A transformer fault diagnosis method, medium and system

Abstract

The application discloses a transformer fault diagnosis method, medium and system, comprising: obtaining a sample set of transformer faults, wherein each sample in the sample set contains the content of different gases dissolved in transformer oil; calculating the ratio of the content of different gases contained in each sample and the relative content of different gases as data features of multiple different feature types of the sample; performing dimension reduction on the data features of each sample by ANOVA variance analysis to obtain a sample after dimension reduction; training an improved support vector machine by the sample after dimension reduction by using a Bonobo Optimizer algorithm and a K-fold cross-validation method; and diagnosing transformer faults by using the trained improved support vector machine. The application can detect transformer abnormalities, fault information and the like in time, and significantly improves the online evaluation and early warning capability of transformer and distribution equipment fault risks.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of transformer fault diagnosis, and in particular to a transformer fault diagnosis method, medium and system. BACKGROUND

[0002] With the construction and development of data platforms, power equipment state monitoring, production management, operation scheduling, environmental meteorological data will gradually be integrated and shared on a unified information platform, thereby promoting the development of power equipment state evaluation, diagnosis and prediction technology towards comprehensive analysis based on panoramic state. However, there are many factors affecting the operation state of power equipment, and the future explosive growth of state monitoring data plus the huge amount of information data related to the state of the device, such as power grid operation and weather environment, the existing method will be difficult to screen and integrate these data, resulting in failure to timely detect transformer faults. SUMMARY

[0003] The embodiments of the present application provide a transformer fault diagnosis method, medium and system to solve the problem that the prior art is difficult to integrate transformer monitoring data, resulting in failure to timely detect transformer faults.

[0004] In a first aspect, a transformer fault diagnosis method is provided, comprising:

[0005] Obtaining a sample set of transformer faults, wherein each sample in the sample set contains the content of different gases dissolved in transformer oil;

[0006] Calculating the ratio of the content of different gases contained in each sample and the relative content of different gases as data features of multiple different feature types of the sample;

[0007] Performing dimensionality reduction on the data features of each sample using ANOVA variance analysis to obtain a reduced sample;

[0008] Training an improved support vector machine using the Bonobo Optimizer algorithm and K-fold cross-validation method through the reduced sample;

[0009] Diagnosing transformer faults using the trained improved support vector machine.

[0010] In a second aspect, a computer readable storage medium is provided, and the computer readable storage medium stores computer program instructions; the computer program instructions are executed by a processor to implement the transformer fault diagnosis method of the first aspect as described above.

[0011] In a third aspect, a transformer fault diagnosis system is provided, comprising the computer readable storage medium of the second aspect as described above.

[0012] Thus, the embodiment of the present application proposes an effective transformer fault diagnosis technical solution from the aspects of model feature input and model parameter optimization, adopts ANOVA variance analysis to perform transformer fault feature screening, inputs the screened features into an improved support vector machine, and outputs a transformer fault type, so that the transformer abnormality, fault and other information can be detected in time, and the online evaluation and early warning capability of the transformer fault risk is significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0013] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the description of the embodiments of the present application will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0014] Figure 1 is a flowchart of the transformer fault diagnosis method of the embodiment of the present application;

[0015] Figure 2 is a comparison diagram of the accuracy rates of different preset numbers of fault diagnosis models;

[0016] Figure 3 is a comparison diagram of model fault diagnosis results of different DGA methods, wherein (a) is a fault diagnosis result based on a Rogers and BO-SVM model, (b) is a fault diagnosis result based on an IEC and BO-SVM model, and (c) is a fault diagnosis result based on an ANOVA and BO-SVM model;

[0017] Figure 4 is a convergence curve diagram of different algorithms. DETAILED DESCRIPTION

[0018] The technical solutions of the embodiments of the present application will be described clearly and completely in combination with the drawings of the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, but not all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0019] The embodiment of the present application discloses a transformer fault diagnosis method. As shown in the figure, Figure 1 the method comprises the following steps:

[0020] Step S101: acquiring a sample set of transformer faults.

[0021] Each sample in the sample set contains the content of different gases dissolved in transformer oil.

[0022] The gas in the embodiment of the application includes: H2, CH4, C2H4, C2H2, C2H6, and each sample contains the content of H2, the content of CH4, the content of C2H4, the content of C2H2 and the content of C2H6 dissolved in transformer oil.

[0023] Step S102: calculating the ratio of the content of different gases and the relative content of different gases contained in each sample as a plurality of different feature types of data features of the sample.

[0024] Specifically, the ratio of the content of different gases includes: CH4 / H2, C2H2 / C2H4, C2H4 / C2H6, C2H6 / CH4, C2H2 / CH4 and C2H2 / H2.

[0025] Specifically, the relative content of different gases includes: CH4 / C1, C2H4 / C1, C2H2 / C1, H2 / C2, CH4 / C2, C2H6 / C2, C2H4 / C2 and C2H2 / C2. Wherein, C1 is the total content of CH4, C2H2 and C2H4, and C2 is the total content of H2, CH4, C2H2, C2H4 and C2H6.

[0026] In this way, 14 feature types are selected in the embodiment of the application.

[0027] Step S103: performing dimension reduction on the data features of each sample by ANOVA variance analysis to obtain the sample after dimension reduction.

[0028] The transformer fault diagnosis method based on the dissolved gas analysis (DGA) technology has been widely applied in engineering practice, however, the evaluation indexes selected by different methods are quite different, and there is incomplete or redundant feature information, which makes it have certain limitations in actual application. Based on this, the embodiment of the application selects the high-dimensional complex dissolved gas features of transformer oil by ANOVA (ANOVA), selects the data features with high correlation and large effect on pattern recognition as the input of the model, which can not only better preserve the original data information, but also avoid the redundancy of input features.

[0029] ANOVA judges whether the categorical independent variable has a significant effect on the numerical dependent variable by testing whether the means of each population are equal. In variance analysis, the object to be tested is the factor, the different performances of the factor are called levels, and the sample data obtained under each factor level are called observation values. When performing variance analysis, the source of data error needs to be investigated, wherein the error of data is represented by sum of squares. The sum of squares reflecting the size of all data errors is called total sum of squares (SST), the difference between the means of each sample is called between-group sum of squares (SSA), and the dispersion of each observation value of each sample is called within-group sum of squares (SSE).

[0030] In the single factor variance analysis of the transformer fault features, the factor refers to the set of all transformer fault types, the level refers to each fault type, and the observation value refers to the specific data features of each fault sample.

[0031] Based on the above analysis, specifically, this step includes the following processes:

[0032] (1) The data features of the same transformer fault type and the same feature type in all samples are divided into the same data set.

[0033] It should be understood that the transformer fault type of the embodiment of the present application includes the transformer normal type, so as to detect whether the transformer is faulty.

[0034] For example, the data features of the transformer fault type of medium and low temperature overheating and the feature type of CH4 / H2 are divided into the same data set.

[0035] (2) The inter-group sum of squares and the intra-group sum of squares of the data set of the same feature type are calculated.

[0036] Specifically, the calculation formula of the inter-group sum of squares is:

[0037] Specifically, the calculation formula of the intra-group sum of squares is:

[0038] Wherein, SSA represents the inter-group sum of squares of the same feature type, SSE represents the intra-group sum of squares of the same feature type, n represents the number of data features in the data set of the i-th transformer fault type under the same feature type, x represents the j-th data feature in the data set of the i-th transformer fault type under the same feature type, represents the mean value of the data features in the data set of the i-th transformer fault type under the same feature type, and represents the mean value of the data features in the data set of all transformer fault types under the same feature type. i ij

[0039] (3) The statistic quantity of the feature type is calculated according to the inter-group sum of squares and the intra-group sum of squares of the data set of the same feature type.

[0040] In order to eliminate the influence of the number of observation values on the size of the error sum of squares, the SSA and the SSE need to be averaged, the degree of freedom of the SSA is k-1, the degree of freedom of the SSE is n-k, and then The statistic quantity is substituted into and ​​​​​The following formula is obtained for calculating the statistical quantity:

[0041]

[0042] wherein F represents the statistical quantity of the same feature type, k-1 represents the degree of freedom of SSA, and n-k represents the degree of freedom of SSE,

[0043] The above process is described with the feature type being CH4 / H2, as shown in Table 1.

[0044] Table 1 CH4 / H2 data feature statistics

[0045]

[0046]

[0047] The mean of the data features in the data set of all transformer fault types under this feature type is:

[0048]

[0049] The inter-group sum of squares of this feature type is:

[0050]

[0051] The transformer fault type is medium and low temperature overheating:

[0052] The transformer fault type is high temperature overheating:

[0053] The transformer fault type is partial discharge:

[0054] The transformer fault type is low energy discharge:

[0055] The transformer fault type is high energy discharge:

[0056] The transformer is normal:

[0057] The above data is added to obtain the intra-group sum of squares of this feature type:

[0058]

[0059] The statistical quantity of this feature type is:

[0060]

[0061] (4) The statistical quantities of all feature types are arranged in descending order, and the data features of the top pre-set number of feature types are selected as the data features of each sample after dimension reduction.

[0062] The greater the value of the statistical quantity F, the more significant the influence of the transformer fault type on the characteristics of the dissolved gas in the transformer oil. Therefore, after arranging the statistical quantities in descending order, the data features of the top pre-set number of feature types are selected. The selection of the pre-set number will greatly affect the diagnostic performance of the model, and the pre-set number can be determined by comparison and simulation.

[0063] Taking 256 groups of samples of a transformer as an example, variance analysis is used to screen and reduce the data features of fault diagnosis, and the data features after dimension reduction are used as the input of the final model. The calculated statistical quantities are shown in Table 2.

[0064] Table 2 Statistical quantities of each feature type

[0065] Feature label Ratio F-value X1 <![CDATA[CH4 / H2]]> 18.159 X2 [C2H2 / C2H4] 49.836 X3 [C2H4 / C2H6] 2.974 X4 [C2H6 / CH4] 8.650 X5 [C2H2 / CH4] 3.960 X6 [C2H2 / H2] 9.215 X7 CH4 / C1 23.674 X8 <![CDATA[C2H4 / C1]]> 28.433 X9 [C2H2 / C1] 65.829 X 10 ]]> [H2 / C2] 69.566 X 11 ]]> CH4 / C2 44.359 X 12 ]]> [C2H6 / C2] 38.446 X 13 ]]> [C2H4 / C2] 58.091 X 14 ]]> [C2H2 / C2] 47.193

[0066] The feature types are arranged in descending order of statistical quantity as follows: X 10 , X9, X 13 , X2, X 14 , X 11 , X 12 , X8, X7, X1, X6, X4, X5, X3.

[0067] The pre-set number is determined by simulation, and it is verified that the fault input after dimension reduction and screening by variance analysis can help improve the diagnostic accuracy of the model. Specifically, the comparison of the accuracy of fault diagnosis models with different pre-set numbers is shown in Table 3. Figure 2 The results show that when the pre-set number is 8 (the data features after dimension reduction are X2, X8, X9, X 10 , X 11 , X 12 , X 13 , X 14 ), the model fault diagnosis accuracy is the highest (91.89%). Compared with the pre-set number of 14, the model diagnosis accuracy of the pre-set number of 8 is improved by 9.46% compared with the accuracy before dimension reduction (82.43%), which verifies the effectiveness of the ANOVA method for screening.

[0068] Step S104: The Bonobo Optimizer algorithm and K-fold cross-validation method are used to train the improved support vector machine through the samples after dimension reduction.

[0069] Support vector machine (SVM) is a binary classification model, given a training sample set D = {(x1, y1), (x2, y2), …, (x m , ym )},y i ∈{-1,+1}, find a partitioning hyperplane w in the sample space based on the training sample set D T x + b = 0, separating samples of different categories. The training sample points closest to the hyperplane are support vectors. The sum of the distances of two different support vectors to the hyperplane is d = 2 / ||w||, which is the margin. To find the dividing hyperplane with the largest margin, we need to find w and b that satisfy the constraints and maximize d, as shown in the following formula:

[0070]

[0071] The above is the basic form of the support vector machine, where w is the normal vector and b is the displacement term.

[0072] The basic model of the support vector machine is based on the assumption that the training samples are linearly separable, that is, there exists a dividing hyperplane that correctly classifies the training samples. However, in transformer fault diagnosis, there is not necessarily a hyperplane in the original sample space that can correctly divide the two types of samples. It is necessary to introduce kernel functions and soft intervals. The kernel function can map the samples from the original space to a higher-dimensional feature space, making the training samples linearly separable in the feature space, while the soft interval allows the hyperplane division to be wrong on some samples. Therefore, the improved support vector machine of the embodiment of the present invention updates the above formula to:

[0073]

[0074] sty i (w T φ(x i )+b)≥1-ξ i .

[0075] ξ i ≥0,i=1,2,...,m

[0076] Where C is the penalty factor; ξ i is the slack variable; φ(x i ) means to convert x i This is a quadratic programming problem, and its dual problem is obtained by the Lagrange multiplier method, as shown in the following formula:

[0077]

[0078]

[0079] KKT requirements are as follows:

[0080]

[0081] Solving the following can be obtained:

[0082]

[0083] Wherein, α i , α j And μ i Indicate the Lagrange multiplier.K(x, x i )Indicate the kernel function.

[0084] In order to ensure that the sample is linearly separable in the feature space, the definition of the feature space is particularly important to the performance of the support vector machine, and the kernel function implicitly defines the feature space, so the selection of the kernel function becomes an important factor affecting the performance.The radial basis kernel function not only has good performance, but also has few parameters and stable performance.Therefore, the kernel function of the embodiment of the application is the radial basis kernel function, as follows:

[0085] K(x i , x j )=exp(-γ||x i -x j || 2 ), γ>0.

[0086] Wherein, γ is the kernel parameter, which determines the bandwidth σ of the radial basis kernel, and the relationship is

[0087] In addition to the selection of the kernel function, the selection of the penalty factor will also affect the generalization ability of the support vector machine.The greater the penalty factor represents the smaller the tolerance of the classifier to the noise points in the boundary, and the higher the classification accuracy, but it is easy to overfit and has poor generalization ability.

[0088] Therefore, the determination of the kernel parameter and the penalty factor is closely related to the improvement of the classification performance of the support vector machine.The embodiment of the application adopts the Bonobo Optimizer algorithm to optimize the parameters of the improved support vector machine, so that the diagnostic performance of the classifier reaches the best.

[0089] Specifically, the step includes the following processes:

[0090] (1) all dimensionality reduced samples are randomly divided into K groups of training data with the same number by K-fold cross validation method, wherein K-1 groups are used as training set, and the remaining 1 group is used as validation set, and the objective function based on the validation set is established.

[0091] Specifically, the objective function is:

[0092]

[0093] Wherein, f indicates the objective function, NT i Indicates the number of samples correctly classified in the validation set, and Ni represents the number of samples in the validation set.

[0094] (2) Initialize the Bonobo population of the Bonobo Optimizer algorithm, set the population size, dimension, and maximum number of iterations.

[0095] wherein each Bonobo individual in the Bonobo population has two dimensions, one dimension is the kernel parameter of the improved support vector machine, and the other dimension is the penalty factor of the improved support vector machine.

[0096] The initialized parameters include: phase probability p p , direction probability p d , positive phase count ppc, negative phase count npc, out-group mating probability p xgm , and the like.

[0097] (3) Train the support vector machine corresponding to the kernel parameter and the penalty factor of each Bonobo individual by K-fold cross-validation method using K groups of training data, and solve the solution of the corresponding objective function, and optimize the solution of the objective function by the Bonobo Optimizer algorithm to obtain the optimal kernel parameter and penalty factor, which are used as the kernel parameter and penalty factor of the trained improved support vector machine.

[0098] The fitness value of the Bonobo Optimizer algorithm is calculated by the fitness function . The fitness function is used as the average classification accuracy of the K-fold cross-validation method.

[0099] During the optimization process, for each group of kernel parameters and penalty factors, the K-fold cross-validation method will cross-validate and repeat K times. Therefore, the fitness function is the average value of K times. Specifically, taking one Bonobo individual as an example, each group of data in K groups of training data is used as a validation set, and the remaining data is used as a training set, which will produce K combinations of training sets and validation sets. The training sets of the K combinations are used to train the improved support vector machine corresponding to the kernel parameter and the penalty factor of the Bonobo individual, so the training is performed K times. After each training, the corresponding validation set is used for verification, and the diagnosis result of the transformer fault is output. In this way, K diagnosis results are obtained, which are substituted into the fitness function to calculate the fitness value, and the objective function is solved.

[0100] The smaller the fitness value is, that is, the larger the objective function is, the closer the Bonobo individual is to the optimal solution. Therefore, during the optimization process of the Bonobo Optimizer algorithm, the dominant alpha Bonobo obtained by each iteration is the Bonobo individual with the maximum objective function (or the smallest fitness value).

[0101] After initialization, the iteration process of the Bonobo Optimizer algorithm is as follows:

[0102] The fission fusion strategy is used to determine p-th Bonobo X p is used to mate with i-th Bonobo X i Mating; according to the size relationship between the random number and the phase probability, different mating strategies are adopted to create new Bonobo individuals; the fitness values of all newly generated Bonobo individuals are calculated to determine the dominant alpha Bonobo X , the algorithm parameters are updated adaptively; it is judged whether the current dominant alpha Bonobo X , the optimal solution of the algorithm, the kernel parameters and the penalty factor of the improved support vector machine reach the convergence standard, if not, the next iteration is performed; if the convergence standard is reached, the kernel parameters and the penalty factor corresponding to the dominant alpha Bonobo are determined as the optimal kernel parameters and the penalty factor. Wherein, the convergence standard is: reaching the maximum iteration number or the dominant alpha Bonobo individual with the same optimal fitness value accounts for 99% of the total population. The above fission fusion strategy; different mating strategies are adopted according to the size relationship between the random number and the phase probability, which are all known strategies of the algorithm, and will not be described here.

[0103] Through this step, the powerful intelligent search characteristics of the Bonobo Optimizer algorithm are used to optimize a certain search range, the algorithm forms several initial solutions in the search space, the iteration is constantly modified according to the target function value, the parameters are updated adaptively, the optimal solution is approached, and finally the combination of (γ, C) that makes the SVM achieve the best performance is found. The improved support vector machine model is established according to the optimal parameter combination of the improved support vector machine.

[0104] Step S105: using the trained improved support vector machine to diagnose transformer faults.

[0105] It should be understood that when diagnosing transformer faults, the contents of the five kinds of gases in the transformer oil are also collected in real time, and the data characteristics corresponding to the reduced feature types are calculated accordingly. These data characteristics are input into the improved support vector machine, and the diagnosis result of the transformer fault is output.

[0106] The embodiment of the application also discloses a computer readable storage medium, which stores computer program instructions; the computer program instructions are executed by a processor to realize the transformer fault diagnosis method described in the above embodiment.

[0107] The embodiment of the application also discloses a transformer fault diagnosis system, which comprises a computer readable storage medium as described in the above embodiment.

[0108] The technical solutions of the application are further described below with a specific application example.

[0109] The performance of the transformer fault diagnosis model based on variance analysis and improved support vector machine (BO-SVM) of the embodiment of the application is tested with the data of a certain transformer, and is compared with other DGA methods (IEC, Rogers method) and other improved support vector machine methods (GA-SVM, PSO-SVM), as follows:

[0110] To illustrate the superiority of the feature screening method based on variance analysis, the traditional DGA-based transformer fault diagnosis methods (Rogers four ratios, IEC three ratios method) are compared. The features of each DGA method are input into the BO-SVM model of the embodiment of the application for fault diagnosis, and the diagnosis results are as shown in Figure 3 , and the comparison results of the model feature input and diagnosis accuracy are as shown in Table 3.

[0111] Table 3 Comparison of the influence of different DGA methods on fault diagnosis performance

[0112]

[0113] From Figure 3 and Table 3, it can be seen that the fault diagnosis accuracy of the embodiment of the application based on ANOVA and BO-SVM is 90.54%, and among the 74 test samples, there are 7 fault types misjudged; and the fault diagnosis accuracy of the BO-SVM with the feature input of the Rogers four ratio method and the IEC three ratio method is 81.08% and 77.03% respectively, and there are 14 and 17 fault types misjudged respectively. The results show that the feature screening method based on ANOVA can significantly improve the fault diagnosis accuracy of the model compared with the traditional method, verifying the superiority of the method.

[0114] To compare the optimization ability of the chimpanzee optimization BO algorithm with the GA and PSO algorithms for the parameters of the transformer fault diagnosis model and the performance of the model based on different optimization algorithms, the features screened based on ANOVA are input into the BO-SVM for classification test, and are compared with the GA and PSO optimized SVM for analysis, and the convergence curves are as shown in Figure 4 , and the convergence results are as shown in Table 4.

[0115] Table 4 Comparison of convergence results of different algorithms

[0116] Method Parameter (C, γ) Convergence algebra Convergence objective function value GA-SVM (169.415,9.775e-04) 119 0.432 PSO-SVM (111.483,2.713e-06) 42 0.135 BO-SVM (196.902,2.325e-06) 14 0.081

[0117] To better distinguish the optimization ability of the chimpanzee optimization BO algorithm from the comparison algorithms, the parameter search range is expanded. The search range of the kernel parameter is (0, 1), and the search range of the penalty factor is (1, 500). The population number is set to 40, and the maximum iteration number is set to 200.Figure 4 In a, GA-SVM convergence curve is indicated, in b, PSO-SVM convergence curve is indicated, in c, BO-SVM convergence curve is indicated, and the convergence speed of the BO-SVM convergence curve is the fastest, and the global optimum is found at the 14th generation, and the fitness function value is 0.081. The PSO-SVM convergence curve has a slower convergence speed, and convergence is completed at the 42nd generation, and the fitness function value is 0.135. The GA-SVM convergence curve completes convergence at the 119th generation, and the fitness function value is 0.432, and obviously, the GA algorithm falls into a local optimum. For a larger search range, the GA algorithm is unstable, and cannot complete the global optimization task, and the PSO and BO algorithms can stably find the global optimum. By comparison between the two algorithms, the chimpanzee optimization BO algorithm is obviously superior to the PSO algorithm in terms of convergence speed and diagnosis accuracy, and the superiority of the BO-SVM is proved.

[0118] The test results show that the transformer fault diagnosis method based on the improved support vector machine under the fault feature screening can accurately make accurate diagnosis on the transformer fault information. Compared with the other two algorithms, the improved support vector machine, the fault diagnosis accuracy of the embodiment of the application is improved by about 5%, and the transformer fault diagnosis has certain practical significance.

[0119] In summary, the embodiment of the application proposes an effective transformer fault diagnosis technical scheme from the aspects of model feature input and model parameter optimization, the transformer fault feature is screened by ANOVA variance analysis, the screened features are input into the improved support vector machine, and the output is a transformer fault type, abnormal information, fault information and the like of the transformer can be found in time, and the online evaluation and early warning capability of the transformer fault risk is significantly improved.

[0120] The above merely describes specific embodiments of the application, but the protection scope of the application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the application, and all should be covered within the protection scope of the application. Therefore, the protection scope of the application should be subject to the protection scope of the claims.

Claims

1. A transformer fault diagnosis method, characterized in that: include: Obtaining a sample set of transformer faults, wherein each sample in the sample set includes the content of different gases dissolved in transformer oil; Calculating the ratio of the contents of different gases and the relative contents of the different gases contained in each sample as data features of multiple different feature types of the sample; ANOVA analysis of variance is used to reduce the dimension of the data features of each sample to obtain a sample after dimension reduction; Improve the support vector machine by training the sample after the dimension reduction using the Bonobo Optimizer algorithm and the K-fold cross validation method; Using the trained improved support vector machine to diagnose transformer faults; The step of improving the support vector machine by training the sample after dimensionality reduction using the Bonobo Optimizer algorithm and the K-fold cross-validation method includes: All the samples after dimensionality reduction are randomly divided into K groups of training data of equal number by K-fold cross validation method, wherein K-1 groups are used as training sets and the remaining group is used as validation sets, and an objective function based on the validation sets is established; Initializing a chimpanzee population of the Bonobo Optimizer algorithm, setting the population size, dimension, and maximum number of iterations, wherein each chimpanzee individual in the chimpanzee population has two dimensions, one dimension being the kernel parameter of the improved support vector machine, and the other dimension being the penalty factor of the improved support vector machine; The K groups of training data are used to train a support vector machine with kernel parameters and penalty factors corresponding to each chimpanzee individual through a K-fold cross-validation method, and the solution of the corresponding objective function is solved. The chimpanzee optimization Bonobo Optimizer algorithm is used to optimize the solution of the objective function to obtain the corresponding optimal kernel parameters and penalty factors, which are used as the kernel parameters and penalty factors of the improved support vector machine after training.

2. The transformer fault diagnosis method according to claim 1, characterized in that: The gases include: H2, CH4, C2H4, C2H2, and C2H6; the ratios of the contents of the different gases include: CH4 / H2, C2H2 / C2H4, C2H4 / C2H6, C2H6 / CH4, C2H2 / CH4, and C2H2 / H2; the relative contents of the different gases include: CH4 / C1, C2H4 / C1, C2H2 / C1, H2 / C2, CH4 / C2, C2H6 / C2, C2H4 / C2, and C2H2 / C2, wherein C1 is the total content of CH4, C2H2, and C2H4, and C2 is the total content of H2, CH4, C2H2, C2H4, and C2H6.

3. The transformer fault diagnosis method according to claim 1, characterized in that: The step of using ANOVA variance analysis to reduce the dimension of the data features of each sample includes: Dividing data features of the same transformer fault type and the same feature type in all the samples into the same data set; Calculate the between-group sum of squares and within-group sum of squares for data sets of the same feature type; The statistics of the feature type are calculated based on the between-group sum of squares and the within-group sum of squares of the data set with the same feature type; Arrange the statistics of all feature types in descending order, and select the data features of a preset number of feature types as the data features after dimensionality reduction for each sample.

4. The transformer fault diagnosis method according to claim 3, characterized in that: The calculation formula for the sum of squares between groups is: ; The calculation formula for the within-group sum of squares is: ; in, SSA represents the sum of squares between groups of the same feature type, SSE represents the within-group sum of squares of the same feature type, Indicates the first i The number of data features in the dataset of class transformer fault types, Indicates the first i The first j Data features, Indicates the first i The mean of the data features in the dataset of class transformer fault types, Represents the mean value of the data feature in the dataset for all transformer fault types under the same feature type.

5. The transformer fault diagnosis method according to claim 4, characterized in that: The calculation formula of the statistic is: ； in, F Represents statistics of the same feature type, k -1 means SSA degrees of freedom, n - k express SSE degrees of freedom, .

6. The transformer fault diagnosis method according to claim 1, characterized in that: The objective function is: ； in, f represents the objective function, Indicates the number of correctly classified samples in the validation set, Indicates the number of samples in the validation set.

7. The transformer fault diagnosis method according to claim 1, characterized in that: During the optimization process of the BonoboOptimizer algorithm, the dominant alpha chimpanzee obtained in each iteration is the chimpanzee individual with the largest objective function.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions; when the computer program instructions are executed by a processor, the transformer fault diagnosis method according to any one of claims 1 to 7 is implemented.

9. A transformer fault diagnosis system, characterized in that: include: The computer-readable storage medium of claim 8.

Images