Federated learning-based cyber attack detection on Electric Vehicles (EVs) in AC/DC hybrid distribution systems

Abstract As increasingly more electric vehicles (EVs) are plugged into modern-day power grids, the integration adds significant cybersecurity risks, particularly within AC/DC hybrid distribution grids with bidirectional power flow and complex control frameworks. Cyber attacks on EVs, such as false data injection (FDI) attacks, may lead to denial-of-charging (DoC), overcharging, or grid instability. To address this challenge, in this paper we propose a new federated learning-based cyber attack detection framework for AC/DC hybrid distribution networks. In contrast to centralized solutions, our solution safeguards data privacy through local detection model training on edge nodes that are distributed without releasing raw data. The constructed framework utilizes real-time voltage, current, and state-of-charge (SOC) measurements to identify anomaly patterns indicative of cyber intrusions. Through micro-PMUs and edge computing, the system detects malicious manipulations of charging profiles and allows for variations in grid conditions. Decentralized and intelligent security implementations are highlighted through the research to be of important value in safely and dependably integrating EVs into future power grid systems. Simulation results demonstrate that the proposed approach achieves high detection accuracy (97.8%) with low false alarm rate (2.5%) and outperforms traditional methods such as SVM and Random Forest under various noise levels.