边缘计算中动态服务器部署与任务卸载联合优化算法

打开文本图片集

中图分类号：TP393 文献标志码：A 文章编号：1001-3695（2025）06-031-1830-08

doi： 10.19734/j. issn.1001-3695.2024.11.0462

Joint optimization algorithm for dynamic server deployment and task offloading in edge computing

BaiWenchao，Lu Xianling† （SchoolofInternetofThings，JiangnanUniversity，WuxiJiangsu214122，China）

Abstract：Inmobileedgecomputing，thefixedlocations of edgeserverdeploymentscanleadtoimbalancedresourceutilizationof edgeservers，resultinginincreasedlatencyandenergyconsumptionduringthetask ofloading process.Toaddressthis issue，thispaperproposedaierarchicalreiforcementlearing-basedjointoptimizationalgorithm.Firstly，itdecomposedthe problemofedge serverplacementand task ofloading and transformedthemintoabi-Markovdecision processThen，itconstructedaglobalintellgentagentmodelforhigher-leveledgeserverdeploymentusingthedeepQ-network，andaccelerated modelconvergencebyintroducingthe K-means algorithmtoprovide high-qualitysamplesforthehigher-layer policy.Itbuilta lower-layer multi-agentmodelfortaskofloadingusingthemulti-agentproximalpolicyoptimizationalgorithm，andimproved trainingstabilitybyintroducingstatenormalizationtoreducethestatesfeaturescalediferencesinthelower-layerpolicy.Finally，itachievedtheultimateoptimizationgoaltroughalternatingoptimizationofthehigher-layerandlower-layerpolicies. Simulationresults indicatethattheproposedalgorithmcanachieveoptimal serverdeploymentandtask ofloading strategies， comparedtorandomstrategiesandotherreinforcementlearningalgorithm，and itdemonstratesgreater benefitsintermsof model training efficiency，target rewards，and load balancing metrics.

Key words：edge computing；task ofloading；edge server deployment；hierarchical reinforcement learning

0 引言

随着5G和6G技术的高速发展，越来越多的计算密集型应用出现[1]，如虚拟现实和面部识别等。（剩余20850字）