Spatio-temporal interactive dynamic learning for traffic flow prediction

As an extensive research field, traffic flow forecasting has far-reaching background and important significance. With the acceleration of urbanization and the popularization of transportation, the urban transportation system is facing more and more serious problems of congestion and efficiency. Under this background, traffic flow forecasting has become one of the important means to improve the efficiency of traffic system management, optimize urban planning and improve the quality of life of residents. In this paper, a traffic flow prediction model (STIDGCN) based on spatiotemporal interactive dynamic graph convolutional network is proposed. Specifically, STIDGCN inherits the advantages of GCN, TCN, and many-chart attention mechanisms. The interactive learning strategy can well mine the dynamic properties of spatio-temporal traffic data, and Gated TCN can further capture complex temporal features. We evaluated the predictive performance of STIDGCN on two real datasets, and the experiments showed that it had the best predictive performance on all datasets compared to the baseline approach.