Abstract
The growth of wind power generation is accelerating worldwide. However, wind turbines need to be carefully controlled to maintain structural stability and optimal power generation. The state of the turbine and power generated is digitized and transmitted over communication networks for physical component monitoring, control, and data management. Motivated by creating power outages, financial loss, and physical wind turbine damage, attackers could craft and inject falsified network traffic to maliciously influence control actions. In this paper, we propose a physics-informed machine learning anomaly detection approach. The mechanism is utilized at each wind turbine, where generated power signals are validated based on laws of physics and incorporated into machine learning algorithm to detect anomalies. We show that the proposed combined physics and machine learning approach improves the accuracy of detecting falsification attacks compared with using machine learning alone.