移动通信2025，Vol.49Issue(6)：35-42,8.DOI:10.3969/j.issn.1006-1010.20250423-0002
面向低轨卫星网络的算力路由策略
Computing Power Routing Strategy for Low Earth Orbit Satellite Networks
摘要
Abstract
The increasing demand for computing power has spurred the development of satellite computing networks,where low earth orbit(LEO)satellite networks complement the service coverage of terrestrial networks.In response to the highly dynamic topology and limited node resources of LEO satellite networks,this paper proposes an inter-satellite computing power routing strategy based on graph neural networks.This graph neural network module enhances the generalization ability of model,allowing it to adapt to these dynamically changing topologies.The proposed algorithm optimizes the selection of the next-hop node to determine the task offloading node,ultimately minimizing the average service time for computing tasks over a specified period.Simulation results demonstrate that,compared to heuristic approaches,the proposed approach increases the overall network throughput by more than 10％and reduces end-to-end transmission delay by nearly 25％.关键词
算力路由/低轨卫星网络/深度强化学习/图采样聚合Key words
computing power routing/low earth orbit satellite networks/deep reinforcement learning/graph sampling and aggregation分类
电子信息工程引用本文复制引用
许柳飞,罗志勇..面向低轨卫星网络的算力路由策略[J].移动通信,2025,49(6):35-42,8.基金项目
国家重点领域研发计划资助项目"面向手机直连的全球运营天地融合总体架构"(2023YFB2904701) （2023YFB2904701）
广东省重点领域研发计划资助项目"卫星互联网NTN一体化融合研究与验证"(2024B0101020006) （2024B0101020006）
广东省区域联合基金重点项目"6G星云算网融合体系关键技术研究"(2023B1515120093) （2023B1515120093）
深圳市自然科学基金重点项目"算网融合体系结构关键问题研究"(JCYJ20220818102209020) （JCYJ20220818102209020）
