面向ET-DQN的卫星网络任务部署算法研究  

Research on Satellite Network Task Deployment Algorithm Based on ET-DQN

作　　者：刘治国[1] 董效奇 汪林[2] 夏清雨 潘成胜[1,3] LIU Zhiguo;DONG Xiaoqi;WANG Lin;XIA Qingyu;PAN Chengsheng(Communication and Network Laboratory,Dalian University,Dalian 116622,China;School of Environmental and Chemical Engineering,Dalian University,Dalian 116622,China;School of Electronics and Information Engineering,Nanjing University of Information Science&Technology,Nanjing 211800,China)

机构地区：[1]大连大学通信与网络重点实验室,辽宁大连116622 [2]大连大学环境与化学工程学院,辽宁大连116622 [3]南京信息工程大学电子与信息工程学院,南京211800

出　　处：《小型微型计算机系统》2024年第2期418-424,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61931004)资助。

摘　　要：卫星边缘计算服务已经成为网络的重要组成部分,特别在地面网络稀疏环境下,可显著的提高用户的服务质量.但卫星网络的资源有限性为边缘计算任务的部署带来了巨大困难.因此,本文针对部署位置和计算资源分配不合理而导致任务处理时延过长、卫星能耗过大的问题,在SDN(Software Defined Network)的卫星-地面联合部署网络架构(SDN-Space-Ground Integrated Network,SSGIN)的系统模型下.提出了ET-DQN(Experience Tournament-DQN)的任务部署算法,该算法通过引入经验竞选机制提高经验池利用率和网络训练的效率,解决了经验池利用率低,而造成的网络过估计、难收敛的问题.仿真结果表明,该方法能够有效降低任务的响应时延以及能耗,很好地改善了传统DQN算法收敛速度慢和样本的利用率低下的问题.Satellite edge computing service has become an important part of the network,especially in the sparse environment of terrestrial network,it can significantly improve the quality of service of users.However,the limited resources of satellite networks bring great difficulties to the deployment of edge computing tasks.Therefore,aiming at the problems of long task processing delay and excessive satellite energy consumption caused by unreasonable deployment location and computing resource allocation,this paper is based on the system model of SDN based satellite ground integrated network(SDN-Space-Ground Integrated Network,SSGIN).A task deployment algorithm based on ET-DQN(Experience Tournament-DQN)is proposed.The algorithm improves the utilization of experience pool and the efficiency of network training by introducing the experience election mechanism,and solves the problems of over estimation and difficult convergence caused by the low utilization of experience pool and the network.Simulation results show that this method can effectively reduce the response delay and energy consumption of tasks,and improve the problems of slow convergence and low sample utilization of traditional DQN algorithm.

关 键 词：软件定义网络 卫星网络 深度强化学习 边缘计算 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]