Electromagnetic voltage transformer ferromagnetic resonance fault identification method, system and device and storage medium

Abstract

The invention relates to the technical field of power distribution network fault detection, in particular to an electromagnetic voltage transformer ferromagnetic resonance fault identification method, system and device and a storage medium. The identification method comprises the following steps: collecting, preprocessing and analyzing a zero-sequence voltage signal of a primary side of a power distribution network system; feature indexes of the low-frequency IMF component in the time domain and the frequency domain are solved and subjected to dimensionality reduction reconstruction, and the most appropriate dimensionality is determined to form a fault recognition feature vector matrix; and establishing a sample set, training a sample subset, building, training and configuring an RF model to construct a fault recognition model, and performing fault recognition on the sample set to judge the fault type of the sample. The improved CEEMD time-frequency signal decomposition method based on the T test and the variance contribution rate is provided for overcoming the defects that the number of IMF components obtained through CEEMD decomposition and the frequency band of the IMF components are not fixed, and a few pseudo-modal IMF components appear in a decomposition result, and KPCA is introduced for feature dimension reduction to solve the problem that redundant information exists due to excessive feature quantities.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network fault detection, and in particular to a method, system, device and storage medium for identifying ferromagnetic resonance fault of an electromagnetic voltage transformer. Background Art

[0002] At present, most distribution networks are equipped with microcomputer detuning devices to detect ferromagnetic resonance faults of electromagnetic voltage transformers. The detection method is to adopt the traditional time-frequency domain signal threshold method, which detects the zero-sequence voltage signal of the distribution network system in real time, quickly calculates its amplitude, spectrum and other time-domain and frequency-domain related characteristic information, and compares it with the pre-set threshold criteria, and then analyzes the system working status and determines whether a single-phase grounding fault or ferromagnetic resonance fault occurs.

[0003] The two core steps of ferromagnetic resonance overvoltage classification and identification technology include the extraction of fault feature quantities and the research of fault classification and identification methods. At present, the feature extraction methods that are widely used in the power field include wavelet transform (WT), empirical mode decomposition (EMD), ensemble empirical mode decomposition (EEMD), complete ensemble empirical mode decomposition (CEEMD), etc.

[0004] Pattern recognition technology is widely used in the field of power system fault identification, such as Gaussian Mixture Model (GMM), Naive Bayes (NB), Decision Tree (DT), K-Nearest Neighbor (KNN) and Random Forest (RF).

[0005] By real-time detection of the zero-sequence voltage signal of the distribution network system, a threshold is set to detect whether ferromagnetic resonance occurs. The threshold criterion of this method is relatively easy to obtain, and there is no special requirement for the detection position. However, due to the nonlinear characteristics of the iron core of the electromagnetic voltage transformer, the resonant voltage signal when a ferromagnetic resonance fault occurs in the distribution network system is complex and changeable. The signal does not simply contain the fundamental frequency component, but produces a hybrid ferromagnetic resonance with different resonant frequencies. The zero-sequence voltage when a single-phase grounding fault occurs also contains fundamental frequency components to varying degrees, which makes it very easy to cause misjudgment between the two types of faults. Therefore, the use of the traditional time-frequency domain signal threshold method has certain limitations.

[0006] For zero-sequence voltage signals in different operating states in the distribution network system, the wavelet basis selection of WT decomposition cannot be changed during the analysis process, and the frequency of each layer of the zero-sequence voltage signal cannot be accurately separated, resulting in the loss of local characteristics of the signal. EMD decomposition is prone to IMF modal aliasing, which causes feature extraction failure and is not conducive to accurate fault identification. The EEMD decomposition result has the problem of excessive residual noise causing the load characteristic information to be submerged. The CEEMD decomposition method will produce invalid pseudo-modal IMF components in its decomposition results. At the same time, the number of IMF components obtained by decomposing different zero-sequence voltage signals is not fixed, and the fault characteristics may exist in different IMF components. Summary of the invention

[0007] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or be understood by implementing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the description and other drawings of the description.

[0008] The purpose of the present invention is to overcome the above-mentioned shortcomings. The present invention provides an identification method and an identification method for ferromagnetic resonance fault of an electromagnetic voltage transformer. The present invention firstly proposes an improved CEEMD time-frequency signal decomposition method based on T test and variance contribution rate to address the shortcomings of low recognition accuracy of the traditional single threshold feature detection method, and then proposes a multi-feature detection method based on time domain and frequency domain. Finally, in order to solve the problem of redundant information due to too many feature quantities, KPCA is introduced to perform feature dimension reduction.

[0009] The technical solution adopted by the present invention to solve the technical problem is: a method for identifying ferromagnetic resonance fault of an electromagnetic voltage transformer, comprising the following steps: Collect and pre-process the zero-sequence voltage signal on the primary side of the distribution network system; Analyze the preprocessed zero-sequence voltage signal to obtain the reconstructed low-frequency IMF component; Solve and reduce the dimension to reconstruct the characteristic index of low-frequency IMF components in time domain and frequency domain, and determine the most suitable dimension to form the characteristic vector matrix of electromagnetic voltage transformer ferromagnetic resonance fault identification based on multiple characteristic quantities in time domain and frequency domain; Establish sample sets and training sample subsets, build, train, and configure RF models to construct an electromagnetic voltage transformer ferromagnetic resonance fault identification model based on WOA optimized RF; The WOA algorithm is used to iteratively search and solve the TreeBagger function; The parameters of the RF model are set, and the ferromagnetic resonance fault identification model of electromagnetic voltage transformer based on WOA optimized RF is used to identify faults in the sample set and determine the fault type of the sample.

[0010] In some embodiments, the preprocessed zero-sequence voltage signal is analyzed to obtain a reconstructed low-frequency IMF component; specifically: The preprocessed zero-sequence voltage signal is analyzed in time domain and frequency domain using the improved CEEMD method based on T test and variance contribution rate to obtain the reconstructed low-frequency IMF component that can characterize the original signal.

[0011] In some embodiments, the specific steps of performing time domain and frequency domain analysis on the pre-processed zero-sequence voltage signal using the improved CEEMD method based on T test and variance contribution rate are as follows: In the original signal x ( t ), add positive and negative Gaussian white noises with opposite numbers to each other. M times, get 2 M A new signal containing positive and negative Gaussian white noise X n,m ( t ), whose expression is: , In the above formula, w m ( t ) is Gaussian white noise, m Indicates the number of times positive (negative) Gaussian white noise is added, m =1, 2, …, M ; (-1) n represents the positive and negative coefficients of Gaussian white noise, n =1, 2; 2 M A signal containing positive (negative) Gaussian white noise X n,m ( t ) Perform EMD decomposition one by one to obtain each signal X n,m ( t ) corresponds to a set of IMF components, defining m The signal is decomposed by EMD to obtain the i The IMF components are f m,i ( t ); 2 M Group signal X n,m ( t ) corresponding to the IMF component fm,i ( t ) and perform the sum and average operation to obtain x ( t ) IMF components of each layer after traditional CEEMD decomposition F i ( t ), whose expression is: , For each layer of IMF components F i ( t ) to conduct a T test, that is, F 1(t) is recorded as the index A 1, F 1( t )+ F 2( t ) is recorded as an indicator A 2, and so on, i indivual F i ( t ) and is recorded as an index A i , respectively calculate the index A 1 to index A i The mean of , and whether the mean is significantly different from 0, assuming ,and a is an integer, if the index A a is significantly not 0, we can get hour, F i ( t ) is the IMF component of the high frequency part; hour, F i ( t ) is the IMF component of the low-frequency part, and the expression of its T test is: , In the above formula, For indicators A i The mean of For indicators A i The standard deviation of n For indicators A i Sample size; For the high frequency IMF component F i ( t ), calculate their variance contribution rates, that is, each IMF componentF i ( t )and x ( t ) signal variance ratio, and then sort the variance contribution rates from high to low. Similarly, the sorting of the variance contribution rates of the IMF components in the low-frequency part is obtained; The high-frequency IMF components ranked first and second in terms of variance contribution rate are reconstructed to obtain the reconstructed high-frequency IMF components. Similarly, the reconstructed low-frequency IMF components are obtained.

[0012] In some embodiments, the characteristic indexes of the low-frequency IMF components in the time domain and the frequency domain are solved and reconstructed by reducing the dimension, and the most suitable dimension is determined to form a characteristic vector matrix for identifying ferromagnetic resonance faults of electromagnetic voltage transformers based on multiple characteristic quantities in the time domain and the frequency domain; specifically: The 33 characteristic indicators of the reconstructed low-frequency IMF components in the time domain and frequency domain are solved, and the KPCA method is used for feature dimensionality reduction. According to the standard that the percentage of variance explained by the first n principal components in the cumulative total variance exceeds 95% for the first time, the most suitable dimension of the new data after dimensionality reduction is determined to be 12, and the eigenvector matrix for ferromagnetic resonance fault identification of electromagnetic voltage transformer based on multiple characteristic quantities in time domain and frequency domain is formed.

[0013] In some embodiments, the KPCA method is used to perform feature dimensionality reduction, comprising the following steps: Normalize the data samples and select the radial basis kernel function for nonlinear mapping; Calculate the kernel matrix, the eigenvectors and eigenvalues ​​of the kernel matrix, calculate the percentage of variance explained by each principal component in the total variance and sort them from large to small, calculate the cumulative percentage of variance explained by the principal component in the total variance, and draw the Pareto chart of KPCA for analysis; The most appropriate data dimension is determined based on the Pareto chart of KPCA. When the cumulative percentage of the variance explained by the principal component accounts for more than 95% of the total variance, it is considered that the data of the corresponding dimension can effectively represent most of the characteristic information of the initial data. The eigenvector of the corresponding dimension is extracted and the projection of the kernel matrix on the vector is calculated to obtain the new data after dimensionality reduction.

[0014] In some embodiments, the KPCA method is used to perform feature dimensionality reduction, specifically: Assume that the initial data is , and zero-mean, n Represents the quantity dimension of the initial data, X Perform nonlinear mapping to high-dimensional space, Represents a nonlinear mapping function, forming a high-dimensional mapping data vector ,in, express dimensional high-dimensional mapping vector, ,and i is an integer, , m is the characteristic dimension of the initial data; Mapping data vectors to high dimensions Perform covariance calculation and calculate The covariance matrix of , and the covariance matrix Performing eigendecomposition operation, we get: , In the above formula, Represents the covariance matrix The characteristic value of express The corresponding dimensional feature vector, ,and j is an integer; The feature vector Mapping vectors in high dimensions Performing linear representation, we get: , In the above formula, express dimensional column vector, express Middle i elements; Combine the above two equations and multiply both sides by ,get: : Define the kernel function matrix , then the above formula can be simplified to: ; According to the above formula, we can get , the high-dimensional mapping data vector In the feature vector The initial data can be obtained by projecting in the direction New data after dimensionality reduction , whose expression is: .

[0015] In some embodiments, the 33 characteristic indicators of the reconstructed low-frequency IMF component in the time domain and the frequency domain include maximum value, minimum value, average value, median value, peak value, rectified average value, variance, standard deviation, kurtosis, skewness, root mean square value, root mean square amplitude, waveform factor, peak factor, pulse factor, margin factor, centroid frequency, mean square frequency, root mean square frequency, frequency variance, frequency standard deviation, mean of spectral kurtosis, standard deviation of spectral kurtosis, skewness of spectral kurtosis, kurtosis of spectral kurtosis, power spectrum entropy, energy entropy, approximate entropy, sample entropy, fuzzy entropy, permutation entropy, and envelope entropy.

[0016] In some embodiments, the preprocessing of the zero-sequence voltage signal is noise reduction preprocessing.

[0017] The present invention also provides a system for identifying ferromagnetic resonance faults of electromagnetic voltage transformers, comprising: An acquisition module, which is used to acquire and pre-process the zero-sequence voltage signal on the primary side of the distribution network system; An analysis module is used to analyze the pre-processed zero-sequence voltage signal to obtain a reconstructed low-frequency IMF component; A solution module is used to solve and reduce the dimension to reconstruct the characteristic indexes of the low-frequency IMF components in the time domain and frequency domain, and determine the most suitable dimension to form a characteristic vector matrix for electromagnetic voltage transformer ferromagnetic resonance fault identification based on multiple characteristic quantities in the time domain and frequency domain; A construction module is used to establish a sample set, a training sample subset, and to build, train, and configure an RF model to construct an electromagnetic voltage transformer ferromagnetic resonance fault identification model based on WOA-optimized RF; The judgment module is used to use the WOA algorithm to iteratively search for the TreeBagger function, set the parameters of the RF model, and use the electromagnetic voltage transformer ferromagnetic resonance fault identification model based on WOA optimized RF to perform fault identification on the sample set to determine the fault type of the sample.

[0018] The present invention provides an electromagnetic voltage transformer ferromagnetic resonance fault identification device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, an electromagnetic voltage transformer ferromagnetic resonance fault identification method as described above is implemented.

[0019] The present invention provides a readable storage medium storing a computer program, which can be executed by a processor of a device locked in the storage medium to implement a method for identifying a ferromagnetic resonance fault of an electromagnetic voltage transformer as described above.

[0020] By adopting the above technical solution, the beneficial effects of the present invention are: The present invention proposes an improved CEEMD time-frequency signal decomposition method based on T test and variance contribution rate, which solves the shortcomings that the number of IMF components and their frequency bands obtained by CEEMD decomposition are not fixed, and a small number of pseudo-modal IMF components appear in the decomposition results. The improved CEEMD method can fix the number of IMF components obtained by decomposing the zero-sequence voltage signal into high-frequency and low-frequency IMF components, and can effectively extract the characteristic information of the zero-sequence voltage signal under different operating conditions of the system.

[0021] The present invention introduces the KPCA method to perform feature dimensionality reduction on the basis of improving the CEEMD time-frequency signal decomposition method, further solving the problem that too many features contain redundant information. The KPCA method significantly improves the cross-aliasing phenomenon existing in the original samples and can effectively reduce the errors caused by redundant information.

[0022] The present invention proposes a WOA algorithm to optimize the RF model to solve the problem of mutual influence between TreeBagger function parameters in the RF model.

[0023] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure.

[0024] Undoubtedly, these and other objects of the present invention will become more apparent after the following detailed description of the preferred embodiment described with reference to various figures and drawings.

[0025] In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, one or several preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation of the present invention.

[0027] In the drawings, the same reference numerals are used for the same components and the drawings are schematic and not necessarily drawn to scale.

[0028] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only one or several embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on such drawings without paying creative work.

[0029] Figure 1 A flow chart of a method for identifying a ferromagnetic resonance fault of an electromagnetic voltage transformer according to some embodiments of the present invention; Figure 2 This is a flow chart of feature dimensionality reduction using the KPCA method in some embodiments of the present invention; Figure 3 A Pareto diagram of KPCA of some embodiments of the present invention; Figure 4 It is a diagram showing the improved CEEMD decomposition effect of zero-sequence voltage in some embodiments of the present invention; Figure 5 A schematic diagram of the cumulative error of the improved CEEMD decomposition of the zero-sequence voltage in some embodiments of the present invention; Figure 6 Spectral distribution diagram of high-frequency and low-frequency IMF components of improved CEEMD in some embodiments of the present invention; Figure 7 A schematic diagram of instantaneous amplitudes of high-frequency and low-frequency IMF components of improved CEEMD according to some embodiments of the present invention; Figure 8 Classification diagram of the optimized RF model for WOA (training set); Fig. 9 Classification of RF model optimized for WOA (test set); Fig.10 Confusion matrix diagram of the optimized RF model for WOA (training set); Fig.11 Confusion matrix plot of the RF model optimized for WOA (test set). DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, but are not used to limit the present invention.

[0031] In addition, in the description of the present invention, it should be understood that the terms "center", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "axial", "radial", "circumferential" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present invention.

[0032] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral body; it can be directly connected or indirectly connected through an intermediate medium, and it can be the internal connection of two elements or the interaction relationship between two elements. However, if it is indicated as a direct connection, it means that the two connected bodies are not connected through a transition structure, but are connected to form a whole through a connecting structure. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances.

[0033] In the present invention, unless otherwise clearly specified and limited, the first feature "on" or "under" the second feature may be that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in an appropriate manner in any one or more embodiments or examples.

[0034] Reference Figure 1 , Figure 1 The present invention is a flow chart of a method for identifying ferromagnetic resonance fault of an electromagnetic voltage transformer according to some embodiments of the present invention.

[0035] According to some embodiments of the present invention, the present invention provides a method for identifying ferromagnetic resonance fault of an electromagnetic voltage transformer. The identification method comprises the following steps: Collect and pre-process the zero-sequence voltage signal on the primary side of the distribution network system; Analyze the preprocessed zero-sequence voltage signal to obtain the reconstructed low-frequency IMF component; Solve and reduce the dimension to reconstruct the characteristic index of low-frequency IMF components in time domain and frequency domain, and determine the most suitable dimension to form the characteristic vector matrix of electromagnetic voltage transformer ferromagnetic resonance fault identification based on multiple characteristic quantities in time domain and frequency domain; Establish sample sets and training sample subsets, build, train, and configure RF models to construct an electromagnetic voltage transformer ferromagnetic resonance fault identification model based on WOA optimized RF; The WOA algorithm is used to iteratively search and solve the TreeBagger function; The parameters of the RF model are set, and the ferromagnetic resonance fault identification model of electromagnetic voltage transformer based on WOA optimized RF is used to identify faults in the sample set and determine the fault type of the sample.

[0036] According to some embodiments of the present invention, optionally, the preprocessed zero-sequence voltage signal is analyzed to obtain a reconstructed low-frequency IMF component; specifically: The preprocessed zero-sequence voltage signal is analyzed in time domain and frequency domain using the improved CEEMD method based on T test and variance contribution rate to obtain the reconstructed low-frequency IMF component that can characterize the original signal.

[0037] When analyzing the fault characteristics of electromagnetic voltage transformer ferromagnetic resonance, the traditional CEEMD decomposition has pseudo-modal phenomenon. The number of IMF components and the frequency band of each IMF component obtained by decomposing different zero-sequence voltage signals are not fixed, and the fault characteristics exist in different IMF components. There is uncertainty in the selection of components. If only one or several IMF components are selected for analysis and used as the basis for fault identification, it may lead to a decrease in the discrimination of the diagnostic threshold and a decrease in the identification accuracy. Therefore, according to some embodiments of the present invention, optionally, the specific steps of using the improved CEEMD method based on T test and variance contribution rate to perform time domain and frequency domain analysis on the pre-processed zero-sequence voltage signal are as follows: In the original signal x ( t ), add positive and negative Gaussian white noises with opposite numbers to each other. M times, get 2 M A new signal containing positive and negative Gaussian white noise X n,m ( t ), whose expression is: , In the above formula, w m ( t ) is Gaussian white noise, m Indicates the number of times positive (negative) Gaussian white noise is added, m =1, 2, …, M ; (-1) n represents the positive and negative coefficients of Gaussian white noise, n =1, 2; 2 M A signal containing positive (negative) Gaussian white noise X n,m ( t ) Perform EMD decomposition one by one to obtain each signal X n,m ( t ) corresponds to a set of IMF components, definingm The signal is decomposed by EMD to obtain the i The IMF components are f m,i ( t ); 2 M Group signal X n,m ( t ) corresponding to the IMF component f m,i ( t ) and perform the sum and average operation to obtain x ( t ) IMF components of each layer after traditional CEEMD decomposition F i ( t ), whose expression is: , For each layer of IMF components F i ( t ) to conduct a T test, that is, F 1(t) is recorded as the index A 1, F 1( t )+ F 2( t ) is recorded as an indicator A 2, and so on, i indivual F i ( t ) and is recorded as an index A i , respectively calculate the index A 1 to index A i The mean of , and whether the mean is significantly different from 0, assuming ,and a is an integer, if the index A a is significantly not 0, we can get hour, F i ( t ) is the IMF component of the high frequency part; hour, F i ( t ) is the IMF component of the low-frequency part, and the expression of its T test is: , In the above formula, For indicators A i The mean of For indicators A i The standard deviation of n For indicators A i Sample size; For the high frequency IMF component F i ( t ), calculate their variance contribution rates, that is, each IMF component F i ( t )and x ( t ) signal variance ratio, and then sort the variance contribution rates from high to low. Similarly, the sorting of the variance contribution rates of the IMF components in the low-frequency part is obtained; The high-frequency IMF components ranked first and second in terms of variance contribution rate are reconstructed to obtain the reconstructed high-frequency IMF components. Similarly, the reconstructed low-frequency IMF components are obtained.

[0038] The present invention uses an improved CEEMD method based on T test and variance contribution rate to perform time domain and frequency domain analysis on the preprocessed zero-sequence voltage signal to obtain a reconstructed low-frequency IMF component that can characterize the original signal, so as to solve the technical problems that the number of IMF components and their frequency bands obtained by CEEMD decomposition are not fixed, and a small number of pseudo-modal IMF components appear in the decomposition result.

[0039] According to some embodiments of the present invention, optionally, the characteristic indexes of the low-frequency IMF components in the time domain and the frequency domain are solved and reconstructed by dimensionality reduction, and the most suitable dimension is determined to form a characteristic vector matrix for identifying ferromagnetic resonance faults of electromagnetic voltage transformers based on multiple characteristic quantities in the time domain and the frequency domain; specifically: The 33 characteristic indicators of the reconstructed low-frequency IMF components in the time domain and frequency domain are solved, and the KPCA method is used for feature dimensionality reduction. According to the standard that the percentage of variance explained by the first n principal components in the cumulative total variance exceeds 95% for the first time, the most suitable dimension of the new data after dimensionality reduction is determined to be 12, and the eigenvector matrix for ferromagnetic resonance fault identification of electromagnetic voltage transformer based on multiple characteristic quantities in time domain and frequency domain is formed.

[0040] Too many features may contain redundant information, which will cause errors in model identification and reduce the recognition accuracy. At the same time, considering that the feature quantities of electromagnetic voltage transformer ferromagnetic resonance fault present a nonlinear correlation, the KPCA method is used for feature dimensionality reduction. The error caused by redundant information is reduced through feature dimensionality reduction, and the recognition accuracy of the model is improved.

[0041] Kernel principal component analysis (KPCA) is a classic dimensionality reduction method with the advantages of simple calculation, applicability to nonlinear data and retention of global data characteristics. This method introduces kernel function technology on the basis of principal component analysis (PCA), nonlinearly maps the initial data to a high-dimensional space, and makes nonlinearly related variables linearly related in the high-dimensional space. PCA is then used to achieve data dimensionality reduction, so that KPCA can effectively process nonlinear data.

[0042] Reference Figure 2 , Figure 2 This is a flow chart of feature dimensionality reduction using the KPCA method in some embodiments of the present invention.

[0043] According to some embodiments of the present invention, optionally, a KPCA method is used to perform feature dimensionality reduction, comprising the following steps: Normalize the data samples and select the radial basis kernel function for nonlinear mapping; Calculate the kernel matrix, the eigenvectors and eigenvalues ​​of the kernel matrix, calculate the percentage of variance explained by each principal component in the total variance and sort them from large to small, calculate the cumulative percentage of variance explained by the principal component in the total variance, and draw the Pareto chart of KPCA for analysis; The most appropriate data dimension is determined based on the Pareto chart of KPCA. When the cumulative percentage of the variance explained by the principal component accounts for more than 95% of the total variance, it is considered that the data of the corresponding dimension can effectively represent most of the characteristic information of the initial data. The eigenvector of the corresponding dimension is extracted and the projection of the kernel matrix on the vector is calculated to obtain the new data after dimensionality reduction.

[0044] According to some embodiments of the present invention, optionally, a KPCA method is used to perform feature dimensionality reduction, specifically: Assume that the initial data is , and zero-mean, n Represents the quantity dimension of the initial data, X Perform nonlinear mapping to high-dimensional space, Represents a nonlinear mapping function, forming a high-dimensional mapping data vector ,in, express dimensional high-dimensional mapping vector, ,and i is an integer, , m is the characteristic dimension of the initial data; Mapping data vectors to high dimensions Perform covariance calculation and calculate The covariance matrix of , and the covariance matrix Performing eigendecomposition operation, we get: , In the above formula, Represents the covariance matrix The characteristic value of express The corresponding dimensional feature vector, ,and j is an integer; The feature vector Mapping vectors in high dimensions Performing linear representation, we get: , In the above formula, express dimensional column vector, express Middle i elements; Combine the above two equations and multiply both sides by ,get: : Define the kernel function matrix , then the above formula can be simplified to: ; According to the above formula, we can get , the high-dimensional mapping data vector In the feature vector The initial data can be obtained by projecting in the direction New data after dimensionality reduction , whose expression is: .

[0045] Reference Figure 3 , Figure 3 1 is a Pareto diagram of KPCA according to some embodiments of the present invention.

[0046] exist Figure 3 In the figure, the horizontal axis represents the number of principal components, the left vertical axis represents the percentage of variance explained by each principal component in the total variance, and the right vertical axis represents the percentage of variance explained by the principal component in the cumulative total variance. Figure 3It can be seen that the percentage of explained variance of different principal components in the Pareto diagram of KPCA is quite different. The percentage of explained variance of the first principal component in the total variance is 37.74%, which is the strongest explanatory power of the data. The percentage of explained variance of the 15th and subsequent principal components in the total variance is less than 1.0%, which is weak in explanatory power of the data and belongs to redundant information, which will cause errors in model recognition and reduce recognition accuracy. The percentage of explained variance of the first 10 principal components in the cumulative total variance exceeds 90% for the first time, and the percentage of explained variance of the first 12 principal components in the cumulative total variance exceeds 95% for the first time. The larger the percentage of explained variance in the cumulative total variance, the more effective the data can be in the dimension corresponding to the value. Considering the factors such as computational efficiency, the ability to represent data and the recognition accuracy, the most suitable data dimension for the initial data after KPCA dimensionality reduction is determined to be 12, and the percentage of explained variance of the principal component in the cumulative total variance is 95.26%.

[0047] According to some embodiments of the present invention, optionally, 33 characteristic indicators of the reconstructed low-frequency IMF component in the time domain and the frequency domain include maximum value, minimum value, mean value, median, peak value, rectified mean value, variance, standard deviation, kurtosis, skewness, root mean square, mean square value, root amplitude, waveform factor, peak factor, pulse factor, margin factor, centroid frequency, mean square frequency, root mean square frequency, frequency variance, frequency standard deviation, mean of spectral kurtosis, standard deviation of spectral kurtosis, skewness of spectral kurtosis, kurtosis of spectral kurtosis, power spectrum entropy, energy entropy, approximate entropy, sample entropy, fuzzy entropy, permutation entropy, and envelope entropy. Since the low-frequency IMF component can effectively characterize the three different working states of the system zero-sequence voltage signal, namely normal operation, single-phase grounding fault and voltage transformer ferromagnetic resonance fault, and considering that when the system has a voltage transformer ferromagnetic resonance fault, mainly frequency division resonance occurs, and the fault characteristics mainly exist in the low-frequency range, the low-frequency IMF component is selected to extract 33 characteristic parameters in the time domain and frequency domain, providing a judgment basis for the detection of voltage transformer ferromagnetic resonance fault.

[0048] According to some embodiments of the present invention, optionally, the preprocessing of the zero-sequence voltage signal is noise reduction preprocessing.

[0049] The present invention uses an improved CEEMD method based on T test and variance contribution rate to perform time domain and frequency domain analysis on the preprocessed zero-sequence voltage signal to obtain a reconstructed low-frequency IMF component that can characterize the original signal, so as to solve the technical problems that the number of IMF components and their frequency bands obtained by CEEMD decomposition are not fixed, and a small number of pseudo-modal IMF components appear in the decomposition results. The details are as follows: Reference Figure 4-7 , Figure 4 It is a diagram showing the improved CEEMD decomposition effect of zero-sequence voltage in some embodiments of the present invention; Figure 5A schematic diagram of the cumulative error of the improved CEEMD decomposition of the zero-sequence voltage in some embodiments of the present invention; Figure 6 Spectral distribution diagram of high-frequency and low-frequency IMF components of improved CEEMD in some embodiments of the present invention; Figure 7 Schematic diagram of instantaneous amplitudes of high-frequency and low-frequency IMF components of improved CEEMD according to some embodiments of the present invention.

[0050] When the arc grounding of phase A cable causes ferromagnetic resonance of voltage transformer in the system, the improved CEEMD proposed above is used to analyze the zero-sequence voltage on the primary side of the system. The decomposition effect is as follows: Figure 4 As shown, the improved CEEMD decomposition cumulative error is as follows Figure 5 shown.

[0051] In order to further verify whether the improved CEEMD method based on T test and variance contribution rate can accurately decompose the high-frequency and low-frequency IMF components of the zero-sequence voltage signal when the voltage transformer ferromagnetic resonance occurs in the system while retaining the fault characteristics, the method is tested by analyzing the spectrum distribution and instantaneous amplitude of the high-frequency and low-frequency IMF components. The analysis results are as follows: Figure 6 and Figure 7 shown.

[0052] Depend on Figure 6 It can be seen that the spectrum distribution of high-frequency and low-frequency IMF components has obvious distinction at 350Hz, and can extract the low-frequency and high-frequency components containing the characteristics of ground fault and resonance fault, indicating that the improved CEEMD method is superior in the high-frequency and low-frequency domain division of zero-sequence voltage signals when the voltage transformer ferromagnetic resonance occurs in the system. Figure 6 It can be seen that there are obvious mutations in the instantaneous amplitudes of the high-frequency and low-frequency IMF components between the normal operation of the zero-sequence voltage signal, the single-phase grounding fault and the voltage transformer ferromagnetic resonance fault. This shows that the improved CEEMD method can avoid errors caused by interference factors such as noise, and can effectively extract the characteristic information of the zero-sequence voltage signal under different operating conditions when the voltage transformer ferromagnetic resonance occurs in the system.

[0053] In summary, the improved CEEMD based on T-test and variance contribution rate can effectively solve the pseudo-modal phenomenon of traditional CEEMD decomposition, and can fix the number of IMF components obtained by decomposing the zero-sequence voltage signal when the system has ferromagnetic resonance of the voltage transformer as high-frequency and low-frequency IMF components, which solves the problem that the diagnostic threshold discrimination is reduced due to the inability to fix the number of IMF components decomposed from different signals. At the same time, this method is suitable for zero-sequence voltage signal analysis under different conditions such as ground fault and ferromagnetic resonance fault, and can accurately decompose the high-frequency and low-frequency IMF components of different zero-sequence voltage signals, and can effectively extract the characteristic information in the current waveform under normal operation, ground fault and resonance fault.

[0054] According to some embodiments of the present invention, the present invention further provides a system for identifying ferromagnetic resonance fault of an electromagnetic voltage transformer, comprising: An acquisition module, which is used to acquire and pre-process the zero-sequence voltage signal on the primary side of the distribution network system; An analysis module is used to analyze the pre-processed zero-sequence voltage signal to obtain a reconstructed low-frequency IMF component; A solution module is used to solve and reduce the dimension to reconstruct the characteristic indexes of the low-frequency IMF components in the time domain and frequency domain, and determine the most suitable dimension to form a characteristic vector matrix for electromagnetic voltage transformer ferromagnetic resonance fault identification based on multiple characteristic quantities in the time domain and frequency domain; A construction module is used to establish a sample set, a training sample subset, and to build, train, and configure an RF model to construct an electromagnetic voltage transformer ferromagnetic resonance fault identification model based on WOA-optimized RF; The judgment module is used to use the WOA algorithm to iteratively search for the TreeBagger function, set the parameters of the RF model, and use the electromagnetic voltage transformer ferromagnetic resonance fault identification model based on WOA optimized RF to perform fault identification on the sample set to determine the fault type of the sample.

[0055] According to some embodiments of the present invention, the present invention provides an identification device for ferromagnetic resonance fault of an electromagnetic voltage transformer, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, it implements a method for identifying ferromagnetic resonance fault of an electromagnetic voltage transformer as described above.

[0056] According to some embodiments of the present invention, the present invention provides a readable storage medium storing a computer program, which can be executed by a processor of a device locked in the storage medium to implement a method for identifying a ferromagnetic resonance fault of an electromagnetic voltage transformer as described above.

[0057] Example This embodiment provides a method for identifying a ferromagnetic resonance fault of an electromagnetic voltage transformer, which specifically includes the following steps: Step 1: collect the zero-sequence voltage signal on the primary side of the distribution network system and perform preprocessing such as noise reduction.

[0058] Step 2: Use the improved CEEMD method based on T test and variance contribution rate to analyze the zero-sequence voltage signal in time domain and frequency domain after noise reduction to obtain the reconstructed low-frequency IMF component that can characterize the original signal.

[0059] Step 3, solve the 33 characteristic indicators in the time domain and frequency domain of the reconstructed low-frequency IMF component, use the KPCA method to reduce the feature dimension, and determine the most suitable dimension of the new data after dimensionality reduction to be 12 based on the standard that the percentage of variance explained by the first n principal components accounts for more than 95% of the cumulative total variance for the first time, thus forming a VT ferromagnetic resonance fault identification characteristic vector matrix based on multiple characteristic quantities in the time domain and frequency domain.

[0060] Step 4: The characteristic vector matrix of the VT ferromagnetic resonance state excited by different factors such as fault excitation type, system-to-ground capacitance value and fault disappearance time described in Chapter 3 is used to form a sample set. The Bootstrap method is used as the sampling principle to obtain the training sample subset, and the CART method is used to build the RF model. The RF model is input for training, and the electromagnetic VT ferromagnetic resonance fault identification model is constructed by configuring the RF model parameters.

[0061] Step 5: Considering the mutual influence among various parameters, the WOA algorithm is used to optimize the five related parameters of the TreeBagger function. The fitness function, whale population size and maximum number of iterations of WOA are set. The optimization search range of each parameter is set as shown in Table 1, and iterative optimization is performed; In step 6, the TreeBagger function obtained by WOA optimization in step 5 is used as the parameter setting of the RF model. The fault identification model based on WOA optimized RF is used to identify faults in the sample set and determine the fault type of the sample.

[0062] Table 1. Search range of relevant parameters of TreeBagger function

[0063] Reference Figure 8-Figure 11 , Figure 8 Classification diagram of the optimized RF model for WOA (training set); Fig. 9 Classification of RF model optimized for WOA (test set); Fig.10 Confusion matrix diagram of the optimized RF model for WOA (training set); Fig.11 Confusion matrix plot of the RF model optimized for WOA (test set).

[0064] The data sample set (1200 samples in the training set and 600 samples in the test set) is input into the fault recognition model optimized by WOA for training and testing. The fault recognition results are as follows: Figure 8-Figure 11 shown.

[0065] Depend on Figure 7The analysis shows that the RF model optimized by WOA achieves high accuracy for both the training set and the test set. The accuracy of the training set is 100%, and the accuracy of the test set is 98.33%.

[0066] analyze Figure 8 From the confusion matrix of the training set of the WOA optimized RF model, we can see that the model has 100% precision and recall for the three working states of normal system operation (labeled as 1), single-phase grounding fault (labeled as 2) and VT ferromagnetic resonance fault (labeled as 3), and the calculated Fl values ​​corresponding to the three working states are also 100%. The results show that after learning and training, the WOA optimized RF model has excellent classification and recognition accuracy and robustness. Analysis Fig. 9 From the confusion matrix of the test set of the RF model optimized by WOA, it can be seen that the Precision and Recall of the model for the normal operation of the system (labeled as 1) are both 99.5%, and the calculated Fl value corresponding to the normal operation is 99.50%; similarly, the Precision, Recall and F1 values ​​of the single-phase grounding fault (labeled as 2) are 99.0%, 97.0% and 98.00% respectively, and the Precision, Recall and F1 values ​​of the VT ferromagnetic resonance fault (labeled as 3) are 99.6%, 98.5% and 99.05% respectively. The F1 value of the classification performance index of the recognition model for the three working states in the distribution network system is above 98%. The results show that the classification effect of the RF model optimized by WOA is good in the test set and its overall performance is superior.

[0067] In order to further verify the effectiveness of the electromagnetic VT ferromagnetic resonance fault identification model based on improved CEEMD feature extraction, KPCA feature dimension reduction and WOA algorithm optimized RF construction, and the necessity of improved CEEMD decomposition method to extract time-frequency domain features and KPCA method for feature dimension reduction, four ablation experiments were designed.

[0068] The first ablation experiment uses improved CEEMD to extract time-frequency domain features and KPCA features for dimensionality reduction. Five different classification models (all optimized by WOA) are set for comparison. The accuracy of the classification results and the training time are used as model evaluation indicators. The test results are shown in Table 2. As shown in Table 2, the fault identification capabilities of the DT, KNN and RF models are excellent, and their accuracy ranks in the top three. Among them, the accuracy of the RF model is 98.33%, and the average training time is 8.6s; the accuracy of the DT model is 92.67%, and the average training time is the shortest at 2.5s; the accuracy of the KNN model is 94.67%, and the average training time is the longest at 8.7s. In terms of comprehensive effect, compared with classification models such as GMM and NB, the DT model has the shortest average training time, but its accuracy is not the best. The average training time of the RF model is slightly increased, but it can still complete the model training in a shorter time and obtain the best accuracy. This verifies that the proposed electromagnetic VT ferromagnetic resonance fault identification model based on improved CEEMD feature extraction, KPCA feature dimensionality reduction and WOA algorithm-optimized RF construction is effective and can more comprehensively realize the identification of three system operating states: normal operation, single-phase grounding fault and VT ferromagnetic resonance fault in the distribution network system.

[0069] Table 2 Test results of different classification models

[0070] The second ablation experiment uses improved CEEMD to extract time-frequency domain features and KPCA to perform feature dimension reduction. Four different algorithms are set to optimize the RF model and compared. The accuracy of the classification results and the average training time are used as evaluation indicators. The experimental results are shown in Table 3. As shown in Table 3, compared with the RF model without optimized parameters, the classification results of the RF model optimized by grid search (GS), particle swarm optimization (PSO) and WOA algorithms are improved to varying degrees, but the average training time of the model is also correspondingly longer. Among them, the classification results of the RF model optimized by the GS algorithm have the highest accuracy of 98.50%, and the average training time is the longest at 132.8s; the classification results of the RF model optimized by the WOA algorithm have the second highest accuracy of 98.33%, and the average training time is only 8.6s. In terms of comprehensive effect, optimizing the RF model using different optimization algorithms can improve the accuracy of the model, indicating that it is necessary to optimize the RF model using a suitable optimization algorithm; optimizing the RF model using the WOA algorithm can quickly complete model training and obtain higher accuracy, indicating that optimizing the RF model using the WOA algorithm is more suitable for the actual engineering application of electromagnetic VT ferromagnetic resonance fault detection in distribution network systems.

[0071] Table 3 Test results of different algorithms to optimize RF model

[0072] The third ablation experiment uses improved CEEMD to extract time-frequency domain features, uses WOA to optimize the RF model, sets 4 different algorithms for feature dimensionality reduction and compares them, and uses the accuracy of the classification results as the evaluation index. The experimental results are shown in Table 4. As shown in Table 4, compared with the RF model without feature dimensionality reduction, the classification results of the RF model are improved to varying degrees after PCA, t-SNE and KPCA algorithms are used for feature dimensionality reduction. Among them, the classification results of the RF model with PCA for feature dimensionality reduction have an accuracy of 3.83%; the classification results of the RF model with t-SNE for feature dimensionality reduction have an accuracy of 7.67%; and the classification results of the RF model with KPCA for feature dimensionality reduction have an accuracy of 8.83%. This ablation experiment further verifies that the use of the KPCA method for feature dimensionality reduction can effectively reduce the errors caused by redundant information, and at the same time improves the accuracy of the electromagnetic VT ferromagnetic resonance fault model, and has excellent recognition capabilities for the three working states of normal operation, single-phase grounding fault and VT ferromagnetic resonance fault in the distribution network system.

[0073] Table 4 Test results under different feature dimensionality reduction methods

[0074] The fourth ablation experiment uses KPCA feature dimension reduction and WOA to optimize the RF model. Four different feature extraction methods are set for comparison. The accuracy of the classification results is used as the evaluation index. The experimental results are shown in Table 5. As shown in Table 5, compared with different feature extraction methods such as traditional EMD, EEMD and CEEMD, the improved CEEMD method has the highest classification result accuracy of 98.33% under the RF model. Compared with the traditional CEEMD method, the accuracy is improved by 3.83%, indicating that the use of improved CEEMD for feature extraction is effective, and can effectively and accurately extract the characteristic information contained in the zero-sequence voltage of the three working states of normal operation, single-phase grounding fault and VT ferromagnetic resonance fault in the distribution network system. At the same time, it can improve the accuracy of the electromagnetic VT ferromagnetic resonance fault model, which is more suitable for the actual engineering application of electromagnetic VT ferromagnetic resonance fault detection in the distribution network system.

[0075] Table 5 Test results under different feature extraction methods

[0076] It should be understood that the embodiments disclosed in the present invention are not limited to the specific processing steps or materials disclosed herein, but should be extended to equivalent substitutions of such features understood by ordinary technicians in the relevant field. It should also be understood that the terms used herein are only used for the purpose of describing specific embodiments and are not meant to be limiting.

[0077] The "embodiment" mentioned in the specification means that the specific features or characteristics described in conjunction with the embodiment are included in at least one embodiment of the present invention. Therefore, the phrases or "embodiment" appearing in various places throughout the specification do not necessarily refer to the same embodiment.

[0078] In addition, the described features or characteristics may be combined in one or more embodiments in any other suitable manner. In the above description, some specific details, such as thickness, quantity, etc., are provided to provide a comprehensive understanding of the embodiments of the present invention. However, those skilled in the relevant art will understand that the present invention can be implemented without one or more of the above specific details or can also be implemented using other methods, components, materials, etc.

Claims

1. A method for identifying ferromagnetic resonance fault of an electromagnetic voltage transformer, characterized in that: The following steps are involved: Collect and pre-process the zero-sequence voltage signal on the primary side of the distribution network system; Analyze the preprocessed zero-sequence voltage signal to obtain the reconstructed low-frequency IMF component; Solve and reduce the dimension to reconstruct the characteristic index of low-frequency IMF components in time domain and frequency domain, and determine the most suitable dimension to form the characteristic vector matrix of electromagnetic voltage transformer ferromagnetic resonance fault identification based on multiple characteristic quantities in time domain and frequency domain; Establish sample sets and training sample subsets, build, train, and configure RF models to construct an electromagnetic voltage transformer ferromagnetic resonance fault identification model based on WOA optimized RF; The WOA algorithm is used to iteratively search and solve the TreeBagger function; The parameters of the RF model are set, and the electromagnetic voltage transformer ferromagnetic resonance fault identification model based on WOA optimized RF is used to identify faults in the sample set and determine the fault type of the sample.

2. The method for identifying ferromagnetic resonance fault of an electromagnetic voltage transformer according to claim 1, characterized in that: Analyze the preprocessed zero-sequence voltage signal to obtain the reconstructed low-frequency IMF component; specifically: The preprocessed zero-sequence voltage signal is analyzed in time domain and frequency domain using the improved CEEMD method based on T test and variance contribution rate to obtain the reconstructed low-frequency IMF component that can characterize the original signal.

3. The method for identifying ferromagnetic resonance fault of an electromagnetic voltage transformer according to claim 2, characterized in that: The specific steps of using the improved CEEMD method based on T test and variance contribution rate to analyze the preprocessed zero-sequence voltage signal in time domain and frequency domain are as follows: In the original signal x ( t ), add positive and negative Gaussian white noises with opposite numbers to each other. M times, get 2 M A new signal containing positive and negative Gaussian white noise X n,m ( t ), whose expression is: ， In the above formula, w m ( t ) is Gaussian white noise, m Indicates the number of times positive (negative) Gaussian white noise is added, m =1, 2, …, M ; (-1) n represents the positive and negative coefficients of Gaussian white noise, n =1, 2; 2 M A signal containing positive (negative) Gaussian white noise X n,m ( t ) Perform EMD decomposition one by one to obtain each signal X n,m ( t ) corresponds to a set of IMF components, defining m The signal is decomposed by EMD to obtain the i The IMF components are f m,i ( t ); 2 M Group signal X n,m ( t ) corresponding to the IMF component f m,i ( t ) and perform the sum and average operation to obtain x ( t ) IMF components of each layer after traditional CEEMD decomposition F i ( t ), whose expression is: ， For each layer of IMF components F i ( t ) to conduct a T test, that is, F 1(t) is recorded as the index A 1, F 1( t )+ F 2( t ) is recorded as an indicator A 2, and so on, i indivual F i ( t ) and is recorded as an index A i , respectively calculate the index A 1 to index A i The mean of , and whether the mean is significantly different from 0, assuming ,and a is an integer, if the index A a is significantly not 0, we can get hour, F i ( t ) is the IMF component of the high frequency part; hour, F i ( t ) is the IMF component of the low-frequency part, and the expression of its T test is: ， In the above formula, For indicators A i The mean of For indicators A i The standard deviation of n For indicators A i Sample size; For the high frequency IMF component F i ( t ), calculate their variance contribution rates, that is, each IMF component F i ( t )and x ( t ) signal variance ratio, and then sort the variance contribution rates from high to low. Similarly, the sorting of the variance contribution rates of the IMF components in the low-frequency part is obtained; The high-frequency IMF components ranked first and second in terms of variance contribution rate are reconstructed to obtain the reconstructed high-frequency IMF components. Similarly, the reconstructed low-frequency IMF components are obtained.

4. The method for identifying ferromagnetic resonance fault of an electromagnetic voltage transformer according to claim 1, characterized in that: Solve and reduce the dimension to reconstruct the characteristic index of low-frequency IMF components in time domain and frequency domain, and determine the most suitable dimension to form the characteristic vector matrix of electromagnetic voltage transformer ferromagnetic resonance fault identification based on multiple characteristic quantities in time domain and frequency domain; Specifically: The 33 characteristic indicators of the reconstructed low-frequency IMF components in the time domain and frequency domain are solved, and the KPCA method is used for feature dimensionality reduction. According to the standard that the percentage of variance explained by the first n principal components in the cumulative total variance exceeds 95% for the first time, the most suitable dimension of the new data after dimensionality reduction is determined to be 12, and the eigenvector matrix for ferromagnetic resonance fault identification of electromagnetic voltage transformer based on multiple characteristic quantities in time domain and frequency domain is formed.

5. The method for identifying ferromagnetic resonance fault of an electromagnetic voltage transformer according to claim 4, characterized in that: The KPCA method is used to reduce the feature dimension, including the following steps: Normalize the data samples and select the radial basis kernel function for nonlinear mapping; Calculate the kernel matrix, the eigenvectors and eigenvalues ​​of the kernel matrix, calculate the percentage of variance explained by each principal component in the total variance and sort them from large to small, calculate the cumulative percentage of variance explained by the principal component in the total variance, and draw the Pareto chart of KPCA for analysis; The most appropriate data dimension is determined based on the Pareto chart of KPCA. When the cumulative percentage of the variance explained by the principal component accounts for more than 95% of the total variance, it is considered that the data of the corresponding dimension can effectively represent most of the characteristic information of the initial data. The eigenvector of the corresponding dimension is extracted and the projection of the kernel matrix on the vector is calculated to obtain the new data after dimensionality reduction.

6. The method for identifying ferromagnetic resonance fault of an electromagnetic voltage transformer according to claim 5, characterized in that: The KPCA method is used for feature dimensionality reduction, specifically: Assume that the initial data is , and zero-mean, n Represents the quantity dimension of the initial data, X Perform nonlinear mapping to high-dimensional space, Represents a nonlinear mapping function, forming a high-dimensional mapping data vector ,in, express dimensional high-dimensional mapping vector, ,and i is an integer, , m is the characteristic dimension of the initial data; Mapping data vectors to high dimensions Perform covariance calculation and calculate The covariance matrix of , and the covariance matrix Performing eigendecomposition operation, we get: ， In the above formula, Represents the covariance matrix The characteristic value of express The corresponding dimensional feature vector, ,and j is an integer; The feature vector Mapping vectors in high dimensions Performing linear representation, we get: ， In the above formula, express dimensional column vector, express Middle i elements; Combine the above two equations and multiply both sides by ,get: ： Define the kernel function matrix , then the above formula can be simplified to: ； According to the above formula, we can get , the high-dimensional mapping data vector In the feature vector The initial data can be obtained by projecting in the direction New data after dimensionality reduction , whose expression is: 。 7. The method for identifying ferromagnetic resonance fault of an electromagnetic voltage transformer according to claim 4, characterized in that: The 33 characteristic indicators of the reconstructed low-frequency IMF components in the time domain and frequency domain include maximum value, minimum value, mean value, median value, peak value, rectified mean value, variance, standard deviation, kurtosis, skewness, root mean square, mean square value, root amplitude, waveform factor, peak factor, pulse factor, margin factor, centroid frequency, mean square frequency, root mean square frequency, frequency variance, frequency standard deviation, mean value of spectral kurtosis, standard deviation of spectral kurtosis, skewness of spectral kurtosis, kurtosis of spectral kurtosis, power spectrum entropy, energy entropy, approximate entropy, sample entropy, fuzzy entropy, permutation entropy, and envelope entropy.

8. The method for identifying ferromagnetic resonance fault of an electromagnetic voltage transformer according to claim 1, characterized in that: The preprocessing of zero-sequence voltage signal is noise reduction preprocessing.

9. A system for identifying ferromagnetic resonance fault of an electromagnetic voltage transformer, characterized in that: include: An acquisition module, which is used to acquire and pre-process the zero-sequence voltage signal on the primary side of the distribution network system; An analysis module is used to analyze the pre-processed zero-sequence voltage signal to obtain a reconstructed low-frequency IMF component; A solution module is used to solve and reduce the dimension to reconstruct the characteristic indexes of the low-frequency IMF components in the time domain and frequency domain, and determine the most suitable dimension to form a characteristic vector matrix for electromagnetic voltage transformer ferromagnetic resonance fault identification based on multiple characteristic quantities in the time domain and frequency domain; A construction module is used to establish a sample set, a training sample subset, and to build, train, and configure an RF model to construct an electromagnetic voltage transformer ferromagnetic resonance fault identification model based on WOA-optimized RF; The judgment module is used to use the WOA algorithm to iteratively search for the TreeBagger function, set the parameters of the RF model, and use the electromagnetic voltage transformer ferromagnetic resonance fault identification model based on WOA optimized RF to perform fault identification on the sample set to determine the fault type of the sample.

10. A device for identifying ferromagnetic resonance fault of an electromagnetic voltage transformer, characterized in that: The method comprises a processor, a memory and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, a method for identifying ferromagnetic resonance fault of an electromagnetic voltage transformer as described in any one of claims 1 to 8 is implemented.

11. A readable storage medium, characterized in that: A computer program is stored, and the computer program can be executed by a processor locked in the storage medium to implement a method for identifying ferromagnetic resonance fault of an electromagnetic voltage transformer as described in any one of claims 1-8.

Images