Enhanced Health Index Prediction of Power Transformers Using Optimized XGBoost and Oil Diagnostic Data

Abstract

Reliable condition assessment of power transformers is crucial for maintaining the stability and efficiency of electrical power systems. This study presents an enhanced data-driven approach for predicting the Transformer Health Index (HI) using optimized Extreme Gradient Boosting (XGBoost) and comprehensive
oil diagnostic parameters, including dissolved gases, moisture, DBDS concentration, and interfacial voltage. A dataset comprising 470 laboratory records was preprocessed through outlier removal, normalization, and feature selection to ensure model robustness. The optimized XGBoost model achieved a high coefficient of determination (R² = 0.876), a mean squared error (MSE = 6.91), and a mean absolute error (MAE = 1.21), outperforming other machine learning baselines. The results confirm the effectiveness of XGBoost in accurately modeling transformer health and provide a reliable foundation for predictive maintenance and asset management in power transformer fleets.