An Input-guided Physics-Informed Neural Network Method for Wind Power Estimation and Coordinated FDIAs Detection in Renewable Energy System

As the IoT-enabled consumer electronic devices and wind power become increasingly integrated into renewable energy system, corresponding security threats like false data injection attacks (FDIAs) are rising. This paper investigates a hidden security vulnerability in renewable energy system, which is coordinated FDIAs exploiting the physical relationship between wind speed and wind power to craft stealthy and disruptive attack vectors. To counter this threat, an input-guided physics-informed neural network (IGPINN)-based wind power estimation model is proposed for real-time monitoring of wind power, which incorporates theoretical wind power as an input to get a good balance between estimation accuracy and interpretability. Building on this model, a detection method is developed by combining IGPINN with support vector machines (SVM), leveraging the physical relationship between wind speed and wind power to construct feature vectors for coordinated FDIA detection. Based on a modified IEEE 39-bus system, case studies show that the proposed wind power estimation model outperforms existing methods in accuracy and interpretability, and the IGPINN-SVM detection method achieves greater performance against coordinated FDIAs than several existing FDIA detection methods.