Study On The Raman Spectral Feature Extraction And Diagnosis Of Oil-paper Insulation Aging

Safe and reliable operation of electrical equipments is the first line of defense against major accidents in power systems,Oil-paper insulation equipment is an important part of power system.A timely and accurate aging diagnosis,especially for old equipment running for more than fifteen years,is one of the key factors to ensure the safe production of electric network and to realize the efficient maintenance of equipment.In the aging process of oil-paper insulation equipment,insulating oil and paper will be decomposed and produce a variety of characteristics,such as furfural,acetone,methanol,CO,CO₂,and dissolved in oil.Which makes insulating oil contains a large amount of aging information about oil-paper insulation.At present,the methods of sampling and detection for aging diagnosis are complex,and most of them depend on a single feature.Laser Raman spectroscopy has advantages in direct detection of mixture systems and state diagnosis.In order to solve many problems in the application of Raman spectroscopy in the aging diagnosis of oil-paper insulation equipment,and lay the foundation for the rapid diagnosis of oil paper insulation equipment in the field of aging.Relying on the project of national major scientific instruments and equipment development.This paper carried out a research on extraction of spectral features and diagnosis of oil-paper insulation.The main research contents and achievements are as follows:?1?A representative large number of oil paper insulation samples in different aging states were prepared by thermal accelerated aging process in laboratory.The change and connection of aging characteristics?polymerization degree,furfural,water,gases and acid value?were analyzed by traditional method.According to the classical oil-paper insulation aging mechanism and typical thermal aging rule,the aging stages of thermal aging samples were defined.Thus,the sample set for Raman spectral analysis were preliminarily established.A Raman spectra detecting platform for aging features was set up.Aiming at the problems as low concentration,strong interference and weak signal.Silver nano SERS substrate with high Raman signal enhancement effect,spatial uniformity and high stability in the process of detection was prepared.Data preprocessing methods such as peak removal,noise reduction and baseline removing are determined.?2?The molecular model of each aging characteristic was constructed and their Raman characteristics were simulated.The identification of the Raman characteristic peaks of each aging characteristic in oil was realized by comparing detection and calculation result.The linear relationship between the concentration of dissolved aging characteristics and the intensity of Raman peak was analyzed.The spectral data with luxuriant information of oil-paper insulation aging were obtained by SERS detecting.Voigt function was used to analyze the original spectra,and the changes of spectra in the aging process were studied.Thus the diagnostic ability of Raman to the aging is preliminarily verified,it provides a basis for the extraction of spectral diagnostic features.?3?The Monte Carlo interaction validation method was employed to screen the small number of outlier samples in the sample datas.The competitive adaptive reweighted sampling algorithm was employed to extract the key variables which are closely related to the aging degree from the Raman signal.By comparing the location of Raman characteristic peaks with the key variables,the discrimination ability of Raman spectra for oil-paper insulation aging is further verified.Principal component analysis was used to extract the features of Raman spectra,the internal relation between the PCs and aging characteristics of oil-paper insulation was analyzed.?4?By calculating the wavelet packet energy entropy,the change of Raman spectral energy in oil-paper insulation was observed from the viewpoint of spectral information entropy.It was found that the wavelet packet energy entropy of Raman spectra increased in the aging process.Thus,kernel entropy component analysis based on information entropy and kernel principal component analysis was proposed.According to the contribution rate of Renyi entropy,kernel entropy component features were extract from Raman spectra.The internal relation between the new features and aging degree of oil-paper insulation was also analyzed.?5?Based on the extracted principal component features and the kernel entropy component features,the BP neural network and the multi-classification support vector machine were established to diagnose the aging state of oil-paper insulation.Genetic algorithm and particle swarm optimization algorithm were used to optimize the two diagnostic models,and the aging diagnosis model were of excellent ability of discrimination.The insulation oil samples taking from typical running transformers were treated as test data.The same spectral analysis method and feature extraction method were used to study the spectra of running insulation oil.The relationship between the new features of Raman spectra and the traditional diagnostic index was analyzed using unsupervised clustering method.The discriminating ability and generalization ability of the diagnostic model was finally verified.The results demonstrated that the method of oil-paper insulation aging diagnosis based on Raman spectroscopy has good ability to diagnose aging stage and application prospects.The research achievement is of great academic value and practical significance to improve the condition based maintenance of oil immersed electrical equipment,and expand the application of Raman spectroscopy in the field of electrical engineering.