GEO-Coordinated Service Handover Across LEO Satellites Using Multi-Agent Deep Reinforcement Learning

Frequent service handover is inevitable in low Earth orbit (LEO) satellite networks, as the rapid movement of LEO satellites limits their visibility duration to ground users. The dynamic and time-varying topology further complicates the maintenance of seamless service continuity and balanced resource utilization. In view of this, we propose a geostationary orbit (GEO)-coordinated service handover scheme, where GEO satellites, remaining stationary relative to ground users, leverage their continuous and comprehensive visibility to make handover decisions among LEO satellites. A multi-agent deep reinforcement learning-based approach is developed to jointly optimize satellite selection and load distribution. Furthermore, service caching and resource backfill mechanisms are introduced to minimize service interruption and improve resource efficiency. Experimental results demonstrate that the proposed scheme achieves lower handover latency, better load balance, and higher resource satisfaction compared with baseline methods.