基于混合深度强化学习的云制造云边协同联合卸载策略

打开文本图片集

Joint offloading strategy for cloud manufacturing based on hybrid deep reinforcement learning in cloud-edge collaboration

Zhang Yaru，Guo Yinzhang† （Colegeof ComputerScience&Technology，Taiyuan UniversityofScience&Technology，TaiyuanO3o024，China）

Abstract：Toaddress theissueofreal-time perceptiondata frommanufacturingresources being dificult toprocess promptlyin acloud-edgecollaborative cloud manufacturingenvironment，considering uncertainfactors suchas the limited computingresources attheedge，dynamicallychanging network conditions，andtaskloads，thispaper proposedacloud-edgecolaborative jointoffloading strategybasedonmixed-baseddeepreinforcementlearning（M-DRL）.Firstly，thisstrategyestablishedajoint ffloadigmodelbycombiningdiscretemodeloffloadinginthecloudwithcontinuoustaskoffloadingattheedge.Secondly， this strategydefinedtheoptimizationproblemasaMDPtominimizethetotalcostofdelayandenergyconsumptionoveraperiod. Finally，thispaper used the M -DRL algorithm，which utilized an integrated exploration strategy of DDPG and DQNand introducedalongshort-termmemory network（LSTM）intothenetworkarchitecture，tosolvethisoptimizationproblem.Simulation results showthatcompared with some existing ofloadingalgorithms，the M-DRL method has goodconvergence and Stability， andsignificantlyreduces thetotalsystemcost.Itprovidesanefectivesolutionforthetimelyprocessingofmanufacturingresource perception data.

Keywords：cloudmanufacturing；cloud-edgecolaboration;jointofloading；LSTMreinforcementlearning；Markovdecision process（MDP）

0 引言

云制造作为一种新兴的生产模式，以前所未有的速度改变着全球制造业的格局。（剩余20442字）