Traffic Flow Prediction Based on Spatiotemporal Feature Fusion for 5G-R Networks

Precise traffic flow prediction is the premise of energy-saving control in 5G networks. However, unlike in public networks, traffic in 5G-R networks has many spikes and closely relates to the train departure direction and timetable, which brings great challenge to the traffic flow prediction. The existing prediction methods for cellular networks cannot achieve satisfactory performance in 5G-R scenario. To address this issue, we propose a spatiotemporal feature fusion based prediction model that enables us to capture the spatiotemporal correlation and periodicity of the traffic flow in 5G-R networks. In the model, time and spatiotemporal features among the traffic are separately extracts via two modules consisting of LSTM and CNN-LSTM networks, respectively. The final predicted value is obtained after feature fusion with attention mechanism. The experimental results show that the proposed model can significantly improve the prediction accuracy of 5G-R traffic compared with baselines. The root mean square error (RMSE) and mean absolute error (MAE) for non-stopping stations are reduced by 76% and 66%, respectively.