移动通信2025，Vol.49Issue(6)：95-102,8.DOI:10.3969/j.issn.1006-1010.20250424-0002
基于联邦学习的空天地一体化网络计算卸载与资源分配方法研究
Federated Learning Driven Computation Offloading and Resource Allocation in Space-Air-Ground Integrated Networks
摘要
Abstract
With the rapid development of unmanned aerial vehicle(UAV)and low earth orbit(LEO)satellites,the space-air-ground integrated network(SAGIN)combined with edge computing is expected to provide seamless and ubiquitous computing power for the Internet of Things(IoT).For computation offloading in SAGIN,energy consumption,delay,and load balancing are critical.This paper proposes a federated learning(FL)based computation offloading and resource allocation approach to jointly optimize the overall task processing latency and energy consumption under the constraint of load balancing.Specifically,the formulated optimization problem is first transformed into a task scheduling problem based on Markov decision process(MDP).Then,based on the deep reinforcement learning(DRL)algorithm,the federated deep Q network(DQN)algorithm is proposed,which allows the DQN agent to interact with the environment to obtain the optimal computation offloading and resource allocation policy.Numerical simulation results the superiority of the proposed scheme in terms of both task processing latency and energy consumption.关键词
空天地一体化网络/计算卸载/资源分配/联邦学习Key words
space-air-ground integrated network/computation offloading/resource allocation/federated learning分类
信息技术与安全科学引用本文复制引用
张恒,马婷,陈光霁,倪艺洋,刘婷婷,李骏..基于联邦学习的空天地一体化网络计算卸载与资源分配方法研究[J].移动通信,2025,49(6):95-102,8.基金项目
国家自然科学基金"面向感通算融合的空天地一体化网络多维资源协同优化研究","面向5G/B5G密集组网的蜂窝车联网高效资源共享方法研究"(62401264,62171217) （62401264,62171217）
