Optimized Handover Management for Reliable Connectivity in GEO-LEO Satellite Networks via Predictive Reinforcement Learning

This paper presents a novel methodology to optimize the handover decision of mobile terminals (MTs) within a cooperative geostationary Earth orbit (GEO) and low Earth orbit (LEO) satellite network. Unlike prior research, which primarily focuses on inter-LEO satellite handovers for stationary terminals, this work addresses the handover challenges associated with MTs in a hybrid GEO-LEO satellite architecture. The proposed framework employs a convolutional neural network (CNN)-long-short-term memory (LSTM) encoder-decoder model to accurately predict the reference signal received power (RSRP) of potential handover target satellites, enabling more informed decision-making. A multi-agent double deep Q-network (MADDQN) algorithm is implemented to determine the optimal handover target, leveraging predicted RSRP data while considering factors such as network load and satellite connection duration. The proposed approach minimizes unnecessary handovers, enhances data throughput for MTs, and ensures stable connectivity and quality of service under diverse network conditions. The proposed method contributes to handover optimization in cooperative GEO-LEO satellite networks, offering valuable insights for next-generation satellite communication systems.