关键词: 电力数据, Paillier同态加密算法, 联邦学习, LSTM神经网络算法

Abstract: The accurate and rapid prediction of power data is not only crucial to the stability and  regular operation of the power system but also has a significant impact on the production and life of the entire society. Therefore, efficient and accurate prediction of power data is an essential work in power data research. Recurrent neural networks have excellent performance in power data prediction problems, but require a large amount of data to train the model. Due to privacy concerns, power companies are reluctant to share their electricity data, making it impossible to train more accurate models. In addition, colossal network resource overhead is incurred while uploading massive amounts of data to a central server to train the federated model. To address these problems, this paper combines federal learning with paillier homomorphic encryption algorithm and proposes a local protection method for power data prediction models based on federal learning and homomorphic encryption. It implements the protection of power data and local model parameters, and cotrains the joint model in a secure manner. We conducted experiments using accurate electricity data, and this method achieved good experimental results.

Key words: power data, Paillier homomorphic encryption algorithm, federated learning, LSTM neural network algorithm