Intelligent Power System Condition Detection Using Optimized XGBoost and SHAP Explainability

Intelligent Power System Condition Detection Using Optimized XGBoost and SHAP Explainability

Forogh Pakrai1

1) Department of electrical engineering, Om.C., Islamic azad university, Omidiyeh, Iran,

IOI Code : AAHD-ABKCAA
Publication : National Conference on Information Technology, Nanotechnology, Artificial Intelligence and Technological Futures Studies(itnaf.ir)
Abstract :
    The rapid development of modern power systems has significantly increased the complexity of monitoring, protection, and fault diagnosis processes. Traditional fault detection approaches are often limited in handling nonlinear relationships and large-scale electrical datasets. Consequently, machine learning techniques have emerged as powerful tools for intelligent condition monitoring and classification in electrical engineering applications. This study proposes an intelligent classification framework based on an optimized Extreme Gradient Boosting (XGBoost) algorithm for identifying power system operating conditions using three-phase electrical measurements. The input dataset consists of six electrical features, including three-phase currents (Ia, Ib, Ic) and three-phase voltages (Va, Vb, Vc). After data preprocessing and feature scaling, three machine learning models, namely XGBoost, Random Forest, and Multi-Layer Perceptron (MLP), were developed and evaluated. To improve the predictive performance of the proposed model, the XGBoost hyperparameters were optimized using the Randomized Search Cross-Validation (RandomizedSearchCV) technique. Experimental results demonstrated that the optimized XGBoost model achieved an accuracy of 99.63%, outperforming or matching the performance of the benchmark classifiers. Furthermore, Explainable Artificial Intelligence (XAI) was incorporated through SHapley Additive exPlanations (SHAP) to interpret the contribution of each electrical feature to the final classification. SHAP analysis revealed that the phase-A current (Ia) was the most influential feature, followed by phase-B current (Ib), indicating that current measurements play a more significant role than voltage measurements in the proposed classification framework. The obtained results demonstrate that integrating optimized ensemble learning with explainable artificial intelligence provides a reliable, accurate, and interpretable solution for intelligent power system condition classification.

Keywords : xgboost machine learning shap explainable artificial intelligence random forest multi layer perceptron (mlp) power system monitoring condition detection