AD-STGCRN: A Model for Traffic Flow Prediction

Abstract Aiming at the problem that the existing prediction models fail to utilize the serial correlation, temporal correlation and spatial correlation of data from multiple perspectives to achieve accurate prediction, a dual-mode spatiotemporal graph convolutional neural network model combined with attention mechanism is proposed. This paper has been tested and validated in very stochastic traffic forecasting. experiments on two publicly available traffic datasets, PeMSD04 and PeMSD08, show that AD-STGCRN has higher prediction accuracy, with the model’s prediction errors reduced by an average of 18%, 16% and 25% for MAE, RMSE and MAPE respectively on PeMSD04 compared to other prediction models. On PeMSD08 they were reduced by an average of 18%, 16% and 20% respectively.