重庆邮电大学学报（自然科学版）2023，Vol.35Issue(6)：1020-1027,8.DOI:10.3979/j.issn.1673-825X.202211020307
面向抽水蓄能电站智能巡检系统的联邦学习隐私保护方法
Privacy-preserving federated learning method for intelligent inspection system in pumped storage power station
摘要
Abstract
To enhance the privacy protection of the federated learning architecture for intelligent inspection system of pumped storage power station,this paper proposes a privacy-preserving federated learning method that can provide high-per-formance differential privacy protection for local model parameters on the server side and during transmission.First,to solve the privacy leakage problem caused by the malicious server matching uploaded parameters and inspection device identities,a random response-based inspection device selection mechanism is designed,which makes the server unable to reason out the identity information of the uploaders based on its actual selection.Second,to solve the privacy leakage problem during the transmission,an adaptive differential privacy technology is developed,which adaptively adds noise to each dimension of the gradient according to its characteristics,thereby reducing model parameter accuracy loss and accelerating convergence with differential privacy protection.Security analyses and simulation results show that this method can effectively avoid the privacy leakage of model parameters and outperforms in model accuracy and training communication overhead.关键词
抽水蓄能电站/智能巡检系统/联邦学习/随机响应/差分隐私Key words
pumped storage power station/intelligent inspection system/federated learning/stochastic response/differenti-al privacy分类
信息技术与安全科学引用本文复制引用
黄建德,何秋,宗悦,王斌,罗远林,吴鹏浩,于尧,郭磊..面向抽水蓄能电站智能巡检系统的联邦学习隐私保护方法[J].重庆邮电大学学报（自然科学版）,2023,35(6):1020-1027,8.基金项目
国家自然科学基金项目(62171113)The National Natural Science Foundation of China(62171113) （62171113）
