Secure and Decentralized Homomorphic Federated Learning for Smart Microgrid Stability

Recently, the Internet-of-Things (IoT)-based smart microgrids are being integrated into the modern power systems, increasing the amount of highly sensitive information exchanged between suppliers and consumers, and its lack of data confidentiality may pose economic damage and threaten the national security. This paper presents a novel decentralized Homomorphic Encrypted Federated Learning (HEFL) framework that solves the security and stability problem of the modern power system. In detail, our HEFL framework enables the microgrids to deploy the encrypted models without decrypting and exposes to the core in server-side aggregation, mitigating security concerns, reducing the network overhead, computation time, and memory consumption. Meanwhile, a verifiable proof mechanism is also designed to ensure the manipulation of the model updates. The experimental results demonstrate that the HEFL method could improve the accuracy of stability prediction by up to 15%, and can reduce the network overhead by 20% along with the computation time by 25%, and can provide a strong solution to ensure the stability and security of the future smart grids.