南京邮电大学学报（自然科学版）2026，Vol.46Issue(1)：56-65,10.DOI:10.14132/j.cnki.1673-5439.2026.01.007
基于多智能体深度强化学习的智能卫星切换策略
Intelligent satellite handover strategy based on multi-agent deep reinforcement learning
摘要
Abstract
The development of the maritime Internet of things(IoT)imposes higher demands on the com-munication quality of ultra-dense low Earth orbit(LEO)satellite networks.Traditional fixed-rule hando-ver strategies struggle to adapt to the high-speed motion of LEO satellites,often leading to frequent han-dovers.This study,referencing the first phase of the Starlink project for constellation modeling,proposes an intelligent handover strategy based on multi-agent deep reinforcement learning.A hierarchical decision-making architecture is constructed to combine traditional threshold-based decision-making with reinforcement learning optimization.Leveraging a multi-terminal collaboration mechanism,it achieves global resource coordination.Multi-dimensional metrics,including carrier-to-noise ratio,elevation angle,and remaining service time,are comprehensively considered to design a comprehensive reward function that guides the agents to learn the optimal handover strategy.Simulation results demonstrate that in an ultra-dense constellation environment,this strategy achieves an average access success rate of 92.4%,an improvement of over 13%compared to traditional methods.Furthermore,it maintains stable performance of 77.6%even under high-load scenarios.Although the handover frequency increases,the continuity of communication is effectively ensured through primary handovers.This provides a novel solu-tion to address the intermittent connectivity issues in the maritime IoT.关键词
超密集低轨卫星网络/切换策略/星座建模/多智能体深度强化学习Key words
ultra-dense low Earth orbit(LEO)satellite networks/handover strategy/constellation mod-eling/multi-agent deep reinforcement learning分类
信息技术与安全科学引用本文复制引用
葛富源,张申虎,闫实..基于多智能体深度强化学习的智能卫星切换策略[J].南京邮电大学学报（自然科学版）,2026,46(1):56-65,10.基金项目
国家自然科学基金联合基金重点项目(U21A20444)和国家自然科学基金(62371067)资助项目 （U21A20444）
