Federated Learning for Personalized Recommendation in Securing Power Traces in Smart Grid Systems

With the proliferation of smart sensors and communication technologies, analyzing power-related data has become increasingly popular in smart grid systems, providing insights into optimal power usage strategies. However, power-related data is often stored and owned by different parties, creating challenges for direct data sharing due to privacy, security, and public safety concerns. In this paper, we propose a novel federated learning framework with personalized recommendations for smart grids that enables collaborative learning of power usage patterns without exposing individual power traces. The proposed framework includes both horizontal and vertical federated learning, which respectively addresses scenarios where data is distributed across the sample space and the feature space. We utilize encoding schemes such as Parlier encoding to ensure lossless and privacy-preserving AI model construction. The proposed framework has promising applications in various aspects of the smart grid, including distributed generation and consumption, electric vehicles, and integrated energy systems. The experimental results demonstrate the effectiveness of our proposed framework in preserving privacy while achieving accurate power usage prediction.