Dynamic Load-Balancing Routing Strategy for LEO Satellite Networks Based on Spatio-Temporal Traffic Prediction

Low earth orbit (LEO) satellite networks have become increasingly important for providing global communication services. However, the highly dynamic topology and uneven load distribution in LEO satellite networks pose significant challenges for load-balancing routing techniques. We propose a dynamic load-balancing routing algorithm (DLBR) based on the multiagent dueling double deep Q-network (D3QN), which adopts a fully distributed local decision-making mode to adapt to the highly dynamic satellite network topology effectively. The algorithm makes optimal routing decisions based on local load information combined with predictive traffic information. To more effectively obtain predictive traffic information and guide the load-balancing routing algorithm in making accurate decisions, we further propose a traffic prediction algorithm based on graph convolutional network (GCN), long short-term memory (LSTM), and Attention mechanism, which fully utilizes the spatial and temporal features of the LEO satellite network traffic for spatiotemporal traffic prediction. Our results demonstrate that the proposed DLBR based on traffic prediction can effectively reduce the maximum and average bandwidth utilization of routing paths, thereby balancing the network load