Research Article
Satellite Traffic Forecast Based on Multi-dimensional Periodic Features
@INPROCEEDINGS{10.1007/978-3-030-93398-2_27, author={Weidong Zhou and Yana Qian and Kanglian Zhao and Wenfeng Li and Fa Chen}, title={Satellite Traffic Forecast Based on Multi-dimensional Periodic Features}, proceedings={Wireless and Satellite Systems. 12th EAI International Conference, WiSATS 2021, Virtual Event, China, July 31 -- August 2, 2021, Proceedings}, proceedings_a={WISATS}, year={2022}, month={1}, keywords={Traffic flow prediction LSTM network Attention mechanism}, doi={10.1007/978-3-030-93398-2_27} }- Weidong Zhou
Yana Qian
Kanglian Zhao
Wenfeng Li
Fa Chen
Year: 2022
Satellite Traffic Forecast Based on Multi-dimensional Periodic Features
WISATS
Springer
DOI: 10.1007/978-3-030-93398-2_27
Abstract
With the development of satellite networks and the increase in business requirements, it is a challenge to better provide high-quality service to users. The accurate prediction of end-to-end traffic can contribute to the realization of congestion control, resource allocation and anomaly detection. Accurate flow forecasts need to consider the short-term and long-term features of the flow. Our paper proposed a model based on deep learning named Multi-Dimensional Temporal Feature Neural Network (MTFNN) to capture both short-term dependencies and long-term dependencies for traffic prediction. MTFNN mainly contains two components: 1) Short-Term Temporal Dependencies which based on Long-Short Term Memory network (LSTM), used to predict the basic trend of traffic. 2) Long-Term Temporal Dynamic Similarity which based on LSTM and Attention mechanisms, used to improve the model’s sensitivity to fluctuations and peak prediction. Experiments performed on real-world public traffic datasets show our proposed model has a smaller prediction error and more accurate peak prediction.
