A Federated Learning Framework for Detecting False Data Injection Attacks in Solar Farms
Liang Zhao, Jiaming Li, Qi Li, Fangyu Li
Abstract
Smart grids face more cyber threats than before with the integration of photovoltaic (PV) systems. Data-driven-based machine learning (ML) methods have been verified to be effective in detecting attacks in power electronics devices. However, standard ML solution requires centralized data collection and processing, which is becoming infeasible in more and more applications due to efficiency issues and increasing data privacy concerns. In this letter, we propose a novel decentralized ML framework for detecting false data injection (FDI) attacks on solar PV dc/dc and dc/ac converters. The proposed paradigm incorporates the emerging technology named federated learning (FL) that enables collaboratively training across devices without sharing raw data. To the best of our knowledge, this work is the first application of FL for power electronics in the literature. Extensive experimental results demonstrate that our approach can provide efficient FDI attack detection for PV systems and is aligned with the trend of critical data privacy regulations.