A Novel Traffic Prediction Method for Dynamic Satellite Networks Based on Graph Attention Networks
Zihan Zhu, Yuzhuo Wang, Ke Wu, Yunpeng Hou, Huasen Hel, Jian Yang · 2025
Satellite networks have been proposed as a vital component in 6G networks for providing global connections. In recent years, satellite network traffic has been increasing. However, the limited onboard resources and inter-satellite link bandwidth make network congestion a critical issue. In order to avoid network congestion and improve quality of service, satellite traffic prediction has received more attention. However, the traditional forecasting models do not fully consider the dynamic topology of satellite networks and the spatial-temporal features of traffic. Thus, we propose a novel traffic prediction method for dynamic satellite networks. Considering that there is traffic correlation between nodes without direct connection, our model introduces a graph generation module to generate adjacency matrices based on dynamic attributes. Moreover, to improve the accuracy of prediction, we further propose the spatial attention module and time sequence processing module to exploit the temporal and spatial correlations of satellite traffic respectively. The experimental results show that our method outperforms the compared algorithms and increases up to 6.84 % prediction accuracy.