Harmonic current signal-based high-voltage cable defect identification method and system

Abstract

The invention discloses a harmonic current signal-based high-voltage cable defect identification method and system. The method comprises the steps of obtaining harmonic current monitoring data of a high-voltage cable metal sheath; analyzing the harmonic current monitoring data by using a principal component analysis method to obtain a principal component characteristic quantity; determining a defect identification initial model based on a self-organizing mapping algorithm, and training and optimizing the defect identification initial model based on the principal component characteristic quantity to obtain a defect identification optimal model; and inputting harmonic data of a to-be-detected high-voltage cable into the defect identification optimal model so as to identify the defect type of the to-be-detected high-voltage cable. Amplitude-frequency characteristics of harmonic current signals are extracted through principal component analysis, abnormal harmonic data generated by different defects are classified and distinguished through a self-organizing mapping algorithm, and therefore defect identification based on high-voltage cable line harmonic signal abnormity is achieved. And reliable technical support can be provided for cable defect identification in a power system in practical application.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of defect identification and state evaluation, and more particularly, to a high-voltage cable defect identification method and system based on harmonic current signals. BACKGROUND

[0002] In recent years, in view of the increasingly difficult problem of offline detection technology application, the research on cable live detection technology at home and abroad has gradually become popular. Although traditional high-voltage cables have relatively mature live detection technologies such as partial discharge and grounding current, these technologies are susceptible to electromagnetic interference and have high technical complexity, and the detection precision and sensitivity need to be further improved, which cannot meet the needs of high-voltage cable line operation state monitoring under the operating conditions of new power systems. Therefore, it is necessary to carry out research on new technologies suitable for new live detection at the present stage. The abnormal harmonic current signal of high-voltage cable has attracted widespread attention, and this technical route has potential advantages such as the ability to realize full-line insulation defect exploration and wide defect identification coverage.

[0003] Existing papers confirm that the abnormal characteristics of cable harmonic current signals are significantly related to insulation deterioration process and potential defects. However, due to the complexity of the content and distribution characteristics of harmonics, the typical characteristic quantities are not clear, and the harmonic content of the typical frequency has deviations in multiple measurements, which is difficult to repeat, resulting in that the existing data analysis method cannot quickly and accurately analyze the cable harmonic content detection results, which restricts the practical application of this technology. Therefore, a new data analysis method needs to be used to process and research the detected harmonic component data.

[0004] Therefore, a high-voltage cable defect identification method based on harmonic current signals is needed. SUMMARY

[0005] The present application proposes a high-voltage cable defect identification method and system based on harmonic current signals to solve the problem of how to efficiently identify high-voltage cable insulation defects.

[0006] In order to solve the above problems, according to one aspect of the present application, a high-voltage cable defect identification method based on harmonic current signals is provided, which comprises:

[0007] Harmonic current monitoring data of the high-voltage cable metal sheath at different measurement time points is obtained;

[0008] The harmonic current monitoring data is analyzed by using principal component analysis to obtain principal component characteristic quantities;

[0009] An initial defect identification model is determined based on a self-organizing mapping algorithm, and the initial defect identification model is trained and optimized based on the principal component characteristic quantities to obtain an optimal defect identification model;

[0010] inputting the harmonic data of the to-be-tested high-voltage cable into the defect identification optimal model to identify the defect type of the to-be-tested high-voltage cable.

[0011] Preferably, wherein the harmonic current monitoring data is analyzed by using principal component analysis to obtain principal component features, comprising:

[0012] determining a data matrix X based on the harmonic current monitoring data; wherein the rows represent different measurement time points, and the columns represent different harmonic components;

[0013] performing standardization processing on the data matrix X, and calculating a covariance matrix based on the data after the standardization processing;

[0014] performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues and eigenvectors, and obtaining principal components based on the eigenvalues and eigenvectors;

[0015] calculating the contribution rate and cumulative contribution rate of each principal component, and determining principal component features based on the contribution rate and cumulative contribution rate.

[0016] Preferably, wherein the principal components are obtained based on the eigenvalues and eigenvectors, comprising:

[0017] by calculating the eigenvalues of the covariance matrix R as λ1≥λ2≥…≥λ n ≥0 and the corresponding eigenvectors as u1,u2,…,u n , obtaining n new index variables composed of eigenvectors as:

[0018]

[0019] wherein u j =(u 1j ,u2 j ,…,u nj ), u nj represents the nth component of the jth eigenvector, y1 is the first principal component, y2 is the second principal component, …, y n is the nth principal component; and are the 2nd, 3rd and (n+1)th harmonic amplitudes measured respectively; n is the number of measured harmonics.

[0020] Preferably, wherein the defect identification initial model is trained and optimized based on the principal component features to obtain a defect identification optimal model, comprising:

[0021] performing format conversion on the harmonic signals corresponding to the principal component features to obtain harmonic conversion data;

[0022] annotating the harmonic conversion data to obtain annotated data;

[0023] inputting the annotated data as training data into the initial defect recognition model, calculating a winner neuron, iteratively updating a weight vector based on the winner neuron, and determining an optimal defect recognition model until a preset convergence condition is reached; wherein after the model converges, each input sample corresponds to a neuron position in a self-organizing mapping (SOM) network, and at this time the SOM network forms a low-dimensional representation reflecting the topology of the input data.

[0024] Preferably, wherein the annotated data is input as training data into the initial defect recognition model, a winner neuron is calculated, and a weight vector is iteratively updated based on the winner neuron, including:

[0025] randomly selecting a sample vector x from the data set i calculating a winner neuron, including: calculating the Euclidean distance between the input sample x and the weight vector w of each neuron in the SOM network i j finding the neuron with the smallest distance as the winner neuron, representing the optimal match of the neuron to the current input, denoted as c:

[0026] c = argmin j ‖x i -w j ‖,

[0027] The Euclidean distance formula is:

[0028]

[0029] updating the weight vector, including: updating the weight vector of the winner neuron and the neurons in its neighborhood, so that the winner neuron is closer to the current input sample x i ; wherein the update formula is:

[0030] w j (t+1) = w j (t) + η(t)·h cj (t)·(x i -w j (t)),

[0031] wherein η(t) is the learning rate, which decreases with the iteration number t, and h cj (t) is a neighborhood function, representing the neighborhood influence degree of neuron j from the winner neuron c; the learning rate η(t) is a key parameter for controlling the weight update speed, which gradually decreases during the training process to ensure that the model gradually converges to a stable state.

[0032] ​According to another aspect of the present application, there is provided a high-voltage cable defect identification system based on harmonic current signals, the system comprising:

[0033] a harmonic current monitoring data acquisition unit configured to acquire harmonic current monitoring data of a high-voltage cable metal sheath at different measurement time points;

[0034] a principal component analysis unit configured to analyze the harmonic current monitoring data by using a principal component analysis method to obtain principal component characteristic quantities;

[0035] a model determination unit configured to determine a defect identification initial model based on a self-organizing mapping algorithm, train and optimize the defect identification initial model based on the principal component characteristic quantities to obtain a defect identification optimal model;

[0036] a defect type identification unit configured to input harmonic data of a high-voltage cable to be tested into the defect identification optimal model to identify a defect type of the high-voltage cable to be tested.

[0037] Preferably, the principal component analysis unit is configured to analyze the harmonic current monitoring data by using a principal component analysis method to obtain principal component characteristic quantities, comprising:

[0038] determining a data matrix X based on the harmonic current monitoring data; wherein rows represent different measurement time points and columns represent different harmonic components;

[0039] performing standardization processing on the data matrix X and calculating a covariance matrix based on the standardization-processed data;

[0040] performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues and eigenvectors, and obtaining principal components based on the eigenvalues and eigenvectors;

[0041] calculating contribution rates and cumulative contribution rates of each principal component, and determining principal component characteristic quantities based on the contribution rates and cumulative contribution rates.

[0042] Preferably, the principal component analysis unit is configured to obtain principal components based on the eigenvalues and eigenvectors, comprising:

[0043] obtaining the eigenvalues λ1≥ λ2≥ … ≥ λn> 0 and corresponding eigenvectors u1, u2, …, unof the covariance matrix R by calculating eigenvalues of the covariance matrix R; n n obtaining n new index variables composed of eigenvectors as:

[0044]

[0045] wherein u1, u2, …, unare eigenvectors of the covariance matrix R; and j 1j u1, u2, …, unare eigenvectors of the covariance matrix R.​​j ,…,u nj ),u nj represents the nth component of the jth feature vector, y1 is the 1st principal component, y2 is the 2nd principal component, …, y n is the nth principal component; and are the measured 2nd, 3rd and n+1th harmonic amplitudes, respectively; n is the number of measured harmonics.

[0046] Preferably, wherein the model determination unit trains and optimizes the defect identification initial model based on the principal component feature quantity to obtain a defect identification optimal model, comprising:

[0047] Converting the harmonic signal corresponding to the principal component feature quantity to obtain harmonic conversion data;

[0048] Labeling the harmonic conversion data for defect type to obtain labeled data;

[0049] Input the labeled data as training data into the defect identification initial model, calculate the winner neuron, update the weight vector based on the winner neuron iteratively until the preset convergence condition is reached, and determine the defect identification optimal model; wherein after model convergence, each input sample corresponds to a neuron position in the self-organizing mapping (SOM) network, at which time the SOM network forms a low-dimensional representation reflecting the topological structure of the input data.

[0050] Preferably, wherein the model determination unit inputs the labeled data as training data into the defect identification initial model, calculates the winner neuron, and iteratively updates the weight vector based on the winner neuron, comprising:

[0051] Randomly select a sample vector x i from the data set, calculate the winner neuron, comprising: calculate the Euclidean distance between the input sample x i and each neuron weight vector w j in the SOM network, find the neuron with the smallest distance as the winner neuron, which represents the optimal match of the current input, denoted as c:

[0052] c=argmin j ‖x i -w j ‖,

[0053] The Euclidean distance formula is:

[0054]

[0055] Updating the weight vector includes: updating the weight vectors of the winning neuron and its neighboring neurons, making the winning neuron more closely resemble the current input sample x. i The update formula is as follows:

[0056] w j (t+1)=w j (t)+η(t)·h cj (t)·(x i -w j (t)),

[0057] Where η(t) is the learning rate, which decreases with the number of iterations t, h cj (t) is the neighborhood function, representing the degree of neighborhood influence of neuron j from the winner neuron c; the learning rate η(t) is the key parameter that controls the speed of weight updates, and it is gradually reduced during training to ensure that the model gradually converges to a stable state.

[0058] This invention provides a method and system for identifying high-voltage cable defects based on harmonic current signals, comprising: acquiring harmonic current monitoring data of the metallic sheath of a high-voltage cable at different measurement time points; analyzing the harmonic current monitoring data using principal component analysis to obtain principal component feature quantities; determining an initial defect identification model based on a self-organizing map algorithm; training and optimizing the initial defect identification model based on the principal component feature quantities to obtain an optimal defect identification model; and inputting the harmonic data of the high-voltage cable under test into the optimal defect identification model to identify the defect type of the high-voltage cable under test. This invention, by first extracting the amplitude-frequency characteristics of the abnormal harmonic current signal of the metallic sheath of the high-voltage cable line using principal component analysis, and then classifying and identifying the abnormal harmonic data generated by different defects using a self-organizing map algorithm, achieves defect identification based on abnormal harmonic signals of high-voltage cable lines, providing reliable technical support for cable defect identification in power systems in practical applications. Attached Figure Description

[0059] Exemplary embodiments of the present invention can be more fully understood by referring to the following figures:

[0060] Figure 1 A flowchart of a high-voltage cable defect identification method 100 based on harmonic current signals according to an embodiment of the present invention;

[0061] Figure 2 This is a schematic diagram of high-voltage cable harmonic signal measurement according to an embodiment of the present invention;

[0062] Figure 3 This is a flowchart of principal component analysis of harmonic signals according to an embodiment of the present invention;

[0063] Figure 4 FIG. 6 is a schematic diagram of a principal component analysis result of the 2nd to 11th harmonics according to an embodiment of the present application;

[0064] Figure 5 FIG. 7 is a schematic diagram of a self-organizing map algorithm identifying a defect type according to an embodiment of the present application;

[0065] Figure 6 FIG. 8 is a schematic diagram of a semiconductive floating defect identification result according to an embodiment of the present application;

[0066] Figure 7 FIG. 9 is a schematic diagram of a metallic floating defect identification result according to an embodiment of the present application;

[0067] Figure 8 FIG. 10 is a schematic diagram of an insulation tape residue defect identification result according to an embodiment of the present application;

[0068] Figure 9 FIG. 11 is a schematic diagram of a semiconductive tip defect identification result according to an embodiment of the present application;

[0069] Figure 10 FIG. 12 is a schematic diagram of a high voltage cable defect identification system 1000 based on a harmonic current signal according to an embodiment of the present application. DETAILED DESCRIPTION

[0070] The exemplary embodiments of the present application will now be described with reference to the accompanying drawings. The present application can, however, be embodied in many different forms and should not be construed as being limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the application to those skilled in the art. Like reference numerals refer to like elements throughout the specification. It will be understood that when a certain term is used in the singular throughout the specification and in the claims, such term also encompasses the plural form unless opposite meaning is clear from the context.

[0071] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0072] Figure 1 FIG. 14 is a flowchart of a method 100 of calibrating a high voltage cable defect identification method based on a harmonic current signal according to an embodiment of the present application. As shown in FIG. 14, the method 100 includes the steps of: measuring a harmonic current signal of a high voltage cable (S110); and identifying a defect type of the high voltage cable based on the harmonic current signal (S120). Figure 1As shown, the high-voltage cable defect identification method based on harmonic current signals provided by the embodiment of the application realizes the defect identification based on the abnormal harmonic signals of the high-voltage cable line, and can provide reliable technical support for the cable defect identification in the power system in actual application. The high-voltage cable defect identification method based on harmonic current signals 100 provided by the embodiment of the application starts from step 101, in step 101, the harmonic current monitoring data of the high-voltage cable metal sheath at different measurement time points is acquired.

[0073] In step 102, the principal component analysis method is used to analyze the harmonic current monitoring data to obtain the principal component characteristic quantity.

[0074] Preferably, wherein the principal component analysis method is used to analyze the harmonic current monitoring data to obtain the principal component characteristic quantity, comprising:

[0075] Based on the harmonic current monitoring data, a data matrix X is determined; wherein the rows represent different measurement time points, and the columns represent different harmonic components;

[0076] The data matrix X is standardized, and a covariance matrix is calculated based on the standardized data;

[0077] The eigenvalue decomposition of the covariance matrix is performed to obtain the eigenvalue and the eigenvector, and the principal component is obtained based on the eigenvalue and the eigenvector;

[0078] The contribution rate and the cumulative contribution rate of each principal component are calculated, and the principal component characteristic quantity is determined based on the contribution rate and the cumulative contribution rate.

[0079] Preferably, wherein the principal component is obtained based on the eigenvalue and the eigenvector, comprising:

[0080] The eigenvalue of the covariance matrix R is calculated as λ1≥λ2≥…≥λ n ≥0 and the corresponding eigenvector is u1,u2,…,u n , the n new index variables composed of the eigenvector are obtained as:

[0081]

[0082] u j =(u 1j ,u2 j ,…,u nj ), u njy1, y2, …, yn represent the n-th component of the j-th eigenvector, y1 is the 1st principal component, y2 is the 2nd principal component, …, and yn is the n-th principal component. n y1, y2, …, yn represent the n-th component of the j-th eigenvector, y1 is the 1st principal component, y2 is the 2nd principal component, …, and yn is the n-th principal component. and are the measured 2nd, 3rd, and (n+1)th harmonic amplitudes, respectively; and n is the number of measured harmonics.

[0083] As shown in FIG. 1, in the present application, the harmonic current monitoring data of the high-voltage metal sheath is obtained through non-contact measurement, aiming at the typical structural characteristics of the double-ended grounding of the high-voltage cable system. Then, the principal component analysis method is used to linearly combine a plurality of harmonic components with certain correlation, to explain as many information in the original data as possible in the least dimension, to reduce the dimension of the harmonic components, and the variables after the dimension reduction are linearly independent of each other, and the finally determined new variable is a linear combination of the original harmonic components. Figure 2 As shown in FIG. 1, the process of principal component analysis includes:

[0084] Figure 3

[0085] (1) Harmonic data arrangement

[0086] First, the cable induced current collection data (containing a plurality of harmonic components) at a plurality of time points are arranged to form a data matrix X, wherein the rows represent different measurement time points, and the columns represent different harmonic components.

[0087] (2) Data standardization

[0088] Since the current amplitudes of different harmonic components can have a large difference in magnitude, the data needs to be standardized first, so that the mean of each feature is 0 and the standard deviation is 1. The calculation formula is as follows:

[0089]

[0090] Where μ is the mean of each column, and σ is the standard deviation of each column.

[0091] (3) Calculate the covariance matrix R

[0092] Next, the covariance matrix of the standardized data is calculated. The covariance matrix R describes the linear relationship between the variables in the data:

[0093]

[0094] Where n is the number of samples.

[0095] (4) Eigenvalue decomposition

[0096] The eigenvalue decomposition of the covariance matrix is performed to obtain the eigenvalues and eigenvectors:

[0097] ​​R = VDV T (3)

[0098] Where V is a matrix containing the eigenvectors of the covariance matrix, these eigenvectors are also called principal components; D is a diagonal matrix, the elements on its diagonal are the corresponding eigenvalues, representing the variance contribution of each principal component.

[0099] The eigenvalues ​​of the covariance matrix R are calculated to be λ1≥λ2≥…≥λ. n ≥0 and the corresponding eigenvectors are u1, u2, ..., u n The resulting n new index variables composed of feature vectors are:

[0100]

[0101] In the formula, u j =(u 1j ,u 2j ,…,u nj ), u nj Let y1 represent the nth component of the j-th eigenvector, where y1 is the first principal component, y2 is the second principal component, ..., y... n It is the nth principal component.

[0102] (5) Analyze the contribution rate of principal components

[0103] Calculate each principal component y j Contribution rate b j (j=1,2,...,n) and y1,y2,…,y n The cumulative contribution rate α of (p≤n) p .

[0104]

[0105] The harmonic detection results, after calculation and analysis, are as follows: Figure 4 As shown, principal component 1 contains the 2nd, 3rd, 5th, 7th, and 11th harmonics; principal component 2 contains the 6th and 9th harmonics; and principal component 3 contains the 8th harmonic. These can be used for further quantitative characterization analysis. The 4th and 10th harmonics have the lowest contribution and the lowest dependence of the common factor on them, and their corresponding harmonics can be eliminated. Therefore, the above three principal components can be selected as feature quantities for further data analysis and research.

[0106] In step 103, an initial defect identification model is determined based on the self-organizing map algorithm, and the initial defect identification model is trained and optimized based on the principal component features to obtain the optimal defect identification model.

[0107] Preferably, the training and optimization of the initial defect identification model based on the principal component features to obtain the optimal defect identification model includes:

[0108] performing format conversion on the harmonic signal corresponding to the principal component feature quantity to obtain harmonic conversion data;

[0109] annotating the harmonic conversion data for defect types to obtain annotation data;

[0110] inputting the annotation data as training data into the initial defect recognition model, calculating a winner neuron, iteratively updating a weight vector based on the winner neuron, and determining an optimal defect recognition model until a preset convergence condition is reached; wherein after model convergence, each input sample corresponds to a neuron position in a self-organizing mapping (SOM) network, and at this time the SOM network forms a low-dimensional representation reflecting the topology of the input data.

[0111] Preferably, wherein the annotation data is input as training data into the initial defect recognition model, a winner neuron is calculated, and a weight vector is iteratively updated based on the winner neuron, including:

[0112] randomly selecting a sample vector x from the data set i , calculating a winner neuron, including: calculating the Euclidean distance between the input sample x i and the weight vector w j of each neuron in the SOM network, finding the neuron with the smallest distance as the winner neuron, representing the optimal match of the neuron to the current input, denoted as c:

[0113] c = argmin j ‖x i -w j ‖,

[0114] The Euclidean distance formula is:

[0115]

[0116] updating the weight vector, including: updating the weight vector of the winner neuron and the neurons in its neighborhood, so that the winner neuron is closer to the current input sample x i ; wherein the update formula is:

[0117] w j (t+1) = w j (t) + η(t)·h cj (t)·(x i -w j (t)),

[0118] wherein η(t) is the learning rate, which decreases with the iteration number t, and h cj(t) is the neighborhood function, representing the degree of neighborhood influence of neuron j from the winner neuron c; the learning rate η(t) is the key parameter that controls the speed of weight updates, and it is gradually reduced during training to ensure that the model gradually converges to a stable state.

[0119] In step 104, the harmonic data of the high-voltage cable under test is input into the defect identification optimal model to identify the defect type of the high-voltage cable under test.

[0120] Self-Organizing Map (SOM) is a neural network model based on unsupervised learning used to map high-dimensional input harmonic data to a low-dimensional (usually two-dimensional) neural network while preserving the data's topological structure. This algorithm iteratively updates the weight vectors of neurons, causing similar data to cluster together in the mapping space, thus achieving harmonic data clustering and visualization. The basic structure of the SOM network is a two-dimensional rectangular or hexagonal grid, with each grid node (i.e., neuron) associated with a weight vector. After training, the input data is mapped onto the SOM grid, where adjacent neurons represent similar harmonic data categories. Figure 5 As shown, the process of determining the optimal model for defect identification includes:

[0121] (1) Initialization

[0122] The principal component information of the harmonic signal obtained by principal component analysis is converted into a format suitable for algorithm processing, represented as a dataset Y = {y1, y2, ..., y}. N}, where each sample y i It is an n-dimensional vector representing a specific harmonic amplitude or phase. The initial weight vector of the SOM network is W = {w1, w2, ..., w...}. m}, where each w j It is an n-dimensional vector, and its initial value is usually randomly assigned.

[0123] (2) Data labeling

[0124] To prepare labeled data for the subsequent training process, labels are added to the harmonic data of known defect types. This data will be used as the training dataset and input into the SOM model.

[0125] (3) Calculate the winner's neuron

[0126] Randomly select a sample vector x from the dataset. i Calculate the winner neuron: Calculate the input sample x i The weight vector w of each neuron in the SOM network jThe Euclidean distance between the input and output is used to find the neuron with the smallest distance (the winner neuron), which represents the best match of that neuron for the current input, denoted as c.

[0127] c = argmin j ||x i -w j ||

[0128] The Euclidean distance formula is:

[0129]

[0130] Update weight vectors: Update the weight vectors of the winner neuron and its neighbors to make them closer to the current input sample x. i The updated formula is:

[0131] w j (t+1)=w j (t)+η(t)·h cj (t)·(x i -w j (t))

[0132] Where η(t) is the learning rate, which decreases with the number of iterations t, h cj (t) is the neighborhood function, representing the degree of neighborhood influence of neuron j from the winning neuron c, and is usually expressed as a Gaussian function:

[0133]

[0134] Where r c and r j ηc and ηj are the positions of neurons c and j in the grid, respectively; σ(t) is the neighborhood width, which decreases over time; η0 is the initial learning rate; and T is the maximum number of iterations.

[0135] Neighborhood function h cj The purpose of (t) is to ensure that adjacent neurons in the SOM network have similar weight changes during the update process, so that the SOM grid can maintain the topological structure of the harmonic input data. This formula shows that as the iteration progresses, the neighborhood width gradually decreases, and eventually only the winning neuron itself will be updated.

[0136] The learning rate η(t) is a key parameter controlling the speed of weight updates. It is usually gradually decreased during training to ensure that the model gradually converges to a steady state.

[0137] (4) Iterative training

[0138] Repeat step (3) above to iteratively update the weight vector until the preset convergence condition is met (such as the weight change being less than the set threshold or the maximum number of iterations is reached).

[0139] (5) Model convergence

[0140] After the SOM model converges, each input sample corresponds to a neuron location in the SOM grid. At this point, the SOM network has formed a low-dimensional representation that reflects the topological structure of the input data.

[0141] In practical applications, the newly acquired harmonic data feature vector is input into the trained SOM model, and the model automatically maps the data to a neuron position in the grid. Adjacent neurons represent similar defect types, so the defect types of the cable can be identified and classified based on the position of the neurons, and the probability of each defect can be quantified.

[0142] In this invention, the visualization capabilities of the SOM model are utilized to present the detection results graphically, helping users intuitively understand the defect distribution of the cable, such as... Figure 6-9 As shown.

[0143] In the graph, the colors of the grid points represent distances, ranging from dark blue to dark red, indicating the distances between different neurons (i.e., the average distance between each neuron and its neighbors). Darker colors (red) indicate larger distances between neurons, typically suggesting these neurons are in different clusters or boundary regions. Lighter colors (blue) indicate smaller distances between neurons, indicating these neurons are clustered together, representing similar data points. The label character on each grid point represents the label of the sample in each neuron after clustering (i.e., the category of the original data): "1" represents "defective cable," and "0" represents "normal cable."

[0144] In an embodiment of the present invention, during field measurements of multiple 50-meter-long 220kV cross-linked polyethylene high-voltage cables, significant differences were observed in the amplitude and frequency of harmonic current signals. After PCA principal component extraction and SOM cluster analysis using the method of the present invention, the defect identification results of cables with typical defects and normal cables are as follows: Figure 6-9 As shown, the clustering accuracy for semi-conductive suspended defect cables reached 91.3%; for metallic suspended defect cables, it reached 91.52%; for cables with residual insulation tape defects, it reached 98%; and for semi-conductive tip defect cables, it reached 84.98%. This verifies the effectiveness and reliability of the method of the present invention, proving that the method of the present invention can provide reliable technical support for cable defect identification in power systems in practical applications.

[0145] Figure 10 This is a schematic diagram of the structure of a high-voltage cable defect identification system 1000 based on harmonic current signals according to an embodiment of the present invention. Figure 10As shown, the high-voltage cable defect identification system 1000 based on harmonic current signals provided by the embodiments of the present invention includes: a harmonic current monitoring data acquisition unit 1001, a principal component analysis unit 1002, a model determination unit 1003, and a defect type identification unit 1004.

[0146] Preferably, the harmonic current monitoring data acquisition unit 1001 is used to acquire harmonic current monitoring data of the metal sheath of the high-voltage cable at different measurement time points.

[0147] Preferably, the principal component analysis unit 1002 is used to analyze the harmonic current monitoring data using principal component analysis to obtain principal component characteristic quantities.

[0148] Preferably, the principal component analysis unit 1002 analyzes the harmonic current monitoring data using principal component analysis to obtain principal component characteristic quantities, including:

[0149] A data matrix X is determined based on the harmonic current monitoring data; where rows represent different measurement time points and columns represent different harmonic components.

[0150] The data matrix X is standardized, and the covariance matrix is ​​calculated based on the standardized data.

[0151] The covariance matrix is ​​decomposed into eigenvalues ​​to obtain eigenvalues ​​and eigenvectors, and principal components are obtained based on the eigenvalues ​​and eigenvectors.

[0152] Calculate the contribution rate and cumulative contribution rate of each principal component, and determine the principal component characteristic quantities based on the contribution rate and cumulative contribution rate.

[0153] Preferably, the principal component analysis unit 1002, which obtains principal components based on the eigenvalues ​​and eigenvectors, includes:

[0154] The eigenvalues ​​of the covariance matrix R are calculated to be λ1≥λ2≥…≥λ. n ≥0 and the corresponding eigenvectors are u1, u2, ..., u n The resulting n new index variables composed of feature vectors are:

[0155]

[0156] Among them, u j =(u 1j u2 j ,…,u nj ), u nj Let y1 represent the nth component of the j-th eigenvector, where y1 is the first principal component, y2 is the second principal component, ..., y... n It is the nth principal component; and These are the measured amplitudes of the 2nd, 3rd, and (n+1)th harmonics, respectively; n is the number of the measured harmonics.

[0157] Preferably, the model determination unit 1003 is used to determine an initial defect identification model based on a self-organizing map algorithm, and to train and optimize the initial defect identification model based on the principal component features to obtain an optimal defect identification model.

[0158] Preferably, the model determination unit 1003 trains and optimizes the initial defect identification model based on the principal component features to obtain the optimal defect identification model, including:

[0159] The harmonic signals corresponding to the principal component feature quantities are format-converted to obtain harmonic conversion data.

[0160] The harmonic conversion data is labeled with defect types to obtain labeled data;

[0161] The labeled data is input as training data into the initial defect identification model. The winner neuron is calculated, and the weight vector is iteratively updated based on the winner neuron until the preset convergence condition is reached, thus determining the optimal defect identification model. After the model converges, each input sample corresponds to a neuron position in the self-organizing map (SOM) network. At this time, the SOM network forms a low-dimensional representation that reflects the topological structure of the input data.

[0162] Preferably, the model determination unit 1003 inputs the labeled data as training data into the initial defect recognition model, calculates the winner neuron, and iteratively updates the weight vector based on the winner neuron, including:

[0163] Randomly select a sample vector x from the dataset. i Calculating the winner neuron includes: calculating the input sample x i The weight vector w of each neuron in the SOM network j The neuron with the smallest Euclidean distance between the input and output is selected as the winner neuron, representing the optimal match of that neuron for the current input, denoted as c.

[0164] c = argmin j ||x i -w j ‖,

[0165] The Euclidean distance formula is:

[0166]

[0167] Updating the weight vector includes: updating the weight vectors of the winning neuron and its neighboring neurons, making the winning neuron more closely resemble the current input sample x. i The update formula is as follows:

[0168] w j (t+1)=w j (t)+η(t)·h cj (t)·(x i -w j (t)),

[0169] Where η(t) is the learning rate, which decreases with the number of iterations t, h cj (t) is the neighborhood function, representing the degree of neighborhood influence of neuron j from the winner neuron c; the learning rate η(t) is the key parameter that controls the speed of weight updates, and it is gradually reduced during training to ensure that the model gradually converges to a stable state.

[0170] Preferably, the defect type identification unit 1004 is used to input the harmonic data of the high-voltage cable under test into the defect identification optimal model in order to identify the defect type of the high-voltage cable under test.

[0171] The high-voltage cable defect identification system 1000 based on harmonic current signals of this invention corresponds to the high-voltage cable defect identification method 100 based on harmonic current signals of another embodiment of this invention, and will not be described again here.

[0172] The present invention has been described with reference to a few embodiments. However, it will be apparent to those skilled in the art that other embodiments besides those disclosed above fall equivalently within the scope of the present invention.

[0173] Generally, all terms used in this invention are interpreted according to their ordinary meaning in the art, unless otherwise expressly defined herein. All references to “a / the / the [device, component, etc.]” ​​are openly interpreted as at least one instance of said device, component, etc., unless otherwise expressly stated. The steps of any method disclosed herein need not be performed in the exact order disclosed unless explicitly stated otherwise.

[0174] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0175] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0176] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0177] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0178] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not 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 modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for identifying defects in high-voltage cables based on harmonic current signals, characterized in that, The method includes: Acquire harmonic current monitoring data of the metal sheath of high-voltage cables at different measurement time points; The harmonic current monitoring data were analyzed using principal component analysis to obtain principal component characteristic quantities. The initial model for defect identification is determined based on the self-organizing map algorithm, and the initial model for defect identification is trained and optimized based on the principal component features to obtain the optimal model for defect identification. The harmonic data of the high-voltage cable under test is input into the defect identification optimal model to identify the defect type of the high-voltage cable under test.

2. The method according to claim 1, characterized in that, The analysis of the harmonic current monitoring data using principal component analysis to obtain principal component characteristic quantities includes: A data matrix X is determined based on the harmonic current monitoring data; where rows represent different measurement time points and columns represent different harmonic components. The data matrix X is standardized, and the covariance matrix is ​​calculated based on the standardized data. The covariance matrix is ​​decomposed into eigenvalues ​​to obtain eigenvalues ​​and eigenvectors, and principal components are obtained based on the eigenvalues ​​and eigenvectors. Calculate the contribution rate and cumulative contribution rate of each principal component, and determine the principal component characteristic quantities based on the contribution rate and cumulative contribution rate.

3. The method according to claim 2, characterized in that, The process of obtaining principal components based on the eigenvalues ​​and eigenvectors includes: The eigenvalues ​​of the covariance matrix R are calculated to be λ1≥λ2≥…≥λ. n ≥0 and the corresponding eigenvectors are u1, u2, ..., u n The resulting n new index variables composed of feature vectors are: Among them, u j =(u 1j u2 j , ...,un j ), un j Let y1 be the nth component of the j-th eigenvector, y2 be the first principal component, y3 be the second principal component, ..., y4 be the nth component. n It is the nth principal component; and These are the measured amplitudes of the 2nd, 3rd, and (n+1)th harmonics, respectively; n is the number of the measured harmonics.

4. The method according to claim 1, characterized in that, The step of training and optimizing the initial defect identification model based on the principal component features to obtain the optimal defect identification model includes: The harmonic signals corresponding to the principal component feature quantities are format-converted to obtain harmonic conversion data. The harmonic conversion data is labeled with defect types to obtain labeled data; The labeled data is input as training data into the initial defect identification model. The winner neuron is calculated, and the weight vector is iteratively updated based on the winner neuron until the preset convergence condition is reached, thus determining the optimal defect identification model. After the model converges, each input sample corresponds to a neuron position in the self-organizing map (SOM) network. At this time, the SOM network forms a low-dimensional representation that reflects the topological structure of the input data.

5. The method according to claim 4, characterized in that, The labeled data is input as training data into the initial defect recognition model, the winner neuron is calculated, and the weight vector is iteratively updated based on the winner neuron, including: Randomly select a sample vector x from the dataset. i Calculating the winner neuron includes: calculating the input sample x i The weight vector w of each neuron in the SOM network j The neuron with the smallest Euclidean distance between the input and output is selected as the winner neuron, representing the optimal match of that neuron for the current input, denoted as c. c=argmin j ‖x i -w j ‖, The Euclidean distance formula is: Updating the weight vector includes: updating the weight vectors of the winning neuron and its neighboring neurons, making the winning neuron more closely resemble the current input sample x. i The update formula is as follows: w j (t+1)=w j (t)+η(t)·h cj (t)·(x i -w j (t)), Where η(t) is the learning rate, which decreases with the number of iterations t, h cj (t) is the neighborhood function, representing the degree of neighborhood influence of neuron j from the winner neuron c; the learning rate η(t) is the key parameter that controls the speed of weight updates, and it is gradually reduced during training to ensure that the model gradually converges to a stable state.

6. A high-voltage cable defect identification system based on harmonic current signals, characterized in that, The system includes: The harmonic current monitoring data acquisition unit is used to acquire harmonic current monitoring data of the metal sheath of high-voltage cables at different measurement time points; The principal component analysis unit is used to analyze the harmonic current monitoring data using the principal component analysis method to obtain principal component characteristic quantities. The model determination unit is used to determine the initial model for defect identification based on the self-organizing map algorithm, and to train and optimize the initial model for defect identification based on the principal component features to obtain the optimal model for defect identification. The defect type identification unit is used to input the harmonic data of the high-voltage cable under test into the defect identification optimal model in order to identify the defect type of the high-voltage cable under test.

7. The system according to claim 6, characterized in that, The principal component analysis unit analyzes the harmonic current monitoring data using principal component analysis to obtain principal component characteristic quantities, including: A data matrix X is determined based on the harmonic current monitoring data; where rows represent different measurement time points and columns represent different harmonic components. The data matrix X is standardized, and the covariance matrix is ​​calculated based on the standardized data. The covariance matrix is ​​decomposed into eigenvalues ​​to obtain eigenvalues ​​and eigenvectors, and principal components are obtained based on the eigenvalues ​​and eigenvectors. Calculate the contribution rate and cumulative contribution rate of each principal component, and determine the principal component characteristic quantities based on the contribution rate and cumulative contribution rate.

8. The system according to claim 7, characterized in that, The principal component analysis unit obtains principal components based on the eigenvalues ​​and eigenvectors, including: The eigenvalues ​​of the covariance matrix R are calculated to be λ1≥λ2≥…≥λ. n ≥0 and the corresponding eigenvectors are u1, u2, ..., u n The resulting n new index variables composed of feature vectors are: Where, u j =(u 1j u2 j ,…,u nj ), u nj Let y1 represent the nth component of the j-th eigenvector, where y1 is the first principal component, y2 is the second principal component, ..., y... n It is the nth principal component; and These are the measured amplitudes of the 2nd, 3rd, and (n+1)th harmonics, respectively; n is the number of the measured harmonics.

9. The system according to claim 6, characterized in that, The model determination unit trains and optimizes the initial defect identification model based on the principal component features to obtain the optimal defect identification model, including: The harmonic signals corresponding to the principal component feature quantities are format-converted to obtain harmonic conversion data. The harmonic conversion data is labeled with defect types to obtain labeled data; The labeled data is input as training data into the initial defect identification model. The winner neuron is calculated, and the weight vector is iteratively updated based on the winner neuron until the preset convergence condition is reached, thus determining the optimal defect identification model. After the model converges, each input sample corresponds to a neuron position in the self-organizing map (SOM) network. At this time, the SOM network forms a low-dimensional representation that reflects the topological structure of the input data.

10. The system according to claim 9, characterized in that, The model determination unit inputs the labeled data as training data into the initial defect identification model, calculates the winner neuron, and iteratively updates the weight vector based on the winner neuron, including: Randomly select a sample vector x from the dataset. i Calculating the winner neuron includes: calculating the input sample x i The weight vector w of each neuron in the SOM network j The neuron with the smallest Euclidean distance between the input and output is selected as the winner neuron, representing the optimal match of that neuron for the current input, denoted as c. c=argmin j ‖x i -w j ‖, The Euclidean distance formula is: Updating the weight vector includes: updating the weight vectors of the winning neuron and its neighboring neurons, making the winning neuron more closely resemble the current input sample x. i The update formula is as follows: w j (t+1)=w j (t)+η(t)·h cj (t)·(x i -w j (t)), Where η(t) is the learning rate, which decreases with the number of iterations t, h cj (t) is the neighborhood function, representing the degree of neighborhood influence of neuron j from the winner neuron c; the learning rate η(t) is the key parameter that controls the speed of weight updates, and it is gradually reduced during training to ensure that the model gradually converges to a stable state.

Images