Privacy-Preserving Anomaly Detection in Wind Turbines using Federated Learning

To combat climate change and meet global sustainability goals, the rapid adoption of wind energy is essential. However, operational challenges with wind turbines, such as unexpected power outages or equipment failures, can lead to safety issues, financial losses, and disruptions in energy production. Traditional centralized monitoring systems, which rely on sending sensitive data to central servers, face concerns related to privacy, latency, and scalability. This study presents a solution to these challenges by using Federated Learning (FL) to develop a decentralized, privacy-preserving anomaly detection system for wind turbines. The proposed approach enables real-time collaboration across dispersed remote terminal units (RTUs) without sharing raw data, ensuring data security while detecting anomalies. The system utilizes advanced machine learning (ML) techniques, including neural networks (NN), random forests (RF), decision trees (DT), and K-nearest neighbors (KNN), to achieve high performance. These ML models had been implemented using python. FL outperformed other approaches in terms of accuracy (0.986), precision (0.979), recall (0.987), and F1-score (0.983), reducing downtime, maximizing energy output, and handling data heterogeneity effectively. The considered attacks are Man-in-the-Middle (MITM), and phishing. The results high-light the system’s potential to enhance wind turbine operations, improving efficiency and reliability while ensuring data privacy.