A Multi-Agent Hierarchical Group-Based Handover Strategy for Massive Users in LEO Satellite Networks

Abstract

In this letter, a multi-agent hierarchical group-based handover strategy is proposed for low Earth orbit (LEO) satellite networks, leveraging the Ward hierarchical clustering-enhanced dueling double deep Q-network (WHC-D3QN) strategy to reduce signaling overhead and enhance throughput. Specifically, the scheme first clusters users via Ward hierarchical clustering and then adopts centralized training with distributed execution to support efficient decision-making for these clustered users. On this basis, a personalized resource allocation network and a multiobjective reward mechanism are designed to alleviate network signaling overhead and load pressure. Simulation results indicate that, under a scenario with 1000 users, the proposed scheme decreases handovers by 61.29% compared to the load least first (LLF) algorithm without decreasing throughput, while also decreasing congestion rate by 46.26% and 55.02% relative to the QoS-aware idle handover (QIH) and Max received signal strength indicator (RSSI) algorithms, respectively.