绝缘油热老化时间及糠醛含量的近红外光谱快速预测方法

Fast Prediction Method of Thermal Aging Time and Furfural Content of Insulating Oil Based on Near-Infrared Spectroscopy

实现变压器油纸绝缘热老化的准确评估是保证电力设备安全运行的重要内容。近红外光谱在石油化工等领域的成功应用,为电气绝缘检测提供了新思路。在130℃、真空箱中进行加速热老化试验,共制备14组老化时间不同的油纸绝缘样品,并利用近红外光谱仪采集绝缘油的光谱,利用液相色谱仪检测油中糠醛含量。原始谱图中, 8 373, 8 264, 7 181, 7 076, 6 981, 5 855, 5 799和5 678 cm^-1处存在明显的吸收峰,具体分析了各吸收峰的归属。采用五点三次多项式Savitzky-Golay卷积平滑算法预处理原始光谱。采用iPLS方法选取关于老化时间的特征谱区为11 209~10 364, 9 087~7 818和7 390~4 424 cm^-1,共1 320个波长点;同时,利用PCA提取该特征谱区的光谱信息,表明前7个主成分累计贡献率达99.78%。在上述基础上,建立了关于老化时间的PCR, PLSR, PCA-BP-ANN预测模型,表明采用共轭梯度算法的PCA-BP-ANN老化时间预测模型表现最优,其RMSEP为18.67,R2为0.997 3。采用iPLS方法选取关于油中糠醛含量的特征谱区为9 107~4 424 cm^-1,共1 210个波长点;同时,利用PCA提取该特征谱区的光谱信息,表明前4个主成分的累计贡献率达99.96%。在上述基础上,建立关于油中糠醛含量的PCR, PLSR和PCA-BP-ANN预测模型,表明采用共轭梯度算法的PCA-BP-ANN糠醛含量预测模型表现最优,其RMSEP为0.134 4,R2为0.987 7。基于绝缘油近红外光谱的老化时间和油中糠醛含量评估具有可行性。

Accurate assessment of transformer oil-paper insulating thermal aging serves as an important part to ensure the safe operation of power equipment.The successful application of Near Infrared Spectroscopy in petrochemicals and other fields provides new ideas for electrical insulation testing.The accelerated thermal aging test has experimented in a vacuum environment of 130℃.Fourteen groups of samples with different aging time are prepared.The spectrum of the aged insulating oil was collected by the Near Infrared Spectroscopy,and the furfural content in transformer oil was detected by high performance liquid chromatography(HPLC).There are obvious absorption peaks at 8373,8264,7181,7076,6981,5855,5799,and 5678 cm^-1 in the original spectrum.This study specifically analyzes the attribution of each absorption peak.The original spectrum was preprocessed using a five-point cubic polynomial Savitzky-Golay convolution smoothing algorithm.The characteristic spectral regions for aging time are selected as 11209~10364,9087~7818,7390~4424 cm^-1,with a total of 1320 wavelength points.At the same time,the spectral information of the characteristic region is extracted by PCA,which indicates that the cumulative contribution rate of the first seven principal components is 99.78%.On the basis of the above,a PCR,PLSR,PCA-BP-ANN prediction model for aging time was established.It is shown that the PCA-BP-ANN aging time prediction model with conjugate gradient algorithm is the best,with RMSEP of 18.67 and R 2 of 0.9973.The characteristic spectral region of the furfural content in the oil is selected from 9107 to 4424 cm-1 for a total of 1210 wavelength points.At the same time,the spectral information of the characteristic region is extracted by PCA,which indicates that the cumulative contribution rate of the first four principal components is 99.96%.On the basis of the above,a PCR,PLSR,PCA-BP-ANN prediction model for the content of furfural in oil was established.It is shown that the PCA-BP-ANN furfural content prediction model with conjugate gradient algorithm performs best,with RMSEP of 0.1344 and R 2 of 0.9877.It is feasible to evaluate the thermal aging time and the furfural content based on near-infrared spectroscopy of insulating oil.