An IGWO-SVM transformer fault diagnosis method based on PCA feature extraction

Abstract

The application discloses an IGWO-SVM transformer fault diagnosis method based on PCA feature extraction, and comprises the following steps: performing dimension reduction processing on original nonlinear transformer fault characteristic data; optimizing the kernel function parameters and the penalty factor of SVM by using an optimization algorithm to obtain an optimal IGWO-SVM model; and inputting the dimension-reduced data into the IGWO-SVM model to perform transformer fault diagnosis. The application firstly selects the features as input variables by using PCA to eliminate redundant, invalid and overlapped information in the fault data set, so that the precision and efficiency of transformer fault diagnosis can be improved; secondly, the traditional GWO is improved by introducing a good point set initialization population strategy, a nonlinear convergence factor adjustment strategy and a position updating strategy based on memory guidance, and performance test comparison is conducted on the IGWO and the traditional algorithm, so that it is proved that the convergence speed and the precision of the IGWO are significantly improved. Finally, the IGWO is used for optimizing the kernel function parameters and the penalty factor of SVM, and an IGWO-SVM fault diagnosis model is constructed.

Description

TECHNICAL FIELD

[0001] The application relates to an IGWO-SVM transformer fault diagnosis method based on PCA feature extraction and belongs to the field of transformer fault diagnosis. BACKGROUND

[0002] A transformer is a key device for realizing power transmission, distribution and use in important places such as power systems, petrochemical industry, transportation and urban construction. Once a fault or anomaly occurs, it may cause power interruption, energy waste, and even major safety accidents, resulting in huge economic losses to individuals and the state. Therefore, efficiently and accurately identifying the fault type of a transformer is of great significance to ensure the normal stability of a power system.

[0003] At present, the dissolved gas analysis (DGA) method is widely used in transformer fault diagnosis, including traditional methods such as IEC three-ratio method, Rogers four-ratio method and code-free ratio method. However, the method has certain limitations in actual engineering application. With the continuous development of artificial intelligence technology, intelligent algorithms can significantly improve the accuracy of transformer fault diagnosis. Common intelligent algorithms include neural networks, decision trees, expert systems and support vector machines. Literature 1 converts the identity mapping learning in the traditional BP neural network into residual learning by stacking multiple residual network modules, solving the problem of poor performance of transformer diagnosis under small sample data. However, artificial neural network model programming is complex and the training time is long. Literature 2 optimizes the decision tree number and maximum split number of Logit Boost-decision tree using whale optimization algorithm, effectively improving the accuracy of transformer fault diagnosis. However, the decision tree algorithm has poor stability and is prone to overfitting. Literature 3 integrates the fault diagnosis rules obtained by traditional three-ratio method, statistical learning method and practical experience method, and establishes an oil dissolved gas composition power transformer fault diagnosis expert system that can expand new rules. However, due to the limitations of learning ability and portability, the expert system is not comprehensive enough in fault diagnosis. Literature 4 uses support vector machine for transformer fault diagnosis, which can better handle small sample and nonlinear data and has strong generalization ability. Compared with other methods, support vector machine (SVM) is particularly suitable for handling small sample and multiple fault types of transformer. However, the classification performance of SVM mainly depends on the kernel function parameters and the penalty factor, and selecting appropriate parameter values is crucial to the diagnosis result. Therefore, it is necessary to study new intelligent algorithms.

[0004] Document 1: ZHAO Wenqing, YAN Hai, ZHOU Zhendong, et al. Transformer fault diagnosis based on residual BP neural network[J]. Electric Power Automation Equipment, 2020, 40(02): 143-148.

[0005] Document 2: ZHANG Guozhi, CHEN Kang, FANG Rongxing, et al. Transformer fault diagnosis method based on DGA and whale algorithm optimized Logit Boost-decision tree[J]. Power System Protection and Control, 2023, 51(07): 63-72.

[0006] Document 3: SHI Ruifeng, SHI Yongfeng, MOO Jun, et al. Oil dissolved gas power transformer fault diagnosis expert system[J]. Power System Technology, 2014, 26(12): 49-54.

[0007] Document 4: YUAN F, ZHAID, WU Q. Blasting vibration velocity prediction based on least squares support vector machine with particle swarm optimization algorithm[J]. Geosystem Engineering, 2019, 22(5): 279-288. SUMMARY

[0008] The application provides an IGWO-SVM transformer fault diagnosis method based on PCA feature extraction, which is used for diagnosing transformer faults.

[0009] The technical scheme of the application is: an IGWO-SVM transformer fault diagnosis method based on PCA feature extraction, comprising: performing dimension reduction processing on original nonlinear transformer fault feature data; using an optimization algorithm to optimize the kernel function parameters and the penalty factor of SVM to obtain an optimal IGWO-SVM model; and inputting the dimension-reduced data into the IGWO-SVM model for transformer fault diagnosis.

[0010] The original nonlinear transformer fault feature data is dimensionally reduced, specifically: the original nonlinear transformer fault feature data is dimensionally reduced by a PCA method; and the first m main features with a cumulative contribution rate exceeding a first preset threshold and a characteristic value less than a second preset threshold are selected as the PCA principal component number.

[0011] The first preset threshold is 90%, and the second preset threshold is 1.

[0012] The optimization algorithm adopts an IGWO optimization algorithm, the IGWO optimization algorithm is improved on the basis of a traditional GWO optimization algorithm, and specifically includes: introducing a good point set initialization population strategy to optimize the selection of an initial population; a nonlinear convergence factor adjustment strategy is proposed; and a position updating strategy based on memory guidance is established.

[0013] The expression of the nonlinear convergence factor adjustment strategy is:

[0014]

[0015] In the formula, c represents a convergence factor, t represents a current iteration number, T represents a maximum iteration number, and c represents a convergence factor. max

[0016] The position updating strategy based on memory guidance is specifically: an inertial weight is introduced to represent the importance of different levels of grey wolves; and according to the importance, a grey wolf position updating expression is determined.

[0017] The inertial weight w is expressed as:

[0018]

[0019] In the formula, the size of the inertial weight w is increased from an initial value w initial to a terminal value w final linearly with the increase of the iteration number; t represents a current iteration number, T represents a maximum iteration number. max

[0020] The grey wolf position updating expression is:

[0021]

[0022] In the formula, x represents a best position of a current solution, u1, u2, e1 and e2 are random numbers uniformly distributed between 0 and 1, e1 and e2 represent an individual learning factor and a global learning factor respectively; x represents a position vector of a grey wolf; α, β and δ represent direction vectors of a next movement of a candidate wolf ω;

[0023] ​​​According to another aspect of the present application, there is provided an IGWO-SVM transformer fault diagnosis system based on PCA feature extraction, comprising: a dimension reduction module for performing dimension reduction processing on original nonlinear transformer fault feature data; a construction module for optimizing kernel function parameters and penalty factors of SVM using an IGWO optimization algorithm to obtain an optimal IGWO-SVM model; and a diagnosis module for inputting the dimension-reduced data into the IGWO-SVM model for transformer fault diagnosis.

[0024] According to another aspect of the present application, there is provided a processor for running a program, wherein the program performs the IGWO-SVM transformer fault diagnosis method based on PCA feature extraction when running.

[0025] The present application has the following advantages: the present application first optimizes features as input variables using principal component analysis (PCA) to eliminate redundant, invalid, and overlapping information in the fault data set, and the fault data processed by PCA dimension reduction can effectively represent the fault features of the fault data set and improve the accuracy and efficiency of transformer fault diagnosis. Second, the traditional grey wolf algorithm (GWO) is improved by introducing a good point set initialization population strategy, a nonlinear convergence factor adjustment strategy, and a position update strategy based on memory guidance, and the improved grey wolf algorithm (IGWO) is compared with GWO, genetic algorithm (GA), and particle swarm optimization (PSO) in performance testing, which proves that the convergence speed and accuracy of IGWO are significantly improved. Finally, IGWO is used to optimize the kernel function parameters and penalty factors of support vector machine (SVM) to construct an IGWO-SVM fault diagnosis model. Simulation results show that compared with PSO-SVM, GA-SVM, and GWO-SVM, the IGWO-SVM fault diagnosis model can effectively improve the accuracy of transformer fault diagnosis, and the feasibility of the proposed method is verified. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is a flowchart of the method of the present application;

[0027] Figure 2 is a graph showing the relationship between the number of different principal components and the eigenvalues obtained using PCA;

[0028] Figure 3 is a graph showing the relationship between the number of different principal components and the cumulative contribution rate obtained using PCA;

[0029] Figure 4 is an optimization convergence curve graph for testing the first test function using different algorithms;

[0030] Figure 5 is an optimization convergence curve graph for testing the second test function using different algorithms;

[0031] Figure 6 This is a graph showing the optimization convergence curves of the third test function tested by different algorithms;

[0032] Figure 7 This is a graph showing the optimization convergence curves of the fourth test function tested by different algorithms;

[0033] Figure 8 This is a classification result diagram of the PSO-SVM fault diagnosis model;

[0034] Figure 9 This is a diagram showing the classification results of the GA-SVM fault diagnosis model;

[0035] Figure 10 This is a classification result diagram of the GWO-SVM fault diagnosis model;

[0036] Figure 11 This is a classification result diagram of the IGWO-SVM fault diagnosis model;

[0037] Figure 12 This is a system block diagram of the present invention. Detailed Implementation

[0038] The invention will be further described below with reference to the accompanying drawings and embodiments, but the scope of the invention is not limited to the description.

[0039] Example 1: As Figures 1-11 As shown, according to one aspect of the present invention, an IGWO-SVM transformer fault diagnosis method based on PCA feature extraction is provided, comprising: performing dimensionality reduction processing on the original nonlinear transformer fault feature data using the PCA method; optimizing the kernel function parameters and penalty factor of the SVM using the IGWO optimization algorithm to obtain the optimal IGWO-SVM model; and inputting the dimensionality-reduced data into the IGWO-SVM model for transformer fault diagnosis.

[0040] Furthermore, the dimensionality reduction processing of the original nonlinear transformer fault feature data specifically involves: performing dimensionality reduction processing on the original nonlinear transformer fault feature data using the PCA method; selecting the top m principal features whose cumulative contribution rate exceeds a first preset threshold and whose feature value is less than a second preset threshold as the number of PCA principal components.

[0041] Furthermore, the first preset threshold is set to 90%, and the second preset threshold is set to 1.

[0042] Further, the IGWO optimization algorithm is improved on the basis of the traditional GWO optimization algorithm, and specifically includes: introducing a good point set initialization population strategy to optimize the selection of the initial population; a nonlinear convergence factor adjustment strategy is proposed; a position updating strategy based on memory guidance is established. Among them, the traditional GWO simulates the hunting behavior of gray wolves based on the social hierarchy system of wolf packs, and the three gray wolves with the best fitness in the wolf pack are marked as alpha, beta and delta, and the remaining gray wolves are marked as omega. GWO can be described as the process of these three wolves leading the wolf pack to surround the prey, chase the prey and attack the prey.

[0043] Further, the expression of the nonlinear convergence factor adjustment strategy is:

[0044]

[0045] In the formula: Convergence factor, t is the current iteration number, T max is the maximum iteration number.

[0046] Further, the position updating strategy based on memory guidance is specifically: introducing an inertia weight to represent the importance of gray wolves of different levels; according to the importance, the position updating expression of the gray wolf is determined.

[0047] Further, the inertia weight w is expressed as:

[0048]

[0049] In the formula, the inertia weight w decreases linearly from the initial value w initial to the final value w final ; t is the current iteration number, T max is the maximum iteration number.

[0050] Further, the position updating expression of the gray wolf is:

[0051]

[0052] In the formula, is the best position of the current solution, u1, u2, e1, e2 are random numbers uniformly distributed between 0 and 1, and e1 and e2 represent individual learning factor and global learning factor, respectively; is the position vector of the gray wolf; represent the direction vector of the candidate wolf omega determined by alpha, beta and delta for the next movement; is the individual historical best position.

[0053] Further, the SVM in the application adopts a one-to-one multi-class SVM method.

[0054] The performance of the present invention will be described below with reference to experimental data:

[0055] I. Performance Comparison Analysis of Introducing PCA into SVM

[0056] Traditional transformer fault diagnosis methods often select five characteristic gases containing hydrogen (H2, CH4, C2H4, C2H2, C2H6) directly as the diagnostic basis, without considering the problem that using a single gas as input will lead to low fault diagnosis accuracy. Therefore, this invention uses a non-coded ratio method to expand the DGA fault characteristic gases, adding the original five gas characteristics, resulting in a total of 14 dimensions of fault characteristics, denoted as T = {T1, T2, ..., T...}. 14 As shown in Table 1. Where C1 is the sum of the contents of alkane gases represented by CH4, C1 = CH4; C2 is the sum of the contents of unsaturated hydrocarbon gases, C2 = C2H2 + C2H4 + C2H6.

[0057] Table 1 Multidimensional Fault Characteristics of Transformers

[0058]

[0059] This invention employs PCA to reduce the dimensionality of the fault feature dataset, using the 14 fault features in Table 1 as input variables and calculating their principal component contribution rates and eigenvalues, such as... Figure 2 , 3 As shown. Based on the magnitude of the principal component eigenvalues ​​and whether the cumulative contribution rate reaches 90%, the correlation between fault characteristic gas data and fault type can be determined. Figure 2 , 3 It can be seen that when the dimension of the fault features is reduced to 7, the eigenvalues ​​are less than 1 and the cumulative principal component contribution rate reaches 96.29%, which already covers most of the effective information. Therefore, the number of input variables after dimensionality reduction is determined to be 7.

[0060] To verify the improvement effect of PCA-reduced fault data on fault diagnosis, traditional SVM was applied to diagnose faults in the data before and after dimensionality reduction, and their running time and diagnostic accuracy were compared. The diagnostic results are shown in Table 2. As can be seen from Table 2, the PCA-based feature extraction method of this invention can improve the SVM fault diagnosis accuracy from 75.25% before dimensionality reduction to 89.57% while removing redundant, invalid, and overlapping information from the dataset.

[0061] Table 2 Comparison before and after feature extraction

[0062]

[0063] II. Performance Testing of IGWO Optimization Algorithm

[0064] In order to verify the performance of IGWO, four kinds of classical test functions are selected for simulation experiment, and are compared and analyzed with GWO, particle swarm optimization algorithm (PSO) and genetic optimization algorithm (GA). The test function information is shown in Table 3. Among them, F1 and F4 are multi-peak functions, mainly used for testing the convergence speed and solution accuracy of the algorithm, F2 and F3 are single-peak functions, mainly used for testing the ability of the algorithm to prevent falling into local optimum. 50 independent experiments are carried out for each test function, and the maximum iteration number is 500, so that the optimization convergence curves of the four algorithms on different test functions can be obtained as shown in Figures 4-7

[0065] Table 3 Test function information

[0066]

[0067] It can be seen from Figure 4 and Figure 6 that for functions F1 and F3, the solution accuracy of IGWO is obviously higher than that of the other three algorithms, and PSO, GA falls into local optimum; it can be seen from Figure 5 and Figure 7 that for functions F2 and F4, the convergence speed of IGWO is obviously accelerated, and the solution accuracy is also higher than that of PSO, GA and GWO. This shows that whether in complex multi-peak function or simple single-peak function, IGWO shows faster convergence speed and solution accuracy, and compared with the other three algorithms, IGWO is more likely to find the optimal solution in the optimization process.

[0068] Third, comparison and analysis of classification results of the present application and different algorithms

[0069] The optimization interval of the SVM kernel function parameter γ is [0, 0.01], and the optimization interval of the penalty coefficient c is [1, 500]. Therefore, the optimal (c, γ) combination is found by using IGWO. The IGWO-SVM fault diagnosis process is shown in Figure 1 , including data preprocessing, IGWO optimization and transformer fault diagnosis.

[0070] The specific steps of IGWO-SVM are as follows:

[0071] S1, obtain the transformer fault feature data and perform PCA dimension reduction processing, and then divide into training set and test set according to the ratio of 8:2.

[0072] S2, normalize the data processed in step S1, set the population size N, the optimization interval of the SVM kernel function parameter γ and the penalty coefficient c, the convergence factor , the coefficient vector and determined by the maximum iteration number T max ​, the grey wolf population is initialized by using the good point set theory;

[0073] S3, input the training set, the SVM machine training according to the training set data, the fitness value of each grey wolf under the current (c, γ) is calculated, the top three are respectively as alpha, beta, delta wolf, and the position of the candidate wolf ω is updated.

[0074] S4, the convergence factor is updated according to the nonlinear convergence factor adjustment strategy, and the traditional update formula is used to update And The position of each grey wolf is updated according to the grey wolf position update expression.

[0075] S5, the fitness of the updated grey wolf position is calculated, and compared with the fitness value of the last iteration. If the new fitness value exceeds the original value, the updated position is regarded as the optimal fitness value; otherwise, the original fitness value remains unchanged.

[0076] S6, judge whether the maximum iteration number is reached. If not, return to step S4; if yes, output the optimal (c, γ) combination and substitute it into the IGWO-SVM model.

[0077] S7, input the test set data into the IGWO-SVM fault diagnosis model with optimal parameters, and output the fault classification result.

[0078] The DGA data used in the application is derived from the online monitoring data of the oil chromatogram of a certain power grid, and there are 1614 data. In order to verify the feasibility and effectiveness of the fault diagnosis method of the application, 640 fault data are randomly extracted for feature extraction by PCA, 7-dimensional feature variables are obtained, and the training and testing are carried out according to the ratio of 8:2, and the specific sample data distribution is shown in table 4. According to the DL / T722-2014 guide, the fault types of transformer are divided into: normal state, high temperature overheating, medium temperature overheating, low temperature overheating, high energy discharge, low energy discharge, partial discharge and arc discharge, and numbered as 0-7.

[0079] Table 4 fault sample distribution

[0080]

[0081] On the MATLAB simulation platform, IGWO-SVM, GWO-SVM, GA-SVM and PSO-SVM fault diagnosis models are constructed, and the transformer fault samples after PCA feature extraction are trained and tested respectively, and the classification results and confusion matrix results of each model are shown in Figures 8-11 As shown in table 5, the fault diagnosis accuracy of different diagnosis models is shown.

[0082] Table 5 comparison of fault diagnosis accuracy of different diagnosis models

[0083]

[0084] By Figure 8 It can be known that 18 errors occur in total when the transformer fault diagnosis is performed by using the PSO-SVM model, which are 2 normal state diagnosis errors, 7 overheat fault diagnosis errors and 9 discharge fault diagnosis errors. Among them, the 7 overheat fault diagnosis errors are 4 high-temperature overheat errors, 1 medium-temperature overheat error and 2 low-temperature overheat errors respectively; the 9 discharge fault diagnosis errors are 5 local discharge errors, 2 low-energy discharge errors and 2 high-energy discharge errors respectively, and the comprehensive fault diagnosis accuracy is 85.94%.

[0085] By Figure 9 It can be known that 11 errors occur in total when the transformer fault diagnosis is performed by using the GA-SVM model, which are 2 normal state diagnosis errors, 1 high-temperature overheat fault diagnosis error and 8 discharge fault diagnosis errors. Among them, the 8 discharge fault diagnosis errors are 5 local discharge errors, 1 low-energy discharge error and 2 high-energy discharge errors respectively, and the comprehensive fault diagnosis accuracy is 91.41%.

[0086] By Figure 10 It can be known that 8 errors occur in total when the transformer fault diagnosis is performed by using the GWO-SVM model, which are 2 normal state diagnosis errors, 1 high-temperature overheat fault diagnosis error and 5 discharge fault diagnosis errors. Among them, the 5 discharge fault diagnosis errors are 2 local discharge errors, 1 low-energy discharge error and 2 high-energy discharge errors respectively, and the comprehensive fault diagnosis accuracy is 93.75%.

[0087] By Figure 11 It can be known that, when the transformer fault diagnosis is performed by using the IGWO-SVM model, only 1 normal state diagnosis error, 2 local discharge diagnosis errors, 1 low-energy discharge diagnosis error, 1 high-temperature overheat fault diagnosis error and 1 high-energy discharge diagnosis error occur, and the comprehensive fault diagnosis accuracy is 95.31%. The four diagnosis methods all have 100% fault diagnosis rate for arc discharge.

[0088] By comparing the IGWO-SVM model of the application with the GWO-SVM, GA-SVM and PSO-SVM models, it can be known that the diagnosis performance of the application is optimal, and the diagnosis accuracy of the application is increased by 1.56%, 2.34% and 5.47% respectively compared with the three diagnosis methods, which indicates that the IGWO-SVM model has better classification performance.

[0089] Through the experimental analysis, it is known that the IGWO-SVM diagnosis model has strong generalization ability, is suitable for the diagnosis of the dissolved gas in the transformer oil, can be applied to the fault classification in other fields, and has certain research value and engineering practical significance.

[0090] According to another aspect of the present application, as shown in Figure 12 The present application provides an IGWO-SVM transformer fault diagnosis system based on PCA feature extraction, comprising: a dimension reduction module for performing dimension reduction processing on original nonlinear transformer fault feature data; a construction module for optimizing the kernel function parameters and the penalty factor of SVM by using an IGWO optimization algorithm to obtain an optimal IGWO-SVM model; and a diagnosis module for inputting the dimension-reduced data into the IGWO-SVM model for transformer fault diagnosis. For parts not described in detail in the above, please refer to the related description of the embodiments.

[0091] According to another aspect of the present application, a processor is provided for running a program, wherein the program performs the IGWO-SVM transformer fault diagnosis method based on PCA feature extraction when running.

[0092] The specific embodiments of the present application are described in detail above in combination with the accompanying drawings, but the present application is not limited to the above embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the purpose of the present application.

Claims

1. A transformer fault diagnosis method based on PCA feature extraction of IGWO-SVM, characterized in that, The method comprises the following steps: dimension reduction is performed on original nonlinear transformer fault feature data; an optimization algorithm is used to optimize kernel function parameters and a penalty factor of SVM to obtain an optimal IGWO-SVM model; dimension-reduced data is input into the IGWO-SVM model for transformer fault diagnosis; the optimization algorithm is an IGWO optimization algorithm, which is improved on the basis of a traditional GWO optimization algorithm and specifically comprises the following steps: a good point set initialization population strategy is introduced to optimize selection of an initial population; a nonlinear convergence factor adjustment strategy is proposed; a position updating strategy based on memory guidance is established; an expression of the nonlinear convergence factor adjustment strategy is as follows: ； In the formula: denotes a convergence factor, is the current iteration number, is the maximum iteration number; the position updating strategy based on memory guidance is specifically as follows: an inertia weight is introduced to represent importance of different levels of grey wolves; a grey wolf position updating expression is determined according to the importance.

2. The IGWO-SVM transformer fault diagnosis method based on PCA feature extraction according to claim 1, characterized in that, The dimension reduction performed on the original nonlinear transformer fault feature data is specifically as follows: PCA is used to perform dimension reduction on the original nonlinear transformer fault feature data; and m main features with a cumulative contribution rate exceeding a first preset threshold and a feature value being less than a second preset threshold are selected as PCA principal component numbers.

3. The IGWO-SVM transformer fault diagnosis method based on PCA feature extraction according to claim 2, characterized in that, The first preset threshold is 90%, and the second preset threshold is 1.

4. The IGWO-SVM transformer fault diagnosis method based on PCA feature extraction according to claim 1, characterized in that, the inertial weight , the expression is: ； where the inertia weight increases from an initial value to a final value ; is the current iteration number, is the maximum iteration number.

5. The IGWO-SVM transformer fault diagnosis method based on PCA feature extraction according to claim 4, characterized in that, The grey wolf position updating expression is as follows: ； wherein, is the best position of the current solution, , are random numbers uniformly distributed between [0, 1], and denote the individual and global learning factors, respectively; is the position vector of the grey wolf; , , represent , , determines the candidate wolf the direction vector of the next move; is the individual historical best position.

6. A transformer fault diagnosis system based on PCA feature extraction of IGWO-SVM, characterized in that, The method comprises the following steps: a dimension reduction module is configured to perform dimension reduction on original nonlinear transformer fault feature data; a construction module is configured to use an IGWO optimization algorithm to optimize kernel function parameters and a penalty factor of SVM to obtain an optimal IGWO-SVM model; a diagnosis module is configured to input dimension-reduced data into the IGWO-SVM model for transformer fault diagnosis; the optimization algorithm is an IGWO optimization algorithm, which is improved on the basis of a traditional GWO optimization algorithm and specifically comprises the following steps: a good point set initialization population strategy is introduced to optimize selection of an initial population; a nonlinear convergence factor adjustment strategy is proposed; a position updating strategy based on memory guidance is established; an expression of the nonlinear convergence factor adjustment strategy is as follows: ； In the formula: denotes a convergence factor, is the current iteration number, is the maximum iteration number; the position updating strategy based on memory guidance is specifically as follows: an inertia weight is introduced to represent importance of different levels of grey wolves; a grey wolf position updating expression is determined according to the importance.

7. A processor, comprising: The processor is configured to run a program, and the program performs the IGWO-SVM transformer fault diagnosis method based on PCA feature extraction when running.

Images