面向物联网安全的域映射增强与对抗增益聚合算法

打开文本图片集

关键词：联邦学习；物联网安全；数据异构；特征增强；权重聚合

中图分类号：TP181;TP309 文献标志码：A 文章编号：1001-3695（2025）12-020-3683-08

doi：10.19734/j.issn.1001-3695.2025.05.0119

Domain mapping enhancement and adversarial gain aggregation algorithm for IoT security

Gu Yue1²，Liu Miao1，2†，Sun Zhenxing³ （1.SchoolofElectroics&InfoationEngnering，Nnjing UniersityfIfoationSciece&Teology，NjingChna;2. Schodoflectcsdtige，uiUesityuiJsuaiadchfe University，QinhuangdaoHebei O66O44，China）

Abstract：Toensuredata securityandexpandtheapplicationof federatedlearning inheterogeneous InternetofThings（IoT） environments，thispaperproposedafederatedlearningframeworkthatintegrateddomainmappingenancementandadversarial gainagregation mechanisms.The framework used domain mappng enhancement to improve feature extraction capabilityand increase featureseparability，therebyimprovingmodelperformance.Italsoapliedanadversarialgainaggegationmethodto dynamicallyadjustclient weights during aggregation，reducing performance degradationof the globalmodel caused byheterogeneity.Compared with FedGen，the frameworkachieveda6.64 percentage pointimprovement in accuracy.The experimental resultsshowthattheproposedframework efectivelymitigatestheimpactofheterogeneity，performswellacrossfourexperimental scenarios，andsignificantlyimprovestheconvergencestabilityandoverallaccuracyofmodelsinIoTapplications.

y words：federated learning；Internet of Things security；data heterogeneity；feature enhancement；weight aggrega

0 引言

物联网（IoT）的快速发展产生了海量数据，为人工智能的发展提供了丰富的资源。（剩余21233字）