移动通信2024，Vol.48Issue(9)：132-140,9.DOI:10.3969/j.issn.1006-1010.20240528-0002
基于PM算法的星间协作计算卸载和任务迁移
Inter-satellite Collaborative Computing Offloading and Task Migration Based on PM Algorithm
摘要
Abstract
In Space-Ground Integrated Networks,the high mobility of LEO satellites poses challenges in maintaining consistent and stable communication links,resulting in suboptimal Quality of Service(QoS)for users.To address this issue,we propose an inter-satellite collaborative computing method based on Deep Reinforcement Learning(DRL).This method minimizes the total weighted energy consumption and latency through computation offloading between terminal users and LEO satellites,as well as task migration among LEO satellites.Furthermore,to address the high-dimensional action space,complex state space,and instability during training in the multi-agent deep deterministic policy gradient(DDPG)algorithm,we introduce an improved approach utilizing a prioritized experience replay mechanism.This approach,termed the cooperative prioritized experience replay migration algorithm,enhances convergence performance.Simulation results validate the effectiveness of the proposed cooperative algorithm.Moreover,compared to the multi-agent twin delayed DDPG algorithm,random algorithms,and local computing schemes,the proposed algorithm significantly reduces the system overall cost.关键词
天地一体化网络/低地球轨道卫星/深度强化学习/星间协作Key words
space-ground integrated network/low earth orbit satellite/deep reinforcement learning/inter-satellite collaboration分类
信息技术与安全科学引用本文复制引用
宁辛,徐飞,刘双煜,申奥祥..基于PM算法的星间协作计算卸载和任务迁移[J].移动通信,2024,48(9):132-140,9.基金项目
陕西省科技厅区域创新能力指导计划(20122qfy01-14) （20122qfy01-14）
咸阳市科技局重点研发项目(2021ZDYF-NY-0019) （2021ZDYF-NY-0019）
