第29卷第1期2024年1月doi:10.13682/j.issn.2095-6533.2024.01.001基于深度Q网络算法的卫星边缘卸载策略西安邮电大学学报JOURNALOFXI'ANUNIVERSITYOFPOSTSANDTELECOMMUNICATIONSVol.29No.1Jan.2024王军选12,王月雯1-2,高阔阔1.2(1.西安邮电大学通信与信息工程学院,陕西西安710121;2.陕西省信息通信网络及安全重点实验室,陕西西安710121)摘要:在星地融合网络中,为了降低用户卸载计算任务的时延和能耗,将边缘计算(MobileEdgeComputing,MEC)技术与星地协同网络结合,提出一种基于深度Q网络(DeepQ-Network,DQN)算法的卫星边缘卸载策略。在卫星网络边缘部署MEC服务器,将中心处理单元(CentralProcessingUnit,CPU)设为可与周围环境交互的智能体,建立任务卸载时延和能耗加权和最小化问题。为求解该非凸优化问题,将其转化为马尔科夫决策过程,确立对应的状态空间、动作空间和奖励函数及策略更新函数,寻求最优解。仿真结果表明,与基于Q学习(Q-learning)策略和基于演员家-评论家(Actor-Critic,AC)策略进行对比,所提策略可以有效地增加系统的平均回报值,降低系统开销。关键词:移动边缘计算;高地球轨道卫星;低地球轨道卫星;深度Q网络;马尔科夫决策过程;第六代移动通信系统中图分类号:TN927AnedgeoffloadingstrategyinsatellitebasedondeepQWANGJunxuanl-2,WANGYuewenl-2,GAOKuokuol.(1.SchoolofCommunicationsandInformationEngineering,Xi'anUniversityofPostsandTelecommunications,Xi'an710121,China;2.ShaanxiKeyLaboratoryofInformationCommunicationNetworkandSecurity,Xi'anAbstract:Inthesatellite-groundfusionnetwork,inordertoreducethedelayandenergyconsump-tionofusers'offloadingcomputingtasks,asatelliteedgeoffloadingcomputingstrategybasedondeepQ-network(DQN)algorithmisproposedbycombiningthemobileedgecomputing(MEC)technologywiththesatellite-groundcollaborativenetwork.TheMECserverisdeployedattheedgeofthesatellitenetwork,andthecentralprocessingunit(CPU)issettobeanintelligentagentthatcaninteractwiththesurroundingenvironment,andthetaskoffloadingdelayandenergyconsump-tionweightingandminimizationproblemsareestablished.Inordertosolvethenon-convexoptimi-zationproblem,itisconvertedtoaMarkovdecisionprocess,andthecorrespondingstatespace,re-wardfunctionandstrategyupdatefun...