A Hybrid Semi-Supervised Transformer Fault Diagnosis Method

Abstract

The present invention provides a hybrid semi-supervised transformer fault diagnosis method, including establishing a semi-supervised iterative transformer fault diagnosis model with filter-based cross-validation feature optimization combined with the XYZ strategy, establishing a transformer fault diagnosis model based on wrapper-based feature optimization, and establishing a hybrid semi-supervised feature optimization transformer fault diagnosis model. The present invention combines the advantages of filter-based and wrapper-based feature optimization to reduce the feature complexity, enhances the discriminability of features compared with a single feature selection method, uses a small amount of DGA labeled data as guidance, and utilizes a large amount of DGA unlabeled data to improve the learning performance, thus solving the transformer fault diagnosis problem in the case of scarce fault labeled samples.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer fault diagnosis, and in particular to a hybrid semi-supervisory transformer fault diagnosis method. Background Art

[0002] Dissolved Gas Analysis (DGA) is the most commonly used method for on-site power transformer fault diagnosis. The various DGA-based transformer fault diagnosis criteria proposed in recent years, both domestically and internationally, are based on field experience. These criteria often result in misjudgments and missed detections due to issues such as incomplete encoding and overly absolute boundaries, causing numerous problems for maintenance. In recent years, with the widespread development of artificial intelligence (AI) technologies in various fields, including support vector machines, decision trees, neural networks, clustering, and fuzzy logic reasoning, have been widely used in transformer fault diagnosis, achieving promising results. However, these technologies require a large amount of DGA data labeled with fault types for model building. Models built with insufficient labeled data often suffer from poor generalization performance and reduced fault diagnosis rates.

[0003] In reality, reliable labeled fault samples available in the field are very limited. However, unlabeled DGA data is readily available. Semi-supervised classification methods can use a small amount of labeled DGA data as guidance while leveraging a large amount of unlabeled DGA data to improve learning performance and increase classification accuracy. This makes them a powerful mathematical model for solving transformer fault diagnosis problems when DGA fault samples are scarce.

[0004] Most existing semi-supervised classification techniques for transformer fault diagnosis use features such as specific gas concentrations, ratios of various gases, or relative percentages. This high feature complexity severely impacts the performance of the initial semi-supervised learning model, causing errors caused by feature redundancy to accumulate during the semi-supervised learning iterations, making it difficult to improve classifier performance. Few semi-supervised transformer fault diagnosis techniques use a single metric or the same type of feature selection method to evaluate the discriminative ability of each feature, but this can easily lead to suboptimal features. Summary of the Invention

[0005] The present invention provides a hybrid semi-supervised transformer fault diagnosis method to overcome the defect that the existing semi-supervised feature selection method for transformer fault diagnosis cannot fairly and impartially utilize all information to evaluate the discriminative ability of features.

[0006] This specification discloses a hybrid semi-supervised transformer fault diagnosis method, comprising:

[0007] S1. Establish a semi-supervised iterative transformer fault diagnosis model based on filtered cross-validation feature selection, as follows:

[0008] Obtain DGA data and divide it into a training set S T and a validation set S V , where the training set S T includes an initial labeled sample set S I and an unlabeled sample set S U ;

[0009] Use the initial labeled sample set S I to train an initial classifier and establish a multi-classification model for transformer fault diagnosis based on support vector machines;

[0010] Use the multi-classification model for transformer fault diagnosis based on support vector machines to predict the fault categories of the unlabeled sample set S U and the validation set S V . Based on one-vs.-many, calculate the multiple posterior probabilities of the multi-classification model for transformer fault diagnosis based on support vector machines on each sample in the unlabeled sample set S U , and then average the posterior probabilities of each sample;

[0011] S2. Establish a semi-supervised iterative model that combines the XYZ strategy for filter-based cross-validation feature selection, as follows:

[0012] In the m th iteration, use the X strategy to select samples with high confidence in the unlabeled sample set S U and integrate them with the labeled sample set S I to reconstitute a labeled training set S L ( m );

[0013] Use the labeled training set S L ( m ) to retrain the multi-classifier of the support vector machine and re-predict the fault categories of the unlabeled sample set S U and the validation set S V ;

[0014] Recalculate the posterior probabilities of each sample in the unlabeled sample set S U ;

[0015] Find the labels predicted for the unlabeled sample set S m in the m th iteration and the U -1th iteration, calculate the number of different labels in the two predictions. If the number of differences is less than 1, stop the iteration; if the number is greater than 1, perform the m +1th iteration until the upper limit value of m is satisfied;

[0016] After the iteration stops, obtain the final classification rate of the validation set S V ;

[0017] According to the new training set S L ( m )'s feature space and the corresponding final labels, use the filtering method to sort the feature space, and select the optimal features using Strategy Y;

[0018] After r * K k-fold cross-validation, calculate the average classification rate on the validation set S V and select the optimal features using Strategy Z;

[0019] S3. Establish a transformer fault diagnosis model based on wrapper feature selection, as follows:

[0020] Establish a feature optimization model combining genetic algorithm and support vector machine, and synchronously optimize the parameters of the support vector machine and the input DGA features;

[0021] Set the parameters of the genetic algorithm, including population, number of iterations, and chromosome length;

[0022] Encode the parameters of the support vector machine and the input DGA features, perform randomized initial population, chromosome decoding, individual fitness ranking, genetic operations, generate offspring individuals, offspring chromosome decoding, and calculate the fitness of offspring chromosome individuals;

[0023] When the stopping condition is reached, output the optimal feature space, the optimal support vector machine parameters, and the cross-validation rate;

[0024] S4. Establish a transformer fault diagnosis model based on hybrid semi-supervised feature selection, as follows:

[0025] Construction of the feature space: Select specific types of gas concentrations, various gas ratios, and relative percentages recommended by IEC or IEEE as the initial feature space;

[0026] Reconstruct the original DGA data, and based on the dataset formed by the initial feature space, use the synthetic minority over-sampling technique to change the data imbalance;

[0027] Dataset preprocessing: Perform arctangent transformation and normalization on the feature space of the dataset with balanced data distribution;

[0028] Establish a semi-supervised iterative model based on filter cross-validation feature selection, use the obtained final classification rate as the value of individual fitness in the transformer fault diagnosis model based on wrapper feature selection, and re-encode the chromosome using the obtained optimal feature space;

[0029] Among them, Strategy X is: If the posterior probability of the predicted classification of the samples in the unlabeled sample set S U is greater than that of the unlabeled sample set SU The average posterior probability of the predicted classification of all samples is considered, and this sample is considered credible. The credible samples are added to the training set S T Reconstitute the labeled training set S L ( m )

[0030] The Y strategy is: according to the sorting, select the feature spaces of each proportion, and based on S L ( m ) for training, and verify on the validation set S V Select the proportion corresponding to the maximum classification rate and replace the encoding with the newly formed feature space;

[0031] The Z strategy is: select from the optimal feature space matrix obtained by r * K k-fold cross-validation, and select the feature space corresponding to the smallest difference from the average classification rate as the optimal feature.

[0032] In summary, the present invention has at least the following beneficial effects:

[0033] The present invention combines the advantages of filter-based and wrapper-based feature selection to reduce the feature complexity, and compared with a single feature selection method, enhances the discriminability of features; uses a small amount of DGA labeled data as guidance and utilizes a large amount of DGA unlabeled data to improve the learning performance, and solves the problem of transformer fault diagnosis in the case of scarce fault-labeled samples. Description of the Drawings

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0035] Figure 1 It is a schematic flow chart of the hybrid semi-supervised transformer fault diagnosis method involved in the present invention. Detailed Embodiments

[0036] In the following, only some exemplary embodiments are simply described. As those skilled in the art can recognize, the described embodiments can be modified in various different ways without departing from the spirit or scope of the embodiments of the present invention. Therefore, the drawings and descriptions are considered to be exemplary rather than restrictive in nature.

[0037] The following disclosure provides many different embodiments or examples for implementing different structures of the embodiments of the present invention. To simplify the disclosure of the embodiments of the present invention, the components and settings of specific examples are described below. Of course, they are only examples and are not intended to limit the embodiments of the present invention. In addition, the embodiments of the present invention may repeat reference numerals and / or reference letters in different examples. This repetition is for the purpose of simplification and clarity, and does not itself indicate the relationship between the various embodiments and / or settings discussed.

[0038] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0039] As Figure 1 shown, this embodiment provides a hybrid semi-supervised transformer fault diagnosis method, including:

[0040] S1. Establish a semi-supervised iterative transformer fault diagnosis model based on filter cross-validation feature selection, specifically as follows:

[0041] Obtain DGA data and divide it into a training set S T and a validation set S V , where the training set S T includes an initial labeled sample set S I and an unlabeled sample set S U ;

[0042] Use the initial labeled sample set S I to train an initial classifier and establish a multi-class transformer fault diagnosis model based on support vector machines;

[0043] Use the multi-class transformer fault diagnosis model based on support vector machines to predict the fault categories of the unlabeled sample set S U and the validation set S V . Calculate the multiple posterior probabilities of the multi-class transformer fault diagnosis model based on support vector machines on each sample in the unlabeled sample set S U using the one-vs.-rest method, and then average the posterior probabilities of each sample;

[0044] S2. Establish a semi-supervised iterative model that combines the XYZ strategy with filter cross-validation feature selection, specifically as follows:

[0045] In the m th iteration, use the X strategy to select samples with high confidence in the unlabeled sample set S U and integrate them with the labeled sample set S I to reconstitute the labeled training set S L ( m );

[0046] Use the labeled training set S L (m ) Retrain the multi-classifier of the support vector machine and re-predict the unlabeled sample set S U and the validation set S V 's fault categories;

[0047] Recalculate the posterior probabilities of each sample in the unlabeled sample set S U ;

[0048] Find the m -th iteration and the m -1-th iteration of the predicted labels for the unlabeled sample set S U , calculate the number of different labels between the two times before and after. If the number of differences is less than 1, the iteration stops; if the number is greater than 1, perform the m +1-th iteration until the upper limit value of m is satisfied;

[0049] After the iteration stops, obtain the final classification rate of the validation set S V ;

[0050] According to the feature space of the new training set S L ( m ), sort the feature space using the filtering method, and select the optimal features using the Y strategy;

[0051] After r * K -fold cross-validation, calculate the average classification rate on the validation set S V , and select the optimal features using the Z strategy;

[0052] S3. Establish a transformer fault diagnosis model based on wrapper feature selection, as follows:

[0053] Establish a feature optimization model combining genetic algorithm and support vector machine, and synchronously optimize the parameters of the support vector machine and the input DGA features;

[0054] Set the parameters of the genetic algorithm, including population, number of iterations, and chromosome length;

[0055] Encode the parameters of the support vector machine and the input DGA features, perform randomized initial population, chromosome decoding, individual fitness ranking, genetic operations, generation of offspring individuals, offspring chromosome decoding, and calculation of offspring chromosome individual fitness;

[0056] When the stop condition is reached, output the optimal feature space, the optimal support vector machine parameters, and the cross-validation rate;

[0057] S4. Establish a transformer fault diagnosis model based on hybrid semi-supervised feature selection, as follows:

[0058] Construction of the feature space: Select the gas concentration of a specific type, various gas ratios, and the relative percentages recommended by IEC or IEEE as the initial feature space;

[0059] Reconstruct the DGA original data, and based on the dataset formed by the initial feature space, use the Synthetic Minority Over-sampling Technique (SMOTE) to change the data imbalance;

[0060] Preprocessing of the dataset: Perform arctangent transformation and normalization on the feature space of the dataset with balanced data distribution;

[0061] Establish a semi-supervised iterative model based on filter cross-validation feature selection. The final classification rate obtained is used as the value of the individual fitness in the transformer fault diagnosis model based on wrapper feature selection. Re-encode the chromosome using the obtained optimal feature space;

[0062] Among them, the X strategy is: If the posterior probability of the predicted classification of the samples in the unlabeled sample set S U is greater than the average posterior probability of the predicted classification of all samples in the unlabeled sample set S U , consider this sample to be credible, and add the credible samples to the training set S T to re-constitute the labeled training set S L ( m );

[0063] The Y strategy is: According to the ranking, select the feature spaces of various proportions, and based on S L ( m ) for training, and perform validation on the validation set S V . Select the proportion corresponding to the maximum classification rate, and replace the encoding with the newly formed feature space;

[0064] The Z strategy is: Select from the optimal feature space matrix obtained by r * K fold cross-validation, and select the feature space corresponding to the smallest difference from the average classification rate as the optimal feature.

[0065] The technical concept of the present invention is as follows:

[0066] 1) Establish a semi-supervised iterative transformer fault diagnosis model based on filter cross-validation feature selection combining XYZ strategies (Filter Cross-validation Method based Semi-supervised, FCM-SS)

[0067] Initialization: (Steps 1 - 3 describe the initialization process)

[0068] Step 1: Divide the dataset, and divide the preprocessed dataset into a training set S T and a validation set SV , the training set is divided into an initial labeled sample set S I and an unlabeled sample set S U . The number of labeled sample sets is much smaller than that of unlabeled sample sets.

[0069] Step 2. Use the initial labeled sample set to train the initial classifier and establish a multi-classification model for transformer fault diagnosis based on the Support Vector Machine (SVM).

[0070] Step 3. Use the multi-classification model for transformer fault diagnosis based on SVM to predict the fault categories of the unlabeled sample set S U and the validation set S V . Based on one-against-all (OAA), calculate the multiple posterior probabilities of the multi-classification SVM on each sample in the unlabeled sample set S U , and then average the posterior probabilities of each sample.

[0071] Iteration process (Steps 4-10 describe the m-th iteration (m = 1, 2,..., m0))

[0072] Step 4 In the m -th iteration, use the X strategy to select the samples with high confidence in S U and integrate them with the labeled sample set S I to reconstitute the labeled training set S L ( m ).

[0073] Step 5 Use the labeled training set S L ( m ) to retrain the SVM multi-classifier and re-predict the fault categories of the unlabeled sample set S U and the validation set S V .

[0074] Step 6 Recalculate the posterior probabilities of each sample in S U .

[0075] Step 7 Find the labels predicted for the unlabeled sample SU in the m -th iteration and the m -1-th iteration, and calculate the number of different labels in the two predictions. If the number of differences is less than 1, the iteration stops. If the number is greater than 1, return to Step 4 for the m +1-th iteration until the upper limit value m0 of m is satisfied.

[0076] Step 8 After the iteration stops, obtain the final classification rate of the validation set S V .

[0077] Step 9: Based on the new training set S L ( m )’s feature space and the corresponding final labels, sorts the feature space using a filtering method, and selects the optimal features using the Y strategy.

[0078] Step 10 r * K fold cross validation, calculate the average classification rate on SV, and use the Z strategy to select the optimal features.

[0079] 2) Establish a transformer fault diagnosis model based on package feature optimization

[0080] Step 11: Establish a feature optimization model combining genetic algorithm with support vector machine, and simultaneously optimize the parameters of the support vector machine and the input DGA features.

[0081] Step 12: Genetic algorithm parameter settings, including population, number of iterations, chromosome length, etc.

[0082] Step 13: Encode the SVM parameters and input DGA features, randomize the initial population, decode the chromosomes, sort the individual fitness, perform genetic operations, generate offspring individuals, decode the offspring chromosomes, and calculate the fitness of the offspring chromosome individuals.

[0083] 3) Establishing a hybrid semi-supervised feature selection transformer fault diagnosis model (HFS-SS)

[0084] Step 14: Construct the feature space. Select the concentration of a specific type of gas, the ratio of various gases, and the relative percentages recommended by IEC or IEEE (see Table 2) as the initial feature space.

[0085] Step 15: Reconstruct the original DGA data. Based on the dataset formed in the initial feature space, use the Synthetic Minority Oversampling Technique (SMOTE) to change the data imbalance.

[0086] Step 16: Dataset preprocessing: perform arctangent transformation (AT) and normalization on the reconstructed DGA data.

[0087] Step 17: Establish a semi-supervised iterative model based on filtered cross-validation feature optimization, and use the final classification rate obtained as the value of individual fitness in each iteration of the transformer fault diagnosis method based on package feature optimization. Use the feature space obtained in each iteration of the genetic algorithm to re-encode the chromosome.

[0088] Step 18: When the preset number of iterations of the genetic algorithm is reached, output the optimal feature space, the optimal SVM parameters, and the cross-validation rate.

[0089] In a specific embodiment, the present invention uses the IEC TC10 public dataset of transformers to establish a transformer fault diagnosis model (HFS-SS) for hybrid semi-supervised feature optimization and train the algorithm, and select the optimal feature subset. The IEC TC10 public dataset contains a total of 167 samples, including six fault types: Low Energy Discharge (LED), High Energy Discharge (HED), Low and Medium Temperature (LMT), High Temperature (HT), Partial Discharge (PD), and Normal Condition (NC). The dissolved gas types in the oil of each sample are measured as H2, CH4, C2H2, C2H4, C2H6, CO, and CO2. Since the amount of partial discharge data is too small, it is not used in the model establishment and algorithm training. The final number of samples and fault types are shown in Table 1.

[0090] Table 1. Number of samples and categories of transformer DGA [[ID=I0]]

[0091]

[0092] Feature space construction. Looking at the research at home and abroad, the related work focuses on the existing DGA ratios. There are obvious differences in the DGA ratios used in the diagnostic models in different literatures, and no unified standard for the gas ratios used in the diagnostic models has been formed. The present invention analyzes the relationship between DGA-related features and faults, and excavates the original gas combination ratio features that are more relevant to transformer DGA fault diagnosis. It is found from the literature that the ratio features related to CO and CO2 have weak relevance in transformer fault diagnosis. Therefore, the original features constructed by the present invention exclude the ratio features related to CO and CO2, and select the features with stronger relevance to transformer fault diagnosis, totaling 34 dimensions, as shown in Table 2. Nevertheless, there are still redundant features among these features, and the feature selection algorithm of the present invention needs to be used for optimization, as shown in Table 2.

[0093] Table 2. Various ratio features

[0094]

[0095] Preprocess the 34-dimensional DGA dataset, including using the Synthetic Minority Over-sampling Technique (SMOTE) to change the data imbalance. The number of balanced samples is shown in Table 1. Use the arctangent transformation (AT) to handle the cases where the denominator or numerator of the characteristic ratio in the balanced data is 0, and perform normalization. The balanced data includes 249 samples.

[0096] ;

[0097] ;

[0098] where x ij and x' ij represent the i th eigenvalue of the j th feature before and after normalization of the x j max and x j min represent the maximum and minimum values of the j th feature of all samples, respectively.

[0099] Use the hybrid semi-supervised feature selection algorithm for fault diagnosis of transformers

[0100] 1) First, set the parameters of the genetic algorithm. The population size and the maximum number of iterations are set to 50 and 200 respectively, the gene length is set to 34, 18, 18, and the ranges of the support vector machine parameters C and γ are [0, 200], and the number of iterations is 200. Then encode the chromosomes. Encoding process: The chromosomes of the genetic algorithm are generated by binary encoding. The chromosome is divided into three segments of genes. The first segment represents the feature space of DGA, with a length consistent with the initial feature dimension. 1 means the feature is selected, and 0 means the feature is not selected. The second and third segments represent the two parameters of SVM, the penalty factor C and the kernel parameter γ , which are composed of binary surfaces, and the true values are the decimal C and γ values.

[0101] 2) Calculate the fitness of chromosome individuals, and use the semi-supervised iterative model of filter-based cross-validation feature selection to calculate the individual fitness.

[0102] Divide the preprocessed dataset into a training set S including 199 samples and a validation set S including 50 samples according to 4 * 5-fold cross-validation. The training set consists of the initial labeled sample set S T and the unlabeled sample set S V , and the training set is composed of the initial labeled sample set S I and the unlabeled sample set SU Composition. The present invention constructs a fault diagnosis model based on different numbers of initial labeled sample sets. (10-fold cross-validation divides the sample set into 5 equal parts, where 4 parts are used as the training set and 1 part is used as the validation set. The sample set is randomly shuffled 10 times to form 50 different combinations of training sets and validation sets).

[0103] Establish a transformer fault diagnosis model based on multi-class SVM. Use the initial labeled sample set to train the initial classifier. SVM is a method suitable for solving small-sample classification problems and is applicable to semi-supervised algorithms. Use OAA to calculate multiple posterior probabilities on each sample of the unlabeled sample set SU of the multi-class SVM f ( x ), and predict the fault categories of the unlabeled sample set and the validation set. The following is the method for obtaining a single posterior probability:

[0104] ;

[0105] ;

[0106] ;

[0107] ;

[0108] where is the Lagrange multiplier; C is the penalty factor, which is used to control the degree of penalty for misclassified samples. The kernel function projects the original feature space into a higher-dimensional space. The present invention selects the radial basis kernel function with better classification effect, and the kernel parameter γ is included in the kernel function, where b is the classification threshold, and the function is used to predict the category of the sample.

[0109] Establish a semi-supervised iterative model for filter cross-validation feature selection. In the m th iteration, use the X strategy to select the samples with high confidence in S U and integrate them with the labeled sample set SI to reconstitute the labeled training set S L ( m ).

[0110] X strategy: If the posterior probability of the fault category predicted by the sample in S U is greater than the average posterior probability of all samples predicted and classified in S U , consider this sample credible and add the credible samples to the training set to reconstitute the labeled training set S L ( m ).

[0111] ;

[0112] Retrain the SVM multi-classifier using the labeled training set SL(m), and re-predict the unlabeled sample set S U and the validation set S V for their fault categories. Recalculate the posterior probabilities of each sample in S U , and find the m th iteration and the m -1th iteration's predicted labels for the unlabeled sample S U . Calculate the number of different labels between the two times. If the number of differences is less than 1, stop the iteration. If the number is greater than 1, return to perform the m +1th iteration until m = 10. After the iteration stops, obtain the final classification rate of the validation set S V . According to the new training set S L ( m )'s feature space and the corresponding final labels, use ReliefF in the filter method to sort the feature space. Select the optimal features using the Y strategy.

[0113] Y strategy: According to the sorting, select the feature space of each proportion, train based on S L ( m ), and then validate on S V . Select the proportion corresponding to the maximum classification rate to form a new feature space to replace the encoding.

[0114] After r * K -fold cross-validation, calculate the average classification rate on D V , and select the optimal features using the Z strategy.

[0115] ;

[0116] Z strategy: Select from the optimal feature space matrix obtained by r * K -fold, and choose the feature space corresponding to the smallest difference from the average classification rate as the optimal feature.

[0117] ;

[0118] 3) Use the obtained final classification rate as the value of the individual fitness in the transformer fault diagnosis method based on wrapper feature selection. Re-encode the chromosome using the obtained optimal feature space, perform genetic operations to generate offspring individuals, decode the offspring chromosomes, and calculate the individual fitness of the offspring chromosome individuals using the semi-supervised cross-validation transformer fault diagnosis method based on the filter method.

[0119] 4) When the requirements for the number of iterations of the genetic algorithm are met, output the optimal feature space, the optimal SVM parameters, and the cross-validation rate on the validation set.

[0120] Based on different initial labeled training sets and the DGA dataset of IEC TC 10, the present invention compares the average correct rates of transformer fault diagnosis on the validation set after 10 * 5-fold cross-validation for hybrid semi-supervised feature selection, hybrid standard feature selection, wrapper semi-supervised feature selection, wrapper standard feature selection, filter semi-supervised feature selection, and filter standard feature selection. The results are shown in Table 3. Among them, since the filter method does not combine the wrapper method, the genetic algorithm is not used. Since the wrapper method does not combine the filter method, the XYZ strategy is not used. As shown in Table 3, based on different initial labeled training sets, for hybrid, wrapper, and filter methods, the semi-supervised method can obtain a higher average correct rate than the standard method (since the standard method does not use unlabeled samples, the X strategy is not used), indicating that the semi-supervised algorithm that uses unlabeled samples to expand the labeled sample set can optimize the model of the standard algorithm and improve the fault diagnosis rate. Based on different initial labeled training sets, whether for the semi-supervised algorithm or the standard algorithm, the algorithm using hybrid feature selection can select more optimized features than the wrapper and filter methods, improving the correct rate of fault diagnosis. Moreover, as the number of initial labeled training sets increases, the correct rates of all algorithms increase in most cases.

[0121] To prove the advantages of the XYZ strategy, since the hybrid semi-supervised feature selection necessarily includes the Z strategy, the present invention compares the hybrid semi-supervised feature selection using the XYZ strategy, the hybrid semi-supervised feature selection using the YZ strategy, the hybrid semi-supervised feature selection using the XZ strategy, and the hybrid semi-supervised feature selection without the XYZ strategy (i.e., wrapper semi-supervised feature selection) based on 90 initial labeled training sets, as shown in Table 4. It is found that the hybrid semi-supervised feature selection using the XYZ strategy obtains the best correct rate, and the hybrid semi-supervised feature selection using the XZ strategy is better than the hybrid semi-supervised feature selection using the YZ strategy, indicating the importance of the X strategy in selecting samples with high confidence in unlabeled samples for improving the model.

[0122] Table 3. Average correct rates of transformer fault diagnosis and optimal feature subsets obtained by different algorithms

[0123]

[0124] Table 4. Influence of the XYZ strategy on the average correct rate of transformer fault diagnosis

[0125]

[0126] The DGA dataset (a total of 312 samples) provided by the local power company is used to test the robustness of the optimal feature subset selected by the proposed method. Table 5 shows the feature subsets selected by different traditional methods, hybrid semi-supervised feature selection, wrapper semi-supervised feature selection, and filter semi-supervised feature selection. After 50 times of training with semi-supervised 10 * 5-fold cross-validation (where the initial training set is 80 samples) on the DGA dataset provided by the power company, the average fault diagnosis rate is obtained. Obviously, the average correct rates obtained by hybrid semi-supervised feature selection, wrapper semi-supervised feature selection, and filter semi-supervised feature selection are much higher than those of other traditional methods. The improvement in the correct rate of hybrid semi-supervised feature selection is the most obvious, indicating the strong robustness of the optimal feature subset selected by the proposed algorithm.

[0127] Table 5 Features selected by different methods and the corresponding semi-supervised average correct rates

[0128]

[0129] The embodiments described above are used to illustrate the present invention, not to limit the present invention. Therefore, changes in the example values or replacement of equivalent elements still belong to the scope of the present invention.

[0130] From the above detailed description, those of ordinary skill in the art can clearly understand that the present invention can indeed achieve the foregoing objectives and has actually met the requirements of the patent law.

[0131] Although the preferred embodiments of the present invention have been described, once those skilled in the art know the basic creative concept, they can make additional changes and modifications to these embodiments. Therefore, the appended claims are intended to be construed as including the preferred embodiments and all changes and modifications falling within the scope of the present invention. The above description is only for the preferred embodiments of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

[0132] It should be noted that the above description of the process is only for illustration and explanation, and does not limit the scope of application of this specification. For those skilled in the art, various corrections and changes can be made to the process under the guidance of this specification. However, these corrections and changes are still within the scope of this specification.

[0133] The basic concepts have been described above. Obviously, for those of ordinary skill in the art after reading this application, the above invention disclosure is only for illustration and does not constitute a limitation to this application. Although not explicitly stated herein, those of ordinary skill in the art may make various modifications, improvements, and corrections to this application. Such modifications, improvements, and corrections are suggested in this application, so such modifications, improvements, and corrections still fall within the spirit and scope of the exemplary embodiments of this application.

[0134] Meanwhile, this application uses specific terms to describe the embodiments of this application. For example, "one embodiment", "an embodiment", and / or "some embodiments" mean a certain feature, structure, or characteristic related to at least one embodiment of this application. Therefore, it should be emphasized and noted that "an embodiment" or "one embodiment" or "an alternative embodiment" mentioned two or more times in different positions in this specification are not necessarily referring to the same embodiment. In addition, certain features, structures, or characteristics in one or more embodiments of this application can be appropriately combined.

[0135] In addition, those of ordinary skill in the art can understand that various aspects of this application can be illustrated and described by several patentable types or situations, including any new and useful process, machine, product, or composition of matter, or any new and useful improvement thereof. Therefore, various aspects of this application can be implemented entirely by hardware, can be implemented entirely by software (including firmware, resident software, microcode, etc.), or can be implemented by a combination of hardware and software. The above hardware or software can all be referred to as "unit", "module", or "system". In addition, various aspects of this application can take the form of a computer program product embodied in one or more computer-readable media, in which computer-readable program code is included.

[0136] The computer program code required for the operations of various parts of this application can be written in any one or more programming languages, including object-oriented programming languages such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C++, C#, VB.NET, Python, etc., conventional procedural programming languages such as C programming language, VisualBasic, Fortran2103, Perl, COBOL2102, PHP, ABAP, dynamic programming languages such as Python, Ruby and Groovy, or other programming languages, etc. This program code can run entirely on the user's computer, or run on the user's computer as an independent software package, or run partially on the user's computer and partially on a remote computer, or run entirely on a remote computer or server. In the latter case, the remote computer can be connected to the user's computer through any network form, such as a local area network (LAN) or a wide area network (WAN), or connected to an external computer (for example, through the Internet), or in a cloud computing environment, or used as a service such as software as a service (SaaS).

[0137] In addition, unless clearly stated in the claims, the order of the processing elements and sequences, the use of numbers and letters, or the use of other names in this application are not used to limit the order of the processes and methods of this application. Although some currently considered useful embodiments of the invention are discussed through various examples in the above disclosure, it should be understood that such details only serve the purpose of illustration, and the appended claims are not limited to the disclosed embodiments. On the contrary, the claims are intended to cover all modifications and equivalent combinations that conform to the essence and scope of the embodiments of this application. For example, although the implementation of the above various components can be embodied in a hardware device, it can also be implemented as a pure software solution, for example, installation on an existing server or mobile device.

[0138] Similarly, it should be noted that, in order to simplify the expression of the disclosure of this application and thus help the understanding of one or more embodiments of the invention, in the description of the embodiments of this application above, sometimes multiple features are merged into one embodiment, drawing or description thereof. However, this method of this application should not be construed as reflecting an intention that the claimed subject matter requires more features than those clearly recited in each claim. On the contrary, the subject matter of the invention should have fewer features than the above single embodiment.

Claims

1. A hybrid semi-supervised transformer fault diagnosis method, characterized in that, Including: S1. Establish a semi-supervised iterative transformer fault diagnosis model based on filter cross-validation feature optimization, specifically as follows: Obtain DGA data and divide it into a training set S T and a validation set S V , the training set S T includes an initial labeled sample set S I and an unlabeled sample set S U ; Utilize the initial labeled sample set S I Train the initial classifier to establish a multi-classification model for transformer fault diagnosis based on support vector machines; Use a multi-classification model for transformer fault diagnosis based on support vector machines to predict the unlabeled sample set S U and the validation set S V of the fault categories. Based on one-vs.-many, calculate the multiple posterior probabilities of the multi-classification model for transformer fault diagnosis based on support vector machines on each sample in the unlabeled sample set S U and then average the posterior probabilities of each sample; S2. Establish a semi-supervised iterative model of filter cross-validation feature optimization combined with the XYZ strategy, specifically as follows: In the m th iteration, the X strategy is used to select samples with high confidence in the unlabeled sample set S U and integrate them with the labeled sample set S I to reconstitute the labeled training set S L ( m ); Using the labeled training set S L ( m ) Retrain the multi-classifier of the support vector machine and re-predict the unlabeled sample set S U and the validation set S V for their fault categories; Recalculate the posterior probabilities of each sample in the unlabeled sample set S U ; Find the m th iteration and the m -1 th iteration of the predicted labels for the unlabeled sample set S U Calculate the number of different labels in the two consecutive iterations. If the number of different labels is less than 1, stop the iteration; If the quantity is greater than 1, perform the m +(1)th iteration until the m upper limit value is satisfied; After the iteration stops, obtain the final classification rate of the validation set S V ; According to the new training set S L ( m )'s feature space and the corresponding final labels, use the filter method to sort the feature space and select the optimal features using Strategy Y; After r * K performing k-fold cross-validation, calculate the average classification rate on the validation set S V and use the Z-strategy to select the optimal features; S3. Establish a transformer fault diagnosis model based on wrapper feature optimization, specifically as follows: Establish a feature optimization model combining genetic algorithm and support vector machine, and synchronously optimize the parameters of the support vector machine and the input DGA features; Set the parameters of the genetic algorithm, including population, number of iterations, and chromosome length; Encode the parameters of the support vector machine and the input DGA features, perform randomized initial population, chromosome decoding, individual fitness ranking, genetic operations, generate offspring individuals, offspring chromosome decoding, and calculate the fitness of offspring chromosome individuals; When the stopping condition is reached, output the optimal feature space, the optimal support vector machine parameters, and the cross-validation rate; S4. Establish a transformer fault diagnosis model of hybrid semi-supervised feature optimization, specifically as follows: Construction of the feature space: Select specific types of gas concentrations, various gas ratios, and relative percentages recommended by IEC or IEEE as the initial feature space; Reconstruct the original DGA data, and based on the dataset formed by the initial feature space, use the synthetic minority over-sampling technique to change the data imbalance; Dataset preprocessing: Perform arctangent transformation and normalization on the feature space of the dataset with balanced data distribution; Establish a semi-supervised iterative model based on filter cross-validation feature optimization, use the obtained final classification rate as the value of individual fitness in the transformer fault diagnosis model based on wrapper feature optimization, and re-encode the chromosome using the obtained optimal feature space; Among them, the X strategy is: if the posterior probability of the predicted classification of the samples in the unlabeled sample set S U is greater than the average posterior probability of the predicted classification of all samples in the unlabeled sample set S U , consider this sample to be credible, add the credible samples to the training set S T and reconstitute the labeled training set S L ( m ); The Y strategy is as follows: according to the sorting, select the feature spaces of various proportions, and based on S L ( m ) for training, and verify on the validation set S V Select the proportion corresponding to the maximum classification rate to form a new feature space to replace the encoding; The Z strategy is as follows: For r * K select from the optimal feature space matrix obtained by k-fold cross-validation, and select the corresponding feature space with the smallest difference from the average classification rate as the optimal feature.

Images