A Privacy-preserving Users' Power Load Prediction Method Based on Federated Learning

The new power system deeply integrates new digital technologies such as cloud computing, big data, Internet of Things, mobile Internet, artificial intelligence, etc., which enables different levels of the electricity grid to develop cooperatively and improves the ability to allocate power grid resources. Particularly, the smart meters located on the electricity consumers' side play an important role in collecting and aggregating users' data. Based on the collected data, users' power load data can be analyzed, which can efficiently allocate power resources and avoid resource waste caused by the limitation of power resource storage. However, the data collected by smart meters involves a large amount of users' private information. How to efficiently and legally use these data for joint data mining without violating users' privacy has become a challenge. In this paper, a privacy-preserving users' power load prediction method based on federated learning technology and RSA-AES encryption system is proposed. Experimental results show that our method can predict users' power load with high accuracy while protecting users' privacy.