Federated Learning for Detecting Cyber Attacks in EVCS Using a Lightweight Neural Network

This research presents a federated learning (FL) framework and employs a lightweight Simple Neural Network (SimpleNN) model to identify cyber-attacks in electric vehicle charging stations (EVCS). Federated learning is instrumental in this scenario because it protects data privacy by storing local data on edge devices while allowing collaborative model training across scattered EVCS. The proposed technique is tested on the IEEE 123-bus system, which has four EVCS dispersed over various buses. Key characteristics such as voltage, frequency, state of charge (SoC), and power are collected and utilized to train the model. The SimpleNN was chosen for its computational efficiency and minimal resource needs, which fit well with the dispersed and resource-constrained nature of FL settings. The findings show a high detection accuracy of almost 95%, demonstrating the FL framework's efficacy in identifying abnormalities. The proposed method uses federated learning to increase anomaly detection performance while ensuring data preservation and reducing communication overhead.