Oiled paper thermal aging stage identification method based on fluorescence optical characteristic parameters

Abstract

A fluorescence optical characteristic parameter-based oiled paper thermal aging stage identification method comprises the following steps: acquiring three-dimensional fluorescence spectra of an oiled paper sample in different thermal aging stages by using a fluorescence spectrometer, and acquiring excitation wavelength, emission wavelength and corresponding fluorescence intensity of the fluorescence spectrometer; extracting a plurality of characteristic parameters from the three-dimensional fluorescence spectrum, wherein the characteristic parameters comprise fluorescence intensity and the position and area of a fluorescence peak; carrying out dimension reduction processing on the characteristic parameters by adopting a principal component analysis method; and S3, training and classifying the dimensionality-reduced characteristic parameters by using an SVM algorithm module, and inputting the three-dimensional fluorescence spectrum of the oil paper sample to be detected into the trained SVM algorithm module after the three-dimensional fluorescence spectrum is processed in the steps S1 to S3 so as to identify the thermal aging stage.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment insulation state identification, and particularly relates to a method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters. Background Art

[0002] Oil-immersed power transformers are the most commonly used type of power transformers. The oil-paper insulation system is the most important way of its internal insulation, and this system largely determines the service life of the transformer. With the development of the power system, the thermal aging problem of the insulation system has become an important factor affecting the operation life and safety of equipment. As a commonly used insulation material in transformers, the accurate evaluation of the thermal aging degree of the oil-paper insulation system is of great significance for the maintenance and operation of equipment. At present, the degree of polymerization (DP) is the main parameter characterizing the thermal aging of the insulation oil-paper system, but problems such as difficult sampling and long detection period limit its practical application in engineering.

[0003] As a non-invasive detection method, fluorescence spectroscopy technology has been widely used in transformer detection in recent years. By analyzing the fluorescence spectrum characteristic parameters, the changes in the microscopic structure and chemical composition of the oil-paper system can be monitored. However, fluorescence spectrum data usually has high dimensionality and strong correlation, and there are certain difficulties in directly using it for the division of the thermal aging stage of the oil-paper system. The existing fluorescence spectrum analysis methods mainly judge the thermal aging state through a single characteristic parameter, lacking the comprehensive utilization of multi-dimensional characteristic parameters, resulting in insufficient accuracy and reliability in the division of the thermal aging stage.

[0004] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present invention, and therefore may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0005] The present invention provides a method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters. Through the extraction of multi-dimensional characteristic parameters and the application of an optimization algorithm, the accurate division of the thermal aging stage of the oil-paper system is realized, which can provide a reference for the state evaluation and on-line monitoring of oil-immersed transformers.

[0006] A method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters includes:

[0007] Step S1: Use a fluorescence spectrometer to collect the three-dimensional fluorescence spectra of the oil-paper sample at different thermal aging stages, and obtain the excitation wavelength and emission wavelength of the fluorescence spectrometer;

[0008] Step S2: Extract multiple characteristic parameters from the three-dimensional fluorescence spectra, and the characteristic parameters include fluorescence intensity, the position and area of the fluorescence peak;

[0009] Step S3: Perform dimensionality reduction on the characteristic parameters using the principal component analysis method;

[0010] Step S4: Use the SVM algorithm module to train and classify the dimensionality-reduced characteristic parameters. Input the three-dimensional fluorescence spectrum of the oil-paper sample to be detected, which has been processed through Steps S1 to S3, into the trained SVM algorithm module to identify the thermal aging stage.

[0011] In the described method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters, the characteristic parameters further include at least one of peak coordinates, peak values, full width at half maximum, total intensity, average intensity, skewness, kurtosis, and standard deviation.

[0012] In the described method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters, in Step S3, the principal component analysis method for performing dimensionality reduction on the characteristic parameters includes:

[0013] Establish a sample set matrix and perform a de-centralization process on the sample set matrix;

[0014] Calculate the covariance matrix of the de-centralized sample set matrix and perform eigen-decomposition to obtain eigenvalues and corresponding eigenvectors;

[0015] Arrange the eigenvalues in descending order, calculate the contribution rate of each principal component, and select the characteristic parameters with a cumulative contribution rate exceeding a preset threshold as the new sample feature space.

[0016] In the described method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters, in Step S2, before extracting the characteristic parameters, preprocess the three-dimensional fluorescence spectrum, which includes:

[0017] Perform smoothing processing on the collected three-dimensional fluorescence spectrum;

[0018] Remove the background signal from the fluorescence spectrum through background baseline subtraction;

[0019] Normalize the fluorescence spectrum data.

[0020] In the described method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters, Step S4 includes:

[0021] Data preparation: Obtain data from the characteristic parameters after dimensionality reduction by principal component analysis. These data are stored in matrix form, with each row representing a sample, each column representing a principal component, and the thermal aging stage label of each sample. The data form a data set, which is divided into a training set used to train the SVM model to learn the mapping relationship between the internal laws and characteristics of the data and the labels, and a test set used to evaluate the performance and generalization ability of the model after training;

[0022] Model construction, initialization of SVM classifier and selection of SVM kernel function, setting optimization objectives and constraints:

[0023] ,

[0024] in, is the normal vector of the hyperplane, b is the bias term, It is a sample The category label is +1 or -1, and n is the number of training samples;

[0025] Model training: Use cross-validation to optimize the hyperparameters of the SVM model. Use the optimized parameters to perform final training on the SVM model on the entire training set. The SVM algorithm solves the convex quadratic programming problem based on the set optimization goals and constraints to find the optimal classification hyperplane and determine the normal vector of the hyperplane. and the value of the bias term b;

[0026] Classification is performed by applying the trained SVM model to the test set or samples processed by the three-dimensional fluorescence spectrum of the oil paper sample to be tested. For each sample to be classified, its feature vector after PCA processing is input into the SVM classifier. The classifier calculates the distance from the sample to the hyperplane based on the learned classification hyperplane, and determines the category to which it belongs based on the sign and size of the distance.

[0027] In the oil paper thermal aging stage identification method based on fluorescent optical characteristic parameters, the SVM algorithm module adopts a stratified 5-fold cross-validation method to perform model training and optimization. The stratified 5-fold cross-validation divides the data set into 5 subsets, each subset is used as a test set in turn, and the remaining 4 subsets are used as training sets, and 5 training and testing are performed.

[0028] In the oil-paper thermal aging stage identification method based on fluorescent optical characteristic parameters, the oil-paper samples pretreated with insulating oil and insulating paperboard are packaged, and 400 mL of insulating oil is quantitatively loaded into a 500 mL high-temperature resistant reagent bottle. The transformer oil and insulating paper samples with an oil-paper ratio of 15:1 are used to construct a transformer solid insulation simulation system. The pure oil sample is set as the control group. 2 Copper flakes were added in a ratio of 1 g / 1 g of copper and insulating oil to simulate the actual operating conditions of the transformer; samples were pretreated to detect whether the moisture content of the samples met the requirements, and samples were taken for polymerization degree detection and fluorescence effect preliminary experiments; thermal aging tests were carried out at 130 ℃, 115 ℃ and 100 ℃ for 32 days, and samples were taken every 4 days to collect the polymerization degree, acid value and moisture chemical parameters of the samples, and at the same time, the three-dimensional fluorescence spectrum was collected.

[0029] In the described method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters, the detection wavelength of the fluorescence spectrometer is 200 - 2050 nm, the spectral bandwidth is 0 - 30 nm, and the wavelength resolution is <0.5 nm.

[0030] In the described method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters, the characteristic parameters also include the full width at half maximum, the integrated intensity of the fluorescence region, and the average intensity of the fluorescence region. Among them,

[0031] The full width at half maximum FWHM is:

[0032] ,

[0033] The integrated intensity of the fluorescence region is:

[0034] ,

[0035] The average intensity of the fluorescence region is:

[0036] ,

[0037] where X p is the emission wavelength corresponding to the fluorescence peak; X l , Xr are the emission wavelengths at the lower half value of the excitation wavelength corresponding to the fluorescence peak; is the fluorescence intensity at the excitation wavelength and the emission wavelength , and are the intervals of the excitation wavelength and the emission wavelength respectively; where represents the total intensity of the fluorescence peak, which can be obtained by integrating the fluorescence peak region; represents the area of the fluorescence peak region, that is, the range of the excitation wavelength and the emission wavelength.

[0038] In the described method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters, the principal component analysis method includes,

[0039] First, establish a data set X with m samples and n characteristic numbers, denoted as the sample set matrix X:

[0040]

[0041] Then, de-center the matrix to obtain a new matrix X, and the calculation formula is as follows:

[0042]

[0043] where, ,

[0044] Calculate the covariance matrix of the new matrix X ,

[0045] Then perform eigendecomposition on the covariance matrix to obtain the eigenvalues and the corresponding eigenvector , sort the eigenvalues ​​from large to small Arrange, calculate vector The corresponding regularized eigenvector As a new sample evaluation dimension matrix, the contribution rate of each principal component to the information content of the original data is calculated, and q variables with cumulative contribution rates exceeding 98% are selected as principal component factors.

[0046] Compared with the prior art, the present invention has the following advantages: it utilizes a variety of fluorescence spectral characteristic parameters to reflect the thermal aging state of the oil-paper system from multiple angles, overcomes the problem of limited information content of a single parameter, and improves the accuracy of thermal aging stage division; it adopts the PCA-SVM algorithm to mine the inherent structure and characteristics of unlabeled big data, eliminate noise and redundant information, utilizes labeled sample training, predicts unknown samples through a classification model, and realizes the extraction of data essential features and accurate classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] By reading the detailed description of the preferred specific embodiments below, various other advantages and benefits of the present invention will become clear to those of ordinary skill in the art. The drawings in the specification are only for the purpose of illustrating the preferred embodiments and are not considered to be limitations of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative work. Moreover, the same reference numerals are used to represent the same components throughout the drawings.

[0048] In the attached picture:

[0049] Figure 1 A schematic diagram of steps of an identification method provided by an embodiment of the present disclosure;

[0050] Figure 2 A mineral oil thermal aging sample diagram provided for one embodiment of the present disclosure;

[0051] Figure 3 A schematic diagram of a three-dimensional fluorescence spectrum example provided for an embodiment of the present disclosure.

[0052] The present invention is further explained below in conjunction with the accompanying drawings and embodiments. DETAILED DESCRIPTION

[0053] Specific embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although specific embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0054] It should be noted that certain terms are used in the specification and claims to refer to specific components. Those skilled in the art should understand that technicians may use different terms to refer to the same component. The specification and claims do not use the difference in terms as a way to distinguish components, but use the difference in the functions of components as the criterion for distinction. As mentioned throughout the specification and claims, "comprising" or "including" is an open-ended term and should be interpreted as "including but not limited to". The subsequent description of the specification is the preferred implementation mode for implementing the present invention, but the description is for the purpose of the general principles of the specification and is not used to limit the scope of the present invention. The protection scope of the present invention shall be determined by the scope defined by the appended claims.

[0055] For the convenience of understanding the embodiments of the present invention, the following will further explain with specific embodiments as examples in conjunction with the accompanying drawings, and each accompanying drawing does not constitute a limitation on the embodiments of the present invention.

[0056] As Figures 1 to 3 shown, the method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters includes the following steps:

[0057] Step S1: Use a fluorescence spectrometer to collect the three-dimensional fluorescence spectra of the oil-paper sample at different thermal aging stages, and obtain the excitation wavelength, emission wavelength, and corresponding fluorescence intensity of the fluorescence spectrometer;

[0058] Step S2: Extract multiple characteristic parameters from the three-dimensional fluorescence spectrum, and the characteristic parameters include fluorescence intensity, the position and area of the fluorescence peak;

[0059] Step S3: Use the principal component analysis method to perform dimensionality reduction processing on the characteristic parameters;

[0060] Step S4: Use the SVM algorithm module to train and classify the dimensionality-reduced characteristic parameters, and input the three-dimensional fluorescence spectrum of the oil-paper sample to be detected after being processed by steps S1 to S3 into the trained SVM algorithm module to identify the thermal aging stage.

[0061] In the preferred implementation mode of the method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters, the characteristic parameters further include at least one of peak coordinates, peak value, full width at half maximum, total intensity, average intensity, skewness, kurtosis, and standard deviation.

[0062] In the preferred implementation of the method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters, in step S3, the principal component analysis for dimensionality reduction of the characteristic parameters includes

[0063] establishing a sample set matrix and performing a de-centralization process on the sample set matrix;

[0064] calculating the covariance matrix of the de-centralized sample set matrix and performing eigenvalue decomposition to obtain eigenvalues and corresponding eigenvectors;

[0065] arranging the eigenvalues from largest to smallest, calculating the contribution rate of each principal component, and selecting the characteristic parameters with a cumulative contribution rate exceeding a preset threshold as the new sample feature space.

[0066] In the preferred implementation of the method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters, in step S2, before extracting the characteristic parameters, the three-dimensional fluorescence spectrum is preprocessed, which includes

[0067] performing a smoothing process on the collected three-dimensional fluorescence spectrum;

[0068] removing the background signal from the fluorescence spectrum through background baseline;

[0069] normalizing the fluorescence spectrum data.

[0070] In the preferred implementation of the method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters, the SVM algorithm module uses a hierarchical 5-fold cross-validation method for model training and optimization.

[0071] In the preferred implementation of the method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters, the hierarchical 5-fold cross-validation divides the data set into 5 subsets, each subset takes turns as the test set, and the remaining 4 subsets are used as the training set for 5 times of training and testing.

[0072] In the preferred implementation of the method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters, the pre-treated oil-paper samples of insulating oil and insulating cardboard are encapsulated. A 500 mL high-temperature resistant reagent bottle is used to quantitatively fill 400 mL of insulating oil, and a transformer solid insulation simulation system is constructed with a transformer oil and insulating paper sample with an oil-paper ratio of 15:1. The pure oil sample is set as the control group, and according to 0.05 cm 2Add copper sheets in the ratio of 1 g of copper to insulating oil to simulate the actual operating conditions of the transformer; conduct pretreatment of the samples, detect whether the moisture content of the samples meets the requirements, take samples for degree of polymerization detection and pre-experiment of fluorescence effect; conduct thermal aging tests at 130 °C, 115 °C and 100 °C for 32 days, sample every 4 days (96 h), collect chemical parameters such as the degree of polymerization, acid value and moisture of the samples, and collect three-dimensional fluorescence spectra at the same time.

[0073] In the preferred embodiment of the method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters, the detection wavelength of the fluorescence spectrometer is 200 - 2050 nm, the spectral bandwidth is 0 - 30 nm, and the wavelength resolution is <0.5 nm.

[0074] In the preferred embodiment of the method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters, the characteristic parameters further include full width at half maximum, integrated intensity of the fluorescence region, and average intensity of the fluorescence region, where

[0075] The full width at half maximum FWHM is:

[0076] ,

[0077] The integrated intensity of the fluorescence region is:

[0078] ,

[0079] The average intensity of the fluorescence region is:

[0080] ,

[0081] where X p is the emission wavelength corresponding to the fluorescence peak; X l , Xr are the emission wavelengths at the lower half value of the excitation wavelength corresponding to the fluorescence peak; is the fluorescence intensity at the excitation wavelength and the emission wavelength , and are the intervals of the excitation wavelength and the emission wavelength respectively; where represents the total intensity of the fluorescence peak, which can be obtained by integrating the fluorescence peak region; represents the area of the fluorescence peak region, that is, the range of the excitation wavelength and the emission wavelength.

[0082] In the preferred embodiment of the method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters, the principal component analysis method includes,

[0083] First, establish a data set X with m samples and n feature numbers, denoted as the sample set matrix X:

[0084]

[0085] Then, de - center the matrix to obtain a new matrix X, and the calculation formula is as follows:

[0086]

[0087] Among them, ,

[0088] Calculate the covariance matrix of the new matrix X ,

[0089] Then, perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues and the corresponding eigenvectors . Arrange the eigenvalues from largest to smallest , and calculate the vector The corresponding normalized eigenvector is used as the new sample evaluation dimension matrix, calculate the contribution rate of the information amount representing the original data of each principal component, and select q variables with a cumulative contribution rate exceeding 98% as the principal component factors.

[0090] The method for identifying the thermal aging stage of the oil - paper system based on fluorescence optical characteristic parameters includes the following steps, as Figure 1 shown:

[0091] S1: Use a fluorescence spectrometer to collect the three - dimensional fluorescence spectra of the oil - paper samples at different thermal aging stages, and obtain data including the excitation wavelength, emission wavelength, and the corresponding fluorescence intensity, etc.;

[0092] S2: Extract multiple characteristic parameters from the collected three - dimensional fluorescence spectra, such as fluorescence intensity, the position and area of the fluorescence peak, etc.;

[0093] S3: Since there are many extracted fluorescence characteristic parameters and they are correlated, use the principal component analysis method to perform dimensionality reduction processing on the characteristic parameters;

[0094] S4: Use the SVM algorithm to train and classify the characteristic parameters processed by PCA.

[0095] Encapsulate the pre - treated oil - paper samples of insulating oil and Wedemann insulating cardboard. Use a 500 mL high - temperature resistant reagent bottle to quantitatively fill 400 mL of insulating oil, and construct a transformer solid insulation simulation system with a transformer oil and insulating paper sample with an oil - paper ratio of 15:1. Set the pure oil sample as the control group, and according to 0.05 cm 2Add copper sheets in the ratio of 1 g of copper to insulating oil to simulate the actual operation of the transformer; conduct pre-treatment of the samples, detect whether the moisture content of the samples meets the requirements, take samples for degree of polymerization detection and fluorescence effect pre-experiment; conduct thermal aging tests at 130 °C, 115 °C and 100 °C for 32 days, sample every 4 days (96 h), and the test samples are as Figure 2 shown; collect chemical parameters such as the degree of polymerization, acid value and moisture of the samples, and collect fluorescence spectra at the same time. The fluorescence spectrometer of S1 is specifically the FLS980 steady-state transient fluorescence spectrometer, as Figure 3 shown. The detection wavelength of the instrument is 200 - 2050 nm, which can cover the normal fluorescence detection range. The parameters of the instrument's detection performance include a spectral bandwidth of 0 - 30 nm and a wavelength resolution of <0.5 nm.

[0096] In the preferred embodiment of the method described above, the fluorescence test method is specifically: first cool the detector to below -20 °C, and then adjust the slit parameters to keep them consistent during each sampling.

[0097] In the preferred embodiment of the method described above, the full width at half maximum, integrated intensity of the fluorescence region, and average intensity of the fluorescence region in the fluorescence characteristic parameters of S2 are specifically:

[0098] The full width at half maximum FWHM is defined as:

[0099]

[0100] Integrated intensity of the fluorescence region is defined as:

[0101]

[0102] Average intensity of the fluorescence region is defined as:

[0103] .

[0104] Where X p is the emission wavelength corresponding to the fluorescence peak; X l , Xr are the emission wavelengths at the lower half value of the excitation wavelength corresponding to the fluorescence peak; is the fluorescence intensity at the excitation wavelength and the emission wavelength , and are the intervals of the excitation wavelength and the emission wavelength respectively; where represents the total intensity of the fluorescence peak, which can be obtained by integrating the fluorescence peak region; represents the area of the fluorescence peak region, that is, the range of the excitation wavelength and the emission wavelength.

[0105] In a preferred embodiment of the method described above, the principal component analysis dimensionality reduction in S3 specifically comprises:

[0106] First, a data set X with m samples and n feature numbers is established and denoted as the sample set matrix X:

[0107]

[0108] Then, the matrix is decentralized to obtain a new matrix X, and the calculation formula is as follows:

[0109]

[0110] where, .

[0111] Calculate the covariance matrix of the new matrix X .

[0112] Then, perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues and the corresponding eigenvectors . Arrange the eigenvalues from largest to smallest . Calculate the vector corresponding to the normalized eigenvector as the new sample evaluation dimension matrix.

[0113] Calculate the contribution rate (%) of each principal component. This contribution rate represents the information volume of the original data, and q variables with a cumulative contribution rate exceeding 98% are selected as the principal component factors.

[0114] In one embodiment: 1. Data preparation:

[0115] (1) Data acquisition and arrangement: Obtain data from the characteristic parameters after the dimensionality reduction processing of the previous principal component analysis (PCA). These data are stored in matrix form, with each row representing a sample and each column representing a principal component. At the same time, prepare the heat aging stage labels corresponding to each sample, namely "weak connection stage" and "amorphous region stage";

[0116] (2) Data division: Divide the arranged data set into two parts, a training set and a test set. The training set is used to train the SVM model so that it learns the mapping relationship between the internal laws and characteristics of the data and the labels; the test set is used to evaluate the performance and generalization ability of the model after the model training is completed to ensure that the model can accurately classify unknown samples in actual applications.

[0117] 2. Model construction:

[0118] (1) Initialize the SVM classifier: Select an appropriate SVM kernel function, such as a linear kernel, polynomial kernel, or Gaussian radial basis function (RBF) kernel, etc. Different kernel functions are suitable for different types of data distributions and problem characteristics. For example, when the data is linearly separable, the linear kernel may perform well; while when the data has complex non-linear relationships, the RBF kernel usually achieves better classification performance. In addition, relevant parameters of the kernel function need to be set, such as the gamma value of the RBF kernel, etc. These parameters will affect the fitting degree and generalization ability of the model to the data, and can be further adjusted through parameter optimization steps later;

[0119] (2) Set the optimization objective and constraints: The core idea of the SVM algorithm is to find an optimal hyperplane that maximizes the margin between samples of different classes on both sides of the hyperplane. For the linearly separable case, the optimization objective is to minimize the reciprocal of the classification margin, while satisfying the constraint that the distance from each class of samples to the hyperplane is greater than or equal to 1. Mathematically, this optimization problem can be expressed as:

[0120]

[0121] where, is the normal vector of the hyperplane, b is the bias term, is the sample 's class label (taking values of +1 or -1), and n is the number of training samples. For the non-linearly separable case, by introducing slack variables and a penalty parameter C, allowing some samples to exist within the margin band or on the side of the hyperplane with misclassification, the optimization objective becomes to balance minimizing the reciprocal of the classification margin and the penalty for misclassified samples.

[0122] 3. Model training:

[0123] (1) Parameter optimization: Use methods such as cross-validation to optimize the hyperparameters of the SVM model. In this invention, the gamma value of the RBF kernel and the penalty parameter C are adopted. By performing multiple trainings and validations on different subsets of the training set, find the parameter combination that makes the model have the best performance on the validation set. In stratified 5-fold cross-validation, the training set is divided into 5 subsets. Each time, 4 of the subsets are used as training data, and the remaining 1 subset is used as validation data. After repeating 5 times, the average performance index is taken as the evaluation criterion to determine the optimal model parameters;

[0124] (2) Model training execution: Use the optimized parameters to perform the final training of the SVM model on the entire training set. During the training process, the SVM algorithm will, according to the set optimization objective and constraints, solve the convex quadratic programming problem to find the optimal classification hyperplane, that is, determine the normal vector of the hyperplane and the value of the bias term b. This process is usually carried out with the help of existing machine learning libraries or tools. In this invention, the scikit-learn library in Python is used, which internally implements efficient optimization algorithms to handle large-scale datasets and complex optimization problems.

[0125] 4. Classification Execution

[0126] Apply the trained SVM model to the test set or actual unknown sample data. For each sample to be classified, input its feature vector after PCA processing into the SVM classifier. The classifier will calculate the distance of the sample to the hyperplane according to the learned classification hyperplane, and determine its belonging class based on the sign and magnitude of the distance. Specifically, if the calculation result is greater than 0, the sample is classified into one class; if it is less than 0, it is classified into another class; when it is equal to 0, it lies on the hyperplane, which is relatively rare. The mathematical expression for the classification decision is:

[0127]

[0128] where x is the feature vector of the sample to be classified, and the sign function is used to take the sign of the calculation result to determine the class label of the sample.

[0129] 5. Result Evaluation:

[0130] (1) Performance Metric Calculation: Use multiple performance metrics to comprehensively evaluate the classification results of the SVM classifier. Commonly used metrics include accuracy, recall, F1-score, and area under the ROC curve (AUC), etc. Accuracy reflects the proportion of correctly classified samples in the total samples; recall focuses on the proportion of correctly classified samples in a certain class among the total samples of that class; the F1-score is the harmonic mean of accuracy and recall, comprehensively considering the relationship between the two; the ROC curve intuitively shows the performance change of the classifier by plotting the true positive rate and false positive rate at different classification thresholds, and the AUC value quantifies the area under the ROC curve, and the value closer to 1 indicates better classifier performance;

[0131] (2) Model analysis and improvement: Based on the results of performance metrics, conduct an in-depth analysis of the classification performance of the SVM model. If the accuracy is low, it may be due to overfitting or underfitting problems in the model. In this case, consider adjusting the model parameters, increasing the amount of training data, or optimizing feature engineering and other measures to improve the model. If the recall rate of a certain category is low, it indicates that the model's recognition ability for samples of this category is insufficient. It may be necessary to further analyze the feature distribution of samples in this category, adjust the classification decision boundary, or adopt specific sampling strategies to improve its classification effect. By continuously evaluating and optimizing the model, achieve the best classification performance in the recognition task of the thermal aging stage of the oil-paper system, providing a reliable basis for the actual transformer condition assessment and maintenance.

[0132] In the preferred embodiment of the described method, the support vector machine classification in S4 is specifically as follows:

[0133] For binary linearly separable samples, to maximize the distance between the two types of samples to the hyperplane Establish a classification function:

[0134]

[0135] According to The size relationship with 0, classify the samples into two types of labels. Therefore, the goal of the classification problem is to find the optimal hyperplane parameters, and the optimization objective and constraint conditions in the search process are:

[0136]

[0137] For non-linearly separable problems, introduce a kernel function to replace the inner product of the classification function and map the classification function in the dual coordinate system to a high-dimensional space. The classification function is:

[0138]

[0139] In the formula Is the Lagrange coefficient, Is the training set sample.

[0140] In the preferred embodiment of the described method, the specific criteria for dividing the thermal aging stage of the recognition method are as follows:

[0141] The cardboard used for transformer insulation can be microstructurally divided into crystalline regions and amorphous regions. Under normal circumstances, during the early and middle stages of the thermal aging of oil-paper, the thermal aging of the cardboard occurs in the amorphous region. When the crystalline region undergoes thermal aging, usually the degree of polymerization of the cardboard no longer changes substantially with the thermal aging time. Therefore, it is considered that the cardboard has entered the late stage of thermal aging. In addition to the crystalline and amorphous regions, there is also a weak connection between cellulose molecules, which is the key factor leading to the initial thermal aging of cellulose. The initial stage of cardboard thermal aging is divided into a weak connection stage and an amorphous region stage, with the DP value divided at 443. Since the embodiments of the present invention focus on the division of the early and middle stages of the thermal aging of the oil-paper system, the thermal aging process of the oil-paper samples is divided into two stages based on the relationship between the degree of polymerization of the samples and the above-mentioned divided DP value. When the DP value is greater than the divided DP value, it is the weak connection stage; when it is less than the divided DP value, it is the amorphous region stage. Therefore, the problem of dividing the thermal aging stage of oil-paper is a binary classification problem.

[0142] In a preferred embodiment of the method described above, the model setting of the identification method is specifically as follows:

[0143] First, conduct 5 repeated experimental samplings on the oil-paper thermal aging samples to obtain sufficient valid data, and then assign corresponding labels according to the division results of the thermal aging stages of different ester-based insulating oils. Calculate multiple dimensions such as the peak coordinates, peak values, half-peak widths, total intensities, average intensities, as well as skewness, kurtosis, and standard deviation of the emission spectrum of each sample to form the initial feature space of the sample. Then, use the principal component analysis method to reduce the dimension of the initial feature space, select the characteristic parameters with a cumulative contribution rate greater than 98% to form a new sample feature space after dimension reduction, and use the SVM algorithm to identify the thermal aging stage of the oil-paper system, and adopt stratified 5-fold cross-validation.

[0144] In a preferred embodiment of the method described above, the identification result of the identification method is shown in Table 1. The identification effect of the PCA-SVM algorithm on the thermal aging stage of mineral oil-paper samples can reach an accuracy rate of more than 90%.

[0145] Table 1 Identification results of the thermal aging stage of oil-paper samples

[0146]

[0147] By combining the PCA and SVM algorithms, the present invention makes full use of the advantages of the two learning modes and realizes the accurate division of the thermal aging stage of the oil-paper system. In specific embodiments, the effectiveness and reliability of the method are verified by experiments on different types of insulating oil-paper samples. The experimental results show that the method of the present invention can significantly improve the accuracy of the thermal aging stage division and provide a new technical means for the condition assessment of the oil-paper insulation system.

[0148] Although the embodiments of the present invention have been described above in conjunction with the accompanying drawings, the present invention is not limited to the above specific embodiments and application fields. The above specific embodiments are merely illustrative and guiding, rather than restrictive. Those of ordinary skill in the art can also make many forms under the inspiration of this specification and without departing from the scope protected by the claims of the present invention, and these all fall within the scope of protection of the present invention.

Claims

1. A method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters, characterized in that The steps include: Step S1: using a fluorescence spectrometer to collect the three-dimensional fluorescence spectra of the oil paper sample at different heat aging stages, and obtaining the excitation wavelength and emission wavelength of the fluorescence spectrometer; Step S2: extracting multiple characteristic parameters from the three-dimensional fluorescence spectrum, the characteristic parameters including fluorescence intensity, position and area of ​​fluorescence peak; Step S3: using principal component analysis to reduce the dimension of characteristic parameters; Step S4: Use the SVM algorithm module to train and classify the feature parameters after dimensionality reduction, and input the three-dimensional fluorescence spectrum of the oil paper sample to be tested into the trained SVM algorithm module after being processed through steps S1 to S3 to identify the thermal aging stage.

2. The method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters according to claim 1, wherein Preferably, the characteristic parameters also include at least one of peak coordinates, peak value, half-peak width, overall intensity, average intensity, skewness, steepness and standard deviation.

3. The method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters according to claim 1, wherein In step S3, the principal component analysis method performs dimension reduction processing on the characteristic parameters, including: Establish a sample set matrix and decentralize the sample set matrix; Calculate the covariance matrix of the decentralized sample set matrix and perform eigendecomposition to obtain the eigenvalues ​​and corresponding eigenvectors; Arrange the eigenvalues ​​from large to small, calculate the contribution rate of each principal component, and select the characteristic parameters whose cumulative contribution rate exceeds the preset threshold as the new sample feature space.

4. A method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters according to claim 1, characterized in that In step S2, before extracting characteristic parameters, the three-dimensional fluorescence spectrum is preprocessed, which includes: Smoothing the collected three-dimensional fluorescence spectrum; The background signal in the fluorescence spectrum is removed by background baseline; The fluorescence spectral data were normalized.

5. A method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters according to claim 1, characterized in that, Step S4 comprises, Data preparation: Data are obtained from the feature parameters after dimension reduction processing by principal component analysis. These data are stored in the form of a matrix, with each row representing a sample, each column representing a principal component, and the thermal aging stage label of each sample. The data constitute a data set, which is divided into a training set for training the SVM model to learn the inherent laws of the data and the mapping relationship between features and labels, and a test set for evaluating the performance and generalization ability of the model after training. Model construction, initialization of SVM classifier and selection of SVM kernel function, setting optimization objectives and constraints: ， Among them, is the normal vector of the hyperplane, b is the bias term, is the sample 's class label, taking values of +1 or -1, and n is the number of training samples; Model training: The hyperparameters of the SVM model are optimized using cross-validation. Using the optimized parameters, the SVM model is finally trained on the entire training set. The SVM algorithm finds the optimal classification hyperplane and determines the normal vector of the hyperplane by solving a convex quadratic programming problem according to the set optimization objectives and constraints. and the value of the bias term b; Classification is performed by applying the trained SVM model to the test set or samples processed by the three-dimensional fluorescence spectrum of the oil paper sample to be tested. For each sample to be classified, its feature vector after PCA processing is input into the SVM classifier. The classifier calculates the distance from the sample to the hyperplane based on the learned classification hyperplane, and determines the category to which it belongs based on the sign and size of the distance.

6. The method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters according to claim 5, wherein The SVM algorithm module uses a stratified 5-fold cross-validation method for model training and optimization. The stratified 5-fold cross-validation divides the data set into 5 subsets, each subset is used as a test set in turn, and the remaining 4 subsets are used as training sets, and training and testing are performed 5 times.

7. A method for identifying the thermal aging stage of oil paper based on fluorescence optical characteristic parameters according to claim 1, wherein, After encapsulating the oil-paper samples pre-treated with insulating oil and insulating paperboard, a 500 mL high-temperature resistant reagent bottle was used to quantitatively fill 400 mL of insulating oil, and a transformer solid insulation simulation system was constructed with a transformer oil and insulating paper sample with an oil-paper ratio of 15:

1. The pure oil sample was set as the control group, and copper sheets were added according to the ratio of 0.05 cm 2 / 1 g of copper and insulating oil to simulate the actual operation condition of the transformer; the samples were pre-treated, and it was detected whether the moisture content of the samples met the requirements, and samples were taken for degree of polymerization detection and fluorescence effect pre-experiment; thermal aging tests were carried out at 130 °C, 115 °C and 100 °C, the thermal aging time was 32 days, samples were taken every 4 days, and the chemical parameters of the degree of polymerization, acid value and moisture of the samples were collected, and three-dimensional fluorescence spectra were collected at the same time.

8. A method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters according to claim 1, characterized in that The detection wavelength of the fluorescence spectrometer is 200-2050nm, the spectral bandwidth is 0-30nm, and the wavelength resolution is <0.5nm.

9. The method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters according to claim 1, characterized in that Characteristic parameters also include half-peak width, volume integral intensity of the fluorescence area, and average intensity of the fluorescence area, among which, The full width at half maximum FWHM is as follows: ， Integrated intensity of the fluorescence region is as follows: ， Average intensity of the fluorescent region is as follows: ， where X p is the emission wavelength corresponding to the fluorescence peak; X l , Xr are the emission wavelengths at the half-peak value of the excitation wavelength corresponding to the fluorescence peak; is at the excitation wavelength and the emission wavelength under the fluorescence intensity, and are the intervals of the excitation wavelength and the emission wavelength respectively; where represents the total intensity of the fluorescence peak, which can be obtained by integrating the fluorescence peak region; represents the area of the fluorescence peak region, that is, the range of the excitation wavelength and the emission wavelength.

10. A method for identifying the thermal aging stage of oil-paper based on fluorescence optical characteristic parameters according to claim 1, characterized in that, The principal component analysis method includes First, a data set X with m samples and n characteristic numbers is established, denoted as the sample set matrix X: ， Then, the matrix is decentralized to obtain a new matrix X, and the calculation formula is as follows: ， Among them, , Calculate the covariance matrix of the new matrix X , Then perform eigendecomposition on the covariance matrix to obtain the eigenvalues and the corresponding eigenvector , sort the eigenvalues ​​from large to small Arrange, calculate vector The corresponding regularized eigenvector As a new sample evaluation dimension matrix, the contribution rate of each principal component to the information content of the original data is calculated, and q variables with cumulative contribution rates exceeding 98% are selected as principal component factors.

Images