Distributed detection and mitigation of FDIAs in smart grids via federated learning

Employment of smart meters in power grids avails efficient data analytics and control over the system. However, the transmission of the measurement data over the communication networks may expose the power system to potential cyberattacks. Among these, false data injection attacks (FDIAs) pose a significant threat to the operation of smart grids. In order to tackle the cyberattacks on smart grids, we propose a federated learning-based method for distributed detection and mitigation of FDIAs. Federated learning facilitates distributed training of machine learning-based attack detectors while preserving privacy of sensitive data. The proposed detection method incorporates a graph autoencoder model that exploits the spatial correlations between the power load profiles of the connected network nodes to efficiently mitigate the effects of FDIAs. Extensive simulations using realistic power load profiles combined with the IEEE-57, 118, and 300 bus test cases corroborate the effectiveness of the proposed approach. • A federated learning-based approach detects and mitigates false data injection attacks (FDIAs). • Graph autoencoder captures spatial correlations in power grid data for enhanced accuracy. • The method ensures data privacy while enabling collaborative cybersecurity in smart grids. • Extensive simulations validate superior detection and mitigation performance over benchmarks. • Tested on IEEE 57, 118, and 300 bus systems using realistic power load profiles.