Physics-guided attention network for smart grid stability prediction

Background Smart grids are experiencing a great deal of instability because the number of decentralized energy sources and the diverse impacts that renewable energy has on grid operations are rising. In response, Decentralized Smart Grid Control (DSGC) frameworks have been created to improve grid resilience through adaptive demand-supply coordination; nevertheless, predicting grid stability is challenging because of the nonlinear interactions between all of the system parameters. Problem statement and research gap Although existing machine learning and deep learning approaches have reached a high degree of accuracy, almost all are purely data-driven and provide limited consideration of physical consistency, interpretability, robustness, and comprehensive statistical validation. Proposed methodology To overcome these limitations, this study proposes the Physics-Guided Attention Network (PGAN) for predicting the stability of smart grids. PGAN uses feature attention learning, residual deep interaction modelling, and DSGC-guided smoothness regularization to learn stable decision boundaries from physically meaningful input variables. Materials The experimental data is the DSGC stability dataset containing 60,000 samples, evaluated based on the testing metrics of accuracy, precision, recall, F1 score, Receiver Operating Characteristics area under the curve (ROC-AUC), and cross-validation. Results PGAN achieved 98.98% classification accuracy on the DSGC dataset, exceeding the performance of many transformer-based network variations. Explanation analysis using Shapley Additive Explanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) indicates that reaction time constants and price elasticity coefficients are the leading parameters associated with stability prediction. Conclusion These results demonstrate that PGAN provides an accurate, interpretable, and computationally efficient framework for smart grid stability monitoring, supporting reliable operation and sustainable integration of renewable energy resources.