Explainable Artificial Intelligence-Based Prediction of Power System Loading Conditions Under Diverse Operating Scenarios

Abstract

Accurate prediction of power system loading conditions is essential for ensuring reliable operation and effective decision-making in modern power networks. This study presents an explainable artificial intelligence (XAI)-based approach for predicting power system loading conditions under diverse operating scenarios using real operational data collected from the Egbin Power Station. The dataset was preprocessed through data cleaning and feature standardization to enhance data quality and model performance. Three machine learning models, namely Support Vector Machine (SVM), Gradient Boosting (GB), and Deep Neural Network (DNN), were developed to classify the loading conditions of four distribution feeders. The models were evaluated using accuracy, precision, recall, and Matthews correlation coefficient (MCC). The results show that the SVM model achieved high predictive performance with accuracies of 0.9963, 0.9863, 0.9838, and 0.985 for Feeders 4, 3, 2, and 1, respectively. However, the Gradient Boosting and Deep Neural Network models demonstrated superior performance, achieving perfect classification results with accuracy, precision, recall, and MCC values of 1 across all feeders. Furthermore, model interpretation techniques were applied to the Gradient Boosting and Deep Neural Network models to enhance transparency and explainability of the predictions. The results indicate that explainable ensemble and deep learning approaches can effectively capture complex nonlinear patterns in power system data while providing interpretable insights for operational decision-making. The proposed framework contributes to the development of reliable, transparent AI-driven tools for the intelligent monitoring and management of power system loading conditions.