M-DRL的低轨道卫星网络计算卸载和任务迁移  

Computing Offloading and Task Migration of LEO Network based on M-DRL

作　　者：徐飞[1] 宁辛 安朔 申奥祥 王泽轩 XU Fei;NING Xin;AN Shuo;SHEN Aoxiang;WANG Zexuan(School of Computer Science and Engineering,Xi’an Technological University,Xi’an 710021,China)

机构地区：[1]西安工业大学计算机科学与工程学院,西安710021

出　　处：《西安工业大学学报》2024年第3期395-404,共10页Journal of Xi’an Technological University

基　　金：咸阳市科技局重点研发项目(2021ZDYF-NY-0019)。

摘　　要：针对无人机网络高时延、低性能、有限带宽、难以解决复杂计算任务问题,提出了一种将低地球轨道卫星和移动边缘计算技术结合形成的MEC辅助LEO卫星网络计算卸载和任务迁移方法。首先通过建立本地计算模型、卸载模型和迁移模型,确定目标优化成本函数。然后为降低模型复杂度,引入多智能体深度强化学习模型,利用多智能体双延迟深度确定性策略梯度(MATD3)算法求解优化问题,降低系统总时延。仿真结果表明,与本地计算及随机迁移算法相比,MATD3算法的任务处理时延分别降低94.55%和83.02%,证明了MATD3算法在计算卸载和任务迁移方面的有效性和可靠性。To solve the problems of high latency,low performance,limited bandwidth,and difficulty in solving complex computing tasks in UAV networks,this paper proposes a MEC assisted LEO satellite network computing offloading and task migration method that combines LEO satellites and Mobile Edge Computing technology.Firstly,the local computing model,offloading model and migration model are established to determine the target optimization cost function.Then,a Multi agent Deep Reinforcement Learning model is introduced to reduce the complexity of the model,and the Multi Agent Twin Delayed Deep Deterministic Policy Gradient algorithm is used to solve the optimization problem and reduce the total system delay.The simulation results show that the task processing latency of MATD3 algorithm is reduced by 94.55%and 83.02%compared with the local computation and random migration algorithms,respectively,which proves the effectiveness and reliability of MATD3 algorithm in computing offloading and task migration.

关 键 词：LEO卫星网络 移动边缘计算 MATD3算法 计算卸载 卫星通信 

分 类 号：TN927.2[电子电信—通信与信息系统]