre-processing of traffic flow information data,and standardizes the neural network models involved in the realization of traffic flow prediction tasks,laying the data and theoretical foundation for the subsequent realization of prediction model construction.Secondly,in order to deeply explore the spatial structure characteristics of the traffic network graph,this paper analyzes the topological structure relationship of the road network,and introduces the restart random walk algorithm to re acquire the spatial relationship of the road network.At the same time,it is verified on the STGCN(Spatial Temporal Graph Convolutional Networks)model,a short-term traffic flow prediction model based on graph convolutional neural network structure,which is a full convolutional structure,Each convolutional module has two time gate convolutional layers and one spatial graph convolutional layer.Through double verification on highways and urban roads,the effectiveness and accuracy of topology reconstruction in this paper are proven,providing a theoretical basis for the subsequent construction of new traffic flow prediction models.Finally,on the basis of the previous research,a new traffic flow prediction model TSTGCN(Traffic Flow Spatial Temporary Graph Convolutional Neural Network Model)is constructed by combining the road network topological structure relationship matrix obtained based on the restart of the random walk algorithm with GCN(Graph Convolutional Networks),LSTM(Long Short Term Memory Networks)and attention mechanism.Among them,GCN can effectively extract features from the spatial dependencies of the road network,analyze the temporal characteristics of traffic flow using LSTM model,and introduce self attention mechanism for feature expression.The results showed that the TSTGCN model performed well in both prediction efficiency and accuracy,with smaller prediction errors and more complete spatiotemporal feature extraction,reflecting the importance of the spatial structure information of the traffic network for traffic flow prediction work. |
