针对差异化设备的任务卸载方法  

Task Offloading Method for Differentiated Devices

作　　者：余楚佳 胡晟熙 林欣郁 陈哲毅 陈星[1,2,3] YU Chujia;HU Shengxi;LIN Xinyu;CHEN Zheyi;CHEN Xing(College of Computer and Data Science,Fuzhou University,Fuzhou 350116,China;Key Laboratory of Spatial Data Mining&Information Sharing,Ministry of Education,Fuzhou 350002,China;Fujian Provincial Key Laboratory of Networking Computing and Intelligent Information Processing(Fuzhou University),Fuzhou 350116,China)

机构地区：[1]福州大学计算机与大数据学院,福州350116 [2]空间数据挖掘与信息共享教育部重点实验室,福州350002 [3]福建省网络计算与智能信息处理重点实验室(福州大学),福州350116

出　　处：《小型微型计算机系统》2024年第8期1816-1824,共9页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(62072108)资助;福建省自然科学基金杰青项目(2020J06014)资助;中央引导地方科技发展资金项目(2022L3004)资助;福建省财政厅科研专项经费项目(83021094)资助.

摘　　要：在边缘计算中,为缓解移动设备计算能力、存储容量受限问题,通常将部分计算密集型任务卸载至边缘服务器.然而,由于移动设备计算能力的差异,无法为所有的移动设备制定统一的卸载方案.若对每个设备均单独进行训练,则无法满足时延需求.针对这一问题,本文提出了一种差异化设备上基于联邦深度强化学习的任务卸载方法.该方法使用环境内已有移动设备的卸载经验,结合深度Q网络和联邦学习框架,构建了一个全局模型.随后,使用新移动设备上少量经验在全局模型上微调以构建个人模型.基于多种场景的大量实验,将本文所提出方法与理想方案、Naive、全局模型和Rule-based算法进行对比.实验结果验证了本文所提出方法针对差异化设备任务卸载问题的有效性,能在花费较短时延的同时得到接近理想方案的卸载方案.In edge computing,to alleviate the constraints of computing power and storage capacity on mobile devices,some computation-intensive tasks might be offloaded to edge servers.However,due to diverse computing power of mobile devices,it is hard to design a unified offloading solution,and the delay requirements cannot be met if devices are trained separately.To address this issue,this paper proposes a task offloading method based on federated deep reinforcement learning on differentiated devices.This method uses the offloading experience of existing mobile devices in the environment,combined with deep Q network and federated learning framework,to construct a global model.Subsequently,fine-tune on the global model with a small amount of experience on new mobile devices to build personal models.Based on a large number of experiments in various scenarios,the proposed method is compared with the ideal scheme,Naive,global model and rule-based algorithms.The experimental results verify the effectiveness of the proposed method for the task offloading problem of differentiated devices,which can obtain a near-optimal offloading scheme while meeting the delay requirements.

关 键 词：边缘计算 任务卸载 依赖感知 深度强化学习 联邦学习 

分 类 号：TP339[自动化与计算机技术—计算机系统结构]