Large scale LEO constellation Q-Learning QoS routing algorithm based on celestial grid

Intelligent QoS routing is a research hotspot and challenge for large-scale LEO constellations. This article focuses on issues such as virtual real topology drift, multi service QoS conflicts, and dynamic load imbalance in LEO constellations, and proposes a Q-Learning QoS routing algorithm based on celestial grids. By integrating non-uniform discretization of the celestial sphere with Beidou grid coding, the problems of frequent link switching and virtual real topology synchronization can be solved. Based on this, a Q-Learning routing algorithm was designed by combining business heat maps, with bandwidth, load, heat level, and hop count as joint optimization objectives. A differentiated QoS reward mechanism was constructed to dynamically avoid congested links through real-time learning. The simulation results show that compared with HLLMR and Dijkstra's algorithm, this algorithm reduces packet loss rate by 4% and 11% respectively, increases throughput by 7% and 15%, and achieves comparable latency to HLLMR. It realizes the collaborative optimization of large-scale LEO constellation QoS guarantee and dynamic load balancing.