Fault diagnosis method of transformer, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a transformer fault diagnosis method, electronic equipment and a storage medium. The method comprises the following steps: acquiring data of gas dissolved in oil of a transformer; according to the data of the dissolved gases in the oil, a first feature set is determined, and the first feature set is used for indicating the absolute content of each dissolved gas, the relative content of multiple dissolved gases and the gas ratio; inputting the first feature set into a transformer fault diagnosis model to obtain a diagnosis result, the transformer fault diagnosis model being obtained by training after joint optimization of a historical feature set and initial parameters of a classification model, the historical feature set being determined by historical dissolved gas data in oil, and the classification model being determined by the historical dissolved gas data in oil; the diagnosis result is used for indicating a fault type corresponding to the data of the dissolved gas in the oil. The method is used for achieving the effect of improving the stability and accuracy of the transformer fault diagnosis model.

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, an electronic device, and a storage medium. Background Art

[0002] As an important device in the power system, once a transformer fails, it will cause serious consequences to the operation of the power system. Therefore, carrying out research on transformer fault diagnosis has important practical significance for improving the power supply reliability of the power system. Traditional fault diagnosis methods based on dissolved gas analysis (DGA) in oil include the IEC three-ratio method, the Roger ratio method, the David triangle method, etc. These methods have simple principles and are easy to implement, but they also have disadvantages such as incomplete ratio coding, overly absolute coding boundaries, and limitations in diagnosing multiple fault types.

[0003] With the development of artificial intelligence and machine learning, theories and methods such as artificial neural network (ANN), expert system, Bayes network, matter-element theory, and support vector machine (SVM) have been widely used in transformer diagnosis technology to improve the accuracy of transformer fault diagnosis models. However, it is difficult to determine the network structure and weights of ANN, the completeness of the knowledge base of the expert system cannot be guaranteed, the Bayes network requires a large amount of sample data, and the matter-element theory has strict requirements for sample consistency and sample size.

[0004] Relatively speaking, SVM (Support Vector Machine) is a classification method with a complete statistical learning theory foundation and excellent learning performance. However, the existing SVM-based transformer fault diagnosis methods are affected by different input features, and there are differences in their classification performance and stability, which may cause problems of conflicting or incorrect diagnostic results in practical applications. In addition, it is difficult to select the optimal parameters of the transformer fault diagnosis model, which may lead to low diagnostic accuracy of the model. Summary of the Invention

[0005] The embodiments of the present application provide a transformer fault diagnosis method, an electronic device, and a storage medium, which are used to solve the problem of the influence of different input features on the transformer fault diagnosis model, and determine the optimal parameters of the transformer fault diagnosis model, so as to achieve the effects of high stability and accuracy of the transformer fault diagnosis model.

[0006] In a first aspect, the embodiments of the present application provide a transformer fault diagnosis method, including: obtaining dissolved gas data in the oil of the transformer;

[0007] According to the dissolved gas data in oil, a first feature set is determined, and the first feature set is used to indicate the absolute content of each dissolved gas, the relative content between multiple dissolved gases, and the gas ratio.

[0008] Input the first feature set into a transformer fault diagnosis model to obtain a diagnosis result. The transformer fault diagnosis model is trained after being jointly optimized by a historical feature set and initial parameters of a classification model. The historical feature set is determined from historical dissolved gas data in oil, and the diagnosis result is used to indicate the fault type corresponding to the dissolved gas data in oil.

[0009] In a possible implementation manner, before obtaining the dissolved gas data in oil of the transformer, the method further includes:

[0010] Obtain historical dissolved gas data in oil and initial parameters of a classification model, and determine a historical feature set based on the historical dissolved gas data in oil;

[0011] Jointly optimize the historical feature set and the initial parameters to obtain a second feature set and target parameters;

[0012] Train the classification model using the second feature set and the target parameters to obtain a transformer fault diagnosis model.

[0013] In a possible implementation manner, the jointly optimizing the historical feature set and the initial parameters to obtain a second feature set and target parameters includes:

[0014] Jointly process the historical feature set and the initial parameters to obtain a plurality of input feature sets, and the plurality of input feature sets constitute an input feature population;

[0015] Determine a first input feature set and a second input feature set from the plurality of input feature sets;

[0016] Determine the fitness of the first input feature set and the second input feature set;

[0017] In a case where the fitness of the first input feature set is less than the fitness of the second input feature set, perform an optimization process on the first input feature set to obtain an optimized first input feature set;

[0018] Update the input feature population based on the optimized first input feature set;

[0019] Determine a second feature set and target parameters based on the updated input feature population.

[0020] In a possible implementation, optimizing the first input feature set to obtain an optimized first input feature set includes:

[0021] Determining the adaptive crossover probability and the adaptive mutation probability of the first input feature set according to the fitness of the first input feature set;

[0022] Based on the adaptive crossover probability and the adaptive mutation probability, performing a crossover mutation operation on the first input feature set to obtain an optimized first input feature set.

[0023] In a possible implementation, determining the fitness of the first input feature set and the second input feature set includes:

[0024] Performing an equal division process on the multiple input feature sets to obtain a preset number of subsets;

[0025] For any one of the preset number of subsets, determining the subset as a test set, and based on the subset, performing a filtering process on the preset number of subsets to obtain a training set;

[0026] Training the classification model with multiple input feature sets in the training set to obtain an initial transformer fault diagnosis model;

[0027] Evaluating the initial transformer fault diagnosis model with the test set to obtain the transformer fault diagnosis accuracy corresponding to each input feature set in the test set;

[0028] Determining the transformer fault diagnosis accuracy as the fitness of the corresponding input feature set;

[0029] Based on the fitness of each input feature set in the test set, determining the fitness of the first input feature set and the second input feature set.

[0030] In a possible implementation, training the classification model with the second feature set and the target parameter to obtain a transformer fault diagnosis model includes:

[0031] Dividing multiple second feature sets into a training sample set and a test sample set, where the second feature set carries a class label, and the class label is used to indicate the fault type of the transformer;

[0032] Sampling the training sample set according to a preset sample weight to obtain a first training sample;

[0033] Determining the standard deviation of the first training sample and determining the standard deviation as the kernel parameter of the first weak classifier of the classification model;

[0034] Train the first weak classifier using the first training sample to obtain the first base classifier and the training error of the first weak classifier;

[0035] Determine the weight of the first base classifier based on the training error of the first weak classifier, and update the sample weights corresponding to the training set;

[0036] Construct an ensemble classifier based on the weight of the first base classifier;

[0037] Optimize the ensemble classifier using the test sample set to obtain a transformer fault diagnosis model.

[0038] In a possible implementation, after determining the weight of the first base classifier based on the training error and updating the sample weights corresponding to the training set, the method further includes:

[0039] When the number of times of updating the sample weights corresponding to the training set has not reached the preset number of updates, sample the training set based on the updated sample weights to obtain a second training sample;

[0040] Train the second weak classifier of the classification model using the second training sample to obtain the second base classifier and the training error corresponding to the second weak classifier;

[0041] Determine the weight of the second base classifier based on the training error of the second weak classifier, and update the sample weights corresponding to the training set;

[0042] Construct an ensemble classifier based on the weight of the first base classifier and the weight of the second base classifier.

[0043] In a second aspect, an embodiment of the present application provides a transformer fault diagnosis device, including:

[0044] An acquisition module, configured to acquire dissolved gas data in transformer oil;

[0045] A determination module, configured to determine a first feature set according to the dissolved gas data in the transformer oil;

[0046] A processing module, configured to input the first feature set into a transformer fault diagnosis model to obtain a diagnosis result.

[0047] In a possible implementation, the device further includes: a training module;

[0048] The acquisition module is further configured to acquire historical dissolved gas data in transformer oil and initial parameters of the classification model, and determine a historical feature set based on the historical dissolved gas data;

[0049] The processing module is further configured to jointly optimize the historical feature set and the initial parameters to obtain a second feature set and target parameters;

[0050] The training module is further configured to train the classification model by using the second feature set and the target parameters to obtain a transformer fault diagnosis model.

[0051] In a possible implementation manner, the processing module is further configured to jointly process the historical feature set and the initial parameters to obtain a plurality of input feature sets, and the plurality of input feature sets form an input feature population;

[0052] The determining module is further configured to determine a first input feature set and a second input feature set from the plurality of input feature sets;

[0053] The determining module is further configured to determine the fitness of the first input feature set and the second input feature set;

[0054] The processing module is further configured to, when the fitness of the first input feature set is less than the fitness of the second input feature set, perform optimization processing on the first input feature set to obtain an optimized first input feature set;

[0055] The processing module is further configured to update the input feature population based on the optimized first input feature set;

[0056] The determining module is further configured to determine a second feature set and target parameters based on the updated input feature population.

[0057] In a possible implementation manner, the determining module is further configured to determine an adaptive crossover probability and an adaptive mutation probability of the first input feature set according to the fitness of the first input feature set;

[0058] The processing module is further configured to perform a crossover and mutation operation on the first input feature set based on the adaptive crossover probability and the adaptive mutation probability to obtain an optimized first input feature set.

[0059] In a possible implementation manner, the apparatus further includes: an evaluation module;

[0060] The processing module is further configured to equally divide the plurality of input feature sets to obtain a preset number of subsets;

[0061] The determining module is further configured to, for any one of the preset number of subsets, determine the subset as a test set;

[0062] The processing module is further configured to filter the preset number of subsets based on the subset to obtain a training set;

[0063] The training module is further configured to train the classification model with multiple input feature sets in the training set to obtain an initial transformer fault diagnosis model;

[0064] The evaluation module is configured to evaluate the initial transformer fault diagnosis model with the test set to obtain the transformer fault diagnosis accuracy corresponding to each input feature set in the test set;

[0065] The processing module is further configured to determine the fitness of the corresponding input feature set based on the transformer fault diagnosis accuracy;

[0066] The processing module is further configured to determine the fitness of the first input feature set and the second input feature set based on the fitness of each input feature set in the test set.

[0067] In a possible implementation manner, the device further includes: a sampling module;

[0068] The processing module is further configured to divide a plurality of the second feature sets into a training sample set and a test sample set;

[0069] The sampling module is further configured to sample the training sample set according to a preset sample weight to obtain a first training sample;

[0070] The determining module is further configured to determine the standard deviation of the first training sample and determine the standard deviation as the kernel parameter of the first weak classifier of the classification model;

[0071] The training module is further configured to train the first weak classifier with the first training sample to obtain a first base classifier and the training error of the first weak classifier;

[0072] The processing module is further configured to determine the weight of the first base classifier based on the training error of the first weak classifier and update the sample weight corresponding to the training set;

[0073] The processing module is further configured to construct an ensemble classifier based on the weight of the first base classifier;

[0074] The processing module is further configured to optimize the ensemble classifier with the test sample set to obtain a transformer fault diagnosis model.

[0075] In a possible implementation manner, the sampling module is further configured to sample the training set based on the updated sample weights to obtain a second training sample when the number of times of updating the sample weights corresponding to the training set does not reach a preset number of updates;

[0076] The training module is further configured to train a second weak classifier of the classification model based on the second training sample to obtain a second base classifier and a training error corresponding to the second weak classifier;

[0077] The determining module is further configured to determine the weight of the second base classifier based on the training error of the second weak classifier;

[0078] The processing module is further configured to update the sample weights corresponding to the training set;

[0079] The processing module is further configured to construct an ensemble classifier based on the weight of the first base classifier and the weight of the second base classifier.

[0080] In a third aspect, an embodiment of the present application provides a fault diagnosis device for a transformer, including: a memory, a processor;

[0081] The memory stores computer-executable instructions;

[0082] The processor executes the computer-executable instructions stored in the memory, so that the processor executes the above first aspect and / or various possible implementation manners of the first aspect.

[0083] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored, and when the computer-executable instructions are executed by a processor, they are used to implement the above first aspect and / or various possible implementation manners of the first aspect.

[0084] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the above first aspect and / or various possible implementation manners of the first aspect.

[0085] A fault diagnosis method, an electronic device, and a storage medium for a transformer provided by an embodiment of the present application obtain dissolved gas data in the transformer oil, determine a first feature set based on the dissolved gas data in the transformer oil, and then input the first feature set into a transformer fault diagnosis model to obtain a diagnosis result. Among them, the transformer fault diagnosis model is trained by jointly optimizing the historical feature set and the initial parameters of the classification model. This method uses a joint optimization method to optimize the input features and initial parameters of the transformer fault diagnosis model, solves the influence of various input features on the stability and accuracy of the transformer fault diagnosis model, determines the optimal parameters of the transformer fault diagnosis model, and achieves the effects of high stability and accuracy of the transformer fault diagnosis model. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] The drawings herein are incorporated into the specification and form a part of this specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application.

[0087] Figure 1 Schematic flowchart of a fault diagnosis method for a transformer provided by the present application Figure 1 ;

[0088] Figure 2 Schematic flowchart of a fault diagnosis method for a transformer provided by the present application Figure 2 ;

[0089] Figure 3 Schematic flowchart of a fault diagnosis method for a transformer provided by the present application Figure 3 ;

[0090] Figure 4 Schematic structural diagram of the input feature set provided by the present application;

[0091] Figure 5 Schematic structural diagram of a fault diagnosis device for a transformer provided by the present application;

[0092] Figure 6 Schematic structural diagram of a fault diagnosis device for a transformer provided by the present application.

[0093] Through the above-mentioned drawings, specific embodiments of the present application have been shown, and there will be more detailed descriptions hereinafter. These drawings and text descriptions are not intended to limit the scope of the concept of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0094] Exemplary embodiments will be described in detail herein, and examples thereof are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0095] First, the terms involved in the present application are explained:

[0096] Dissolved gases in oil: When a transformer fails, gases are generated inside the transformer and dissolved in the oil. Different types of transformer failures result in different dissolved gases and their contents in the oil.

[0097] Fitness: In a genetic algorithm, fitness is the main indicator describing the performance of an individual. Based on the fitness value, individuals are selected for survival of the fittest.

[0098] Transformer failures can be divided into two categories: external failures and internal failures. External failures are easily detected and eliminated by the staff. However, due to the isolation of the transformer shell, internal failures are difficult to detect, which not only reduces the power supply reliability but may also cause a major paralysis of the power system. Therefore, the fault diagnosis of transformers is extremely important for the power system.

[0099] When a failure occurs, various gases are generated inside the transformer and dissolved in the oil. Different types of failures result in different dissolved gases and their contents in the oil. Therefore, detecting and analyzing the dissolved gases in transformer oil can determine the type of transformer failure. Existing methods for diagnosing transformer failure types based on DGA data include the IEC three-ratio method, the Roger ratio method, the ANN network, the SVM, etc. However, the IEC three-ratio method and the Roger ratio method have disadvantages such as incomplete ratio coding, overly absolute coding boundaries, and limitations in diagnosing multiple failure types. It is difficult to determine the network structure and weights of the ANN.

[0100] SVM is a classification method with a complete statistical learning theory foundation and excellent learning performance. However, the existing SVM-based transformer fault diagnosis models have significant differences in fault diagnosis performance for various input features, which may cause conflicts or errors in the diagnosis results in practical applications. Moreover, it is difficult to select the optimal parameters of the transformer fault diagnosis model, seriously affecting the accuracy of transformer fault diagnosis.

[0101] In view of the above problems existing in the prior art, the present application provides a fault diagnosis method for a transformer. By obtaining historical dissolved gas data in oil and initial parameters of a classification model, and determining a historical feature set based on the historical dissolved gas data in oil, then jointly optimizing the historical feature set and the initial parameters to obtain a second feature set and target parameters. Among them, the second feature set is the preferred input feature, and the target parameters are the optimal parameters of the classification model. This method uses the second feature set and target parameters to train the classification model to obtain a transformer fault diagnosis model, which not only solves the problem of the performance impact of various types of input features on the transformer fault diagnosis model, but also determines the optimal parameters of the transformer fault diagnosis model, improving the stability and accuracy of the transformer fault diagnosis model.

[0102] The following will specifically describe the technical solutions of the present application and how the technical solutions of the present application solve the above technical problems with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below with reference to the accompanying drawings.

[0103] Figure 1 Schematic flow of a fault diagnosis method for a transformer provided by the present application Figure 1 , as Figure 1 shown, the method includes:

[0104] S101. Obtain the dissolved gas data in the oil of the transformer.

[0105] Among them, the dissolved gas data in the oil includes the content of each dissolved gas in the oil. The dissolved gas in the oil includes gases generated by the decomposition of insulating oil and other insulating materials. For example, hydrocarbon gases such as CH4, C2H6, C2H4, and C2H2, and H2. Optionally, on-line monitoring technology can be used to obtain the dissolved gas data in the oil of the transformer in real time.

[0106] S102. Determine a first feature set according to the dissolved gas data in the oil.

[0107] Among them, the first feature set is used to indicate the absolute content of each dissolved gas in the oil, the relative content between multiple dissolved gases, and the gas ratio.

[0108] Optionally, analyze and organize the obtained dissolved gas data in oil, and determine the absolute content, relative content of each dissolved gas in oil, as well as the total gas amount and total hydrocarbon gas amount as the first feature set. For example, the first feature set corresponding to the five dissolved gases in oil CH4, C2H6, C2H4, C2H2, and H2 includes the absolute content of dissolved gases in oil f3(C2H6), f4(C2H4), f5(C2H2), f2(CH4), f1(H2), and the relative content between the dissolved gases in oil f6(CH4 / H2), f7(C2H2 / C2H4), f8(C2H2 / CH4), f9(C2H6 / C2H2), f 10 (C2H4 / C2H6). Additionally, considering the total gas amount and total hydrocarbon gas amount, the first feature set can also include the total amount of dissolved gases in oil and the relative mass fractions of C2H4, C2H2, and CH4, f 11 (CH4 / S)%, f 12 (C2H2 / S)%, f 13 (C2H4 / S)%, where S = C2H4 + C2H2 + CH4. Furthermore, the first feature set can also include the relative content of each dissolved gas in oil to the total gas amount. For example, f 14 (f1 / f 19 ), f 15 (f2 / f 19 ), f 16 (f3 / f 19 ), f 17 (f4 / f 19 ), f 18 (f5 / f 19 ), where f 19 (TG) is the total amount of the above five hydrocarbon gases.

[0109] Optionally, construct the first feature set based on the input features corresponding to the above 20 dissolved gases in oil.

[0110] S103. Input the first feature set into the transformer fault diagnosis model to obtain a diagnosis result.

[0111] Among them, the transformer fault diagnosis model is obtained by jointly optimizing and training the historical feature set and the initial parameters of the classification model. The historical feature set is determined from historical dissolved gas data in oil, and the diagnosis result is used to indicate the fault type corresponding to the dissolved gas data in oil.

[0112] Specifically, input the first feature set into the transformer fault diagnosis model, and use the improved genetic algorithm (IGA) to optimize the first feature set. Based on the results of the optimization process, determine the fault type of the transformer.

[0113] Optionally, construct the transformer fault diagnosis model based on the IGA-AdaBoost-SVM model. Obtain the historical feature set by preprocessing the historical dissolved gas data in oil. Use the IGA model to jointly optimize the historical feature set and the initial parameters of the AdaBoost-SVM model to obtain the optimal parameters of the classification model AdaBoost-SVM and the optimal historical feature set. Then, use the above optimal parameters and the optimal historical feature set to train the classification model to obtain the transformer fault diagnosis model. This method jointly optimizes the input feature set and the initial parameters, improving the diagnostic performance of the transformer fault diagnosis model.

[0114] It should be noted that the AdaBoost-SVM model is obtained by combining multiple SVM classifiers into a strong classifier using the Adaboost (Adaptive Boosting) algorithm. Among them, the Adaboost algorithm is an ensemble learning algorithm. This method uses the AdaBoost-SVM model to combine multiple weak classifiers into a strong classifier, improving the classification accuracy and making the diagnostic accuracy of the obtained transformer fault diagnosis model higher.

[0115] A transformer fault diagnosis method provided by an embodiment of the present application obtains the dissolved gas data in the oil of the transformer, determines the first feature set based on the dissolved gas data in the oil, and then inputs the first feature set into the transformer fault diagnosis model to obtain a diagnostic result. Among them, the transformer fault diagnosis model is obtained by training after jointly optimizing the historical feature set and the initial parameters of the classification model. This method uses the method of joint optimization to optimize the input features and the initial parameters of the transformer fault diagnosis model, solving the influence of various input features on the stability and accuracy of the transformer fault diagnosis model, and determining the optimal parameters of the transformer fault diagnosis model, achieving the effect of improving the stability and accuracy of the transformer fault diagnosis model.

[0116] Figure 2 The flowchart of a transformer fault diagnosis method provided by the present application Figure 2 , such as Figure 2 shown. On the basis of the Figure 1 embodiment, a possible training process of the transformer fault diagnosis model is supplemented and described. The method includes:

[0117] S201. Obtain historical dissolved gas data in oil and initial parameters of the classification model, and determine a historical feature set based on the historical dissolved gas data in oil.

[0118] Specifically, obtain historical dissolved gas data in oil and initial parameters of the classification model, analyze and preprocess the historical dissolved gas data in oil to obtain a historical feature set. Optionally, construct a classification model based on the AdaBoost-SVM model. The initial parameters of this classification model include the parameter C of the penalty degree, the kernel parameter, and the learning rate Learning_rate. In addition, adopt the method described in step S102 to determine the absolute content and relative content of various dissolved gases in oil, and construct a historical feature set based on the absolute content and relative content.

[0119] Optionally, after obtaining the historical data of the transformer, aiming at the problem of sample imbalance where the number of normal state samples in the transformer historical data is much larger than the number of fault state samples, use the K-nearest neighbor upsampling method to expand the fault class samples to obtain the historical dissolved gas data in oil.

[0120] S202. Jointly optimize the historical feature set and the initial parameters to obtain a second feature set and target parameters.

[0121] Optionally, encode the historical feature set and the initial parameters onto one chromosome to obtain a chromosome population. Use the IGA model to select two chromosomes from the population, and determine the fitness values corresponding to the above two chromosomes. According to the comparison of the fitness values of the above two chromosomes, determine one chromosome that needs to be optimized from the two chromosomes, perform crossover and mutation operations on it, and then put the chromosome and the chromosome that does not need to be optimized back into the population. This method enables the optimal chromosome to be always retained in the population, which is conducive to the population converging to the global optimum. Select the optimal input feature set and initial parameters from the historical feature set and the initial parameters as the second feature set and target parameters.

[0122] It should be noted that the input features of the classification model and the model parameters influence each other. According to different input features, the optimized model parameters obtained are also different. Therefore, this method jointly optimizes the input features and the model parameters to obtain the optimal model parameters, thereby improving the diagnostic effect of the transformer fault diagnosis model.

[0123] S203. Train the classification model using the second feature set and the target parameters to obtain a transformer fault diagnosis model.

[0124] Specifically, substitute the target parameters into the classification model, and use the second feature set to train the classification model to obtain a transformer fault diagnosis model. Optionally, divide multiple second feature sets into a training set and a test set, use the training set to train the classification model to construct an ensemble classifier to obtain an initial transformer fault diagnosis model, and then use the test set to optimize the transformer fault diagnosis model to obtain a transformer fault diagnosis model.

[0125] A transformer fault diagnosis method provided by an embodiment of the present application obtains historical dissolved gas data in oil and initial parameters of a classification model, determines a historical feature set based on the historical dissolved gas data in oil, then jointly optimizes the historical feature set and the initial parameters to obtain a second feature set and target parameters, and further trains the classification model according to the second feature set and the target parameters to obtain a transformer fault diagnosis model. This method trains the classification model by using the second feature set and optimal parameters obtained through joint optimization processing to construct a transformer fault diagnosis model, improving the stability and accuracy of the transformer fault diagnosis model.

[0126] Figure 3 It is a flow schematic of a transformer fault diagnosis method provided by the present application Figure 3 , such as Figure 3 shown. Based on the embodiment of Figure 1 or Figure 2 embodiment, a possible implementation manner of the training process of the transformer fault diagnosis model is described in detail. The method includes:

[0127] S301. Obtain historical dissolved gas data in oil and initial parameters of the classification model, and determine a historical feature set based on the historical dissolved gas data in oil.

[0128] The explanation of this step S301 is similar to that of the above step S201 and will not be repeated here.

[0129] S302. Jointly process the historical feature set and the initial parameters to obtain a plurality of input feature sets.

[0130] Among them, the plurality of input feature sets constitute an input feature population, and the initial parameters include a penalty parameter, a learning rate, and the number of weak classifiers.

[0131] Optionally, Figure 4It is a structural schematic diagram of the input feature set provided for this application. The historical feature set and the initial parameters are encoded onto the same chromosome to obtain the input feature set. For example, the 20 initial fault features listed in the above step S102 are determined as the historical feature set, and the penalty factor C, learning rate learning_rate, and the number of weak classifiers n_estimators of the classification model AdaBoost-SVM are determined as the initial parameters, and the above historical feature set and initial parameters are encoded onto one chromosome in binary form, and the encoding result is as Figure 4 shown. For historical features, if the binary number is 1, it means that the fault is selected, and if the binary number is 0, it means that the historical feature is not selected. L1 is the length of the historical feature set, and L1 = 20 is taken. The model parameters are finally decoded into decimal numbers. The lengths of L2, L3, and L4 determine the accuracy of the decimal numbers after decoding. L2 = L3 = L4 = 16 is taken, and the scale of the entire chromosome population is 100. The optimization space range of C is [2 -12 , 3000], the optimization space range of learning_rate is [2 -12 , 1], n_estimators takes positive integers, and the optimization space range is [2 -12 , 30].

[0132] S303. Among the multiple input feature sets, determine the first input feature set and the second input feature set; determine the fitness of the first input feature set and the second input feature set.

[0133] Optionally, in the input feature population determined in the above step S303, randomly select two chromosomes to obtain the first input feature set and the second input feature set, use the first input feature set and the second input feature set to train the classification model respectively, and use the test samples in the multiple input feature sets to test the classification model, and determine the accuracy rates obtained from the tests as the fitness of the first input feature set and the second input feature set respectively.

[0134] S304. In the case where the fitness of the first input feature set is less than the fitness of the second input feature set, perform optimization processing on the first input feature set to obtain the optimized first input feature set.

[0135] Among them, the optimization processing includes crossover processing and mutation processing.

[0136] Specifically, compare the fitness sizes of the first input feature set and the second input feature set, and determine the input feature set with a fitness less than the other party as the input feature set to be optimized.

[0137] Optionally, according to the fitness of the first input feature set, determine the adaptive crossover probability and the adaptive mutation probability of the first input feature set; based on the adaptive crossover probability and the adaptive mutation probability, perform a crossover and mutation operation on the first input feature set to obtain an optimized first input feature set.

[0138] Optionally, in the case where the fitness of the first input feature set is less than the fitness of the second input feature set, determine the first input feature set as the input feature set to be optimized, calculate the adaptive crossover probability and the adaptive mutation probability of the first input feature set according to the fitness of the first input feature set, and then perform crossover and / or mutation processing on the first input feature set according to the adaptive crossover probability and the adaptive mutation probability to obtain an optimized first input feature set. Additionally, no processing is performed on the second input feature set whose fitness is greater than the first input feature set.

[0139] S305. Update the input feature population based on the optimized first input feature set.

[0140] Specifically, after obtaining the optimized first input feature set, return the second input feature set and the optimized first input feature set to the input feature population, and update the input feature population so that the chromosome with the highest fitness is always retained in the input feature population, facilitating the search for the optimal input feature set and the optimal parameters of the classification model.

[0141] S306. Determine the second feature set and the target parameters based on the updated input feature population.

[0142] Optionally, for multiple input feature sets in the updated input feature population, decode the encoding corresponding to the historical feature set in a binary manner to obtain multiple decoded input features, and determine the input features with a binary number of 1 among the multiple input features as the second feature set. Additionally, decode the encoding corresponding to the initial feature set in hexadecimal to obtain the optimal parameters, i.e., the target parameters.

[0143] S307. Divide the multiple second feature sets into a training sample set and a test sample set.

[0144] Among them, the second feature set carries a class label, and the class label is used to indicate the fault type of the transformer. The fault types of the transformer include thermal faults, partial discharges, high-energy discharges, low-energy discharges, and normal states. For different fault types, the gases dissolved in the oil are different, and the contents of the gases dissolved in the oil are also different.

[0145] Optionally, divide multiple second feature sets into a training sample set and a test sample set according to a preset ratio. Among them, the training sample set is X, and the class label is Y = {-1, 1}. A set of training sample sets can be, for example, D = {(x1, y1), ……, (x n , y n )}, x i ∈X, y i ∈Y.

[0146] It should be noted that before dividing multiple second feature sets into a training sample set and a test sample set, the multiple second features can also be normalized to reduce the mutual exclusivity between sample data, thereby improving the fault diagnosis accuracy of the transformer.

[0147] S308. Sample the training sample set according to a preset sample weight to obtain a first training sample.

[0148] Optionally, after dividing the second feature set, initialize the sample weight and the number of iterations, and sample according to the sample weight in the training sample set to determine the first training sample for training the first weak classifier. Among them, when initializing the sample weight, each sample weight can be made the same. For example, W1(i) = 1 / n, i = 1, 2, ……, n, and sample according to W1(i) in the training sample set D determined in the above step S307 to obtain the first training sample d t of the first weak classifier C t .

[0149] S309. Determine the standard deviation of the first training sample, and determine the standard deviation as the kernel parameter of the first weak classifier of the classification model.

[0150] Optionally, calculate the average value of the first training sample, and for each sample data in the first training sample, determine the deviation of the sample data from the average value. Then, determine the sum of squares of the deviations according to the average value corresponding to multiple sample data, determine the standard deviation of the first training sample according to the sum of squares and the average value, and determine the standard deviation as the kernel parameter of the first weak classifier. This method determines the standard deviation of each training sample as the kernel parameter of the corresponding weak classifier, which ensures the difference of each weak classifier while improving the accuracy of the ensemble classifier.

[0151] S310. Train the first weak classifier with the first training sample to obtain a first base classifier and the training error of the first weak classifier; determine the weight of the first base classifier based on the training error of the first weak classifier, and update the sample weight corresponding to the training set.

[0152] Optionally, the first weak classifier is trained using the first training sample to obtain the first base classifier and the training error of the first weak classifier. Based on this training error, the weights of the misclassified training samples are increased, and the weights of the correctly classified training samples are decreased, thereby updating the sample weights of the training set. During the subsequent process of training weak classifiers, the weights of the base learners with smaller error rates are increased, and the weights of the base learners with larger error rates are decreased.

[0153] Optionally, after determining the weight of the first base classifier based on the training error and updating the sample weights corresponding to the training set, the method further includes:

[0154] In the case where the number of times of updating the sample weights corresponding to the training set has not reached the preset number of updates, based on the updated sample weights, the training set is sampled to obtain a second training sample; the second weak classifier of the classification model is trained using the second training sample to obtain the second base classifier and the training error corresponding to the second weak classifier; the weight of the second base classifier is determined based on the training error of the second weak classifier, and the sample weights corresponding to the training set are updated; an ensemble classifier is constructed based on the weight of the first base classifier and the weight of the second base classifier.

[0155] Optionally, the number of times of updating the sample weights corresponding to the training set is the current iteration number of model training. In the case where the current iteration number has not reached the preset iteration number, the second weak classifier is trained based on the training set after updating the sample weights, and this iterative training is performed until the preset iteration number is reached, and multiple base classifiers are obtained.

[0156] S311. Construct an ensemble classifier based on the weight of the first base classifier;

[0157] Optionally, the decision function of the first base classifier is determined according to the weight of the first base classifier, and an ensemble classifier is constructed according to this decision function.

[0158] Optionally, according to the multiple base classifiers obtained through multiple iterations, the decision function of the ensemble classifier is determined using the following formula:

[0159]

[0160] where T is the preset number of updates, α t is the weight of the t-th base classifier, h t is the t-th base classifier, and x is the sample data.

[0161] It should be noted that the larger the weight of the base classifier, the more important it reflects its importance in the ensemble classifier.

[0162] S312. Optimize the integrated classifier using the test sample set to obtain a transformer fault diagnosis model.

[0163] Optionally, evaluate the integrated classifier using the test sample set, and adjust the structure and parameters of the integrated classifier according to the evaluation results, so that the obtained transformer fault diagnosis model has higher accuracy and stability.

[0164] A fault diagnosis method for a transformer provided by an embodiment of the present application obtains multiple input feature sets by jointly processing a historical feature set and initial parameters, optimizes an input feature population based on the fitness of the input feature sets, and then trains a classification model according to the optimized second feature set and target parameters, constructs an integrated classifier, and further optimizes the performance of the integrated classifier based on a training set to obtain a transformer fault diagnosis model. This method compares the fitness of two input feature sets, so that the chromosome with the highest fitness is always retained in the input feature population, which is conducive to finding the optimal input feature set and the optimal parameters of the classification model. In addition, this method uses the standard deviation of the training samples as the kernel parameter of the weak classifier of the classification model, taking into account the accuracy and difference of each weak classifier, and improving the diagnostic accuracy and stability of the transformer fault diagnosis model.

[0165] In a possible implementation manner, based on the above Figures 2 - 4 , the method for calculating the fitness of the first input feature set and the second input feature set is described in detail. The method includes:

[0166] Divide the multiple input feature sets into equal parts to obtain a preset number of subsets; for any one of the preset number of subsets, determine the subset as a test set, and filter the preset number of subsets based on the subset to obtain a training set; train the classification model using the multiple input feature sets in the training set to obtain an initial transformer fault diagnosis model; evaluate the initial transformer fault diagnosis model using the test set to obtain the transformer fault diagnosis accuracy corresponding to each input feature set in the test set; determine the transformer fault diagnosis accuracy as the fitness of the corresponding input feature set; based on the fitness of each input feature set in the test set, determine the fitness of the first input feature set and the second input feature set.

[0167] Optionally, five subsets are preset, and the multiple input feature sets are evenly divided into five equal parts. Then, any one subset is arbitrarily selected as the test set, and the other four subsets are used as the training set. This method is iterated five times, and the selected test set is different each time. The classification model is trained using the training set to obtain an initial transformer fault diagnosis model. The obtained initial transformer fault diagnosis model is evaluated using the test set. According to the fault diagnosis accuracy corresponding to each input feature set in the test set, the fitness of each input feature set in the test set is determined. This method is iterated five times, and different test sets are selected each time to obtain the fitness corresponding to all input feature sets, and then the fitness of the first input feature set and the second input feature set is determined.

[0168] A fault diagnosis method for a transformer provided by an embodiment of the present application is based on a preset number of subsets to equally divide multiple input feature sets. Among the preset number of subsets obtained by the equal division process, any one subset is arbitrarily selected as the test set, and the remaining four are used as the training set to train the classification model. The classification model is iteratively trained in this way until all subsets have been used as the test set to evaluate the initially obtained transformer fault diagnosis model, so as to determine the accuracy corresponding to each input feature set, and then determine the fitness of each input feature set, and obtain the fitness of the first input feature set and the second input feature set therefrom. This method uses multiple different combinations of training sets and test sets to train and evaluate the classification model, reduces the overfitting of the transformer fault diagnosis model to a specific training set, improves the generalization ability of the transformer fault diagnosis model, and improves the accuracy of the evaluation results by using multiple different test sets to evaluate the transformer fault diagnosis model, making the determined fitness of the first input feature set and the second input feature set more accurate.

[0169] Figure 5 The following is a schematic structural diagram of a fault diagnosis device for a transformer provided by the present application, as Figure 5 shown. A fault diagnosis device 50 provided in this embodiment includes:

[0170] An acquisition module 501, configured to acquire the dissolved gas data in the transformer oil;

[0171] A determination module 502, configured to determine a first feature set according to the dissolved gas data in the transformer oil;

[0172] A processing module 503, configured to input the first feature set into a transformer fault diagnosis model to obtain a diagnosis result.

[0173] In a possible implementation manner, the device further includes: a training module 504;

[0174] The obtaining module 501 is further configured to obtain historical dissolved gas in oil data and initial parameters of a classification model, and determine a historical feature set based on the historical dissolved gas in oil data;

[0175] The processing module 503 is further configured to jointly optimize the historical feature set and the initial parameters to obtain a second feature set and target parameters;

[0176] The training module 504 is further configured to train the classification model by using the second feature set and the target parameters to obtain a transformer fault diagnosis model.

[0177] In a possible implementation manner, the processing module 503 is further configured to jointly process the historical feature set and the initial parameters to obtain a plurality of input feature sets, and the plurality of input feature sets form an input feature population;

[0178] The determining module 502 is further configured to determine a first input feature set and a second input feature set from the plurality of input feature sets;

[0179] The determining module 502 is further configured to determine the fitness of the first input feature set and the second input feature set;

[0180] The processing module 503 is further configured to, when the fitness of the first input feature set is less than the fitness of the second input feature set, perform optimization processing on the first input feature set to obtain an optimized first input feature set;

[0181] The processing module 503 is further configured to update the input feature population based on the optimized first input feature set;

[0182] The determining module 502 is further configured to determine a second feature set and target parameters based on the updated input feature population.

[0183] In a possible implementation manner, the determining module 502 is further configured to determine an adaptive crossover probability and an adaptive mutation probability of the first input feature set according to the fitness of the first input feature set;

[0184] The processing module 503 is further configured to perform a crossover and mutation operation on the first input feature set based on the adaptive crossover probability and the adaptive mutation probability to obtain an optimized first input feature set.

[0185] In a possible implementation manner, the device further includes: an evaluation module 505;

[0186] The processing module 503 is further configured to equally divide the multiple input feature sets to obtain a preset number of subsets;

[0187] The determining module 502 is further configured to, for any one of the preset number of subsets, determine the subset as a test set;

[0188] The processing module 503 is further configured to perform filtering processing on the preset number of subsets based on the subset to obtain a training set;

[0189] The training module 504 is further configured to train the classification model by using the multiple input feature sets in the training set to obtain an initial transformer fault diagnosis model;

[0190] The evaluation module 505 is configured to evaluate the initial transformer fault diagnosis model by using the test set to obtain the transformer fault diagnosis accuracy corresponding to each input feature set in the test set;

[0191] The processing module 503 is further configured to determine the transformer fault diagnosis accuracy as the fitness of the corresponding input feature set;

[0192] The processing module 503 is further configured to determine the fitness of the first input feature set and the second input feature set based on the fitness of each input feature set in the test set.

[0193] In a possible implementation manner, the device further includes: a sampling module 506;

[0194] The processing module 503 is further configured to divide the multiple second feature sets into a training sample set and a test sample set;

[0195] The sampling module 506 is further configured to sample the training sample set according to a preset sample weight to obtain a first training sample;

[0196] The determining module 502 is further configured to determine the standard deviation of the first training sample and determine the standard deviation as the kernel parameter of the first weak classifier of the classification model;

[0197] The training module is further configured to train the first weak classifier by using the first training sample to obtain a first base classifier and the training error of the first weak classifier;

[0198] The processing module 503 is further configured to determine the weight of the first base classifier based on the training error of the first weak classifier and update the sample weight corresponding to the training set;

[0199] The processing module 503 is further configured to construct an ensemble classifier based on the weights of the first base classifier;

[0200] The processing module 503 is further configured to optimize the ensemble classifier by using the test sample set to obtain a transformer fault diagnosis model.

[0201] In a possible implementation manner, the sampling module 506 is further configured to sample the training set based on the updated sample weights to obtain a second training sample when the number of times of updating the sample weights corresponding to the training set does not reach a preset number of update times;

[0202] The training module 504 is further configured to train the second weak classifier of the classification model based on the second training sample to obtain a second base classifier and the training error corresponding to the second weak classifier;

[0203] The determining module 502 is further configured to determine the weight of the second base classifier based on the training error of the second weak classifier;

[0204] The processing module 503 is further configured to update the sample weights corresponding to the training set;

[0205] The processing module 503 is further configured to construct an ensemble classifier based on the weights of the first base classifier and the weights of the second base classifier.

[0206] The fault diagnosis device of a transformer provided in this embodiment can execute the method provided in the above method embodiment, and its implementation principle and technical effect are similar, which will not be elaborated here in this embodiment.

[0207] Figure 6 It is a schematic structural diagram of a fault diagnosis device of a transformer provided in this application. As Figure 6 shown, the electronic device 60 provided in this embodiment includes: at least one processor 601 and a memory 602. Optionally, the device 60 further includes a communication component 603. Wherein, the processor 601, the memory 602 and the communication component 603 are connected through a bus 604.

[0208] In a specific implementation process, at least one processor 601 executes the computer execution instructions stored in the memory 602, so that at least one processor 601 executes the above method.

[0209] The specific implementation process of the processor 601 can refer to the above method embodiment, and its implementation principle and technical effect are similar, which will not be elaborated here in this embodiment.

[0210] In the above embodiments, it should be understood that the processor may be a central processing unit (CPU for short), or other general-purpose processors, digital signal processors (DSP for short), application specific integrated circuits (ASIC for short), etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the method disclosed in combination with the invention can be directly embodied as being executed and completed by a hardware processor, or executed and completed by a combination of hardware and software modules in the processor.

[0211] The memory may include a high-speed memory (Random Access Memory, RAM), and may also include a non-volatile memory (Non-volatile Memory, NVM), such as at least one disk memory.

[0212] The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, the buses in the drawings of this application are not limited to only one bus or one type of bus.

[0213] This application also provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0214] This application also provides a computer-readable storage medium, in which computer-executable instructions are stored, and when the processor executes the computer-executable instructions, the above method is implemented.

[0215] The above-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, a magnetic disk or an optical disc. The readable storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0216] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be part of the processor. The processor and the readable storage medium can be located in an Application Specific Integrated Circuits (ASIC). Of course, the processor and the readable storage medium can also exist as discrete components in a device.

[0217] The division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Additionally, the couplings or direct couplings or communication connections shown or discussed between each other can be through some interfaces, and the indirect couplings or communication connections of devices or units can be in electrical, mechanical, or other forms.

[0218] The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0219] Furthermore, in each embodiment of the present invention, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0220] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or this part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0221] Those of ordinary skill in the art will understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments; and the foregoing storage medium includes: various media such as ROM, RAM, magnetic disk, or optical disk that can store program code.

[0222] Finally, it should be noted that: After considering the specification and practicing the invention disclosed herein, those skilled in the art will readily conceive of other embodiments of the present invention. The present invention is intended to cover any variations, uses, or adaptations of the present invention, which follow the general principles of the present invention and include known common knowledge or conventional technical means in the technical field not disclosed by the present invention. It is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present invention is only limited by the appended claims.

Claims

1. A fault diagnosis method for a transformer, characterized in that, Including: Obtaining the dissolved gas data in the oil of the transformer; Determining a first feature set according to the dissolved gas data in the oil, where the first feature set is used to indicate the absolute content of each dissolved gas, the relative content between multiple dissolved gases, and the gas ratio; Inputting the first feature set into a transformer fault diagnosis model to obtain a diagnosis result, where the transformer fault diagnosis model is trained after being jointly optimized by a historical feature set and initial parameters of a classification model, the historical feature set is determined from historical dissolved gas data in the oil, and the diagnosis result is used to indicate the fault type corresponding to the dissolved gas data in the oil.

2. The method according to claim 1, wherein Before obtaining the dissolved gas data in the oil of the transformer, the method further includes: Obtaining historical dissolved gas data in the oil and initial parameters of a classification model, and determining a historical feature set based on the historical dissolved gas data in the oil; Jointly optimizing the historical feature set and the initial parameters to obtain a second feature set and target parameters; Training the classification model using the second feature set and the target parameters to obtain a transformer fault diagnosis model.

3. The method according to claim 2, wherein The jointly optimizing the historical feature set and the initial parameters to obtain a second feature set and target parameters includes: Jointly processing the historical feature set and the initial parameters to obtain a plurality of input feature sets, and the plurality of input feature sets form an input feature population; Determining a first input feature set and a second input feature set from the plurality of input feature sets; Determining the fitness of the first input feature set and the second input feature set; In the case where the fitness of the first input feature set is less than the fitness of the second input feature set, performing an optimization process on the first input feature set to obtain an optimized first input feature set; Updating the input feature population based on the optimized first input feature set; Determining a second feature set and target parameters based on the updated input feature population.

4. The method according to claim 3, characterized in that, The performing an optimization process on the first input feature set to obtain an optimized first input feature set includes: Determining the adaptive crossover probability and the adaptive mutation probability of the first input feature set according to the fitness of the first input feature set; Performing a crossover and mutation operation on the first input feature set based on the adaptive crossover probability and the adaptive mutation probability to obtain an optimized first input feature set after the optimization process.

5. The method according to claim 3, wherein The determining the fitness of the first input feature set and the second input feature set includes: Performing an equal division process on the plurality of input feature sets to obtain a preset number of subsets; For any one of the preset number of subsets, determining the subset as a test set, and filtering the preset number of subsets based on the subset to obtain a training set; Training the classification model using the plurality of input feature sets in the training set to obtain an initial transformer fault diagnosis model; Evaluating the initial transformer fault diagnosis model using the test set to obtain the transformer fault diagnosis accuracy corresponding to each input feature set in the test set. Determine the accuracy rate of the transformer fault diagnosis as the fitness of the corresponding input feature set; Based on the fitness of each input feature set in the test set, determine the fitness of the first input feature set and the second input feature set.

6. The method according to claim 2, wherein The training of the classification model using the second feature set and the target parameter to obtain a transformer fault diagnosis model includes: Divide a plurality of the second feature sets into a training sample set and a test sample set, where the second feature set carries a class label, and the class label is used to indicate the fault type of the transformer; Sample the training sample set according to a preset sample weight to obtain a first training sample; Determine the standard deviation of the first training sample, and determine the standard deviation as the kernel parameter of the first weak classifier of the classification model; Train the first weak classifier using the first training sample to obtain a first base classifier and the training error of the first weak classifier; Determine the weight of the first base classifier based on the training error of the first weak classifier, and update the sample weight corresponding to the training set; Construct an ensemble classifier based on the weight of the first base classifier; Optimize the ensemble classifier using the test sample set to obtain a transformer fault diagnosis model.

7. The method according to claim 6, characterized in that, After determining the weight of the first base classifier based on the training error and updating the sample weight corresponding to the training set, the method further includes: In the case where the number of times of updating the sample weight corresponding to the training set does not reach the preset number of updates, sample the training set based on the updated sample weight to obtain a second training sample; Train the second weak classifier of the classification model using the second training sample to obtain a second base classifier and the training error corresponding to the second weak classifier; Determine the weight of the second base classifier based on the training error of the second weak classifier, and update the sample weight corresponding to the training set; Construct an ensemble classifier based on the weight of the first base classifier and the weight of the second base classifier.

8. A fault diagnosis device for a transformer, characterized in that, Including: An acquisition module, configured to acquire the dissolved gas data in the transformer oil; A determination module, configured to determine a first feature set according to the dissolved gas data in the transformer oil; A processing module, configured to input the first feature set into a transformer fault diagnosis model to obtain a diagnosis result.

9. A fault diagnosis device for a transformer, characterized in that, Including: A memory, a processor; The memory stores computer execution instructions; The processor executes the computer execution instructions stored in the memory, so that the processor executes the method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer execution instructions, and when the computer execution instructions are executed by a processor, they are used to implement the method according to any one of claims 1-7.

Images