Explainable Artificial Intelligence for Real-Time Stability Assessment in Smart Power Systems
Abstract
INTRODUCTION: Smart grid stability analysis is necessary for reliable power system operations in the presence of increased renewable energy penetration and distributed energy resources.
Objectives
This study proposes a novel Graph Transformer–Temporal Fusion Transformer (GT-TFT) with an Explainable Artificial Intelligence (XAI) framework for real-time smart grid stability prediction.
Methods
A feature-correlation graph is constructed and processed using a Graph Transformer (GT) to capture spatial dependencies, while a Temporal Fusion Transformer (TFT) learns temporal dynamics from smart grid measurements. The extracted features are fused through a feature fusion layer and classified using a neural network classifier. SHapley Additive exPlanations (SHAP) is employed to improve model interpretability.
Results
Experimental evaluation on the Smart Grid Stability dataset demonstrates superior performance, achieving an accuracy of 0.9879, precision of 0.9886, recall of 0.9926, F1-score of 0.9905, and AUC of 0.9999. Comparative analysis confirms that the proposed framework outperforms existing baseline models.
Conclusion
The suggested GT-TF-XAI framework offers a solution to real-time smart grid stability assessment that is accurate, reliable, and explainable to enable effective decision-making in current smart power systems.
