电讯技术2025，Vol.65Issue(7)：1050-1059,10.DOI:10.20079/j.issn.1001-893x.250122005
面向智能电网的可验证隐私保护联邦学习方法
A Verifiable Privacy-preserving Federated Learning Method for Smart Grids
摘要
Abstract
The smart grid upgrades the traditional power system into a more efficient,reliable,and sustainable one by integrating advanced information and communication technologies.However,centralized data processing models face challenges related to privacy breaches and data security is designed.Therefore,a verifiable privacy-preserving federated learning method is proposed for the smart grid and a secure data aggregation scheme based on secret sharing and its homomorphic properties is designed.Before uploading gradient information,users locally add masks to their gradients and share the mask values with other users through secret sharing.Upon receiving the privacy-processed gradients,the server utilizes the homomorphic properties to recover the sum of the masks,ensuring secure aggregation and reducing communication overhead.Additionally,during data transmission,data integrity verification is conducted based on verifiable secret sharing and authenticated encryption techniques to ensure the authenticity and completeness of data transmitted between the client and the server.Simulation results demonstrate that this scheme maintains superior model performance and lower communication costs while ensuring user privacy and data integrity,achieving model accuracies of 99.2%and 99.5%on the MNIST and CIFAR datasets,respectively.关键词
智能电网/隐私保护/联邦学习/可验证秘密共享/数据聚合Key words
smart grid/privacy protection/federated learning/verifiable secret sharing/data aggregation分类
信息技术与安全科学引用本文复制引用
艾徐华,银源,董贇,谭期文,韦宗慧,黄依婷..面向智能电网的可验证隐私保护联邦学习方法[J].电讯技术,2025,65(7):1050-1059,10.基金项目
广西电网公司科技项目(046100KC23040002) （046100KC23040002）
