基于雙池DDPG的邊緣計(jì)算卸載延時(shí)性能研究

打開文本圖片集

中圖分類號：TN929.5-34 文獻(xiàn)標(biāo)識碼：A 文章編號：1004-373X（2025）17-0117-07

DOI：10.16652/j.issn.1004-373x.2025.17.018引用格式：，等.基于雙池DDPG的邊緣計(jì)算卸載延時(shí)性能研究[J].現(xiàn)代電子技術(shù)，2025，48（17）：117-123.

Abstract：Withtheproliferationofcomputation-intensiveandlatency-sensitiveservices，itisdificultforthetraditional cloudcomputing modeltomettherequirementsoflowlatencyandhigh-qualityservices.Taskscanbeofloaded todgesubnets toreducelatencyandaleviatenetworkcongestionbythedistributededgecomputing mode.Inviewof theoffoadingproblemof collaborativecomputingonedgecomputingservers，acollaborativecomputing resourcealocationalgorithmbasedondep reinforcementlearningisproposed.Firstly，theoptimizationobjectiveisdeterminedaccordingtotheestablishedheterogneous edge network odel，thetasklatencyisinimied，andtheewardfunctionisiven.SecondlyaccordingtothetraditioalG （deepdeterministicpolicygradient）algorithm，thedual-polisintroducedtoclassifyandstorethedatatoimprovethequalityof empiricaldata.Finally，theagentistrainedtofindoutanoptimalstrategytoaapttotheenvironmenttoachieveboththeload balancingandtheresourceefectivealocationamongservers.Simulationresultsshowthat theimprovedD3PG（dual-poolDDPG） algorithmfullyconsiderstheheterogeneityofedgesubnetsandthemobilityofedgedevices，nditstask latencyisreducedby 3.8%～24.8%in comparison with that of the traditional learningalgorithms.

Keywords：heterogeneousnetwork；edge computing;computationofloading；task latency；deepreinforcementlearning; D3PG

0 引言

隨著無線通信技術(shù)的發(fā)展，計(jì)算密集型和延遲敏感型邊緣應(yīng)用業(yè)務(wù)不斷涌現(xiàn)，這對網(wǎng)絡(luò)邊緣側(cè)的數(shù)據(jù)、算力、存儲等需求大幅提高。（剩余11306字）