基于深度Q网络算法的卫星边缘卸载策略
An edge offloading strategy in satellite based on deep Q network algorithm
王军选 1王月雯 1高阔阔1
作者信息
- 1. 西安邮电大学 通信与信息工程学院,陕西 西安 710121;陕西省信息通信网络及安全重点实验室,陕西 西安 710121
- 折叠
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(Mobile Edge Computing,MEC)技术与星地协同网络结合,提出一种基于深度 Q网络(Deep Q-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(Central Processing Unit,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.
Abstract
In the satellite-ground fusion network,in order to reduce the delay and energy consump-tion of users'offloading computing tasks,a satellite edge offloading computing strategy based on deep Q-network(DQN)algorithm is proposed by combining the mobile edge computing(MEC)technology with the satellite-ground collaborative network.The MEC server is deployed at the edge of the satellite network,and the central processing unit(CPU)is set to be an intelligent agent that can interact with the surrounding environment,and the task offloading delay and energy consump-tion weighting and minimization problems are established.In order to solve the non-convex optimi-zation problem,it is converted to a Markov decision process,and the corresponding state space,re-ward function and strategy update function are established,and the optimal solution is found.Simu-lation results show that compared with the Q-learning strategy and the actor-critic(AC)strategy,the proposed strategy can effectively increase the average return of the system and reduce the sys-tem overhead.
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Key words
mobile edge computing/high earth orbit satellite/low earth orbit satellite/deep Q net-work/Markov decision process/the 6th generation of mobile communication system引用本文复制引用
出版年
2024
NETL
NSTL
维普
万方数据