Traffic flow prediction based on the spatiotemporal attention

Long-term accurate traffic flow prediction has always been a challenging task. Existing frameworks usually only consider local spatiotemporal relationships, ignoring global spatiotemporal dependencies. In response to this problem, this paper proposes a traffic flow prediction model based on spatiotemporal attention. The network consists of a spatiotemporal attention module and an auxiliary module, which integrates global and local spatiotemporal relations. The attention module can capture global and local spatiotemporal features at the same time, and the gating-spatial attention-one-dimensional time convolution module is used as an auxiliary prediction. Integrating the two modules into a unified layer, the spatiotemporal attention network can capture more spatiotemporal dependencies. Experimental results on multiple public transportation datasets show that the proposed method achieves more advanced performance.