Research on Traffic Flow Forecasting of Spatio-temporal Convolutional Networks with Auto-correlation

Spatial–temporal graph modeling is a significant assignment to analyze the temporal–spatial correlation of traffic flow forecasting models. However, most of the existing methods capture the spatial dependence through the fixed graph structure. The explicit graph structure can hardly reflect the real dependency, and the real relationship may be lost due to incomplete connections in the data. In addition, the existing methods cannot effectively capture the dynamic spatio-temporal correlation and periodicity. To overcome this problem, a traffic flow forecasting model Auto-correlation Spatio-temporal Convolutional Network (Auto-STCN) is proposed, and models the temporal dependence, spatial correlation, and periodicity of traffic flow, respectively. Temporal Convolution Network and Graph Convolution Network are used to extract the temporal dependence and spatial correlation of traffic flow. Auto-correlation is used to calculate the Auto-correlation of the sequence to capture the periodic dependence of traffic flow, and similar subsequences are aggregated by time delay aggregation. An adaptive adjacency matrix is constructed in the Auto-STCN model, and it is learned by node embedding to extract the spatial features of traffic flow dynamics. Experiments on two datasets show that the Auto-STCN model has better prediction performance.