Power Data Federated Learning Privacy Policy Based on Optimized Homomorphic Encryption

This is the authors' abstract. We don't add key points for this paper.

With the advent of big data in electricity power, concerns about privacy and security have led most power stations to avoid sharing their electricity data, resulting in “data islands.” Federated learning emerged to address this, but it still poses privacy risks during training. Balancing privacy protection and data utilization is a current challenge. Federated learning involves exchanging intermediate results, and sending plaintext results can risk privacy breaches. Encryption during training increases computational and communication overhead, affecting efficiency, and is challenging in resource-limited environments. To address these issues, a federated learning model combining differential privacy with optimized Paillier homomorphic encryption has been proposed, ensuring data protection and improving training accuracy.