Fault diagnosis method, device and equipment of current transformer and storage medium

Abstract

The invention relates to the technical field of fault diagnosis, and discloses a fault diagnosis method and device for a current transformer, equipment and a storage medium. The method comprises the following steps: performing current signal acquisition and complementary set empirical mode decomposition on a current transformer to obtain a plurality of intrinsic mode function components; performing time alignment and weighted reconstruction on the intrinsic mode function component to obtain a current signal prediction sequence; performing Markov transfer field conversion on the current signal prediction sequence to generate a two-dimensional Markov transfer field feature image; inputting the two-dimensional Markov transfer field feature image into a fringe pooling convolutional neural network, extracting remote spatial features through a fringe pooling module, and performing weight adaptive adjustment on a feature channel through an SE attention mechanism module to obtain a fault type and a fault degree; according to the invention, the problems of nonlinearity and non-stationary characteristics of the signal are effectively solved, and the fault diagnosis accuracy of the current transformer is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a fault diagnosis method, device, equipment and storage medium for a current transformer. Background Art

[0002] As an important measuring device in the power system, the operating status of the current transformer directly affects the safe and stable operation of the power grid. With the continuous expansion of the power grid and the in-depth promotion of smart grid construction, the difficulty of current transformer fault diagnosis is increasing. Traditional fault diagnosis methods mainly rely on manual experience judgment and simple mathematical and statistical analysis. It is difficult to accurately identify the type and degree of faults under complex working conditions, and there are problems such as low diagnostic accuracy and poor reliability.

[0003] In recent years, with the development of artificial intelligence technology, fault diagnosis methods based on deep learning have been widely used. However, due to the high reliability and long-term stability of current transformers, it is difficult to obtain fault samples and the sample imbalance problem is serious, which affects the training effect of deep learning models. At the same time, the fault characteristics of current transformers under different operating conditions are complex and changeable, and traditional deep learning methods are difficult to effectively extract and utilize these feature information. In addition, the fault characteristics of current transformers are nonlinear and non-stationary, and a single signal processing method is difficult to fully capture fault information. Existing diagnostic methods often ignore the time-frequency characteristics and state transition characteristics of signals, and lack the ability to quantitatively evaluate the degree of faults, which seriously restricts the accuracy and practicality of fault diagnosis. Summary of the invention

[0004] The present invention provides a fault diagnosis method, device, equipment and storage medium for a current transformer, which effectively solves the problem of nonlinear and non-stationary characteristics of signals and improves the fault diagnosis accuracy of the current transformer.

[0005] In a first aspect, the present invention provides a method for diagnosing a fault of a current transformer, the method comprising: Perform current signal acquisition and complementary set empirical mode decomposition on the current transformer to obtain multiple intrinsic mode function components; Performing time alignment and weighted reconstruction on the intrinsic mode function components to obtain a current signal prediction sequence; Performing Markov transfer field conversion on the current signal prediction sequence to generate a two-dimensional Markov transfer field characteristic image; The two-dimensional Markov transfer field feature image is input into the stripe pooling convolutional neural network, the long-range spatial features are extracted through the stripe pooling module, and then the weight of the feature channel is adaptively adjusted through the SE attention mechanism module to obtain the fault type and fault degree.

[0006] In a second aspect, the present invention provides a fault diagnosis device for a current transformer, the fault diagnosis device for a current transformer comprising: An acquisition module is used to collect current signals of the current transformer and to perform complementary set empirical mode decomposition to obtain multiple intrinsic mode function components; A reconstruction module, used for performing time alignment and weighted reconstruction on the intrinsic mode function components to obtain a current signal prediction sequence; A conversion module, used for performing Markov transfer field conversion on the current signal prediction sequence to generate a two-dimensional Markov transfer field characteristic image; The extraction module is used to input the two-dimensional Markov transfer field feature image into the stripe pooling convolutional neural network, extract the long-range spatial features through the stripe pooling module, and then adaptively adjust the weights of the feature channels through the SE attention mechanism module to obtain the fault type and fault degree.

[0007] The third aspect of the present invention provides a current transformer fault diagnosis device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the current transformer fault diagnosis device executes the above-mentioned current transformer fault diagnosis method.

[0008] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned current transformer fault diagnosis method.

[0009] In the technical solution provided by the present invention, the original current signal is adaptively decomposed by introducing the complementary set empirical mode decomposition algorithm, which effectively solves the problem of nonlinear and non-stationary characteristics of the signal and improves the accuracy of feature extraction. The augmented Dickey-Fuller test and dual-model prediction strategy are adopted to deal with the stationary and non-stationary components in a targeted manner, thereby enhancing the reliability of signal reconstruction. The Markov transition field is used for the two-dimensional mapping of fault features, which fully explores the state transition characteristics of the signal and improves the feature expression ability. The deep learning network structure of the stripe pooling module and the SE attention mechanism is designed to enhance the ability to capture remote spatial features and highlight the weights of key feature channels. A fault diagnosis mechanism based on the entropy value of the probability distribution is established, which realizes the accurate determination of the fault type and degree and provides a reliable basis for diagnosis. Through the multi-level feature extraction and fusion strategy, the problem of insufficient samples is effectively overcome and the generalization ability of the model is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0011] Figure 1 A schematic flow chart of a fault diagnosis method for a current transformer provided in an embodiment of the present application; Figure 2 A schematic block diagram of the structure of a fault diagnosis device for a current transformer provided in an embodiment of the present application; Figure 3 A schematic block diagram of the structure of a fault diagnosis device for a current transformer provided in an embodiment of the present application. DETAILED DESCRIPTION

[0012] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0013] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change based on actual conditions.

[0014] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0015] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0016] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0017] See also Figure 1 , Figure 1 A flow chart of a fault diagnosis method for a current transformer provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the fault diagnosis method of the current transformer provided in the embodiment of the present application includes steps S100 to S600.

[0018] Step S100, collecting current signals of the current transformer and performing complementary set empirical mode decomposition to obtain a plurality of intrinsic mode function components; It is understandable that the execution subject of the present invention may be a fault diagnosis device for a current transformer, or may be a terminal or a server, which is not specifically limited herein. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0019] Specifically, the original current signal generated during the operation of the current transformer is collected to obtain complete and high-precision original data. Since the original current signal is subject to pulse interference from the external environment during the acquisition process, it is subjected to median filtering to effectively remove the pulse interference in the signal and obtain the current signal without pulse interference. Outliers are detected and eliminated for the current signal without pulse interference. Outliers in the current signal are caused by abnormal events or noise, and these abnormal data will affect the accuracy of subsequent analysis. Outliers are detected and eliminated through appropriate algorithms to improve the credibility of the signal. Wavelet transform is performed on the current signal without outliers to decompose the signal into wavelet coefficients of different frequency bands. Wavelet transform can effectively extract the local features of the signal, so that high-frequency noise and the detailed features of the signal itself can be separated. Soft threshold processing is performed on the wavelet coefficients to suppress noise. Soft threshold processing sets a threshold, attenuates or even sets the coefficients less than the threshold, and retains the coefficients greater than the threshold, thereby minimizing the impact of noise while maintaining the main features of the signal as much as possible. The noise-reduced current signal is processed in segments, and the overall signal is divided into segments within several time periods so that each segment of the signal can be analyzed separately. The mean and variance of the segmented current signal are calculated. The mean and variance of each segment of the signal are calculated and used to standardize the signal to eliminate the deviation introduced by different acquisition conditions or equipment characteristics, and obtain a standardized segmented current signal. The standardized segmented signal is reorganized into a complete standardized current signal through time domain splicing. The standardized current signal is subjected to complementary set empirical mode decomposition, and the problem of modal aliasing that is prone to occur in traditional empirical mode decomposition is solved by introducing the complementary characteristics of white noise. During the decomposition process, the signal is decomposed layer by layer into several intrinsic mode function (IMF) components, and each IMF component represents the characteristics of a specific frequency range of the signal.

[0020] Gaussian white noise with an amplitude of 0.2 times the standard deviation is added to the standardized current signal to generate N groups of auxiliary noise signals. The introduction of Gaussian white noise helps to solve the problem of modal aliasing in traditional empirical mode decomposition, making the decomposition process more stable and more robust. The auxiliary noise signal is superimposed on the standardized current signal to obtain a group of noisy signals. These noisy signals retain the main features of the original signal, and at the same time achieve signal diversification by adding different noises. Local extreme point detection is performed on each group of noisy signals to identify the maximum and minimum points of the signal. Cubic spline interpolation is performed on the maximum and minimum points to obtain the upper and lower envelopes. Cubic spline interpolation can accurately fit the envelope while ensuring smoothness, ensuring that it can closely fit the upper and lower boundaries of the signal. By calculating the average value of the upper and lower envelopes, the mean envelope is generated to reflect the local trend of the signal. The mean envelope is subtracted from the standardized current signal to obtain the candidate intrinsic mode function. The candidate function is evaluated by standard deviation calculation to see if it meets the conditions of the intrinsic mode function. The intrinsic mode function needs to meet two points: first, the difference between the number of extreme points and the number of zero crossing points of the signal in the global range is at most 1; second, at any time point, the mean envelope of the signal should be close to zero. If the candidate function does not meet these conditions, it is used as a new input signal, and the process of generating envelopes and calculating mean envelopes by cubic spline interpolation is repeated until the conditions are met. Once the conditions are met, the first intrinsic mode function component is obtained. The first intrinsic mode function component is subtracted from the standardized current signal to obtain the first decomposition residual signal. This residual signal is regarded as a new signal to be decomposed, and the above steps are repeated, that is, from the detection of local extreme points to the extraction of candidate intrinsic mode functions, until the intrinsic mode function conditions are met again, and the next intrinsic mode function component is obtained. After each decomposition, the residual signal is updated to the new signal to be decomposed, and this process is repeated until the residual signal becomes a monotonic function or its amplitude is lower than the preset threshold. Under this termination condition, the decomposition process ends, and all intrinsic mode function components and the final residual components are obtained. In order to enhance the accuracy of decomposition, the intrinsic mode function components obtained by decomposing N groups of auxiliary noise signals are averaged. By superimposing and averaging multiple decomposition results, the random errors introduced by noise are eliminated, and the stability and credibility of the decomposition results are improved. After ensemble averaging, multiple intrinsic mode function components are finally obtained.

[0021] Step S200, performing time alignment and weighted reconstruction on the intrinsic mode function components to obtain a current signal prediction sequence; Specifically, a preliminary analysis is performed on multiple intrinsic mode function components, and their mean, variance, kurtosis coefficient and skewness coefficient are calculated. These statistical characteristics can describe the overall distribution characteristics of the signal. At the same time, by extracting the frequency domain features of the signal, indicators such as spectral power and main frequency components are obtained, so that the characteristics of the time domain and frequency domain are integrated to form a time-frequency feature vector. Based on the time-frequency feature vector, an augmented Dickey-Fuller test is performed on each intrinsic mode function component to determine the stationarity of the signal. The augmented Dickey-Fuller test divides the intrinsic mode function components into a stationary component set and a non-stationary component set by calculating the test statistic and comparing it with the critical value. For the stationary component set, the delay characteristics of the signal are analyzed by the autocorrelation function and the partial autocorrelation function, and the optimal order of the autoregressive moving average model (ARMA model) is determined by combining the AIC criterion (Akaike Information Criterion). After determining the model order, the least squares method is used to solve the parameters of the ARMA model, thereby realizing the modeling and prediction of the stationary component. Through this process, the future value of the stationary component is estimated to form the predicted value of the stationary component. For the set of non-stationary components, the non-stationary components are input into the K-means clustering module, the number of clusters is set to k, and the Euclidean distance is used as the similarity measure to calculate the distance from each sample point to each cluster center, and finally determine the k cluster center points. Based on these cluster center points, the Gaussian kernel function value of the non-stationary component is calculated, and the hidden layer nodes of the radial basis function neural network (RBFNN) are constructed based on this. In this way, the nonlinear characteristics of the non-stationary component are effectively captured. After the hidden layer nodes are constructed, the orthogonal least squares method is used to calculate its output matrix to obtain the weight parameters of the output layer. These weight parameters are used to predict the non-stationary components to obtain the predicted values ​​of the non-stationary components. The predicted values ​​of the stationary components and the predicted values ​​of the non-stationary components are rearranged according to the original time series to ensure that the predicted sequence can accurately correspond to the time points in the original sequence to avoid information dislocation. Since the predicted values ​​have unequally spaced sampling points, cubic spline interpolation is used to process the sequence to achieve smooth alignment of irregular time intervals and obtain the aligned predicted sequence. The aligned predicted sequence is weighted to reconstruct the energy component. The energy ratio of each component is calculated and applied as weighted coefficients to sequence reconstruction. The purpose of weighted reconstruction is to adjust the predicted sequence according to the energy contribution of each component to make it closer to the characteristics of the original signal and finally generate a current signal prediction sequence.

[0022] Step S300, performing Markov transfer field conversion on the current signal prediction sequence to generate a two-dimensional Markov transfer field feature image; Specifically, a time delay parameter τ is set for the current signal prediction sequence, where τ=1. The time delay parameter τ determines the reconstruction step size of the signal in the phase space, and τ=1 is to maximize the continuity of the time series. On this basis, the one-dimensional current signal prediction sequence is converted into a state matrix with a reconstruction dimension of 10 by the phase space reconstruction method. The state matrix contains the embedded representation of the original time series in the multidimensional space. The state matrix is ​​discretized and the state matrix of continuous values ​​is converted into discrete values ​​for easy statistics and probability calculation. Each data point is evenly divided according to the maximum and minimum value intervals of the state matrix, and each data point is mapped to a predefined discrete state interval to obtain a set of discrete state sequences. The discrete state sequence simplifies the complex signal representation into a finite set of states. The adjacent state pairs in the discrete state sequence are counted, and the number of occurrences of transitions from state i to state j is calculated to generate a state transition statistical matrix, in which each element represents the transition frequency of a specific state pair. The state transition statistical matrix reflects the dynamic relationship between different states in the sequence. The transition statistics matrix is ​​normalized, and the transition frequency of each row is divided by the total number of transitions in the row to obtain the Markov transition probability matrix. Each element of the Markov transition probability matrix represents the conditional probability of transitioning from one state to another. The Markov transition probability matrix is ​​probability normalized so that all probability values ​​are evenly distributed in the range of 0 to 1 to obtain the normalized probability matrix. Based on the normalized probability matrix, a two-dimensional image matrix is ​​constructed, in which each transition probability value is mapped to the pixel value of the corresponding position, and the initial image of the Markov transition field is obtained, which reflects the transition probability distribution of the signal between different states and is an important graphical representation of the time series signal. The initial image of the Markov transition field is subjected to histogram equalization. Histogram equalization redistributes the original pixel values ​​by calculating the cumulative distribution function of the pixel values, so that the overall contrast of the image is improved. The equalized image can better highlight the hidden feature patterns in the signal. The enhanced feature image is smoothed with a 5×5 Gaussian kernel. Gaussian kernel smoothing effectively removes high-frequency noise in the image by calculating the weighted average of local pixels, while retaining the main characteristic structure, and finally generates a two-dimensional Markov transfer field feature image.

[0023] Step S400: Input the two-dimensional Markov transfer field feature image into the stripe pooling convolutional neural network, extract the long-range spatial features through the stripe pooling module, and then adaptively adjust the weights of the feature channels through the SE attention mechanism module to obtain the fault type and fault degree.

[0024] Specifically, the two-dimensional Markov transfer field feature image is input into the first convolution layer of the stripe pooling convolutional neural network. The first convolutional layer uses 32 convolution kernels of size 3×3 and the step size is set to 1. The initial features in the image are extracted through convolution operation. The convolutional feature map is processed by the ReLU activation function. The nonlinear mapping enhances the network's ability to express features. At the same time, batch normalization is used to standardize the feature map to reduce the impact of internal covariate shift and ensure the consistency of feature distribution. The first layer feature map is input into the stripe pooling module of the stripe pooling convolutional neural network. The stripe pooling module uses a stripe kernel of size 1×3 and a sliding window with a step size of 1 in the horizontal direction to aggregate the feature map. This design can fully capture the long-range dependency of the feature map in the horizontal direction without destroying the feature integrity in the vertical direction. The maximum value pooling operation is performed in the sliding window to retain the feature value with the strongest response in the window and generate the stripe feature. The stripe feature is input into the second convolution layer of the stripe pooling convolutional neural network for feature extraction. In this layer, 64 convolution kernels of size 3×3 and stride 1 are used to extract high-level features of the image through convolution operation. The convolved feature map is nonlinearly mapped through the ReLU activation function to enhance the network's ability to capture complex feature patterns. At the same time, the distribution of the feature map is adjusted through batch normalization to ensure the stability and convergence of network training, and the second-layer feature map is obtained. The second-layer feature map is input into the squeeze layer of the SE attention mechanism module. The squeeze layer compresses the H×W dimension of the original feature map into a 1×1×C feature vector by performing a global average pooling operation on each channel to generate channel descriptors. These channel descriptors are condensed expressions of global information for each channel and capture the importance of features within the channel. The channel descriptors are passed to the first fully connected layer of the SE attention mechanism module for dimensionality reduction. By reducing the redundancy of features, more compact intermediate features are extracted. The intermediate features are input into the second fully connected layer for dimensionality increase, restored to the feature dimension consistent with the original number of channels, and the channel weight coefficients are generated through the Sigmoid activation function. These channel weight coefficients represent the importance of each channel and are used for subsequent feature recalibration. In the feature recalibration stage, each channel of the second-layer feature map is multiplied element-by-element by the corresponding channel weight coefficient to enhance the feature response of important channels while suppressing the contribution of minor channels to generate fault feature mapping data. The fault feature mapping data combines the long-range spatial features extracted by the stripe pooling module with the channel-level weight information adaptively adjusted by the SE module, and is highly representative and discriminative. The probability distribution of each fault type is calculated based on the fault feature mapping data. Through the classification layer of the network, the probability value corresponding to each fault type is extracted, and the uncertainty of the fault feature is quantitatively analyzed in combination with the entropy value of the probability distribution. The high and low entropy values ​​assist in judging the degree of fault. High entropy values ​​correspond to complex or severe faults, while low entropy values ​​represent simple or minor fault types.

[0025] The softmax function is used to perform nonlinear transformation on each component of the fault feature mapping data. The role of the softmax function is to convert the input high-dimensional feature data into a normalized probability value between 0 and 1, while ensuring that the sum of all components is 1, and obtain a fault type probability vector with n fault types. Each component in this probability vector represents the normalized probability value of the corresponding fault type. Each component in the fault type probability vector is logarithmically transformed to provide a nonlinear measure of the probability component for the subsequent information entropy calculation, thereby reflecting the uncertainty of the probability distribution of each type. According to the information entropy formula, the transformed components are weighted and summed to calculate the probability distribution entropy value of the diagnosis result. The information entropy value measures the uncertainty of the system. When the entropy value is high, it indicates that the probability distribution is more uniform and the uncertainty of the diagnosis result is higher; while the low entropy value indicates that the probability distribution is more concentrated and the diagnosis result of the fault type is more reliable. Based on the fault type probability vector, the fault type corresponding to the component with the largest probability value is selected as the candidate fault type. In order to ensure the reliability of the diagnosis result, the probability value is compared with the preset confidence threshold. If the probability value exceeds the confidence threshold, the candidate fault type is directly used in the preliminary diagnosis result; otherwise, the diagnosis result needs to be optimized in combination with further analysis. After the preliminary diagnosis result is determined, the corresponding probability distribution entropy value is analyzed. If it is found that the probability distribution entropy value is greater than the preset entropy value threshold, it indicates that the system has a large uncertainty in the diagnosis result of the fault type, triggering the secondary diagnosis mechanism to further improve the accuracy of the diagnosis. In the secondary diagnosis mechanism, the fault type probability vector is matched and analyzed with the pre-stored fault feature library. Using Euclidean distance as the similarity measurement index, the distance between the fault type probability vector and each sample in the feature library is calculated, and the feature template with the smallest distance is selected as the reference sample. By comparing historical data with typical features, the diagnosis result is corrected to obtain the corrected fault type. The probability distribution entropy value is interval mapped to quantify the severity of the fault. By defining a piecewise function, the entropy value is divided into three intervals, corresponding to mild, moderate and severe fault levels. A low entropy value corresponds to a mild fault with a high degree of certainty in the diagnosis result; a medium entropy value indicates that the system has a certain degree of uncertainty in the diagnosis process, corresponding to a moderate fault; and a high entropy value corresponds to a complex or severe fault type that is difficult for the system to clearly classify. Through this step, a complete diagnosis result of the fault type and fault degree is finally obtained.

[0026] In the embodiment of the present invention, the complementary set empirical mode decomposition algorithm is introduced to adaptively decompose the original current signal, which effectively solves the problem of nonlinear and non-stationary characteristics of the signal and improves the accuracy of feature extraction. The augmented Dickey-Fuller test and dual model prediction strategy are adopted to deal with the stationary and non-stationary components in a targeted manner, thereby enhancing the reliability of signal reconstruction. The Markov transition field is used for the two-dimensional mapping of fault features, which fully explores the state transition characteristics of the signal and improves the feature expression ability. The deep learning network structure of the stripe pooling module and the SE attention mechanism is designed to enhance the ability to capture remote spatial features and highlight the weights of key feature channels. A fault diagnosis mechanism based on the entropy value of the probability distribution is established, which realizes the accurate determination of the fault type and degree and provides a reliable basis for diagnosis. Through the multi-level feature extraction and fusion strategy, the problem of insufficient samples is effectively overcome and the generalization ability of the model is improved.

[0027] In a specific embodiment, the process of executing step S100 may specifically include the following steps: The current signal of the current transformer is collected to obtain the original current signal, and the original current signal is processed by median filtering to obtain the current signal with pulse interference removed; Performing outlier detection and elimination on the current signal from which pulse interference is removed to obtain the current signal from which outliers are removed, and performing wavelet transformation on the current signal from which outliers are removed to obtain wavelet coefficients; Performing soft threshold processing on the wavelet coefficients to obtain a current signal after noise reduction, and performing segment processing on the current signal after noise reduction to obtain a segmented current signal; Calculating the mean and variance of the segmented current signal to obtain a standardized segmented current signal, and splicing the standardized segmented current signal in the time domain to obtain a standardized current signal; Complementary ensemble empirical mode decomposition is performed on the standardized current signal to obtain multiple intrinsic mode function components.

[0028] Specifically, the original current signal of the current transformer is obtained in real time through sensors or acquisition equipment. ,in Indicates time. The original signal is disturbed by the external environment and contains non-stationary components such as high-frequency pulse noise, which requires median filtering. The mathematical expression of median filtering is: ; in, is the signal after median filtering, and the window length is Through median filtering, pulse interference is effectively removed and the main trend of the signal is retained. Perform outlier detection and removal. Outliers are detected by statistical methods, such as the outlier removal method based on standard deviation: ; in, is the signal mean, is the signal standard deviation, is the removal threshold (usually 2 or 3). After removing the outliers, we get a new signal , its noise interference is further reduced. Wavelet transform is performed to extract multi-resolution features. Wavelet transform is expressed as: ; in, is the scaled and translated form of the mother wavelet function, and Represents the scale factor and time translation factor respectively. Wavelet transform decomposes the signal into wavelet coefficients of different frequency bands. , which is used to analyze the local characteristics of the signal. In order to remove the residual noise, the wavelet coefficients are subjected to soft threshold processing, and the formula is: ; in, is the processed wavelet coefficient, is the threshold. Soft threshold processing effectively reduces the impact of noise on the signal. Segment processing is performed. Assume that the signal is divided into segments, each segment length is , then The segment signal is represented as: ; The mean and variance of each signal are calculated, and the standardization formula is: ; in, and Respectively After segment normalization, all segment signals are restored to standardized current signals through time domain splicing. . For the standardized signal Perform complementary set empirical mode decomposition. Add Gaussian white noise : ; in, is the noise amplitude. Then Perform empirical mode decomposition into several intrinsic mode function (IMF) components By superimposing the decomposition results of multiple groups of noise and taking the average, the final intrinsic mode function components are obtained: ; in, is the number of noise groups, For the The decomposition of the group For example, if the original signal is a sine wave superimposed with noise, it can be decomposed into a pure low-frequency sine signal and a high-frequency noise component through complementary set empirical mode decomposition.

[0029] In a specific embodiment, the process of performing the complementary set empirical mode decomposition on the standardized current signal to obtain multiple intrinsic mode function components may specifically include the following steps: M1: Add Gaussian white noise with an amplitude of 0.2 times the standard deviation to the standardized current signal to generate N groups of auxiliary noise signals; M2: Superimpose the auxiliary noise signal with the standardized current signal to obtain a noisy signal group, and perform local extreme point detection on the noisy signal group to obtain the maximum and minimum points of the signal; M3: Perform cubic spline interpolation on the maximum and minimum points to obtain the upper and lower envelopes, and calculate the average values ​​of the upper and lower envelopes to obtain the mean envelope; M4: Subtract the mean envelope from the standardized current signal to obtain the candidate intrinsic mode function, and calculate the standard deviation of the candidate intrinsic mode function to obtain the decomposition residual; M5: Determine whether the candidate intrinsic mode function meets the eigenmode function condition based on the decomposition residual. If not, repeat steps M3 to M4 until the condition is met to obtain the first eigenmode function component. M6: subtract the first intrinsic mode function component from the standardized current signal to obtain the first decomposition residual signal, and repeat steps M2 to M5 using the first decomposition residual signal as a new signal to be decomposed to obtain the next intrinsic mode function component; M7: Repeat steps M2 to M6 until the decomposed residual signal becomes a monotonic function or is less than a preset threshold, and all intrinsic mode function components and final residual components are obtained; M8: performing ensemble averaging on the intrinsic mode function components obtained by decomposing N groups of auxiliary noise signals to obtain multiple intrinsic mode function components.

[0030] Specifically, for the standardized current signal Gaussian white noise is added to enhance the stability of the signal. Assume that the standard deviation of the standardized current signal is , the generated Gaussian white noise satisfy , that is, the mean is zero and the standard deviation is 0.2 times the standard deviation of the original signal. The white noise generated by the group is independently obtained. Group auxiliary noise signal , its mathematical expression is: ; in, represents the number of the noise group, is a time variable. Perform local extreme point detection to identify the maximum points in the signal and minimum point This process is completed through derivative judgment, that is, satisfying And the second-order derivative When it is negative, it is a maximum value, and when it is positive, it is a minimum value. After the extreme points are detected, cubic spline interpolation is performed on these extreme points to generate the upper envelope of the signal. and lower envelope The interpolation function is expressed as: ; The mean envelope is obtained by calculating the average of the upper envelope and the lower envelope : ; Normalize the current signal Subtract mean envelope , and obtain the candidate eigenmode functions : ; For the candidate eigenmode functions Calculate standard deviation , and get the decomposition residual : ; in, is the total signal duration, yes If the decomposition residual If the preset threshold is exceeded or the candidate intrinsic mode function does not meet the conditions of the intrinsic mode function (such as the difference between the number of zero crossings and the number of extreme values ​​is not greater than 1, and the mean envelope is close to zero), the envelope calculation and subtraction operations need to be re-executed until the conditions are met, and finally the first intrinsic mode function component is obtained. . Subtract the normalized current signal from , and get the first decomposition residual signal : ; by As a new signal to be decomposed, repeat the process of extreme point detection, envelope calculation, subtraction operation and intrinsic mode function condition judgment to extract subsequent intrinsic mode function components in turn. , until the decomposed residual signal becomes a monotonic function or the amplitude is less than the preset threshold. After completing the above decomposition, the corresponding intrinsic mode function component sets are obtained. The intrinsic mode function components of the group noise signal are averaged to obtain the final intrinsic mode function components: ; in, Indicates The first eigenmode function components.

[0031] In a specific embodiment, the process of executing step S200 may specifically include the following steps: The mean, variance, kurtosis coefficient and skewness coefficient of multiple intrinsic mode function components are calculated, and the frequency domain features of the intrinsic mode function components are extracted to obtain the time-frequency feature vector; Based on the time-frequency eigenvector, the augmented Dickey-Fuller test is performed on the intrinsic mode function components, and the intrinsic mode function components are divided into a stationary component set and a non-stationary component set according to the test statistic; For the stationary component set, the delay order of the autocorrelation function and the partial autocorrelation function is calculated, and the optimal order of the autoregressive moving average model is determined by combining the AIC criterion. The model parameters are solved by the least squares method to obtain the predicted value of the stationary component. Input the non-stationary component set into the K-means clustering module, set the number of clusters to k, calculate the distance between the sample and the cluster center by Euclidean distance, and obtain k cluster center points; Based on k cluster center points, the Gaussian kernel function value is calculated for the non-stationary component set, the hidden layer nodes of the radial basis function neural network are constructed, and the hidden layer output matrix is ​​obtained; The output layer weights are calculated for the hidden layer output matrix by using the orthogonal least squares method, and the non-stationary components are predicted by the output layer weights to obtain the predicted values ​​of the non-stationary components; The predicted values ​​of the stationary component and the predicted values ​​of the non-stationary component are rearranged according to the original time series, and the unequally spaced sampling points are processed using cubic spline interpolation to obtain the aligned prediction series; The energy ratio of each component of the aligned prediction sequence is calculated, and the energy ratio is used as a weighting coefficient to reconstruct the sequence to obtain the current signal prediction sequence.

[0032] Specifically, for the multiple intrinsic mode function components obtained by decomposition Perform basic statistical feature calculations, including mean ,variance , Kurtosis coefficient and the skewness coefficient These characteristics reflect the basic statistical properties of the signal, and their calculation formulas are: ; ; in, is the length of the time series, For the The value of the eigenmode function component. Extract frequency domain features and obtain spectrum characteristics through fast Fourier transform. Frequency domain features include main frequency and spectrum power : ; in, for The Fourier transform of is the upper frequency limit. Combining the above statistical characteristics and frequency domain characteristics, the time-frequency feature vector is constructed Based on the time-frequency feature vector ,right An augmented Dickey-Fuller (ADF) test is performed to determine its stationarity. The null hypothesis of the ADF test is that the signal is non-stationary, and its test statistic is calculated by the following regression model: ; in, , is the regression coefficient, Represents the trend item, is the coefficient of the delay order, is the error term. When the t-statistic value of is significantly less than zero, the signal is judged to be stable. It is divided into a stationary component set and a non-stationary component set. For the stationary component set, the autocorrelation function (ACF) and partial autocorrelation function (PACF) are used to determine the optimal order of the ARMA model. . The best model order is selected by the AIC criterion: ; in, is the likelihood estimate of the model. After fitting the ARMA model, the least squares method is used to solve the model parameters and obtain the predicted value of the stationary component. For the non-stationary component set, input it into the K-means clustering module and set the number of clusters , calculate the sample by Euclidean distance With cluster center Distance: ; Select Cluster centers are obtained and radial basis function neural networks are constructed based on these centers. The Gaussian kernel function is calculated as follows: ; in, is the kernel width. Construct the hidden layer output matrix , the output layer weights are calculated by orthogonal least squares method , and make predictions for the non-stationary components: ; The predicted values ​​of the stationary component and the predicted values ​​of the non-stationary component are rearranged in the order of the original time series, and the unequally spaced sampling points are processed using cubic spline interpolation to generate an aligned prediction sequence. The energy ratio of each component is calculated by the aligned prediction sequence , the formula is ; Will As the weighted coefficient, the sequence is reconstructed to obtain the final current signal prediction sequence: .

[0033] In a specific embodiment, the process of executing step S300 may specifically include the following steps: A time delay parameter τ is set for the current signal prediction sequence, where τ=1, and the current signal prediction sequence is reconstructed in phase space to obtain a state matrix with a reconstruction dimension of 10; Discretize the state matrix, divide each data point evenly according to the interval range of maximum and minimum values, and obtain a discrete state sequence; Statistics are taken for the state pairs at adjacent moments in the discrete state sequence, the number of occurrences of state i transitioning to state j is calculated, and a state transition statistics matrix is ​​established; Based on the state transition statistics matrix, each state transition frequency is normalized, and the transition frequency of each row is divided by the total number of transitions to obtain the Markov transition probability matrix; The Markov transition probability matrix is ​​probability normalized to obtain a normalized probability matrix, and a two-dimensional image matrix is ​​constructed based on the normalized probability matrix, and each transition probability value is used as a pixel value at a corresponding position to obtain an initial image of the Markov transition field; The Markov transfer field initial image is histogram equalized, and the pixel values ​​are redistributed by calculating the cumulative distribution function to obtain an enhanced feature image. A 5×5 Gaussian kernel smoothing process is performed based on the enhanced feature image to obtain a two-dimensional Markov transfer field feature image.

[0034] Specifically, the current signal prediction sequence Setting the time delay parameters ,make , and based on this, the signal is reconstructed in phase space. Phase space reconstruction is a method of embedding a one-dimensional time series into a high-dimensional space, the purpose of which is to capture the dynamic characteristics of the signal. Assume that the reconstruction dimension is , the reconstructed state matrix Defined as: ; in, is the total length of the sequence, is the time delay, is the embedding dimension. Each row represents a state vector, describing the local characteristics of the signal at different time points. Discretize the state and divide its value range into a finite number of discrete intervals. Assume that the minimum and maximum values ​​of the state matrix are and , divide the value range evenly into intervals, the width of each interval is: ; For any element , mapping it to discrete states : ; in, , indicating the The state vector Through discretization, the continuous-valued signal is represented as a sequence of discrete states. . Count the state pairs at adjacent moments in the discrete state sequence and construct the state transition statistics matrix .set up Indicates from the state Transfer to state Frequency: ; After the statistics are completed, the frequency of each row is normalized to obtain the Markov transition probability matrix : ; in, Indicates from the state Transfer to state The conditional probability of Based on the Markov transition probability matrix , perform probability normalization and obtain the normalized probability matrix . Map all transition probability values ​​to the range of [0,1] to enhance the contrast of the probability distribution: ; Normalize the probability matrix Mapping to a two-dimensional image matrix , whose pixel values ​​correspond one to one with the normalized probability values: ; Generate a two-dimensional Markov transition field initial image to reflect the transition probability distribution between states. In order to enhance the image features, the initial image is processed by histogram equalization. By calculating the cumulative distribution function of pixel values , the original pixel value Mapping to new pixel values : ; After histogram equalization, the contrast of the image is enhanced, which is more conducive to subsequent feature extraction. The enhanced image is smoothed by Gaussian kernel, using a size of The Gaussian kernel matrix : ; in, is the standard deviation of the Gaussian kernel. Smoothing is done via a convolution operation: ; The final generated two-dimensional Markov transition field feature image Smoother and clearer, it can effectively represent the dynamic behavior pattern of the current signal.

[0035] In a specific embodiment, the process of executing step S400 may specifically include the following steps: The two-dimensional Markov transfer field feature image is input into the first convolution layer of the stripe pooling convolutional neural network, and features are extracted through 32 convolution kernels of size 3×3 and convolution operation with a step size of 1. The feature map is then activated by ReLU and batch normalized to obtain the first-layer feature map. The first layer feature map is input into the stripe pooling module of the stripe pooling convolutional neural network, and a stripe kernel of 1×3 and a horizontal sliding window with a step size of 1 are used for feature aggregation. The maximum value pooling operation is performed on the features in the window to obtain the horizontal stripe features; The stripe features are input into the second convolutional layer of the stripe pooling convolutional neural network for processing. 64 3×3 convolution kernels and a convolution operation with a step size of 1 are used to extract high-level features, and ReLU activation and batch normalization are performed to obtain the second-layer feature map. The second layer feature map is input into the squeeze layer of the SE attention mechanism module, and a global average pooling operation is performed on each feature channel to compress the feature map of H×W dimension into a 1×1×C feature vector to obtain a channel descriptor; The channel descriptor is input into the first fully connected layer of the SE attention mechanism module for dimensionality reduction to obtain the intermediate features, and the intermediate features are input into the second fully connected layer of the SE attention mechanism module for dimensionality increase to obtain the channel weight coefficient; Each channel of the second-layer feature map is multiplied element by element by the corresponding channel weight coefficient to perform channel-level feature recalibration to obtain fault feature mapping data; The probability distribution of each fault type is calculated based on the fault feature mapping data, and the fault type and fault degree are determined based on the probability distribution entropy value.

[0036] Specifically, the two-dimensional Markov transfer field feature image Input to the first convolutional layer of the striped pooling convolutional neural network. The first convolutional layer uses 32 size The convolution kernel , the step size is set to 1, and the local features of the input image are extracted through convolution operation. The mathematical expression of convolution operation is: ; in, Indicates The convolution kernels are located at The output eigenvalues ​​of is the bias, is the weight parameter of the convolution kernel. The convolution operation can capture local feature patterns, such as edges and textures. Apply the ReLU activation function to the convolution result: ; The ReLU activation function can introduce nonlinear characteristics, suppress negative values, and improve the expressiveness of the model. The activation results are standardized through batch normalization, and the formula is: ; in, and are the mean and variance of the mini-batch data, and is a learnable parameter, is a smoothing term to prevent the denominator from being zero. After convolution, activation and normalization, the first layer feature map is obtained. . The first layer feature map Input to the stripe pooling module, which uses The stripe kernel slides the window in the horizontal direction with a step size of 1, and performs the maximum pooling operation on the feature values ​​in each window. The calculation formula of stripe pooling is: ; This operation can aggregate the feature information in the horizontal direction while maintaining the integrity of the features in the vertical direction, generating horizontal stripe features. . The stripe features Input to the second convolutional layer. The second convolutional layer uses 64 size The convolution kernel , the step size is also 1, and higher-level feature extraction is performed. After the same convolution, ReLU activation and batch normalization operations as the first layer, the second layer feature map is obtained. . The second layer feature map The input is sent to the squeeze layer of the SE attention mechanism module. The squeeze layer performs global average pooling on each feature channel. The spatial dimension is compressed to generate a channel descriptor : ; in, It is a channel The channel descriptor z is input into the first fully connected layer of the SE module to complete the dimensionality reduction operation: ; in, and are the weight matrix and bias of the first fully connected layer, is the activation function. The intermediate features after dimensionality reduction It is passed to the second fully connected layer for dimensionality increase to generate the channel weight coefficient s: ; The dimension of s is the same as the number of original channels. Each channel of is multiplied element by element with the corresponding channel weight coefficient to achieve channel-level feature recalibration and obtain fault feature mapping data : ; Mapping data based on fault signatures Calculate the probability distribution of each fault type. Use the softmax function to map the features to the probability space: ; in, It is The probability of a failure type, is the classification weight, is the number of fault types. Combined with the entropy value of the probability distribution: ; Entropy It indicates the uncertainty of the prediction, and the severity of the fault can be further determined by comparing it with the set threshold. For example, when the entropy value is low and the probability of a certain fault type is significantly higher than other types, it is determined to be this type of fault; when the entropy value is high, it indicates that there is uncertainty in the diagnosis result.

[0037] In a specific embodiment, the process of calculating the probability distribution of each fault type according to the fault feature mapping data and determining the fault type and fault degree in combination with the probability distribution entropy value may specifically include the following steps: Each component of the fault feature mapping data is nonlinearly transformed by the softmax function, and the normalized probability values ​​of n fault types are calculated to obtain the fault type probability vector; Perform logarithmic transformation on each component in the fault type probability vector, and perform weighted summation on the transformed components according to the information entropy formula to obtain the probability distribution entropy value of the diagnosis result; Based on the fault type probability vector, the fault type with the largest probability value is selected as the candidate fault type, and the probability value is compared with the preset confidence threshold to obtain a preliminary diagnosis result; Analyze the probability distribution entropy value corresponding to the preliminary diagnosis result. When the probability distribution entropy value is greater than the preset entropy value threshold, trigger the secondary diagnosis mechanism. The fault type probability vector is calculated with the pre-stored fault feature library through Euclidean distance calculation, and the feature template with the smallest distance is selected as the reference sample to obtain the corrected fault type; The entropy value of the probability distribution is mapped to intervals and divided into three intervals by a piecewise function, corresponding to the mild, moderate and severe fault degrees, respectively, to obtain the fault type and fault degree.

[0038] Specifically, the fault feature mapping data Input the softmax function, for each component Perform nonlinear transformation and calculate The normalized probability value of each fault type forms a fault type probability vector The mathematical expression of the softmax function is: ; in, It is The probability of a failure type, is the component value corresponding to the fault feature, is the total number of fault types. The role of the softmax function is to convert feature values ​​into probability values, ensuring that all satisfy and For each component in the probability vector Perform logarithmic transformation to obtain , and perform weighted summation according to the information entropy formula to calculate the probability distribution entropy value of the diagnosis result The formula for information entropy is: ; in, It measures the uncertainty of the probability distribution. When it is close to zero, it means that the probability distribution is highly concentrated and the system is highly certain about the diagnosis result; When it is larger, it means that the distribution is more dispersed and the diagnosis result has greater uncertainty. , select the fault type with the largest probability value as the candidate fault type. Suppose the maximum probability value is , the corresponding fault type is ,but: ; Will With preset confidence threshold Compare, if , the preliminary diagnosis result is directly adopted As a diagnostic type; otherwise, the corresponding entropy value needs to be further analyzed .when Greater than the preset entropy threshold When , it indicates that the uncertainty of the diagnosis result is high, triggering the secondary diagnosis mechanism. In the secondary diagnosis mechanism, the fault type probability vector The template vector in the pre-stored fault feature library Perform Euclidean distance calculation, select the feature template with the smallest distance as the reference sample, and correct the fault type. The formula for Euclidean distance is: ; in, It is The template vector The probability value of the components. So that: ; Corresponding template It is the reference sample, and its type is used as the corrected fault type. Perform interval mapping and divide it into three intervals through piecewise function, corresponding to mild, moderate and severe fault levels respectively. Suppose the entropy value interval is , the mapping function is: ; The fault type after correction and the fault degree corresponding to the entropy value are combined to finally generate the fault diagnosis result.

[0039] See also Figure 2 , Figure 2 A schematic block diagram of the structure of a fault diagnosis device 200 for a current transformer provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the fault diagnosis device 200 of the current transformer includes: The acquisition module 210 is used to collect the current signal of the current transformer and perform complementary set empirical mode decomposition to obtain multiple intrinsic mode function components; A reconstruction module 220 is used to perform time alignment and weighted reconstruction on the intrinsic mode function components to obtain a current signal prediction sequence; A conversion module 230, used for performing Markov transfer field conversion on the current signal prediction sequence to generate a two-dimensional Markov transfer field feature image; The extraction module 240 is used to input the two-dimensional Markov transfer field feature image into the stripe pooling convolutional neural network, extract the long-range spatial features through the stripe pooling module, and then adaptively adjust the weights of the feature channels through the SE attention mechanism module to obtain the fault type and fault degree.

[0040] Through the synergy of the above components, the complementary set empirical mode decomposition algorithm is introduced to adaptively decompose the original current signal, effectively solving the nonlinear and non-stationary characteristics of the signal and improving the accuracy of feature extraction. The augmented Dickey-Fuller test and dual model prediction strategy are used to deal with the stationary and non-stationary components in a targeted manner, enhancing the reliability of signal reconstruction. The Markov transition field is used for the two-dimensional mapping of fault features, fully mining the state transition characteristics of the signal and improving the feature expression ability. The deep learning network structure of the stripe pooling module and SE attention mechanism is designed to enhance the ability to capture remote spatial features and highlight the weight of key feature channels. A fault diagnosis mechanism based on the entropy value of the probability distribution is established, which realizes the accurate determination of the fault type and degree and provides a reliable diagnosis basis. Through the multi-level feature extraction and fusion strategy, the problem of insufficient samples is effectively overcome and the generalization ability of the model is improved.

[0041] See also Figure 3 , Figure 3A schematic block diagram of the structure of a current transformer fault diagnosis device 300 provided in an embodiment of the present application, wherein the current transformer fault diagnosis device 300 includes a processor 301 and a memory 302, wherein the processor 301 and the memory 302 are connected via a device bus 303, wherein the memory 302 may include a non-volatile storage medium and an internal memory.

[0042] The non-volatile storage medium can store a computer program. The computer program includes program instructions, and when the program instructions are executed by the processor 301, the processor 301 can execute any of the above-mentioned current transformer fault diagnosis methods.

[0043] The processor 301 is used to provide computing and control capabilities to support the operation of the entire current transformer fault diagnosis device 300 .

[0044] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor 301, the processor 301 can execute any of the above-mentioned current transformer fault diagnosis methods.

[0045] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a partial structure related to the scheme of the present application, and does not constitute a limitation on the fault diagnosis device 300 of the current transformer involved in the scheme of the present application. The specific fault diagnosis device 300 of the current transformer may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0046] It should be understood that the processor 301 may be a central processing unit (CPU), and the processor 301 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0047] It should be noted that those skilled in the art can clearly understand that, for the convenience and simplicity of description, the specific working process of the current transformer fault diagnosis device 300 described above can refer to the corresponding process of the aforementioned current transformer fault diagnosis method, and will not be repeated here.

[0048] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by one or more processors, the one or more processors implement the fault diagnosis method of the current transformer provided in the embodiment of the present application.

[0049] The computer-readable storage medium may be an internal storage unit of the current transformer fault diagnosis device 300 of the aforementioned embodiment, such as a hard disk or memory of the current transformer fault diagnosis device 300. The computer-readable storage medium may also be an external storage device of the current transformer fault diagnosis device 300, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc., equipped with the current transformer fault diagnosis device 300.

[0050] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0051] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0052] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A fault diagnosis method for a current transformer, characterized in that: include: Perform current signal acquisition and complementary set empirical mode decomposition on the current transformer to obtain multiple intrinsic mode function components; Performing time alignment and weighted reconstruction on the intrinsic mode function components to obtain a current signal prediction sequence; Performing Markov transfer field conversion on the current signal prediction sequence to generate a two-dimensional Markov transfer field characteristic image; The two-dimensional Markov transfer field feature image is input into the stripe pooling convolutional neural network, the long-range spatial features are extracted through the stripe pooling module, and then the weight of the feature channel is adaptively adjusted through the SE attention mechanism module to obtain the fault type and fault degree.

2. The fault diagnosis method of current transformer according to claim 1, characterized in that: The current transformer is subjected to current signal acquisition and complementary set empirical mode decomposition to obtain multiple intrinsic mode function components, including: The current signal of the current transformer is collected to obtain an original current signal, and the original current signal is subjected to median filtering to obtain a current signal with pulse interference removed; Performing outlier detection and elimination on the current signal from which the pulse interference is removed to obtain the current signal from which the outliers are removed, and performing wavelet transformation on the current signal from which the outliers are removed to obtain wavelet coefficients; Performing soft threshold processing on the wavelet coefficients to obtain a current signal after noise reduction, and performing segment processing on the current signal after noise reduction to obtain a segmented current signal; Calculating the mean and variance of the segmented current signals to obtain standardized segmented current signals, and splicing the standardized segmented current signals in the time domain to obtain standardized current signals; Complementary set empirical mode decomposition is performed on the standardized current signal to obtain a plurality of intrinsic mode function components.

3. The fault diagnosis method of current transformer according to claim 2, characterized in that: The method of performing complementary set empirical mode decomposition on the standardized current signal to obtain a plurality of intrinsic mode function components includes: M1: Adding Gaussian white noise with an amplitude of 0.2 times the standard deviation to the standardized current signal to generate N groups of auxiliary noise signals; M2: superimposing the auxiliary noise signal with the standardized current signal to obtain a noisy signal group, and performing local extreme point detection on the noisy signal group to obtain the maximum and minimum points of the signal; M3: performing cubic spline interpolation on the maximum point and the minimum point to obtain an upper envelope and a lower envelope, and calculating the average values ​​of the upper envelope and the lower envelope to obtain a mean envelope; M4: subtracting the mean envelope from the standardized current signal to obtain a candidate intrinsic mode function, and performing standard deviation calculation on the candidate intrinsic mode function to obtain a decomposition residual; M5: judging whether the candidate intrinsic mode function satisfies the intrinsic mode function condition based on the decomposition residual, and if not, repeatedly executing steps M3 to M4 until the condition is satisfied, to obtain the first intrinsic mode function component; M6: subtracting the first intrinsic mode function component from the standardized current signal to obtain a first decomposition residual signal, and repeating steps M2 to M5 with the first decomposition residual signal as a new signal to be decomposed to obtain a next intrinsic mode function component; M7: Repeat steps M2 to M6 until the decomposed residual signal becomes a monotonic function or is less than a preset threshold, and obtain all intrinsic mode function components and final residual components; M8: performing collective averaging on the intrinsic mode function components obtained by decomposing the N groups of auxiliary noise signals to obtain multiple intrinsic mode function components.

4. The fault diagnosis method of current transformer according to claim 3, characterized in that: The step of performing time alignment and weighted reconstruction on the intrinsic mode function components to obtain a current signal prediction sequence includes: Calculating the mean, variance, kurtosis coefficient and skewness coefficient of the multiple intrinsic mode function components, and extracting frequency domain features of the intrinsic mode function components to obtain time-frequency feature vectors; Performing an augmented Dickey-Fuller test on the intrinsic mode function components based on the time-frequency eigenvectors, and dividing the intrinsic mode function components into a stationary component set and a non-stationary component set according to the test statistics; The stationary component set is determined by calculating the delay order of the autocorrelation function and the partial autocorrelation function, combining the AIC criterion to determine the optimal order of the autoregressive moving average model, and solving the model parameters by the least squares method to obtain the predicted value of the stationary component; The non-stationary component set is input into the K-means clustering module, the number of clusters is set to k, and the distance between the sample and the cluster center is calculated by Euclidean distance to obtain k cluster center points; Calculating Gaussian kernel function values ​​for the non-stationary component set based on the k cluster center points, constructing hidden layer nodes of the radial basis function neural network, and obtaining a hidden layer output matrix; Calculating the output layer weights of the hidden layer output matrix by using the orthogonal least squares method, and predicting the non-stationary component by using the output layer weights to obtain the predicted value of the non-stationary component; The predicted values ​​of the stationary component and the predicted values ​​of the non-stationary component are rearranged according to the original time series, and the unequally spaced sampling points are processed by cubic spline interpolation to obtain an aligned prediction sequence; The energy ratio of each component of the aligned prediction sequence is calculated, and the energy ratio is used as a weighting coefficient to reconstruct the sequence to obtain a current signal prediction sequence.

5. The fault diagnosis method of current transformer according to claim 4, characterized in that: The performing Markov transfer field conversion on the current signal prediction sequence to generate a two-dimensional Markov transfer field feature image includes: Setting a time delay parameter τ for the current signal prediction sequence, where τ=1, and performing phase space reconstruction on the current signal prediction sequence to obtain a state matrix with a reconstruction dimension of 10; Discretize the state matrix, evenly divide each data point according to the interval range of the maximum value and the minimum value, and obtain a discrete state sequence; Counting the state pairs at adjacent moments in the discrete state sequence, calculating the number of occurrences of state i transferring to state j, and establishing a state transfer statistical matrix; Based on the state transition statistical matrix, each state transition frequency is normalized, and the transition frequency of each row is divided by the total number of transitions to obtain a Markov transition probability matrix; Performing probability normalization on the Markov transition probability matrix to obtain a normalized probability matrix, and constructing a two-dimensional image matrix based on the normalized probability matrix, taking each transition probability value as a pixel value at a corresponding position, and obtaining an initial image of the Markov transition field; The Markov transfer field initial image is subjected to histogram equalization, pixel values ​​are reallocated by calculating the cumulative distribution function to obtain an enhanced feature image, and a 5×5 Gaussian kernel smoothing process is performed based on the enhanced feature image to obtain a two-dimensional Markov transfer field feature image.

6. The fault diagnosis method of current transformer according to claim 5, characterized in that: The two-dimensional Markov transfer field feature image is input into the stripe pooling convolutional neural network, the long-range spatial features are extracted through the stripe pooling module, and the weight of the feature channel is adaptively adjusted through the SE attention mechanism module to obtain the fault type and fault degree, including: Inputting the two-dimensional Markov transfer field feature image into the first convolution layer of the stripe pooling convolutional neural network, performing feature extraction through 32 convolution kernels of 3×3 size and convolution operation with a step size of 1, and performing ReLU activation and batch normalization processing on the feature map to obtain a first layer feature map; Inputting the first layer feature map into the stripe pooling module of the stripe pooling convolutional neural network, using a stripe kernel of 1×3 and a horizontal sliding window with a step size of 1 to perform feature aggregation, performing a maximum pooling operation on the features in the window, and obtaining horizontal stripe features; Inputting the stripe features into the second convolution layer of the stripe pooling convolutional neural network for processing, using 64 convolution kernels of 3×3 size and a convolution operation with a step size of 1 to extract high-level features, and performing ReLU activation and batch normalization processing to obtain a second-layer feature map; The second layer feature map is input into the squeeze layer of the SE attention mechanism module, and a global average pooling operation is performed on each feature channel to compress the feature map of H×W dimension into a 1×1×C feature vector to obtain a channel descriptor; Input the channel descriptor into the first fully connected layer of the SE attention mechanism module for dimensionality reduction to obtain intermediate features, and input the intermediate features into the second fully connected layer of the SE attention mechanism module for dimensionality increase to obtain channel weight coefficients; Each channel of the second layer feature map is multiplied element by element by the corresponding channel weight coefficient to perform channel-level feature recalibration to obtain fault feature mapping data; The probability distribution of each fault type is calculated according to the fault feature mapping data, and the fault type and fault degree are determined in combination with the probability distribution entropy value.

7. The fault diagnosis method of a current transformer according to claim 6, characterized in that: The calculating the probability distribution of each fault type according to the fault feature mapping data, and determining the fault type and fault degree in combination with the probability distribution entropy value, includes: Performing a nonlinear transformation on each component of the fault feature mapping data by using a softmax function, calculating normalized probability values ​​of n fault types, and obtaining a fault type probability vector; Performing a logarithmic transformation on each component in the fault type probability vector, and performing weighted summation on the transformed components according to an information entropy formula to obtain a probability distribution entropy value of the diagnosis result; Based on the fault type probability vector, the fault type with the largest probability value is selected as the candidate fault type, and the probability value is compared with a preset confidence threshold to obtain a preliminary diagnosis result; Analyze the probability distribution entropy value corresponding to the preliminary diagnosis result, and trigger a secondary diagnosis mechanism when the probability distribution entropy value is greater than a preset entropy value threshold; Calculate the Euclidean distance between the fault type probability vector and the pre-stored fault feature library, select the feature template with the smallest distance as a reference sample, and obtain the corrected fault type; The probability distribution entropy value is interval mapped, and the entropy value is divided into three intervals by a piecewise function, corresponding to light, moderate and severe fault degrees, respectively, to obtain the fault type and fault degree.

8. A fault diagnosis device for a current transformer, characterized in that: A method for diagnosing a fault of a current transformer according to any one of claims 1 to 7, comprising: An acquisition module is used to collect current signals of the current transformer and to perform complementary set empirical mode decomposition to obtain multiple intrinsic mode function components; A reconstruction module, used for performing time alignment and weighted reconstruction on the intrinsic mode function components to obtain a current signal prediction sequence; A conversion module, used for performing Markov transfer field conversion on the current signal prediction sequence to generate a two-dimensional Markov transfer field characteristic image; The extraction module is used to input the two-dimensional Markov transfer field feature image into the stripe pooling convolutional neural network, extract the long-range spatial features through the stripe pooling module, and then adaptively adjust the weights of the feature channels through the SE attention mechanism module to obtain the fault type and fault degree.

9. A fault diagnosis device for a current transformer, characterized in that: The fault diagnosis device for the current transformer comprises: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory to enable the fault diagnosis device of the current transformer to execute the fault diagnosis method of the current transformer according to any one of claims 1-7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the fault diagnosis method for the current transformer according to any one of claims 1 to 7 is implemented.

Images