Research on traffic flow prediction of dynamic graph convolution network based on spatio-temporal attention mechanism

Accurate traffic flow prediction is the basis of urban traffic guidance and control, which is of great significance to intelligent traffic management and control. The complex spatial and temporal dependencies of traffic flow, as well as the periodicity and spatial heterogeneity of capturing traffic data, make accurate traffic flow prediction still a major challenge. To address the above problems, this article proposes DGCN-STA, a dynamic graph convolutional network for traffic flow prediction based on a spatio-temporal attention mechanism. Specifically, in the time dimension, this article constructs a new self-attention mechanism that can utilize local contexts, specifically for numerical sequence representation transformation, so that the prediction model is able to capture the temporal correlation of traffic flow, which is conducive to long-term prediction. In the spatial dimension, a dynamic graph convolution method is constructed to capture spatial correlations in a dynamic manner using self-attention, modeling for the periodicity of traffic flows and capturing spatial heterogeneity through embedding modules. The spatio-temporal correlation of long time series is then modeled using the spatio-temporal attention mechanism. Many experiments are conducted on four datasets, and the experimental results show that the DGCN-STA model proposed in this article has better prediction performance compared with the baseline method.