Spatio-Temporal Traffic Flow Prediction Based on Coordinated Attention

Traffic flow prediction can provide effective support for traffic management and control and plays an important role in the traffic system. Traffic flow has strong spatio-temporal characteristics, and existing traffic flow prediction models tend to extract long-term dependencies of traffic flow in the temporal and spatial dimensions individually, often ignoring the potential correlations existing between spatio-temporal information of traffic flow. In order to further improve the prediction accuracy, this paper proposes a coordinated attention-based spatio-temporal graph convolutional network (CVSTGCN) model for simultaneously and dynamically capturing the long-term dependencies existing between the spatio-temporal information of traffic flows. CVSTGCN is composed of a full convolutional network structure, which combines coordinate methods to specify the influence degrees of different feature information in different spatio-temporal dimensions, and the spatio-temporal information of different spatio-temporal dimensions by the graph convolutional network. In addition, the hard-swish activation function is introduced to replace the Rectified Linear Unit (ReLU) activation function in the prediction of traffic flow. Finally, evaluation experiments are conducted on two real datasets to demonstrate that the proposed model has the best prediction performance in both short-term and long-term forecasting.