基于博弈优化边缘学习的物联网入侵检测研究  

Leveraging edge learning and game theory for intrusion detection in Internet of things

作　　者：梁浩然 伍军[1] 赵程程 李建华[1] LIANG Haoran;WU Jun;ZHAO Chengcheng;LI Jianhua(Shanghai Jiao Tong University,Shanghai 200240,China;Muroran Institute of Technology,Muroran 050-8585,Japan)

机构地区：[1]上海交通大学网络安全技术研究院,上海200240 [2]室兰工业大学,日本室兰050-8585

出　　处：《物联网学报》2021年第2期37-47,共11页Chinese Journal on Internet of Things

基　　金：国家自然科学基金资助项目(No.U20B2048,No.61972255)。

摘　　要：随着5G的商用和6G开始布局,海量物联网设备正在加速接入互联网,为新一代信息物理融合系统提供决策数据。物联网设备的高度异构及分布式特性使得物联网面临复杂威胁,这些威胁可使信息物理融合系统的关键决策失效。传统技术难以在保护节点隐私的前提下进行入侵检测且容易形成单点故障,同时缺乏协同入侵检测激励机制。因此,基于博弈优化边缘学习,研究了面向物联网的入侵检测系统。基于联邦学习在网络边缘构建了分布式隐私保护物联网入侵检测框架。在此基础上,基于多主多从博弈优化边缘学习过程,激励可信的入侵检测服务器及边缘设备参与边缘联邦学习。仿真实验证明了所提出的联网入侵检测系统的安全性和有效性。With the commercialization of 5G and the development of 6G,more and more Internet of things(IoT)devices are linked to the novel cyber-physical system(CPS)to support intelligent decision making.However,the highly decentralized and heterogeneous IoT devices face potential threats that may mislead the CPS.Traditional intrusion detection solutions cannot protect the privacy of IoT devices,and they have to deal with the single point of failure,which prevents these solutions from being deploying in IoT scenarios.The edge learning and game theory based intrusion detection for IoT was proposed.Firstly,an edge learning based intrusion detection framework was proposed to detect potential threats in IoT.Moreover,a multi-leader multi-follower game was employed to motivate trusted parameter servers and edge devices to participate in the edge learning process.Experiments and evaluations show the security and effectiveness of the proposed intrusion detection framework.

关 键 词：物联网 边缘学习 博弈论 入侵检测 

分 类 号：TN915.08[电子电信—通信与信息系统]