Spatio-Temporal Multi-View Based Short-Term Traffic Forecasting for Incomplete Time Series in LEO Satellite Networks
Liang Peng, Jie Yan, Binquan Guo, Xiaoxiang Wang · 2024
Accurate short-term traffic forecasting is essential to improve the efficiency of data transmission in Low Earth Orbit (LEO) satellite networks. Due to collector failures, transmission errors and memory failures in the complex space environment, the traffic value absence phenomenon may occur. However, in-complete traffic time series can significantly reduce the accuracy of traffic forecasting. To overcome this problem, in this work, we propose a novel Spatio- Temporal Multi-view based Short-term Traffic Forecasting (STMV-STF) model for incomplete time series to improve the accuracy of traffic prediction by combining the unique spatio-temporal correlation of satellite network traffic. Specifically, this model utilizes the time-lagged pearson correlation equation to select the$k$most correlated time series to impute the missing values from a spatial view. Meanwhile, this model utilizes a proposed gated recurrent unit (GRU) with a memory decay gate (MG-GRU) to impute the missing values from a temporal view. Finally, we aggregate the missing values imputed from these two views and design the training process of this model to enable online imputation of missing traffic values and real-time output of predicted traffic values. Experiments on real traffic datasets show that the STMV-STF model achieves 19.86% to 33.46% error reduction under different missing rate conditions compared to the baseline model in terms of root mean square error (RMSE) metric.