A tactical Internet incorporating multi-source dynamic spatio-temporal characteristics traffic prediction model

Abstract

Tactical Internet traffic has extremely dynamic spatio-temporal features and is closely related to external features such as weather and elevation, existing network traffic prediction models can not extract its complex features well, a tactical Internet traffic prediction model that fuses multi-source dynamic spatio-temporal features is proposed. Firstly, external features are fused with traffic features as multi-source features; then the spatial features of network traffic at the current moment are extracted and the convolution weights over time are iteratively updated to obtain the spatial feature information under different time slices; finally, the spatial information of the current and historical moments are aggregated by the temporal convolution layer to predict the multi-source dynamic spatio-temporal traffic at the next moment. Compared with the single base model, the proposed method is better in all three evaluation metrics, namely, mean absolute error (MAE), root mean square error (RMSE) and coefficient of determination (R2).

Keywords

• network traffic prediction|tactical internet|multi-source dynamic spatial and temporal characteristics|neural network