Feature extraction method and device for GIS ultra-high frequency partial discharge signal

Abstract

The present invention discloses a method and device for extracting features from ultra-high frequency partial discharge (UHF) signals from a GIS, relating to the field of electrical data processing. The method comprises: optimizing the number of modal components, K, of a VMD algorithm using a center frequency method to obtain an optimal K value; decomposing the UHF partial discharge signal using the VMD algorithm with the optimal K value as the modal component number to obtain K intrinsic mode function (IMF) components; performing different types of entropy calculations on each IMF component to obtain values ​​of different types of characteristic variables in each IMF component and constructing a feature matrix; inputting the feature matrix into a dynamic network marker model based on a sample covariance matrix to extract dominant characteristic variables and their corresponding key network markers, calculating quantitative correlation indices between the dominant characteristic variables, and constructing the extracted key features based on the key network markers and the quantitative correlation indices. The present invention addresses issues such as insufficient robustness of feature expression.

Description

Technical Field

[0001] The present invention relates to the field of electrical data processing, and in particular to a method and device for extracting features of GIS ultra-high frequency partial discharge signals. Background Art

[0002] In recent years, my country has been actively promoting the construction of a new power system based primarily on renewable energy. Compared to traditional power systems, these new power systems, with their power source structure primarily based on renewable energy, exhibit significant randomness, volatility, and intermittency. These systems also place increasing demands on flexibility, controllability, safety, and stability, posing significant challenges to the operation and maintenance of the high-voltage power equipment that forms the core architecture of the system. The risk of large-scale power outages caused by failure or damage to critical power equipment is increasing. As a high-voltage power device in power systems, the condition of gas-insulated switchgear (GIS) is crucial to the safe operation of these systems. During the production, transportation, installation, and operation of GIS, defects that affect the equipment's insulation are inevitable. After a period of high-voltage, high-current operation, these defects can cause partial discharge (PD) within the GIS. Various insulation defects caused by partial discharge (PD) are the primary culprits for equipment accidents and serve as a warning sign of insulation hazards. These tiny discharges can corrode the insulation structure. If not promptly detected and addressed, the deterioration can expand, eventually leading to insulation breakdown, impacting the normal operation of power equipment and, in severe cases, potentially paralyzing the entire power system. At the same time, different insulation defects have different degrees of harm to the insulation status of the equipment. Therefore, accurately judging the type of insulation defect that causes PD phenomenon is of great significance for guiding maintenance work and timely reducing its potential risks.

[0003] Extracting characteristic variables is key to PD type identification methods, and their quality directly impacts the performance of classification algorithms. Currently, high requirements exist for the analysis, processing, and feature extraction of UHF partial discharge (PD) signals from GIS, as these signals have complex, nonstationary, and nonlinear characteristics. Traditional time-domain or frequency-domain analysis struggles to accurately extract fault characteristics from signals. However, time-frequency analysis methods, with their advantages of time and frequency localization and multi-scale analysis, can characterize the relationship between signal frequency and time series, making them suitable for the analysis and processing of nonstationary signals. Wavelet transforms offer advantages such as low entropy, multi-resolution, decorrelation, and flexible basis selection, but they are limited by their limited adaptability and energy leakage. Empirical mode decomposition (EMD) can effectively extract information from noisy, nonlinear, and nonstationary processes, but suffers from modal aliasing and endpoint effects. Ensemble empirical mode decomposition (EEMD) overcomes the shortcomings of the EMD method, but this method is computationally intensive and currently difficult to implement on edge computing gateways for online monitoring devices. Compared to these methods, variational mode decomposition (VMD) is an adaptive multi-resolution technique that can reduce computational complexity while maintaining signal decomposition quality. It has excellent performance in processing non-stationary and nonlinear signals, can improve the reliability of signal decomposition, and avoid redundant features. However, this method has certain limitations, as the modal decomposition parameters significantly affect performance.

[0004] At the same time, most feature extraction methods focus on extracting a single feature in isolation or processing multiple independent features in parallel. Taking the typical variational mode decomposition-entropy analysis method as an example, traditional processing workflows typically only extract the independent entropy value features of each modal component. These single entropy features exhibit significant limitations in complex noisy environments: single metrics such as power spectral entropy are highly sensitive to noise perturbations, and the stability performance of different entropy measures varies significantly, resulting in insufficient robustness in feature representation. Although integrating multiple entropy features can partially compensate for the shortcomings of single metrics, the parallel processing of independent features generates a high-dimensional, highly correlated, redundant feature space, which can easily lead to the curse of dimensionality when directly input into classification models. More importantly, this processing approach ignores the inherent correlation mechanism between different entropy features in the modal components. Specifically, various entropy metrics essentially reflect characteristic properties of different dimensions of the same signal, and their coordinated variations contain rich complementary information. Therefore, currently used correlation methods still have certain limitations in mining the nonlinear coupling relationships between multidimensional features, which to some extent affects the completeness and accuracy of complex signal feature representation. Summary of the Invention

[0005] The purpose of this application is to propose a feature extraction method and device for GIS ultra-high frequency partial discharge signals in response to the above-mentioned technical problems.

[0006] In a first aspect, the present invention provides a method for extracting features of a GIS ultra-high frequency partial discharge signal, comprising the following steps:

[0007] Acquire GIS UHF signals and pre-process them to obtain UHF partial discharge signals;

[0008] Obtain training data and use the center frequency method to optimize the number of modal components K of the VMD algorithm to obtain the optimal K value;

[0009] The ultra-high frequency partial discharge signal is decomposed using the VMD algorithm with the optimal K value of modal components to obtain K intrinsic mode function components. Different types of entropy calculations are performed on each intrinsic mode function component to obtain the values ​​of different types of characteristic variables in each intrinsic mode function component. A characteristic matrix is ​​constructed based on the values ​​of different types of characteristic variables in each intrinsic mode function component.

[0010] The feature matrix is ​​input into the dynamic network marker model based on the sample covariance matrix to extract the dominant feature variables and their corresponding index mark set. The set of index marks is used as the key network marker, and the correlation quantitative index between the dominant feature variables is calculated. The extracted key features are constructed based on the key network markers and the correlation quantitative index.

[0011] As a preference, training data is obtained and the center frequency method is used to optimize the number of modal components K of the VMD algorithm to obtain the optimal K value, specifically including:

[0012] Collect ultra-high frequency partial discharge signals corresponding to different partial discharge types of GIS devices and form training data;

[0013] For each partial discharge type, the central frequency values ​​of each intrinsic mode function component obtained by the VMD algorithm under different modal component numbers K are calculated for the ultra-high frequency partial discharge signal;

[0014] For each partial discharge type, the relative rate of change of the center frequency value of the last intrinsic mode function component under adjacent K values ​​is calculated as shown in the following formula:

[0015] ;

[0016] in, It represents the center frequency value of the K+1th intrinsic mode function component when the number of modal components is K+1. It represents the center frequency value of the Kth intrinsic mode function component when the number of modal components is K. Indicates the relative rate of change when the number of modal components is K;

[0017] The number K of modal components corresponding to each partial discharge type when the relative change rate is less than or equal to the threshold is counted, and the number K of modal components corresponding to the most frequent occurrences of all partial discharge types when the relative change rate is less than or equal to the threshold is taken as the optimal K value.

[0018] Preferably, different types of entropy calculations include calculations of power spectrum entropy, sample entropy, fuzzy entropy, permutation entropy and distribution entropy, and the result of each type of entropy calculation corresponds to a characteristic variable.

[0019] Preferably, the feature matrix is ​​input into a dynamic network marker model based on a sample covariance matrix to extract the dominant feature variables and their corresponding index tag sets, specifically including:

[0020] Performing standardization on the feature matrix to obtain a standardized feature matrix;

[0021] A sample covariance matrix of the standardized feature matrix is ​​constructed, and the sample covariance matrix is ​​subjected to eigendecomposition. The dominant feature variables are screened out according to the eigenvectors and element rankings corresponding to the maximum eigenvalues ​​of the sample covariance matrix, and the set of index marks corresponding to the dominant feature variables is extracted.

[0022] As an optimal method, the characteristic decomposition process of the sample covariance matrix is ​​as follows:

[0023] ;

[0024] Where C represents the sample covariance matrix, The first eigenvalues is a diagonal matrix with diagonal elements, and the eigenvalues ​​of the sample covariance matrix C To sort from large to small, the first eigenvalue of the sample covariance matrix C is obtained is the maximum eigenvalue of the sample covariance matrix C, is the unit eigenvector is an orthogonal matrix of columns, , n is the number of characteristic variables;

[0025] The element in the i-th row and j-th column of the sample covariance matrix C Expressed as:

[0026] ;

[0027] in, Represents the first Characteristic vector of the column Hedi Characteristic vector of the column The covariance between , represents the maximum eigenvalue of the sample covariance matrix, and are the maximum eigenvalues ​​of the sample covariance matrix The corresponding unit eigenvector No. elements and elements, corresponding to the first element in the feature matrix after normalization. Characteristic vector of the column Hedi Characteristic vector of the column ; is the first order of the sample covariance matrix eigenvalues, and are the first eigenvalues The corresponding unit eigenvector No. elements and elements;

[0028] Compare the eigenvectors corresponding to the largest eigenvalues The n elements in the matrix are selected as the leading feature variables, and the index mark set of the leading feature variables in the standardized feature matrix is ​​used as the key network mark. .

[0029] As a preference, for the dominant characteristic variables, use the correlation quantitative index It is used to quantify the correlation between the dominant feature variables, as shown in the following formula:

[0030] ;

[0031] Among them, m is the key network marker The number of elements in , the value of m is set to satisfy and , P value is a positive real number in [0,1], n is the number of characteristic variables, Represents the first Characteristic vector of the column Hedi Characteristic vector of the column The covariance between Indicates the absolute value, represents the L2 norm;

[0032] The key features are vectors composed of key network markers and quantitative correlation indicators.

[0033] In a second aspect, the present invention provides a feature extraction device for a GIS ultra-high frequency partial discharge signal, comprising:

[0034] a preprocessing module configured to acquire and preprocess the GIS UHF signal to obtain an UHF partial discharge signal;

[0035] A K value optimization module is configured to obtain training data and optimize the number K of modal components of the VMD algorithm using a center frequency method to obtain an optimal K value;

[0036] The eigendecomposition module is configured to decompose the ultra-high frequency partial discharge signal using a VMD algorithm with an optimal K value for the number of modal components to obtain K intrinsic mode function components; perform different types of entropy calculations on each intrinsic mode function component to obtain values ​​of different types of eigenvalues ​​in each intrinsic mode function component; and construct a characteristic matrix based on the values ​​of different types of eigenvalues ​​in each intrinsic mode function component;

[0037] The key feature extraction module is configured to input the feature matrix into the dynamic network marker model based on the sample covariance matrix to extract the dominant feature variables and their corresponding set of index tags, use the set of index tags as the key network tags, calculate the correlation quantitative indicators between the dominant feature variables, and construct the extracted key features based on the key network tags and the correlation quantitative indicators.

[0038] In a third aspect, the present invention provides an electronic device comprising one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by one or more processors, the one or more processors implement the method described in any implementation manner in the first aspect.

[0039] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any implementation manner in the first aspect.

[0040] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, which implements the method described in any implementation manner in the first aspect when the computer program is executed by a processor.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] (1) The feature extraction method of GIS UHF partial discharge signal proposed in this invention is to effectively mine the internal features of different types of GIS partial discharge UHF signals. First, the center frequency method is used to optimize the modal decomposition number K of the VMD algorithm, and the UHF partial discharge signal is decomposed into intrinsic mode function components with different frequency scale characteristics using the VMD algorithm. Then, multiple entropy values ​​are calculated for each intrinsic mode function component obtained by decomposition, and the characteristic attributes of different dimensions of the same signal are further mined, thereby obtaining the corresponding feature matrix and performing normalization processing. In addition, the nonlinear coupling relationship between different entropy features is considered, and the feature matrix is ​​input into the dynamic network marker model based on the sample covariance matrix to extract the index mark set of the dominant feature variable and its corresponding correlation quantitative index, thereby enhancing the discrimination ability and robustness of the GIS UHF partial discharge signal feature extraction.

[0043] (2) The feature extraction method of GIS ultra-high frequency partial discharge signal proposed in the present invention constructs key features based on the index tag set of the extracted dominant characteristic variables and their corresponding correlation quantitative indicators, reveals the dynamic evolution of different partial discharge types of GIS, takes into account the nonlinear coupling relationship between characteristic variables, does not cause the dimensionality curse problem, and enhances the distinguishability of different partial discharge types of GIS.

[0044] (3) The key features extracted by the feature extraction method of GIS ultra-high frequency partial discharge signal proposed in the present invention can be further input into the classification device corresponding to different partial discharge types for identification to determine the corresponding partial discharge type. The classification using the key features can significantly improve the recognition accuracy of the partial discharge type. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0046] Figure 1 Schematic diagram of the flow of a feature extraction method for GIS ultra-high frequency partial discharge signals according to an embodiment of the present application;

[0047] Figure 2 A flowchart of a method for extracting features of a GIS ultra-high frequency partial discharge signal according to an embodiment of the present application;

[0048] Figure 3 The time domain diagram and spectrum diagram of the tip discharge UHF signal and its VMD decomposition result in the feature extraction method of the GIS UHF partial discharge signal in the embodiment of the present application;

[0049] Figure 4 The time domain diagram and spectrum diagram of the air gap discharge ultra-high frequency signal and its VMD decomposition result in the feature extraction method of the GIS ultra-high frequency partial discharge signal in the embodiment of the present application;

[0050] Figure 5 The time domain diagram and spectrum diagram of the suspended discharge UHF signal and its VMD decomposition result in the feature extraction method of the GIS UHF partial discharge signal in the embodiment of the present application;

[0051] Figure 6 This is a schematic diagram of key network markings for three types of partial discharges in the feature extraction method for GIS ultra-high frequency partial discharge signals according to an embodiment of the present application;

[0052] Figure 7 This is a quantitative index diagram of correlation between three types of partial discharges in the feature extraction method of GIS ultra-high frequency partial discharge signals according to an embodiment of the present application;

[0053] Figure 8 Schematic diagram of a feature extraction device for GIS ultra-high frequency partial discharge signals according to an embodiment of the present application;

[0054] Figure 9A schematic diagram of the hardware structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0055] To make the objectives, technical solutions, and advantages of the present invention more apparent, the present invention will be further described in detail below with reference to the accompanying drawings. It is apparent that the embodiments described are only some, not all, of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort are intended to fall within the scope of protection of the present invention.

[0056] Figure 1 A method for extracting features of a GIS ultra-high frequency partial discharge signal provided by an embodiment of the present application is shown, comprising the following steps:

[0057] S1, obtains the GIS ultra-high frequency signal and pre-processes it to obtain an ultra-high frequency partial discharge signal.

[0058] Specifically, refer to Figure 2 The pre-processing process includes interception and filtering. In one embodiment, the local discharge related signal in the ultra-high frequency signal is intercepted and filtered using the five-layer db4 wavelet decomposition to obtain the ultra-high frequency local discharge signal.

[0059] S2, obtain training data and use the center frequency method to optimize the number of modal components K of the VMD algorithm to obtain the optimal K value.

[0060] In a specific embodiment, obtaining training data and optimizing the number K of modal components of the VMD algorithm using the center frequency method to obtain the optimal K value specifically includes:

[0061] Collect the pre-processed ultra-high frequency partial discharge signals corresponding to different partial discharge types of the GIS device and form training data;

[0062] For each partial discharge type, the central frequency values ​​of each intrinsic mode function component obtained by the VMD algorithm under different modal component numbers K are calculated for the ultra-high frequency partial discharge signal;

[0063] For each partial discharge type, the relative rate of change of the center frequency value of the last intrinsic mode function component under adjacent K values ​​is calculated as shown in the following formula:

[0064] ;

[0065] in, It represents the center frequency value of the K+1th intrinsic mode function component when the number of modal components is K+1. It represents the center frequency value of the Kth intrinsic mode function component when the number of modal components is K. Indicates the relative rate of change when the number of modal components is K;

[0066] The number K of modal components corresponding to each partial discharge type when the relative change rate is less than or equal to the threshold is counted, and the number K of modal components corresponding to the most frequent occurrences of all partial discharge types when the relative change rate is less than or equal to the threshold is taken as the optimal K value.

[0067] Specifically, for the extraction of key features of GIS ultra-high frequency partial discharge signals, the first thing to consider is the selection of the number of modal components K. A K value that is too large will lead to modal repetition, while a value that is too small will lead to under-decomposition of the modal. The embodiment of the present application uses the center frequency method to determine the K value, and tries it in order from small to large. In one embodiment, the threshold value can be 10%. When the relative change rate of the center frequency value of the last intrinsic mode function (IMF) component under adjacent K values ​​(i.e., the judgment accuracy) is ) is less than or equal to 10%, it can be considered that the change of the center frequency value is not significant, and the K value at this time is the optimal K value. As a judgment of accuracy.

[0068] In the embodiment of the present application, a UHF partial discharge signal acquisition test platform for a typical defective electrode model of a GIS device is built to collect UHF partial discharge signals. From the large number of collected UHF partial discharge signals, 150 UHF partial discharge signals of each of three different discharge types are randomly selected as verification samples for the feature extraction method of the embodiment of the present application. The programming environment is Matlab2022b.

[0069] Considering that the GIS UHF partial discharge signal is a non-stationary and complex time series, the VMD algorithm is used to decompose the original UHF signal into K different intrinsic mode function components. The determination of the K value has a great influence on the decomposition result. In one embodiment of the present application, the center frequency method is used to determine the optimal K value. In one embodiment, the penalty factor of the VMD algorithm is set to , convergence criterion tolerance The relationship between K value and center frequency value obtained through multiple groups of experiments is shown in Table 1 to Table 3.

[0070] Table 1 Center frequency values ​​of IMF components under different K values ​​of tip discharge:

[0071]

[0072] Table 2 Center frequency values ​​of IMF components under different K values ​​of air gap discharge:

[0073]

[0074] Table 3 Center frequency values ​​of IMF components under different K values ​​of suspended discharge:

[0075]

[0076] When judging accuracy When K=6, K=7, and K=6, the change of the center frequency value is not significant, and the K value at this time is the optimal K value. , after which the center frequency value has no significant change, so the optimal K value is selected as 6.

[0077] The optimal K value is determined based on the training data. The VMD algorithm with the optimal K value can also be used to perform feature decomposition on the subsequent processing of ultra-high frequency partial discharge signals of unknown partial discharge types.

[0078] S3, using the VMD algorithm with the optimal K value of modal components to decompose the ultra-high frequency partial discharge signal, and obtain K intrinsic mode function components; performing different types of entropy calculations on each intrinsic mode function component, respectively, to obtain different types of characteristic variables in each intrinsic mode function component; constructing a characteristic matrix according to the values ​​of different types of characteristic variables in each intrinsic mode function component.

[0079] In a specific embodiment, different types of entropy calculations include calculations of power spectrum entropy, sample entropy, fuzzy entropy, permutation entropy, and distribution entropy, and the result of each type of entropy calculation corresponds to a feature variable.

[0080] Specifically, the VMD algorithm with 6 modal components is used to decompose the ultra-high frequency partial discharge signals of three typical partial discharge types, and the ultra-high frequency partial discharge signals of three typical partial discharge types and the VMD decomposition results are given as follows: Figure 3-Figure 5 As shown in the figure, the decomposition results of the UHF PD signal show that the IMF components obtained after the VMD algorithm decomposes the UHF PD signal have certain frequency and amplitude characteristics, effectively solving the modal aliasing problem existing in the EMD algorithm. Furthermore, the detailed features of the components obtained from the decomposition of the UHF PD signal of different PD types are also significantly different. This shows that after the VMD algorithm decomposes the UHF PD signal, its internal features are effectively mined, enhancing the distinguishability of the UHF PD signal. For UHF PD signals of unknown PD type, the corresponding PD type can also be determined by the frequency and amplitude characteristics of the IMF components obtained after the VMD algorithm decomposes the UHF PD signal.

[0081] In order to quantitatively analyze the IMF components after VMD decomposition, the power spectrum entropy (SE), sample entropy (SampEn), fuzzy entropy (FE), permutation entropy (PE) and distribution entropy (DistEn) are calculated for the IMF components after VMD decomposition to obtain the corresponding entropy features. These entropy features are mapped to nodes of the dynamic network, and a feature matrix containing multi-scale complexity information is constructed. Each column of entropy features in the feature matrix constitutes a feature variable.

[0082] The feature matrix is ​​shown as follows:

[0083] ;

[0084] in, and Respectively represent characteristic variables and The value of the characteristic variable at the kth eigenmode function component, k=1,2,…,K, , n is the number of characteristic variables;

[0085] Specifically, three types of ultra-high frequency partial discharge signals are randomly extracted and their corresponding characteristic matrices are calculated as shown in Table 4-6.

[0086] Table 4 Tip discharge characteristic matrix:

[0087]

[0088] Table 5 Air gap discharge characteristic matrix:

[0089]

[0090] Table 6 Suspension discharge characteristic matrix:

[0091]

[0092] S4, input the feature matrix into the dynamic network marker model based on the sample covariance matrix to extract the dominant feature variables and their corresponding index mark set, use the set of index marks as the key network markers, calculate the correlation quantitative index between the dominant feature variables, and construct the extracted key features based on the key network markers and the correlation quantitative index.

[0093] In a specific embodiment, the feature matrix is ​​input into a dynamic network marker model based on a sample covariance matrix to extract the dominant feature variables and their corresponding index tag sets, specifically including:

[0094] Performing standardization on the feature matrix to obtain a standardized feature matrix;

[0095] A sample covariance matrix of the standardized feature matrix is ​​constructed, and the sample covariance matrix is ​​subjected to eigendecomposition. The dominant feature variables are screened out according to the eigenvectors and element rankings corresponding to the maximum eigenvalues ​​of the sample covariance matrix, and the set of index marks corresponding to the dominant feature variables is extracted.

[0096] In a specific embodiment, the characteristic decomposition process of the sample covariance matrix is ​​shown in the following formula:

[0097] ;

[0098] Where C represents the sample covariance matrix, The first eigenvalues is a diagonal matrix with diagonal elements, and the eigenvalues ​​of the sample covariance matrix C To sort from large to small, the first eigenvalue of the sample covariance matrix C is obtained is the maximum eigenvalue of the sample covariance matrix C, is the unit eigenvector is an orthogonal matrix of columns, , n is the number of characteristic variables;

[0099] The element in the i-th row and j-th column of the sample covariance matrix C Expressed as:

[0100] ;

[0101] in, Represents the first Characteristic vector of the column Hedi Characteristic vector of the column The covariance between , represents the maximum eigenvalue of the sample covariance matrix, and are the maximum eigenvalues ​​of the sample covariance matrix The corresponding unit eigenvector No. elements and elements, corresponding to the first element in the feature matrix after normalization. Characteristic vector of the column Hedi Characteristic vector of the column ; is the first order of the sample covariance matrix eigenvalues, and are the first eigenvalues The corresponding unit eigenvector No. elements and elements;

[0102] Compare the eigenvectors corresponding to the largest eigenvalues The n elements in the matrix are selected as the leading feature variables, and the index mark set of the leading feature variables in the standardized feature matrix is ​​used as the key network mark. .

[0103] In a specific embodiment, for the dominant characteristic variable, a correlation quantitative index is used. It is used to quantify the correlation between the dominant feature variables, as shown in the following formula:

[0104] ;

[0105] Among them, m is the key network marker The number of elements in , the value of m is set to satisfy and , P value is a positive real number in [0,1], n is the number of characteristic variables, Represents the first Characteristic vector of the column Hedi Characteristic vector of the column The covariance between Indicates the absolute value, represents the L2 norm;

[0106] The key features are vectors composed of key network markers and quantitative correlation indicators.

[0107] Specifically, each feature of the feature matrix is ​​Z-Score standardized so that the mean of each feature is 0 and the standard deviation is 1, eliminating the influence of the dimensional differences of each feature on the construction of the covariance matrix. Then, it is input into the dynamic network markers based on sample covariance matrices (SDNM) model, with a P value of 0.7. The set of feature variables corresponding to the first m elements of the eigenvector corresponding to the maximum eigenvalue in the standardized feature matrix is ​​selected as the dynamic network marker (DNM), and the key network markers are obtained. and correlation indicators , the following are the results of 20 randomly selected samples from each category and like Figure 6 and Figure 7 As shown. It can be seen that the key network markers of the tip discharge , correlation index Mainly concentrated between 0.91 and 0.94. Key network markers for air gap discharge , correlation index Mainly concentrated between 0.85 and 0.9. Key network markers for suspended discharge , correlation index It is mainly concentrated between 0.94 and 0.97. Therefore, the key network markers and correlation indicators It has good cohesion and separability, so the embodiment of the present application can select key network tags and correlation indicators It has the key characteristics of the UHF partial discharge signal. Specifically, the key network can be marked and correlation indicators The concatenated vector is used as the key feature.

[0108] To verify the advantages of the methods proposed in the examples of this application for feature extraction of UHF partial discharge signals from GIS, the key features extracted in the examples of this application were input into a support vector machine (SVM) for classification, along with features extracted using EMD and EEMD decomposition and the SDNM model, features extracted using VMD decomposition and separate SE or PE features, and features extracted using VMD decomposition and calculation of multiple entropy (ME). For all models, the first 80% of the training data was used as the training set, and the last 20% as the test set. The recognition results for each model are shown in Table 7.

[0109] Table 7 Accuracy of partial discharge type identification using different feature extraction methods:

[0110]

[0111] The following conclusions can be drawn from the test results:

[0112] 1) The recognition results of the three partial discharge type recognition methods (VMD-SDNM-SVM, EMD-SDNM-SVM, and EEMD-SDNM-SVM) show that the recognition accuracy of the UHF partial discharge signal decomposition using the VMD algorithm is improved by 16.4% and 10.95% respectively compared with the two signal modal decomposition algorithms (EMD and EEMD). This shows that the decomposition of the VMD algorithm can better realize the mining of the internal features of the UHF partial discharge signal, and the obtained components can better represent the original signal.

[0113] 2) By comparing the recognition results of the three partial discharge type recognition methods, VMD-SDNM-SVM, VMD-PE-SVM and VMD-SE-SVM, it can be seen that the SDNM model combines the complementary information of various entropy indicators and considers the nonlinear coupling relationship between various entropies. Compared with only extracting the independent entropy value features of each modal component, it has stronger robustness and distinguishability.

[0114] 3) Comparing the recognition results of the VMD-SDNM-SVM and VMD-ME-SVM diagnostic models, it can be seen that the SDNM model not only fully integrates the complementary information of various entropy indicators and considers the nonlinear coupling relationship between various entropies, but also avoids the dimensionality curse problem, thereby enhancing the distinguishability of different GIS partial discharge types, thereby effectively classifying GIS partial discharge defects.

[0115] Further references Figure 8 As an implementation of the methods shown in the above figures, the present application provides an embodiment of a feature extraction device for GIS ultra-high frequency partial discharge signals. Figure 1 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0116] The present application provides a device for extracting features of a GIS ultra-high frequency partial discharge signal, including:

[0117] The preprocessing module 1 is configured to obtain the GIS UHF signal and perform preprocessing to obtain the UHF partial discharge signal;

[0118] The K value optimization module 2 is configured to obtain training data and optimize the number K of modal components of the VMD algorithm using a center frequency method to obtain an optimal K value;

[0119] The eigendecomposition module 3 is configured to decompose the ultra-high frequency partial discharge signal using a VMD algorithm with an optimal K value for the number of modal components to obtain K intrinsic mode function components; perform different types of entropy calculations on each intrinsic mode function component to obtain values ​​of different types of eigenvalues ​​in each intrinsic mode function component; and construct a characteristic matrix based on the values ​​of different types of eigenvalues ​​in each intrinsic mode function component;

[0120] The key feature extraction module 4 is configured to input the feature matrix into the dynamic network marker model based on the sample covariance matrix to extract the dominant feature variables and their corresponding index mark sets, use the set of index marks as the key network marks, calculate the correlation quantitative indicators between the dominant feature variables, and construct the extracted key features based on the key network marks and the correlation quantitative indicators.

[0121] Furthermore, a GIS partial discharge type recognition model is constructed and trained to obtain a trained partial discharge type recognition model, which includes the above-mentioned feature extraction device and classification device; the key features are input into the trained partial discharge type recognition model, the corresponding key features are first extracted by the feature extraction device, and then the key features are input into the classification device, so that the corresponding partial discharge type can be identified.

[0122] Specifically, training data can be constructed using key features of different partial discharge types. A classifier can be trained using this training data for each partial discharge type, resulting in a trained classifier. This classifier can use a classifier such as an SVM. The key features corresponding to the UHF partial discharge signal to be predicted are input into the trained classifier corresponding to each partial discharge type to identify the corresponding partial discharge type.

[0123] Figure 9 Schematic diagram of the hardware structure of the electronic device provided by the embodiment of the present invention. Figure 9 As shown, the electronic device of this embodiment includes: a processor 901 and a memory 902; wherein the memory 902 is configured to store computer-executable instructions; and the processor 901 is configured to execute the computer-executable instructions stored in the memory to implement the various steps performed by the electronic device in the above-described embodiment. For details, please refer to the relevant description of the aforementioned method embodiment.

[0124] Optionally, the memory 902 may be independent or integrated with the processor 901 .

[0125] When the memory 902 is independently provided, the electronic device further includes a bus 903 for connecting the memory 902 and the processor 901 .

[0126] An embodiment of the present invention further provides a computer storage medium, in which computer-executable instructions are stored. When the processor 901 executes the computer-executable instructions, the above method is implemented.

[0127] An embodiment of the present invention further provides a computer program product, including a computer program. When the computer program is executed by the processor 901, the above method is implemented.

[0128] In the embodiments provided herein, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the module division is merely a logical functional division. In actual implementation, other division methods may be used. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not implemented. In addition, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interface, device or module, which may be electrical, mechanical or other forms.

[0129] Modules described as separate components may or may not be physically separate, and components shown as modules may or may not be physical units, that is, they may be located in one place or distributed across multiple network elements. Some or all of these modules may be selected to implement the solution of this embodiment based on actual needs.

[0130] In addition, the functional modules in various embodiments of the present invention may be integrated into a single processing unit, each module may exist physically separately, or two or more modules may be integrated into a single unit. The units formed by the above modules may be implemented in the form of hardware or hardware plus software functional units.

[0131] The above-mentioned integrated module implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) or processor 901 to perform some steps of the methods of various embodiments of the present application.

[0132] It should be understood that the processor 901 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), or application-specific integrated circuits (ASIC). A general-purpose processor may be a microprocessor, or the processor 901 may be any conventional processor 901. The steps of the method disclosed in the present invention may be directly implemented as being executed by the hardware processor 901, or may be implemented by a combination of hardware and software modules in the processor 901.

[0133] The memory 902 may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk memory, and may also be a USB flash drive, a mobile hard disk, a read-only memory, a magnetic disk, or an optical disk.

[0134] Bus 903 can be an Industry Standard Architecture (ISA), a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. Bus 903 can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, the bus 903 in the drawings of this application is not limited to only one bus 903 or only one type of bus 903.

[0135] The storage medium may be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The storage medium may be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0136] An exemplary storage medium is coupled to the processor 901, so that the processor 901 can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be an integral part of the processor 901. The processor 901 and the storage medium can be located in an application-specific integrated circuit (ASIC). Of course, the processor 901 and the storage medium can also exist as discrete components in an electronic device or a main control device.

[0137] Those skilled in the art will appreciate that all or part of the steps in the above-described method embodiments can be implemented using hardware associated with program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0138] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A feature extraction method for GIS ultra-high frequency partial discharge signals, characterized in that: The following steps are involved: Acquire GIS UHF signals and pre-process them to obtain UHF partial discharge signals; Obtain training data and use the center frequency method to optimize the number of modal components K of the VMD algorithm to obtain the optimal K value, specifically including: Collect ultra-high frequency partial discharge signals corresponding to different partial discharge types of GIS devices and form training data; For each partial discharge type, the central frequency values ​​of each intrinsic mode function component obtained by the VMD algorithm under different modal component numbers K are calculated for the ultra-high frequency partial discharge signal; For each partial discharge type, the relative rate of change of the center frequency value of the last intrinsic mode function component under adjacent K values ​​is calculated as shown in the following formula: Among them, ω K+1 Indicates the center frequency value of the K+1th intrinsic mode function component when the number of modal components is K+1, ω K Indicates the center frequency value of the Kth eigenmode function component when the number of modal components is K, ε K Indicates the relative rate of change when the number of modal components is K; Counting the number K of modal components corresponding to each partial discharge type when the relative change rate is less than or equal to a threshold, and taking the number K of modal components corresponding to the relative change rate less than or equal to the threshold that appears the most times among all partial discharge types as the optimal K value; Decomposing the ultra-high frequency partial discharge signal using a VMD algorithm with the optimal K value for the number of modal components to obtain K intrinsic mode function components; performing different types of entropy calculations on each intrinsic mode function component to obtain values ​​of different types of characteristic variables in each intrinsic mode function component; and constructing a characteristic matrix based on the values ​​of the different types of characteristic variables in each intrinsic mode function component; The feature matrix is ​​input into a dynamic network marker model based on a sample covariance matrix to extract a set of dominant feature variables and their corresponding index tags, specifically including: performing a normalization process on the feature matrix to obtain a normalized feature matrix; A sample covariance matrix of the standardized feature matrix is ​​constructed, and the sample covariance matrix is ​​subjected to eigendecomposition. The dominant feature variables are screened out according to the eigenvector corresponding to the maximum eigenvalue of the sample covariance matrix and the ranking of its elements, and a set of index marks corresponding to the dominant feature variables is extracted; the set of index marks is used as the key network mark, and the correlation quantitative index between the dominant feature variables is calculated, and the extracted key features are constructed based on the key network mark and the correlation quantitative index.

2. The feature extraction method of GIS ultra-high frequency partial discharge signal according to claim 1 is characterized in that: The different types of entropy calculations include calculation of power spectrum entropy, calculation of sample entropy, calculation of fuzzy entropy, calculation of permutation entropy and calculation of distribution entropy, and the result of each type of entropy calculation corresponds to a characteristic variable.

3. The feature extraction method of GIS ultra-high frequency partial discharge signal according to claim 1 is characterized in that: The characteristic decomposition process of the sample covariance matrix is ​​shown in the following formula: C=VΛV T ; Where C represents the sample covariance matrix, Λ∈R n×n The lth eigenvalue λ of the sample covariance matrix C is Cl is a diagonal matrix with diagonal elements, and the eigenvalue λ of the sample covariance matrix C Cl To sort from large to small, the first eigenvalue λ of the sample covariance matrix C is obtained C1 is the maximum eigenvalue of the sample covariance matrix C, V∈R n×n The unit eigenvector v l is an orthogonal matrix of columns, l=1,2,...,n, n is the number of characteristic variables; The element C in the i-th row and j-th column of the sample covariance matrix C ij Expressed as: Among them, C ij Represents the eigenvector X of the i-th column in the normalized feature matrix i and the eigenvector X of the jth column j The covariance between i,j=1,2,...,n,λ C1 represents the maximum eigenvalue of the sample covariance matrix, v i1 and v j1 are the maximum eigenvalue λ of the sample covariance matrix respectively C1 The i-th element and the j-th element of the corresponding unit eigenvector v1 correspond to the eigenvector X of the i-th column in the normalized feature matrix. i and the eigenvector X of the jth column j ;λ Cl is the lth eigenvalue of the sample covariance matrix, v il and v jl are the lth eigenvalue λ of the sample covariance matrix respectively Cl The corresponding unit eigenvector v l The i-th element and the j-th element of ; Compare the n elements in the eigenvector v1 corresponding to the maximum eigenvalue, select the eigenvalues ​​of the corresponding columns in the standardized feature matrix corresponding to the top m elements as the dominant eigenvalues, and the set of index marks of the dominant eigenvalues ​​in the standardized feature matrix as the key network mark J D .

4. The feature extraction method of GIS ultra-high frequency partial discharge signal according to claim 3 is characterized in that: For the dominant characteristic variables, the correlation quantitative index C is used D It is used to quantify the correlation between the dominant feature variables, as shown in the following formula: Among them, m is the key network label J D The number of elements in , the value of m is set to satisfy And 2≤m≤n, P value is a positive real number in [0,1], n is the number of characteristic variables, C pq represents the eigenvector X of the pth column in the normalized feature matrix p and the eigenvector X of the qth column q The covariance between , | | represents the absolute value, || || represents the L2 norm; The key feature is a vector formed by concatenating the key network tag and the correlation quantitative index.

5. A feature extraction device for GIS ultra-high frequency partial discharge signals, characterized in that: include: a preprocessing module configured to acquire and preprocess the GIS UHF signal to obtain an UHF partial discharge signal; The K value optimization module is configured to obtain training data and optimize the number of modal components K of the VMD algorithm using the center frequency method to obtain the optimal K value. Specifically, it includes: Collect ultra-high frequency partial discharge signals corresponding to different partial discharge types of GIS devices and form training data; For each partial discharge type, the central frequency values ​​of each intrinsic mode function component obtained by the VMD algorithm under different modal component numbers K are calculated for the ultra-high frequency partial discharge signal; For each partial discharge type, the relative rate of change of the center frequency value of the last intrinsic mode function component under adjacent K values ​​is calculated as shown in the following formula: Among them, ω K+1 Indicates the center frequency value of the K+1th intrinsic mode function component when the number of modal components is K+1, ω K Indicates the center frequency value of the Kth eigenmode function component when the number of modal components is K, ε K Indicates the relative rate of change when the number of modal components is K; Counting the number K of modal components corresponding to each partial discharge type when the relative change rate is less than or equal to a threshold, and taking the number K of modal components corresponding to the relative change rate less than or equal to the threshold that appears the most times among all partial discharge types as the optimal K value; The eigendecomposition module is configured to decompose the ultra-high frequency partial discharge signal using a VMD algorithm with the modal component number being the optimal K value to obtain K intrinsic mode function components; perform different types of entropy calculations on each intrinsic mode function component to obtain values ​​of different types of eigenvalues ​​in each intrinsic mode function component; and construct a characteristic matrix based on the values ​​of the different types of eigenvalues ​​in each intrinsic mode function component; The key feature extraction module is configured to input the feature matrix into the dynamic network marker model based on the sample covariance matrix to extract the dominant feature variables and their corresponding index tags, specifically including: performing a normalization process on the feature matrix to obtain a normalized feature matrix; A sample covariance matrix of the standardized feature matrix is ​​constructed, and the sample covariance matrix is ​​subjected to eigendecomposition. The dominant feature variables are screened out according to the eigenvector corresponding to the maximum eigenvalue of the sample covariance matrix and the ranking of its elements, and a set of index marks corresponding to the dominant feature variables is extracted; the set of index marks is used as the key network mark, and the correlation quantitative index between the dominant feature variables is calculated, and the extracted key features are constructed based on the key network mark and the correlation quantitative index.

6. An electronic device comprising: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.

8. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.

Images