Physics-Informed Learning for Predicting Transient Voltage Angles in Renewable Power Systems Under Gusty Conditions

As renewable energy penetration and extreme weather events increase, accurately predicting power system behavior is essential for reducing risks and enabling timely interventions. This study presents a physics-informed learning approach to forecast transient voltage angles in power systems with integrated wind energy under gusty wind conditions. We developed a simulation framework that generates wind power profiles with significant gust-induced variations over a one-minute period. We evaluated the effectiveness of physics-informed neural networks (PINNs) by integrating them with LSTM (long short-term memory) and GRU (gated recurrent unit) architectures and compared their performance to standard LSTM and GRU models trained using only mean squared error (MSE) loss. The models were tested under three wind energy penetration scenarios—20%, 40%, and 60%. Results show that the predictive accuracy of PINN-based models improves as wind penetration increases, and the best-performing model varies depending on the penetration level. Overall, this study highlights the value of physics-informed learning for dynamic prediction under extreme weather conditions and provides practical guidance for selecting appropriate models based on renewable energy integration levels.