EFedDSA: An Efficient Differential Privacy-Based Horizontal Federated Learning Approach for Smart Grid Dynamic Security Assessment

Abstract

Enhanced by machine learning (ML) techniques, data-driven dynamic security assessment (DSA) in smart cyber-physical grids has attracted significant research interest in recent years. However, the current centralized ML architectures have limited scalability, are vulnerable to privacy exposure, and are costly to manage. To resolve these limitations, we propose a novel effective and secure distributed DSA method based on horizontal federated learning (HFL) and differential privacy (DP), namely EFedDSA. It leverages local system operating data to predict and estimate the system stability status and optimize the power systems in a decentralized fashion. In order to preserve the privacy of the distributed DSA operating data, EFedDSA incorporates Gaussian mechanism into DP. To reduce the computational burden from multiple transmission communication rounds, a discounting method for the total communication round is proposed to reduce the total transmission rounds. Theoretical analysis on the Gaussian mechanism of EFedDSA provides formal DP guarantees. Extensive experiments conducted on the New England 10-machine 39-bus testing system and the synthetic Illinois 49-machine 200-bus testing system demonstrate that the proposed EFedDSA method can achieve advantageous DSA performance with fewer communication rounds, while protecting the privacy of the local model information compared to the state of the art.