Prediction Of Short-Term Traffic Flow Based On BiGRU-TCN And Graph Neural Network

With the continuous expansion of urban scale,the number of road sensors is also increasing,resulting in massive traffic data,which poses a huge challenge to the study of traffic prediction model.As typical spatio-temporal graph data,traffic data has complex temporal and spatial correlations.However,most of the previous research on traffic prediction has focused on capturing the temporal characteristics of traffic data,ignoring the inherent temporal causal relationship between the spatial dynamics of traffic data and traffic events.Therefore,this thesis combines the advantages of graph neural network(GNN),bidirectional gated neural network(Bi-GRU)and time convolutional network(TCN)to establish a traffic flow prediction model,excavate the spatio-temporal characteristics of traffic data,and accurately predict short-term traffic flow.The main research contents are as follows:(1)A BiGRU-TCN model with a temporal convolution module is proposed,which consists of a Bi-GRU and a causal temporal convolutional network(TCN)embedded in capturing traffic data,which is responsible for extracting temporal features in traffic road networks.It captures the overall temporal correlation through the context analysis of traffic events in the time series to reveal the potential causal relationship between traffic events.Compared with other traffic prediction methods on three real traffic dataset,the proposed BiGRU-TCN model is proved to be effective in extracting time features.(2)This thesis combines the graph neural network(GNN)for extracting spatial features with the BiGRU-TCN model for extracting temporal features to build a spatiotemporal combination model that can capture temporal features.Experiments on Pe MS highway dataset show that the prediction effect of spatial-temporal combination model is better than that of the model dealing with time or space features alone,which verifies the necessity of extracting spatial features in traffic flow prediction task.(3)In view of the fact that the general spatio-temporal combination model cannot fully extract the spatio-temporal features in the traffic network,an improved spatiotemporal combination model is proposed in this thesis.The original traffic data is input into the graph neural network(GNN)and the bidirectional gated time convolutional network(BiGRU-TCN)in parallel,so as to extract the spatio-temporal features in the traffic data to the maximum extent.Simulation experiments are carried out on the SZtaxi dataset of urban traffic network,and it is verified that the improved spatio-temporal combination model in this thesis achieves better results in predicting traffic tasks.