基于深度确定性策略梯度的星地融合网络可拆分任务卸载算法  

Split task offloading algorithm for satellite-ground integrated networks based on deep deterministic policy gradient

作　　者：宋晓勤[1,2] 吴志豪 赖海光 雷磊[1] 张莉涓 吕丹阳 郑成辉 SONG Xiaoqin;WU Zhihao;LAI Haiguang;LEI Lei;ZHANG Lijuan;LYU Danyang;ZHENG Chenghui(College of Electronics and Information Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 210016,China;National Mobile Communications Research Laboratory,Southeast University,Nanjing 210096,China;Nanjing Cowave Communication Technology Co.,Ltd.,Nanjing 211135,China)

机构地区：[1]南京航空航天大学电子信息工程学院,江苏南京210016 [2]东南大学移动通信全国重点实验室,江苏南京210096 [3]南京控维通信科技有限公司,江苏南京211135

出　　处：《通信学报》2024年第10期116-128,共13页Journal on Communications

基　　金：国家自然科学基金资助项目(No.62371232);东南大学移动通信全国重点实验室开放研究基金资助项目(No.2024D13);未来网络科研基金资助项目(No.FNSRFP-2021-ZD-4)。

摘　　要：为解决低轨卫星网络中星地链路任务卸载时延长的问题,提出了一种基于深度确定性策略梯度(DDPG)的星地融合网络可拆分任务卸载算法。针对不同地区用户建立了星地融合网络的多接入边缘计算结构模型,通过应用多智能体DDPG算法,将系统总服务时延最小化的目标转化为智能体奖励收益最大化。在满足子任务卸载约束、服务时延约束等任务卸载约束条件下,优化用户任务拆分比例。仿真结果表明,所提算法在用户服务时延和受益用户数量等方面优于基线算法。To address the prolonged task offloading delay in low earth orbit satellite networks,a split task offloading algorithm based on deep deterministic policy gradient(DDPG)was proposed for satellite-ground integrated networks.A multi-access edge computing structural model of the satellite-ground integrated network was established for users in different regions.By applying a multi-agent DDPG algorithm,the objective of minimizing total system service delay was transformed into maximizing agent reward returns.Under the constraints of sub-task offloading,service delay,and other task offloading conditions,the user task splitting ratio was optimized.Simulation results indicate that the proposed algorithm outperforms baseline algorithms in terms of user service delay and the number of benefited users.

关 键 词：星地融合网络 深度确定性策略梯度 资源分配 多接入边缘计算 

分 类 号：TN92[电子电信—通信与信息系统]