Random Forest-Based Aging Diagnosis Method for Oil-Impregnated Paper Insulation and Thermalization Tank

Abstract

A method and a thermal aging tank for diagnosing the aging of oil-paper insulation based on random forest, comprising the following steps: simulating the aging state of a transformer, obtaining oil-paper insulation samples with different oil-paper ratios and different aging states, and collecting Raman spectra of the oil-paper insulation samples; establishing an oil-paper insulation Raman spectrum database; extracting Raman spectral feature quantities with a high degree of correlation with the aging state from the Raman spectrum database based on random forest; establishing a relationship model between the Raman spectral feature quantities and the aging state, and judging the aging degree of the oil-paper insulation with an unknown aging state. The method provided by the present invention greatly reduces the input parameters for subsequent modeling and the training time of the model, improves the diagnostic ability for aging samples of different oil-paper insulation types, and provides a new method for effectively and accurately realizing the aging stages of different oil-paper insulation types and different aging states of oil-paper insulation equipment.

Description

Technical Field

[0001] The present invention belongs to the field of safety status assessment of power equipment, and particularly relates to an aging diagnosis method and a thermalization tank for oil-paper insulation based on random forest. Background Art

[0002] Accurately and effectively diagnosing and evaluating the aging state of operating oil-paper insulation equipment is of great significance for ensuring the safe and stable operation of power equipment. At present, the aging of oil-paper insulation equipment can be evaluated according to the degree of polymerization of insulating paper. However, this method is greatly affected by the equipment and the operation is complex, making it difficult to be applied in engineering practice. Therefore, in actual operation, the aging of oil-paper insulation equipment is often indirectly evaluated by detecting insulating oil. Three independent aging degree judgment indexes, namely the degree of polymerization of insulating paper, the concentration of furfural in oil, and the concentration ratio of carbon monoxide to carbon dioxide, are specified in relevant national standards. Among them, the measurement of the degree of polymerization of insulating paper is relatively accurate and direct, but it requires operations such as power outage and lifting the transformer hood of the transformer, and it is a destructive test. Generally, this method is not adopted unless necessary; the standard stipulates that when measuring the concentration of furfural in oil, it is recommended to use a high-performance liquid chromatography instrument, which requires the configuration of a mobile phase and a series of complex processes such as extraction, and it takes a long time; when using the concentration ratio of carbon monoxide to carbon dioxide to judge the aging degree, due to the diversity of gas sources, the standard also clearly states that the accuracy of this method is not high and it is only for reference. In addition, among the above three methods, large-scale equipment such as an oscillator and a viscosity tester are required for the measurement of the degree of polymerization, a high-performance liquid chromatography instrument is required for the measurement of the furfural concentration, and a gas chromatography instrument and a degassing device are also required for the measurement of the concentrations of carbon monoxide and carbon dioxide. Therefore, most of the existing mature technologies can only be completed in the laboratory and have certain limitations, and do not have the ability of rapid and accurate diagnosis on site. It is very necessary to study a rapid, effective and on-site applicable oil-paper insulation aging evaluation method.

[0003] Prior art document 1 (CN 112485609 B) discloses a Raman spectroscopy diagnosis method for transformer oil-paper insulation aging. The deficiency of prior art document 1 is that: effective aging feature extraction is not carried out on the Raman spectrum of oil-paper insulation aging, and the entire spectrum data is directly used as the input of the diagnosis model. When the types, quantities, etc. of test samples increase, if the aging diagnosis of samples is carried out again, the performance of the diagnosis model will be weakened, that is, the prediction ability of the model will decline. Summary of the Invention

[0004] In order to solve the deficiencies in the prior art, the purpose of the present invention is to provide an aging diagnosis method and a thermalization tank for oil-paper insulation based on random forest. Based on the established Raman spectrum database, random forest is used to extract Raman features, and the aging state of oil-paper insulation equipment can be effectively and accurately evaluated.

[0005] The present invention adopts the following technical solution.

[0006] An oil-paper insulation aging diagnosis method based on random forest includes the following steps:

[0007] Step 1, simulate the aging state of the transformer, obtain oil-paper insulation samples with different oil-paper ratios and different aging states, and collect the Raman spectra of the oil-paper insulation samples;

[0008] Step 2, establish an oil-paper insulation Raman spectrum database by using the Raman spectra obtained in Step 1;

[0009] Step 3, screen multiple Raman spectral feature quantities related to the aging state of the oil-paper insulation in the Raman spectrum database based on random forest;

[0010] Step 4, establish a relationship model between the Raman spectral feature quantities and the aging state, and judge the aging degree of the oil-paper insulation with unknown aging state.

[0011] In Step 1, the oil-paper ratios include 10:1, 15:1, and 20:1; the aging states include good insulation, early aging, middle aging, and late aging;

[0012] In Step 1, oil-paper insulation samples with aging times of 1, 3, 5, 10, 15, 20, 25, and 30 days are obtained respectively to obtain oil-paper insulation samples with different aging states.

[0013] Step 1 further includes preprocessing the Raman spectra;

[0014] The preprocessing includes spike identification based on derivative spectra, spike removal method based on cubic curves, smoothing and denoising method based on the median of three-point cyclic fast Fourier transform, and spectrum normalization based on the maximum and minimum values.

[0015] Step 3 includes:

[0016] Step 3.1, use the nth decision tree to classify the data in the OOB dataset to obtain the classification accuracy of the aging state before perturbation n = 1, 2,..., 100;

[0017] Step 3.2, perturb the values of the Raman spectral feature quantity X j , j in the OOB dataset, where j is the number of Raman spectral dimensions, j = 1, 2,..., 1024, and use the nth decision tree to classify the data in the OOB dataset again to obtain the classification accuracy of the aging state after perturbation

[0018] Step 3.3, repeat Step 3.1 and Step 3.2, and calculate the Raman spectral feature quantity X of each decision tree jThe classification accuracy of the aging state before and after being disturbed. Regarding the Raman spectral feature quantity X j The importance for classifying the aging state is defined as

[0019] Step 3 further includes ranking the importance of each Raman spectral feature quantity for classifying the aging state according to the calculation result, and screening out the Raman spectral feature quantity with importance.

[0020] M takes the value of 0.3.

[0021] Step 4 includes: dividing the oil-paper insulation samples obtained in Step 1 into training samples and test samples; constructing a relationship model between the Raman spectral feature quantity and the aging state; training the relationship model with the training samples, and testing the trained relationship model with the test samples; collecting the oil-paper insulation with unknown aging state.

[0022] The relationship model includes an input layer, a hidden layer, and an output layer; among them, the data input into the input layer is the Raman spectral feature quantity screened out in Step 3; the output layer is the aging state of the oil-paper insulation; the number of hidden layers is equal to the number of training samples.

[0023] The hidden layer adopts the function ψ(X) = exp(-||X - X p || / 2σ 2 )), where X is the screened Raman spectral feature quantity, X p is the center of the basis function ψ; σ is the width parameter of the basis function ψ; the output layer is expressed as: where L is the number of hidden layers.

[0024] The thermalization tank for diagnosing the aging of oil-paper insulation based on random forest, which is used for the method for diagnosing the aging of oil-paper insulation based on random forest

[0025] An oil inlet pump, an oil outlet pump, an oil valve, and a gas valve are arranged at the top of the thermalization tank. The outside of the thermalization tank is immersed in dimethyl silicone oil. The internal filling materials are insulating oil and nitrogen from bottom to top in sequence. Multiple copper bars are suspended in the thermalization tank. The bottom inside the thermalization tank is insulating paper. The copper bars and the insulating paper are completely immersed in the insulating oil.

[0026] The beneficial effects of the present invention are as follows. Compared with the prior art, it can effectively and accurately classify the aging stages of different oil-paper types, and the prediction results for different oil-paper types are relatively high. The present invention identifies the aging states of samples in multiple stages, so it can be well used for diagnosing the aging state of oil-paper insulation of different oil-paper types. In addition, the present invention selects three oil-paper insulation types with oil-paper mass ratios of 10:1, 15:1, and 20:1, and obtains aging samples of different insulation types based on the oil-paper insulation accelerated thermal aging experimental platform. Compared with the 10:1 oil-paper mass ratio insulation type commonly studied in the prior art research, the present invention expands the types of original samples of oil-paper insulation aging and expands the diagnostic objects for equipment of different oil-paper insulation types. To improve the aging diagnosis ability for different oil-paper insulation types, the present invention effectively extracts multiple Raman spectral characteristic quantities that can reflect the aging state of oil-paper insulation from a large amount of Raman spectrogram data through the random forest algorithm. Therefore, the method provided by the present invention greatly reduces the input parameters for subsequent modeling and the training time of the model, improves the diagnostic ability for aging samples of different oil-paper insulation types, and provides a new method for effectively and accurately realizing the aging stages of different oil-paper insulation types and different aging states of oil-paper insulation equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a flowchart of the oil-paper insulation aging diagnosis method based on random forest of the present invention;

[0028] Figure 2 is a flowchart of the pre-treatment of oil-paper insulation thermal aging of the present invention;

[0029] Figure 3 is a schematic structural diagram of the oil-paper insulation thermal aging tank of the present invention;

[0030] Figure 4 is a graph showing the variation law of the degree of polymerization of insulating paper with time under three different oil-paper ratios of the present invention;

[0031] Figure 5 is the importance of each characteristic variable calculated based on the OBB random forest of the present invention;

[0032] Figure 6 is a three-dimensional distribution diagram of characteristic points of samples in different aging stages with an oil-paper ratio of 10:1 of the present invention.

[0033] Reference numerals are: 1 - inlet oil pump, 2 - air valve, 3 - oil valve, 4 - outlet oil pump, 5 - nitrogen, 6 - aging tank, 7 - copper bar, 8 - insulating oil, 9 - insulating paper, 10 - aging box, 11 - dimethyl silicone oil. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0034] The present application will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention and cannot be used to limit the protection scope of the present application.

[0035] The present application provides an oil-paper insulation aging diagnosis method based on random forest, as Figure 1 shown, which includes the following steps:

[0036] Step 1, simulate the aging state of the transformer, obtain oil-paper insulation samples with different oil-paper ratios and different aging states, and collect the Raman spectra of the oil-paper insulation samples; the oil-paper ratios in Step 1 include 10:1, 15:1, 20:1; the aging states include good insulation, early aging, mid-aging, and late aging;

[0037] In Step 1, oil-paper insulation samples with aging times of 1, 3, 5, 10, 15, 20, 25, and 30 days are respectively obtained to obtain oil-paper insulation samples with different aging states. Step 1 also includes preprocessing the Raman spectra; the preprocessing includes spike identification based on derivative spectra, spike removal method based on cubic curves, smoothing and denoising method based on the median of three-point cyclic fast Fourier transform, and spectrum normalization based on maximum and minimum values.

[0038] Step 2, establish an oil-paper insulation Raman spectrum database using the Raman spectra obtained in Step 1;

[0039] Step 3, extract Raman spectral feature quantities with high correlation with the aging state from the Raman spectrum database based on random forest; specifically, Step 3 includes: [[ID=**********]]

[0040] Step 3.1, use the nth decision tree to classify the data in the OOB dataset to obtain the classification accuracy of the aging state before perturbation n = 1, 2,..., 100;

[0041] In the present application, OOB refers to Out of bag: in random forest, m training samples will be sampled n times by bootstrap (random sampling with replacement), and each sampling generates a sampling set with m samples, which enter n parallel decision trees.

[0042] Step 3.2, perturb the value of the Raman spectral feature quantity X j in the OOB dataset, where j is the number of Raman spectral dimensions, j = 1, 2,..., 1024, and use the nth decision tree to classify the data in the OOB dataset again to obtain the classification accuracy of the aging state after perturbation

[0043] Step 3.3, repeat Step 3.1 and Step 3.2, and calculate the Raman spectral feature quantity X of each decision treej The classification accuracy rate of the aging state before and after being disturbed. For this Raman spectral feature quantity X j The importance for classifying the aging state is defined as

[0044] Step 3 also includes, according to the calculation results, sorting the importance of each Raman spectral feature quantity for classifying the aging state, and screening the Raman spectral feature quantities with importantness

[0045] M takes the value of 0.3, and in this embodiment, M = 6.

[0046] Step 4, establish a relationship model between the Raman spectral feature quantity and the aging state, and judge the aging degree of the oil-paper insulation with unknown aging state.

[0047] Step 4 includes: dividing the oil-paper insulation samples obtained in Step 1 into training samples and test samples; constructing a relationship model between the Raman spectral feature quantity and the aging state; training the relationship model with the training samples, and testing the trained relationship model with the test samples; collecting the oil-paper insulation with unknown aging state.

[0048] The relationship model includes an input layer, a hidden layer, and an output layer; wherein the data input into the input layer is the Raman spectral feature quantity screened in Step 3; the output layer is the aging state of the oil-paper insulation; the number of hidden layers is equal to the number of training samples.

[0049] The hidden layer adopts the function ψ(X) = exp(-||X - X p || / 2σ 2 ), where X is the screened Raman spectral feature quantity, X p is the center of the basis function ψ; σ is the width parameter of the basis function ψ; the output layer is expressed as: where L is the number of hidden layers.

[0050] Example 1.

[0051] Step 1, use the device of the present invention Figure 2 to more realistically simulate the aging state of the transformer, obtain more realistic oil-paper ratios and oil-paper insulation samples with different aging states, and collect the Raman spectra of all samples;

[0052] Specifically, in this embodiment, the aging state of the real transformer is simulated according to the IEEE guidelines in the accelerated thermal aging method in a sealed system. The specific experimental steps are as Figure 2As shown: Before the aging test, the samples need to be dried. To remove the moisture in the insulating paper and insulating oil, the insulating paper and insulating oil are first placed in a vacuum drying oven and dried continuously for 48 hours with the temperature stabilized at 90 °C. The oil-paper ratios of oil-paper insulation equipment of different models are concentrated between 10:1 and 20:1. To obtain databases of different oil-paper insulation types and aging states, a total of three types with different oil-paper ratios (10:1, 15:1, 20:1) are designed, and the insulating paper samples are immersed in the insulating oil. To remove the air in it, the oil-impregnated insulating paper is then placed in a vacuum drying oven and dried for another 48 h. The pretreated experimental samples are placed in the thermal aging tank provided by the present invention for accelerated thermal aging at 130 °C. As Figure 3 shown, insulating oil samples are obtained on the 1st day, 3rd day, 5th day, 10th day, 15th day, 20th day, 25th day, and 30th day respectively. Each type is sampled 10 times each time, and 80 groups of samples are obtained in total after sampling 8 times. A total of 240 groups are obtained for the three types. According to the national standard detection standard, the degree of polymerization (DP) of the oil-paper insulation aging samples is detected, and the oil-paper insulation aging samples are divided into four aging states: good insulation I: DP≥900; early aging II: (500≤DP<900); middle aging III: (250≤DP<500); end aging IV (DP<250); Since the aging effects of oil-paper insulation of three different insulation types are different, according to the measured degree of polymerization of the insulating paper in the laboratory (see Figure 4 ), the aging samples of the three insulation types are divided stage by stage according to the division requirements, and the specific division is shown in Table 1.

[0053] Table 1 Division of aging samples of three insulation types

[0054]

[0055] Pretreat the Raman spectra and measure the degree of polymerization of the insulating paper of all samples; for the pretreatment of the Raman spectra, the methods of spike recognition based on derivative spectra, spike removal based on cubic curves, smoothing and denoising based on the median of three-point cyclic fast Fourier transform, and spectrum normalization based on the maximum and minimum values are respectively adopted.

[0056] Step 2, establish an oil-paper insulation Raman spectrum database by using the Raman spectra obtained in Step 1;

[0057] It is divided into four aging states according to the degree of polymerization (DP) of the insulating paper: State I indicates good insulation (DP≥900), State II indicates the early stage of aging (500≤DP<900), State III indicates the middle stage of aging (250≤DP<500), and State IV indicates the final stage of aging (DP<250). A Raman spectrum database is established based on the preprocessed Raman spectra and labels (aging states); the Raman spectrum database includes the Raman spectra of all samples and the corresponding aging states of the samples.

[0058] Step 3: Use the random forest (RF) supervised feature extraction algorithm to extract the Raman spectral feature quantities with a high degree of relevance to the aging state in the established database. The Raman feature quantities screened by the RF algorithm of the present invention are less than 20 to improve the prediction performance of the diagnostic model;

[0059] The calculation method of the RF feature selection in step (4) of the present invention is: perform importance evaluation according to OOB (out of bag), where OOB means that assuming there are n (the maximum value is 100) decision trees in the random forest model. First, let n = 1. And use the nth decision tree to classify the data in the OOB dataset, and the correct classification rate of its aging state is recorded as For the Raman spectral feature X j (where j is the number of Raman spectral dimensions, j = 1, 2,..., 1024) in the OOB dataset, perturb its value, and then use the nth decision tree to classify the data in the OOB dataset again. The correct classification rate of its aging state is recorded as Repeat the above two steps to calculate the correct classification rates of the aging state before and after the perturbation of X for each decision tree. j The importance of the feature X based on OOB for the random forest model is defined as j

[0060] Step 4: Establish a relationship model between the Raman spectral feature quantities and the aging state;

[0061] Specifically, use the extracted Raman spectral feature quantities as the input of the relationship model and the aging state as the output of the model, and use the training set samples to complete the parameters of the model, and finally successfully establish the relationship model.

[0062] Substitute the test set samples into the input of the model to obtain the results of their aging states.

[0063] ① Use the training samples screened by the random forest as the input of the relationship model and the four aging states as the output of the relationship model, and save the parameters of its training model;

[0064] ②After training, import the trained model, select the Raman spectral features of the test samples with the same feature quantity, and use them as the input of the relationship model;

[0065] ③Output through the relationship model to obtain the aging state of the test sample.

[0066] The present invention uses the random forest algorithm to evaluate the importance of each feature in the spectrum according to the out-of-bag data error rate index, performs normalized mapping on the importance, calculates the comprehensive feature importance, sorts according to the importance, and screens out Raman spectral features with an importance ≥ 0.3. In this embodiment, finally 6 Raman spectral features are selected for subsequent diagnosis. These 6 features can well distinguish different aging categories, reducing the feature dimension while retaining the aging information contained in the original spectrum.

[0067] In this embodiment, the relationship model is divided into three layers: the input layer, the hidden layer and the output layer. Among them, the input layer is the feature quantity of the Raman spectrum of the oil-paper insulation selected; the function expression used in the hidden layer is ψ(X) = exp(-||X - X p || / 2σ 2 ), where X is the Raman spectral feature quantity selected, X p and σ are the center and width parameters of the basis function ψ. The third layer is the output layer, which is the response to the input pattern. The output layer is the aging state of the oil-paper insulation, and its aging state can be expressed as where L is the number of hidden layers, and the number of hidden layers is equal to the number of training samples (in the present invention, L = 12). Divide the number of samples of each aging state above into a training set and a test set according to a ratio of 7:3. Use the training set to establish a relationship model between the Raman spectral features and the aging state, and use the test set to verify the model. The verification results of the final model are shown in Tables 2, 3 and 4.

[0068] Table 2 Prediction results of the aging state of test samples with an oil-paper ratio of 10:1

[0069]

[0070] Table 3 Prediction results of the aging state of test samples with an oil-paper ratio of 15:1

[0071]

[0072] Table 4 Prediction results of the aging state of test samples with an oil-paper ratio of 20:1

[0073]

[0074] The present invention also provides a heat treatment tank for diagnosing the aging of oil-paper insulation based on random forest, such as Figure 3As shown in the figure, an oil inlet pump 1, an oil outlet pump 4, an oil valve 3, and a gas valve 2 are provided at the top of the heat treatment tank. The outside of the heat treatment tank is immersed in dimethyl silicone oil 11 in an aging box 10. The internal filling materials are insulating oil 8 and nitrogen 5 from bottom to top in sequence. A plurality of copper bars 7 are suspended in the heat treatment tank, and insulating paper 9 is provided at the bottom inside the heat treatment tank. The copper bars 7 and the insulating paper 9 are completely immersed in the insulating oil.

[0075] The oil inlet pump and the oil outlet pump prevent air from entering the aging tank during the sampling process, thus affecting the subsequent aging rate; the heat treatment tank provided by the present invention more realistically simulates the sampling process of a real transformer. Through such improvement, the obtained samples are more real and reliable.

[0076] Example 2

[0077] In this example, the data of standard oil samples of three types of oil-paper insulation with different aging states configured by the State Key Laboratory of Power Transmission Equipment & System Security and New Technology, Chongqing University were collected. There were 80 samples for each type, totaling 240 samples. The Raman spectra of the oil samples were measured using a Raman spectrometer; the degree of polymerization of the insulating paper was measured using a viscosity tester.

[0078] For the Raman spectral data of 240 samples, methods such as spike recognition based on derivative spectra and spike removal based on cubic curves, smoothing denoising method based on the median of three-point cyclic fast Fourier transform, and spectrum normalization method based on maximum and minimum values were used to remove the noise and redundant information of the Raman signal.

[0079] The spectral feature information was extracted using a random forest supervised feature extraction algorithm, and the importance of each Raman feature was calculated. Figure 5 (The features have been arranged in descending order) is the importance of each Raman feature. From Figure 5 the Raman features with an importance greater than 0.3 were selected as the input variables for the subsequent model. There were a total of 6 feature variables highly correlated with the aging state here. Figure 6 is the three-dimensional distribution diagram of the top three highly important feature points. It can be seen that through the three features, the sample points in different aging states are not completely clustered together in space, and different aging states can be divided in space. Based on this, the 6 selected feature variables were used as the input of the relationship model between the Raman spectral feature variables and the aging state. Through model training with 56 training samples, a relationship model between the Raman spectral feature variables and the aging state was established. The remaining oil-paper insulation type samples can all be trained and diagnosed using this model, and the diagnosis accuracy rate reaches over 90%. The diagnosis results are shown in Tables 1, 2, 3, and 4.

[0080] The beneficial effects of the present invention are as follows. Compared with the prior art, it can effectively and accurately classify the aging stages of different oil-paper types, and the prediction results for different oil-paper types all reach more than 90%. The present invention has excellent aging state recognition ability for samples in each stage, so it can be well used for the aging state diagnosis of oil-paper insulation of different oil-paper types. In addition, the present invention selects three oil-paper insulation types with oil-paper mass ratios of 10:1, 15:1, and 20:1, and obtains aging samples of different insulation types based on the oil-paper insulation accelerated thermal aging experimental platform. Compared with the 10:1 oil-paper mass ratio insulation type commonly studied in the prior art research, the present invention expands the types of original samples of oil-paper insulation aging and expands the diagnostic objects for equipment of different oil-paper insulation types. To improve the aging diagnosis ability for different oil-paper insulation types, the present invention effectively extracts several Raman spectral characteristic quantities that can reflect the aging state of oil-paper insulation from a large amount of Raman spectrogram data through the random forest algorithm. Therefore, the method provided by the present invention greatly reduces the input parameters for subsequent modeling and the training time of the model, improves the diagnostic ability for aging samples of different oil-paper insulation types, and provides a new method for effectively and accurately realizing the aging stages of different oil-paper insulation types and different aging states of oil-paper insulation equipment.

[0081] The applicant of the present invention has made a detailed description and illustration of the embodiments of the present invention in combination with the accompanying drawings of the specification. However, those skilled in the art should understand that the above embodiments are only the preferred implementation schemes of the present invention, and the detailed description is only to help readers better understand the spirit of the present invention, rather than a limitation on the protection scope of the present invention. On the contrary, any improvement or modification made based on the spirit of the present invention should fall within the protection scope of the present invention.

Claims

1. A diagnosis method for the aging of oil-paper insulation based on random forest, characterized in that, It includes the following steps: Step 1: Simulate the aging state of the transformer, obtain oil-paper insulation samples with different oil-paper ratios and different aging states, and collect the Raman spectra of the oil-paper insulation samples; Step 2: Establish a Raman spectrum database of oil-paper insulation using the Raman spectra obtained in Step 1; Step 3: Based on random forest, screen out multiple Raman spectral feature quantities related to the aging state of oil-paper insulation in the Raman spectrum database; Step 3.1, Use the nth decision tree to classify the data in the OOB dataset to obtain the classification accuracy of the aging state before perturbation n = 1, 2,..., 100; Step 3.2, perturb the Raman spectral feature quantity X in the OOB dataset j where j is the number of Raman spectral dimensions, j = 1, 2, …… 1024, and classify the data in the OOB dataset again using the nth decision tree to obtain the classification accuracy of the aging state after perturbation Step 3.3, repeat Step 3.1 and Step 3.2 to calculate the Raman spectral feature quantity X of each decision tree j The classification accuracy rate of the aging state before and after being perturbed, and use this Raman spectral feature quantity X j The importance for classifying the aging state is defined as Rank the importance of each Raman spectral feature quantity for classifying the aging state according to the calculation results, and screen the Raman spectral feature quantity with importance of 0.3; M takes the value of 0.3 Step 4, establish a relationship model between Raman spectral characteristic quantities and aging states, and judge the aging degree of oil-paper insulation with unknown aging states; divide the oil-paper insulation samples obtained in Step 1 into training samples and test samples; construct a relationship model between Raman spectral characteristic quantities and aging states; train the relationship model with the training samples, and test the trained relationship model with the test samples; collect oil-paper insulation with unknown aging states; the relationship model includes an input layer, a hidden layer and an output layer; among them, the data input into the input layer is the Raman spectral characteristic quantities screened in Step 3; the output layer is the aging state of the oil-paper insulation; the number of hidden layers is equal to the number of training samples; the hidden layer adopts the function Ψ(X)=exp(-||X-X p || / 2σ 2 ), where X is the screened Raman spectral characteristic quantity, X p is the center of the basis function ψ; σ is the width parameter of the basis function ψ; the output layer is expressed as: where L is the number of hidden layers.

2. The method for diagnosing the aging of oil-paper insulation based on random forest according to claim 1, wherein the oil-paper ratios in Step 1 include 10:1, 15:1, and 20:1; the aging states include good insulation, early aging, middle aging, and late aging; In Step 1, oil-paper insulation samples with aging times of 1, 3, 5, 10, 15, 20, 25, and 30 days are respectively obtained to obtain oil-paper insulation samples with different aging states.

3. The method for diagnosing the aging of oil-paper insulation based on random forest according to claim 2, wherein Step 1 further includes preprocessing the Raman spectra; the preprocessing includes spike identification based on derivative spectra, spike removal method based on cubic curves, smoothing denoising method based on the median of three-point cyclic fast Fourier transform, and spectrum normalization based on maximum and minimum values.

Images