The accuracy of predicting the degree of polymerization (DP) of insulating paper using Raman and FTIR fusion spectroscopy is affected by the initial moisture content. To address this issue, this paper constructs a robust spectral analysis framework. This method integrates Raman and infrared spectra, adopts an improved mRMR algorithm with an implicit moisture penalty term to achieve anti-interference feature extraction, and conducts efficient prediction based on the Matern 5/2 Gaussian process regression model optimized by L-BFGS. The R2 of M52-GPR optimized by L-BFGS reached 99.01%. The root mean square error (RMSE) and mean absolute error (MAE) decreased to 15.14 and 11.36 respectively. This paper verified the superiority of the mRMR algorithm in spectral feature selection in water-containing environments, providing a high-precision regression analysis method for the aging assessment of insulating materials.
26 May 2026 in IEEE Transactions on Dielectrics and Electrical Insulation
The equilibrium behavior of methanol (MeOH) between the oil and paper phases plays a critical role in evaluating the aging state of oil–paper insulation. Existing studies generally assume a constant oil–paper equilibrium coefficient for MeOH; however, this assumption neglects the effects of insulation paper aging on its microstructure and chemical properties. In this study, eight insulation paper samples with different aging degrees were prepared to investigate the variation in MeOH equilibrium behavior. Experimental results show that aging significantly alters the paper microstructure and reduces hydroxyl group content, leading to a substantial reduction in MeOH adsorption capacity. Specifically, the adsorption capacity of aged paper decreases by up to 90.22% compared with that of new paper. Based on these findings, a variable equilibrium coefficient model dependent on the degree of polymerization (DP) is proposed. Furthermore, thermal aging experiments demonstrate that when the DP decreases to approximately 460, the total MeOH concentration declines due to esterification reactions. These results demonstrate the proposed model and provide improved support for accurate assessment of oil–paper insulation aging.