Abstract: Currently, traffic flow prediction does not completely consider the correlation feature information between traffic flow and density and traffic flow and speed. In addition, it also ignores multi-scale temporal features. Thus, in this study, we propose a double-branch spatial-temporal graph convolutional neural network for traffic flow prediction. First, we categorize traffic data into two temporal granularity data inputs, i.e., near-term input and periodic input, according to the periodicity of traffic data. Second, we use an associative gated linear unit to extract the correlation feature information between traffic flow and density and traffic flow and speed in each input branch. Third, we use the graph convolutional layer and multi-scale temporal convolutional layer to extract the spatial-temporal context information of the correlation feature. We also use the prediction convolutional layer to obtain the near-term and periodic double-branch prediction results. Finally, we achieve an accurate prediction of the traffic flow by fusing the prediction results through the gating mechanism. Our experimental results show that this model is superior to other models in terms of the accuracy and stability of traffic flow prediction.