Study On The Raman Spectral Diagnosis Of Transformer Oil-Paper Insulation Aging

Oil-immersed transformer is an important high-voltage equipment in the power system,and its reliability is related to the operating efficiency of the whole power system.The ability to timely monitor the aging of the transformer oil-paper insulation,quickly and accurately diagnose the aging degree of the transformer is of great significance to understand the operating status of the transformer.Laser Raman spectroscopy technology is a very effective method in the aging detection of oil-paper insulation.It does not directly contact with the oil-aging samples during the whole process,and has high detection efficiency and repeatability.The samples obtained from a single batch of oil-paper insulation aging tests are limited.In order to achieve the on-site transformer aging diagnostic criteria,it is urgent to expand the oil-paper insulation aging sample database.In this paper,based on the existing research on Raman spectroscopy of characteristic aging substances,the diagnostic ability of transformer oil-paper insulation aging diagnosis method under multi-batch and fine classification was considered,and the following related research was carried out:(1)Sample acquisition and Raman spectral pretreatment.A sample of oil-paper insulation similar to the insulation structure in the on-site transformer was constructed,and the heat aging treatment process of the sample was set according to the IEEE guidelines standard;The local Pearson correlation coefficient was used to calculate the Raman spectra of oil samples with different aging days,and the abnormal samples were eliminated;According to the standard of national guidelines,the sample polymerization degree of different aging days was measured,and the aging degree of oil samples with different aging days was calibrated;The combined pretreatment of baseline correction and noise removal of Raman spectra was carried out by compound sparse derivative modeling.(2)Feature analysis and feature selection of sample Raman spectra based on variance normalization.The variance analysis of the Raman intensity corresponding to the single Raman shift under the whole sample and the different aging degree sample subsets were carried out respectively,and the two different variances of the single Raman shift were fused with the normalization idea of removing the influence of dimension;The Raman shift of the spectral center of the aging characteristic substance with large Raman intensity variation was selected to analyze the correlation between the Raman intensity variation trend of each characteristic substance peak and the material content in the aging process of oil-paper insulation;According to the difference weight order difference after variance fusion,the Raman spectrum of the whole sample was selected several times.(3)The classification of transformer oil-paper insulation aging diagnosis methods and the comparison analysis of the diagnosis effect of each method under various conditions.According to the principles of various types of diagnostic methods,the diagnostic methods were divided into supervised and unsupervised algorithms,and supervised algorithms were subdivided into single classifier and combined classifier algorithms;Representative algorithms for different types of diagnostic methods were selected: K-means clustering algorithm,Fisher algorithm and Random Forest algorithm.The three methods were used to analyze the adaptability of aging diagnosis model for multi-feature,multi-batch and multi-classification of all samples.(4)Analysis on diagnostic effect of combination model of heterogeneous classifier.According to the advantages of different types of diagnostic models,a variety of heterogeneous classifiers were combined according to the model serial fusion method;By analyzing the diagnosis results of heterogeneous classifier combination model and Random Forest model,the overall diagnostic accuracy of heterogeneous classifier combination model was 93.59% and that of Random Forest model was 91.03%.It was verified that the heterogeneous classifier combination model with multi-type algorithm has more generalization ability under multiple batches and subdivision classes;Based on the evaluation factors such as lifting degree and Kappa coefficient,the advantages of model fusion for the diagnosis of oil-paper insulation aging were determined.The results show that Raman spectroscopy is an effective method for the diagnosis of oil-paper insulation aging.By combining with the data analysis method,the field diagnosis level of transformer aging state can be greatly improved.It lays a foundation for accurate and stable diagnosis of transformer oil-paper insulation aging.