基于联邦学习的网络多敏感属性数据并行加密  

Parallel encryption of network multi sensitive attribute data based on federated learning

作　　者：王晨飞 赵文华 安业腾 朱青 马建勋 WANG Chenfei;ZHAO Wenhua;AN Yeteng;ZHU Qing;MA Jianxun(State Grid Customer Service Center,Tianjin 300300,China;State Grid Siji Testing Technology(Beijing)Co.,Ltd.,Beijing 100192,China)

机构地区：[1]国家电网有限公司客户服务中心,天津300300 [2]国网思极检测技术(北京)有限公司,北京100192

出　　处：《电子设计工程》2024年第22期166-169,174,共5页Electronic Design Engineering

基　　金：天津市电力公司科技项目(KJ08-2-9)。

摘　　要：网络多敏感属性数据量较大,且噪声数据占比较高,导致数据加密效果不佳。因此,提出基于联邦学习的网络多敏感属性数据并行加密方法。在去中心化联邦架构下,构造公钥密钥和私钥密钥。选取两个维度数据构成网络多敏感属性数据集合,将多敏感属性数据作为子密文对数据进行训练,并将输入的随机误差向量作为基础密文。构建基于联邦学习的并行加密框架,结合重加密方法完成多敏感属性数据的并行加密。实验结果表明,该方法加密过程中的最大平均绝对误差仅达到0.2,能够有效剔除全部噪声数据,并保证数据不被泄露,加密效果好。The network has a large amount of sensitive attribute data and a high proportion of noisy data,resulting in poor data encryption performance.Therefore,a parallel encryption method for network multi sensitive attribute data based on federated learning is proposed.Under the decentralized federated architecture,construct public key and private key keys.Select two dimensions of data to form a network multi sensitive attribute data set,train the multi sensitive attribute data as sub ciphertext,and use the input random error vector as the basic ciphertext.Build a parallel encryption framework based on Federated learning,and complete parallel encryption of multi sensitive attribute data with re encryption method.The experimental results show that the maximum mean absolute error in the encryption process of this method is only 0.2,which can effectively eliminate all noise data and ensure that the data is not leaked,and the encryption effect is good.

关 键 词：联邦学习 多敏感 属性数据 并行加密 

分 类 号：TN99[电子电信—信号与信息处理]