摘要： 大数据时代，数据安全性和隐私性受到越来越多的关注和重视。联邦学习被视为是一种隐私保护的可行技术，允许从去中心化的数据中训练深度模型。针对电力投资系统中各部门因担心数据隐私信息泄露而带来的数据孤岛和隐私保护问题，提出了一种隐私保护的联邦学习框架，允许各部门自有数据在不出本地的情况下，联合训练模型。首先，提出了联邦学习的架构，支持分布式地训练模型；其次，引入同态加密技术，提出了隐私保护的联邦平均学习流程，在数据隐私保护的情况下，实现联合训练模型；最后，实验结果表明，该框架具有较好的收敛性，而且联合训练得到的模型具有较好的精度。

中图分类号： TN711.1；TP311
文献标识码： A
DOI：10.16157/j.issn.0258-7998.211828
中文引用格式： 杨东宁，谢潇睿，吉志坤，等. 一种隐私保护的联邦学习框架[J].电子技术应用，2022，48(5)：94-97，103.
英文引用格式： Yang Dongning，Xie Xiaorui，Ji Zhikun，et al. A privacy-preserving federated learning framework[J]. Application of Electronic Technique，2022，48(5)：94-97，103.

A privacy-preserving federated learning framework

Abstract： In the era of big data, more and more attention has been paid to data security and privacy. Federated learning is regarded as a promising privacy-preserving technology, which allows training a deep model from decentralized data. To solve the problem of isolated data island and privacy protection caused by the fear of data privacy information leakage in the power investment system，this paper proposes a privacy-preserving federal learning framework, which allows departments to jointly train the model without releasing their local data. Firstly, a federated learning architecture is proposed to support distributed training model. Secondly, homomorphic encryption technology is introduced and a federal average learning process is proposed to realize the joint training model while the privacy of data are protected. Finally, the experimental results show that the framework has good convergence and the joint training model has good accuracy.

Key words : data privacy；federated learning；deep learning；homomorphic encryption；convolutional neural network