Prediction of Degree of Polymerization of Insulating Paper in Transformer Hotspot Region Based on Dual Chemical Markers

Abstract

Insulating paper in transformer winding hotspots represents the most vulnerable component of solid insulation systems, with its aging condition being critical for fault prediction. As key chemical markers for assessing the degree of polymerization (DP) of insulating paper, furfural and methanol concentrations (MCs) were measured through nonuniform thermal aging experiments. To address the limitations of conventional DP prediction models in capturing dynamic aging characteristics and achieving stable model optimization, a DP prediction framework is proposed by integrating dynamic dual chemical markers with an extreme gradient boosting (XGBoost) model optimized via a multistrategy enhanced escape (MSESC) algorithm. The concentrations and change rates of furfural and methanol, together with winding temperature, are used as model inputs, while the proposed MSESC algorithm improves global search stability and convergence efficiency in hyperparameter optimization. Experimental validation on 110 samples demonstrated the proposed model’s superiority, achieving a determination coefficient (R ${}^{{2}}\text {)}$ of 0.99. SHAP analysis further identified MC, furfural concentration, and winding temperature as dominant factors influencing DP. The MSESC-XGBoost model significantly advances DP prediction accuracy and provides robust support for transformer insulation condition assessment.