Traffic Prediction Of Smart City Based On Multi-feature Spatio-temporal Map Network

Intelligent traffic systems(ITS)are complex and comprehensive systems,which are mainly composed of traffic data information collection system,data processing and analysis system and informationtrelease system.The traffic data information is collected mainly through sensors or GPS positioning system,and then is collected and transmitted to the traffic information processing and analysis center system.The data is uniformly processed and mined through relevant algorithms or models to analyze and master the road traffic operation status,and then is used to judge the congestion status and whether there are emergencies,which can help the relevant personnel of the traffic management department make road dispatching decisions in time and improve people’s travel experience.Traffic flow prediction is an indispensable part of intelligent transportation system.If the traffic flow can be accurately predicted,it is of great significance to traffic planning and vehicle scheduling.The complex spatial dependence,nonlinear dynamic variability(factors such as changes in road conditions or major emergencies)and difficulty in obtaining relevant data information(road fork structure,public transport facilities,regional functional attributes,etc.)in the road network has led to great challenges to the prediction accuracy.This dissertation proposes a new traffic flow prediction model and runs the model on two real data sets collected.The results verified by real data show that the prediction accuracy of the model proposed in this dissertation is significantly improved compared with the existing prediction models.The research contents and main contributions are summarized as follows:1.The traffic association information between nodes is mapped into a graph network to construct a multi-attribute fusion GCN module(mf-gcn).Then we transfer the graph network containing multiple features(such as road structure,average occupancy rate,etc.)into the graph neural network(GCN)module to construct a traffic flow prediction model,and fully mine the structural information of traffic map,so as to capture the spatial correlation between nodes,and improve the accuracy of traffic flow prediction.2.A multi fusion graph network dynamic spatiotemporal traffic flow prediction(DGAT)based on graph attention is constructed.The multi-attribute fusion GCN(mf-gcn)is used to capture the spatial correlation in traffic flow prediction,the graph attention captures the dynamic transformation of traffic flow data,and the convolution neural network is used to capture the time dimensiontinformation.The results show that the prediction performance of the traffic flow prediction model DGAT proposed in this dissertation is better than other existing methods.3.A multi fusion graph network traffic flow prediction model(mfg-eca)based on multichannel attention is constructed.The multi-channel attention model(ECA)is introduced to effectively realize the local cross-channel interaction without dimensionality reduction through one-dimensional convolution,which can avoid dimensionality reduction while learning channel attention.The time dimension is input into the ECA module as a channel,combined with the multi-attribute fusion GCN(mf-gcn),to learn the correlation,spatial correlation and dynamic variability of traffic data between different time dimensions.The results verified by real data show that the prediction accuracy of the proposed model is significantly improved compared with DGAT model and other comparative models...