A method for identifying the aging degree of insulation paper based on improved linear discriminant analysis

Abstract

The present invention discloses a method for identifying the aging degree of insulating paper based on improved linear discriminant analysis, comprising the following steps: (1) collecting insulating paper samples with different aging degrees, using a near-infrared spectrometer to detect near-infrared spectrum data of the samples, and obtaining a near-infrared spectrum data matrix X of insulating paper with different aging degrees; (2) classifying the samples into different aging types according to their degree of polymerization, and the spectrum data of each type of sample are respectively X k ; (3) Insulation paper samples with different aging states are grouped into two groups with different aging degrees; (4) For each group of samples, a binary classification model of insulation paper aging state based on spectral data is established using the linear discriminant analysis method; (5) The aging state recognition models of each group of insulation paper are integrated using a one-to-one strategy to obtain a comprehensive recognition model. The method of the present invention can effectively identify the aging state of insulation paper and provide theoretical support for the field application of oil-paper insulation aging state assessment technology.

Description

Technical Field

[0001] The invention belongs to the technical field of aging assessment of solid insulation of power equipment, and in particular relates to a method for identifying the aging degree of insulation paper based on improved linear discriminant analysis. Background Art

[0002] Large power transformers are key equipment in power grids, and their main insulation structure is oil-paper insulation. Oil-paper insulation will age under the action of various stresses such as heat and electricity, which increases the risk of equipment failure sharply. Since insulating oil is easy to replace, while insulating paper is difficult to replace and aging is inevitable, the life of the oil-paper insulation structure mainly depends on the aging of the insulating paper. There are many methods for evaluating the aging of insulating paper, such as methods for analyzing the furfural content in oil, methods for measuring the degree of polymerization of insulating paper, and so on. Among them, the degree of polymerization is the most direct parameter for the aging of standard insulating paper. However, the traditional method for measuring the degree of polymerization is destructive and difficult to implement in on-site transformer testing. A method that can accurately evaluate the aging of oil-paper insulation has great engineering value. Near infrared spectroscopy (NIRS) has been applied to transformer aging status assessment as a rapid non-destructive detection method, but the current assessment model has low accuracy. For example, the traditional linear discriminant analysis method is difficult to directly apply to the aging assessment of insulating paper. Therefore, the present invention proposes a method for identifying the degree of aging of insulating paper based on improved linear discriminant analysis, which can effectively overcome the above-mentioned shortcomings and provide theoretical support for improving the reliability of transformer aging status assessment. Summary of the invention

[0003] The purpose of the present invention is to propose a method for identifying the aging degree of insulating paper based on improved linear discriminant analysis, so as to achieve safe, non-destructive, rapid and convenient identification of the aging degree of insulating paper.

[0004] The present invention is implemented by the following technical solutions:

[0005] A method for identifying the aging degree of insulating paper based on improved linear discriminant analysis comprises the following steps:

[0006] 1) Collecting insulation paper samples with different aging degrees, using a near-infrared spectrometer to detect near-infrared spectrum data of the samples, and obtaining a near-infrared spectrum data matrix X of insulation paper with different aging degrees;

[0007] 2) The samples are divided into different aging types according to their degree of polymerization. The spectral data of each type of sample are X k ;

[0008] 3) Group the insulating paper samples in different aging states into groups;

[0009] 4) For each group of samples, a binary classification model of insulation paper aging status based on spectral data was established using the linear discriminant analysis method;

[0010] 5) Use a one-to-one strategy to integrate the aging status identification models of each group of insulation paper to obtain a comprehensive identification model.

[0011] A further improvement of the present invention is that in step 2), the samples are divided into different aging types according to their degree of polymerization, and the spectral data of each type of sample are respectively X k , k = 1, 2, 3, 4, specifically:

[0012] If the sample aggregation value is less than 350, it is classified into Class I; if the sample aggregation value is between 350-600, it is classified into Class II; if the sample aggregation value is between 600-850, it is classified into Class III; if the sample aggregation value is greater than 850, it is classified into Class IV, and the corresponding spectral data matrices are X1, X2, X3, and X4 respectively;

[0013] A further improvement of the present invention is that in step 3), the insulating paper samples in different aging states are divided into two groups, and a total of six groups of samples are divided into:

[0014] {X1, X2}; {X1, X3}; {X1, X4}; {X2, X3}; {X2, X4}; {X3, X4}.

[0015] A further improvement of the present invention is that in step 4), a linear discriminant analysis method is used to establish a binary classification model of the insulation paper aging state based on spectral data, specifically:

[0016] 1) Calculate the average vector μ of each type of sample spectral data k :

[0017]

[0018] In the formula, μ k is the average vector of the k-th class samples, x k (i) is the spectral data of the k-th sample;

[0019] 2) Calculate the inter-class spectral divergence matrix S respectively B and the intra-class spectral divergence matrix S W :

[0020]

[0021]

[0022] In the formula, m i 、m j are the average vectors of the i-th and j-th samples respectively, m is the average vector of all samples, x i is the spectral data of the i-th sample, and N is the sample capacity;

[0023] 3) Solve the matrix S W -1 S B Eigenvalues ​​and eigenvectors, and select the eigenvector with the largest eigenvalue to form the characteristic matrix W:

[0024] S W -1 S B ν=λν

[0025] In the formula, v is the eigenvector, λ is the eigenvalue, and the eigenvector with the larger eigenvalue is selected to form the characteristic matrix W;

[0026] 4) Based on the matrix W, the original spectral data and the average vector are transformed to obtain the converted spectral data and the center point, and the aging state of the sample is classified, specifically:

[0027] Y=X×W

[0028] centers=mean×W

[0029] Where Y is the converted spectral data, centers is the converted sample center point, and according to the Euclidean distance between the converted sample spectrum and the center point, it is judged as the closest aging class.

[0030] A further improvement of the present invention is that in step 5), each group of binary classification models of insulation paper aging status is integrated using a one-to-one strategy to obtain a comprehensive recognition model, specifically: each group of binary classification aging recognition models established is used to evaluate unknown samples, and the final recognition result is obtained according to the evaluation result of each model by voting.

[0031] The present invention has the following beneficial technical effects:

[0032] When identifying the aging degree of insulating paper, the present invention does not need to collect insulating paper samples of the test transformer, will not damage the insulating paper, and does not need to perform special pre-processing on the test insulating paper. The spectrum detection time only takes a few seconds, and has the technical characteristics of safety, non-destructive, fast and convenient. At the same time, the traditional linear discriminant analysis method is improved, and the accuracy of sample aging state identification is improved by establishing two classification models separately and then integrating them, providing technical support for the accurate selection of oil-paper insulation aging state assessment model, which is conducive to the diagnosis and assessment of the aging state of power transformers. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a flow chart for establishing a comprehensive aging recognition model;

[0034] Figure 2This is the near infrared spectra of insulation paper with different aging degrees;

[0035] Figure 3 It is the linear discriminant analysis flow chart;

[0036] Figure 4 It is a flow chart of the one-to-one strategy integration binary classification model. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0038] like Figure 1 As shown, the present invention provides a method for identifying the aging degree of insulating paper based on improved linear discriminant analysis, comprising the following steps:

[0039] 1. Collect commonly used insulation paper of power transformers with different aging degrees, and use near-infrared spectrometer to detect the near-infrared spectrum of insulation paper. Figure 2 In this example, a total of 136 insulation paper samples were selected, of which 100 samples were used to form an insulation paper type discrimination database, and the remaining 36 samples were used to test the accuracy of the method.

[0040] 2. The samples are divided into different aging types according to their degree of polymerization. The spectral data of each type of sample are X k , k = 1, 2, 3, 4, specifically:

[0041] If the sample aggregation value is less than 350, it is classified into Class I; if the sample aggregation value is between 350-600, it is classified into Class II; if the sample aggregation value is between 600-850, it is classified into Class III; if the sample aggregation value is greater than 850, it is classified into Class IV, and the corresponding spectral data matrices are X1, X2, X3, and X4 respectively;

[0042] 3. The insulation paper samples in different aging states were divided into two groups, and the total number of samples was divided into six groups, specifically:

[0043] {X1, X2}; {X1, X3}; {X1, X4}; {X2, X3}; {X2, X4}; {X3, X4}.

[0044] 4. For each group of samples, a binary classification model of insulation paper aging status based on spectral data is established using the linear discriminant analysis method. Figure 3 As shown, specifically:

[0045] 1) Calculate the average vector μ of each type of sample spectral data k :

[0046]

[0047] In the formula, μ k is the average vector of the k-th class samples, x k (i) is the spectral data of the k-th sample;

[0048] 2) Calculate the inter-class spectral divergence matrix S respectively B and the intra-class spectral divergence matrix S W :

[0049]

[0050]

[0051] In the formula, m i 、m j are the average vectors of the i-th and j-th samples respectively, m is the average vector of all samples, x i is the spectral data of the i-th sample, and N is the sample capacity;

[0052] 3) Solve the matrix S W -1 S B Eigenvalues ​​and eigenvectors, and select the eigenvector with the largest eigenvalue to form the characteristic matrix W:

[0053] S W -1 S B ν=λν

[0054] In the formula, v is the eigenvector, λ is the eigenvalue, and the eigenvector with the larger eigenvalue is selected to form the characteristic matrix W;

[0055] 4) Based on the matrix W, the original spectral data and the average vector are transformed to obtain the converted spectral data and the center point, and the aging state of the sample is classified, specifically:

[0056] Y=X×W

[0057] centers=mean×W

[0058] Where Y is the converted spectral data, centers is the converted sample center point, and according to the Euclidean distance between the converted sample spectrum and the center point, it is judged as the closest aging class.

[0059] 5. Use a one-to-one strategy to integrate the binary classification models of each group of insulation paper aging status to obtain a comprehensive recognition model, specifically:

[0060] Each set of binary classification aging recognition models established was used to evaluate unknown samples. The final recognition results were obtained by voting based on the evaluation results of each model. The discrimination results in this example showed that the recognition accuracy was 97% for the 100 samples in the established sample set and 69.4% for the 36 samples in the validation set.

Claims

1. A method for identifying the aging state of insulating paper based on improved linear discriminant analysis, characterized in that: The following steps are involved: 1) Collecting insulation paper samples with different aging degrees, using a near-infrared spectrometer to detect near-infrared spectrum data of the samples, and obtaining a near-infrared spectrum data matrix X of insulation paper with different aging degrees; 2) The samples are divided into different aging types according to their degree of polymerization, and the spectral data of each type of sample are Xk; 3) Group the insulating paper samples in different aging states into groups; 4) For each group of samples, a binary classification model of the insulation paper aging state based on spectral data was established using the linear discriminant analysis method; The linear discriminant analysis method is used to establish a binary classification model of insulation paper aging state based on spectral data, specifically: 1) Calculate the average vector μk of each type of sample spectral data: In the formula, μk is the average vector of the k-th sample, and xk(i) is the spectral data of the k-th sample; 2) Calculate the inter-class spectral divergence matrix SB and the intra-class spectral divergence matrix SW respectively: In the formula, mi and mj are the average vectors of the i-th and j-th samples respectively, m is the average vector of all samples, xi is the spectral data of the i-th sample, and N is the sample capacity; 3) Solve the eigenvalue and eigenvector of the matrix SW-1SB, and select the eigenvector with the largest eigenvalue to form the characteristic matrix W: SW-1SBν=λν In the formula, v is the eigenvector, λ is the eigenvalue, and the eigenvector with the larger eigenvalue is selected to form the characteristic matrix W; 4) Based on the matrix W, the original spectral data and the average vector are transformed to obtain the converted spectral data and the center point, and the aging state of the sample is classified, specifically: Y=X×W centers=mean×W In the formula, Y is the converted spectral data, centers is the converted sample center point, and according to the Euclidean distance between the converted sample spectrum and the center point, it is judged as the closest aging class; 5) Use a one-to-one strategy to integrate the aging status identification models of each group of insulation paper to obtain a comprehensive identification model.

2. According to the method for identifying the aging state of insulating paper based on improved linear discriminant analysis in claim 1, in step 2, the samples are divided into different aging types according to their degree of polymerization values, and the spectral data of each type of sample are Xk, k=1, 2, 3, 4, specifically: If the sample aggregation value is less than 350, it is classified into Class I; if the sample aggregation value is between 350-600, it is classified into Class II; if the sample aggregation value is between 600-850, it is classified into Class III; if the sample aggregation value is greater than 850, it is classified into Class IV. The corresponding spectral data matrices are X1, X2, X3, and X4 respectively.

3. According to the method for identifying the aging state of insulating paper based on improved linear discriminant analysis in claim 1, in step 3, the insulating paper samples in different aging states are grouped in pairs, and a total of six groups of samples are divided into: {X1, X2}, {X1, X3}, {X1, X4}, {X2, X3}, {X2, X4}, {X3, X4}.

4. The method for identifying the aging state of insulating paper based on improved linear discriminant analysis according to claim 1, characterized in that: In step 5), the binary classification models of the aging status of each group of insulating paper are integrated using a one-to-one strategy to obtain a comprehensive recognition model. Specifically, each group of binary classification aging recognition models established is used to evaluate the unknown samples, and the final recognition result is obtained according to the evaluation result of each model by voting.

Images