Research On Quantitative Detection Of Furfural Content In Transformer Oil Based On Near Infrared Spectroscopy

The oil-immersed transformer is the core equipment to ensure the smooth and safe operation of the power grid.The oil-paper insulation system is mainly composed of insulating oil and insulating paper.During long-term operation,it will gradually age due to multiple factors such as temperature,moisture,and oxygen.As a result,its insulation performance is reduced,which may eventually lead to electrical accidents.Achieving accurate assessment of the aging characteristics of transformer oil-paper insulation is an important part of ensuring the safe operation of power equipment.At present,the application of near-infrared spectroscopy in agricultural production,petrochemical and other fields has gradually become mature;the role of near-infrared spectroscopy in the field of power detection has also gradually been paid attention to.In this paper,a single-factor accelerated thermal aging experiment was carried out in a vacuum-sealed aging test box at a constant temperature of 130°C,and oil-paper insulation thermal aging samples with different aging times were prepared.Using high performance liquid chromatography,the furfural content in the oil was measured;and the near-infrared spectra of all insulating oil samples were collected by the transmission method.Through the density functional simulation calculation and the energy level formula of anharmonic vibration,the assignment of the absorption peak of furfural and the assignment of the near-infrared spectral absorption peak of insulating oil are studied.Based on the near-infrared spectroscopy of insulating oil,a full-spectrum quantitative analysis model of furfural content in oil was established.Using the near-infrared absorption peak assignment of insulating oil and the iPLS algorithm,respectively,the quantitative analysis model of the full spectrum area was optimized,and the characteristic wavelength variables related to the furfural content in the oil were screened out.Based on the characteristic wavelength variables,the PCA-BP model is established.The specific research results are as follows.（1）Based on the B3LYP hybrid functional theory of density functional theory and the anharmonic vibrational energy level formula,the absorption peaks of furfural in the near-infrared spectrum are calculated,and the attribution of the absorption peaks in the near-infrared spectrum of insulating oil is further studied.Using the 6-311G（d,p）basis set in the Gaussian 09 program,the vibration frequency and IR spectrum of furfural were calculated.Among them,the absorption peak of stretching vibration of C=O is1760 cm^-1;the absorption peak of stretching vibration of C-H group sharing carbon atom with C=O is 2903 cm^-1,and the absorption peak of bending vibration is 1396 cm^-1.Combining the energy level formula of anharmonic vibration,it is calculated that:the combined frequency of C=O stretching vibration and the double frequency of the in-plane bending vibration of CH is around 4590 cm^-1;C=O stretching vibration and CH stretching vibration The combined frequency of is around 4660 cm^-1;the double frequency and triple frequency of C=O stretching vibration are around 5088 cm^-1 and6690 cm^-1 respectively.Finally,analyze the first combined frequency zone,O-H and C=O zone,the first frequency doubled zone,the second combined frequency zone,the second doubled frequency zone,the third combined frequency zone,and the third doubled zone of the insulating oil near infrared spectrum.The attribution of each absorption peak in the frequency region.（2）Based on the near-infrared spectroscopy of insulating oil,a full-spectrum quantitative analysis model of furfural content in oil was established.PCA-MD was used to identify and eliminate 5 abnormal samples with high leverage among 145insulating oil samples.Using PCA-K-S,105 samples were selected to form the calibration set,and 35 samples were used to form the verification set.The normalization and first-order differential preprocessing with a difference width of 3 are performed on the near-infrared spectrum of insulating oil.Using the PLS algorithm,a quantitative analysis model for the full spectrum of the near-infrared spectrum of the insulating oil was established.The RMSECV was 0.3228,the RMSEP was 0.3347,and the corresponding R² was 0.92 and 0.91,respectively.（3）Based on the research on the attribution of the near-infrared spectrum absorption peaks of insulating oil,the full spectrum area quantitative analysis model is optimized,and three characteristic spectrum areas for the quantitative analysis of furfural content in oil are selected:C=O and C-H combined frequency(4710-4540 cm^-1),C-H first-level multiplication area(6000-5400 cm^-1),second-level multiplication area(8800-8100 cm^-1).Studying the pros and cons of the three characteristic spectral regions,it is concluded that the combined spectral regions of 6000-5400 cm^-1 and 8800-8100cm^-1 have the best modeling performance.The R² and RMSEP of the NIR-PLSR model for quantitative analysis of furfural content in oil established based on the combined spectrum are 0.98 and 0.1888,respectively.Compared with the prediction results of the full-spectrum quantitative analysis model,the prediction results are greatly improved.（4）Using iPLS to optimize the quantitative analysis model of the full spectrum area,the characteristic spectrum area is 7390-6970 cm^-1.Based on this characteristic spectrum area,the established iPLS-PLSR model for quantitative analysis of furfural content in oil has R²=0.98 and RMSEP=0.692.Finally,the iPLS-PCA-BP model was established with the characteristic spectrum area of 7390-6970 cm^-1.When using 3principal components and conjugate gradient learning algorithm,the iPLS-PCA-BP model is the best,and its R² and RMSEP are 0.1466 respectively.The deviations of the predicted values of furfural content in the oil of the optimal model were all within±0.2mg/L.