Research On Traffic Flow Forecast Method Based On Attention Mechanism

In recent years,the problem of traffic congestion has become more and more serious.In order to reduce the economic losses and social impacts caused by traffic congestion,it is necessary to use the Intelligent Transportation System(ITS)to manage and improve traffic.Traffic flow prediction is the basis of ITS,which predicts future traffic conditions by analyzing historical traffic data.Accurate and real-time traffic flow prediction not only provides a scientific basis for the management of relevant departments,but also improves the efficiency and safety of public travel.Traffic flow data has complex temporal and spatial dependence,which is challenging for its accurate prediction.At present,most of the methods are based on the traffic flow time series feature information for prediction,and some of the prediction methods consider the spatial characteristics of the traffic network,but tend to extract the static spatial correlation of the traffic network structure,ignoring the dynamic spatial-temporal correlation of traffic flow data.In view of the above problems,,this dissertation proposes to use the spatio-temporal attention mechanism to predict traffic flow.The main research contents and results are as follows:(1)In order to solve the problem that traditional methods cannot fully extract the spatio-temporal characteristics of traffic flow data,this dissertation proposes a combined model based on spatio-temporal convolutional network(GCN-TCN).The model includes multiple spatio-temporal convolutional layers.which are composed of a gated temporal convolutional network for extracting temporal features and a graph convolutional network for spatial features.By stacking spatio-temporal convolutional layers,spatio-temporal features of traffic flow data can be effectively extracted simultaneously.(2)To solve the problem that it is difficult to extract the dynamic spatio-temporal correlation of traffic flow,this dissertation proposes a combined model based on spatiotemporal attention mechanism(GAT-Att TCN).Among them,the temporal attention mechanism is based on the gated temporal convolution network with attention layer,and the spatial attention mechanism is based on the graph attention network.The combination of temporal and spatial attention mechanisms effectively extracts the dynamic spatio-temporal correlation of traffic flow data.(3)The two combined models were simulated on California highway data set Pe MSD7 and Shenzhen taxi GPS data set(SZ-taxi).The experimental results show that the prediction effect of the two combined models is better than the baseline models.Among them,the average accuracy of the GCN-TCN combined model reached 92.3%,while the training of GAT-Att TCN combined model took more time,but its prediction accuracy was further improved,with the average accuracy rate reaching 93.4%.Experiments show that the GAT-Att TCN combined model can effectively mine the dynamic spatio-temporal correlation of traffic flow data,and has higher prediction performance and better interpretability.