A Deep Ensemble Network Model for Refined Traffic Volume Prediction Considering Spatial-Temporal Features 2020-01-5191
Robust and accurate short-term traffic volume prediction methods are indispensable in driving assistance and active traffic control and management. With the popularity of deep learning, the hybrid methods play an important role in improving the prediction accuracy. To fully cover the spatial-temporal characteristics of traffic flow, this paper proposes an attention-based spatial and temporal model (AST) through combining convolutional neural network (CNN), gated recurrent unit (GRU). Besides, the attention mechanism is also introduced after GRU to further improve the prediction accuracy. The experiments which are carried on the Beijing expressway traffic volume data indicate that the AST model has better performance than the baseline models in terms of prediction accuracy. Compared with ARIMA, SVR, CNN, and GRU, the MAE of AST is reduced by 21.6%, 20.9%, 11.0%, and 9.9%; the MAPE is reduced by 14.4%, 15.1%, 10.7%, and 10.3%; the RMSE is reduced by 22.6%, 20.3%, 11.0%, and 10.8%. Finally, the comparison of prediction accuracy between AST and the sub-components verifies the necessity of each part of the model.