Research on Traffic Forecasting Based on Graph Structure Generation

Predicting the future state of a transportation system is a challenging problem with very complex spatiotemporal dependencies. Our research focuses on the use of graph structure learning to obtain graph structures that better fit the specific problem in order to better perform the prediction task. In this paper, we improve on previous graph structure generation methods and propose a graph structure generation model. The model is able to obtain specific graph structures based on sequence dependencies. The output of this model is used as the base graph structure, and then Graph Convolution Recurrent Network (GCRN) is used to perform the next step of prediction. We have conducted extensive experiments on real-world datasets. The results show that our model performs the prediction task well and outperforms other graph generation methods.