基于深度强化学习的隐私保护任务卸载策略
Privacy-preserving task offloading strategy based on deep reinforcement learning
王亚林 1王康 1张博文 1王茂励1
作者信息
- 1. 曲阜师范大学网络空间安全学院,273165,山东省曲阜市
- 折叠
摘要
针对多移动边缘计算(MEC)服务器网络中计算卸载任务特征隐私泄露的问题,提出了一种基于深度强化学习(DQL)的隐私保护的在线计算卸载新方法.该方法首先通过对计算卸载特征任务的隐私泄露进行度量,设定隐私泄露约束阈值,保护用户卸载的隐私信息;然后以最小化系统能耗为目标,将该优化问题转变为马尔可夫决策过程;最后利用深度Q网络(DQN)算法求解出满足隐私约束和最小化能耗目标的接近最优的卸载决策.实验结果表明,该算法能够在有效降低用户设备平均能耗的同时,使任务卸载频率始终满足隐私约束.
Abstract
Existing deep reinforcement learning-based computational offloading approaches protect only usage pattern privacy and location privacy.In this paper,we consider a new privacy problem in multi-MEC server networks,i.e.,computational offloading task feature privacy leakage,which is lacking in current re-search.To address this issue,this paper proposes a new DQN-based privacy-preserving online computation-al offloading approach in MEC networks.The method will set the privacy leakage threshold by measuring the privacy leakage of the computational offloading feature task,which can protect the privacy information of user offloading;then the optimization problem is transformed into a Markov decision process with the objective of minimizing system energy consumption;finally the optimal offloading decision satisfying the privacy constraint and minimizing energy consumption objective is solved by the DQN-based algorithm.The experimental results show that the algorithm can effectively reduce the total energy consumption of the system while effectively reducing user privacy leakage.
关键词
任务卸载/多MEC服务器/隐私泄漏量/DQNKey words
task offloading/multi-MEC server/privacy leakage/DQN引用本文复制引用
基金项目
山东省自然科学基金(ZR20211260301)
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据