Deep Reinforcement Learning Based Multi-Objective Handover for LEO Satellite Networks

Satellite communication is widely recognized as a solution to achieve global coverage. However, due to the high mobility of the Low Earth Orbit (LEO) satellites, user equipment (UE) needs to handover frequently to maintain a stable connection. Furthermore, the complex topology of satellite networks exacerbates the challenge of user mobility management. This paper leverages artificial intelligence (AI) to optimize the satellite handover process, proposing a multi-objective optimization strategy to minimize handover frequency, balance satellite load, and maximize system throughput. A novel reward function is designed to enhance the stability of the user service, and a lightweight multilayer neural network is introduced to reduce computational complexity. The simulation results show that the proposed method outperforms state-of-the-art approaches in improving system throughput and reducing average handover times.