Traffic Flow Prediction Based on Graph Convolutional Networks

In order to accurately extract the temporal and spatial correlations of nodes, a traffic flow prediction model based on graph convolutional networks is proposed. The graph convolution network is used to model the spatial structure of traffic flow, and the Convolutional neural network is used to capture the time characteristics of traffic flow; On this basis, add a spatiotemporal attention mechanism to adaptively capture the temporal and spatial correlations of traffic flow; Weighted fusion is performed on recent time periods, daily time periods, and weekly time periods to obtain the final prediction results, in order to achieve traffic prediction for specific time windows. Experiments were carried out on four real data sets of PEMS, using Mean absolute error, Root-mean-square deviation, and average ab-solute percentage error as evaluation indicators. By comparing with other types of traffic flow prediction models, this model performs better on all three evaluation indicators and has better prediction performance.