Spatio-temporal Traffic Prediction-Assisted Interruption-Tolerant Backup Routing Algorithm for Large-Scale LEO Satellite Networks

Large-scale Low Earth Orbit (LEO) satellite networks offer the potential to provide global broadband access and are regarded as a key enabler for future 6G communications. However, LEO satellite networks face two major challenges, including unevenly distributed terrestrial traffic and frequent topology changes caused by inter-satellite link (ISL) interruptions or satellite failures. Moreover, traditional single-path routing algorithms suffer from ineffective load balancing, high sensitivity to ISL interruptions, and low resource utilization. Meanwhile, existing multi-path routing algorithms struggle to achieve accurate and optimal traffic allocation. To address the above challenges, we propose a traffic prediction-assisted interruption-tolerant backup routing (TPBR) scheme. Specifically, we first construct an AIdriven routing control architecture to manage routing efficiently while decoupling traffic prediction from routing decisions. Then, we develop a spatio-temporally adaptive traffic prediction model that combines graph convolutional networks with temporal convolutional networks to accurately predict the original traffic distribution on each ISL. Furthermore, we design a link-disjoint route precalculation method to calculate multiple backup paths, and propose a dynamic backup routing algorithm to enable accurate and dynamic traffic allocation. Finally, we comprehensively evaluate the proposed TPBR algorithm in various scenarios. The simulation results show that TPBR significantly outperforms existing algorithms and achieves excellent load balancing and interruption tolerance, especially in dynamic ISL interruption scenarios, where the average delay and packet loss ratio are reduced by up to 27.31% and 67.23%, respectively, and the throughput is improved by up to 40.96%.