一种多智能体协同的接人网切片资源分配方法

打开文本图片集

中图分类号：TN929.5 文献标志码：A文章编号：1003-3114（2025）04-0780-09

Abstract：Network slicingtechologyneeds toachievend-to-endon-demandsegmentationof physicalnetworks，efectivelyllo cateavailableatesaongdinttenants，ndlaeuslyimncessaryesoucstoachscetotdietd requirements.Amult-agentreinforcementlearing methodforresourcealocationinRadioAcess Network（RAN）isproposedtoaddressthisisue.irstly，alayeredframeworkisdesignedtosetupeuralnetworkagentsatachtenant，hichdynamicalljustre sourceallocationsrategiesasedonchangesinunderlingtraffc.henanagegationlaerissetupomanagethenetworkpara tersofeachitelliententandusteinggrihseusdtoageateentokparamesofntellgtagntssilar needfrom tenants，whichareusedtoupdatethelocalmodelsofinteligentagentsateachtenant，achievingfastertrainingandreducingcommunicationcostswhileachievingtheglobaloptimalsolution.Simulationresultsshowthattherewardvalueofthisalgorithmis improved by about 15.7% compared to other existing algorithms，and the Service Level Agreement （SLA） satisfaction of tenants and the overall resource utilization of the system can be maintained at a high level，with good scalability.

Keywords：5Gnetworkslicing；RANslices；deepreinforcementlearning；multi-agentreinforcementlearning;radio resource allocation

0引言

网络切片是5G通信技术的关键技术之一[1]允许基础设施提供商创建多个端到端逻辑网络，并在每个逻辑网络中部署特定的存储、网络和计算资源，再提供给租户。（剩余11800字）