CN111103517A - Vacuum degree partial discharge pulse group classification and identification method - Google Patents

Vacuum degree partial discharge pulse group classification and identification method Download PDF

Info

Publication number
CN111103517A
CN111103517A CN202010062693.2A CN202010062693A CN111103517A CN 111103517 A CN111103517 A CN 111103517A CN 202010062693 A CN202010062693 A CN 202010062693A CN 111103517 A CN111103517 A CN 111103517A
Authority
CN
China
Prior art keywords
classification
partial discharge
discharge
discharge pulse
class
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202010062693.2A
Other languages
Chinese (zh)
Inventor
杨永旭
张学敏
党军朋
罗洪宏
石岩
陶伟
刘校豪
李天福
陈嘉骏
牛双成
钱威
和强宇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Yuxi Power Supply Bureau of Yunnan Power Grid Co Ltd
Original Assignee
Yuxi Power Supply Bureau of Yunnan Power Grid Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Yuxi Power Supply Bureau of Yunnan Power Grid Co Ltd filed Critical Yuxi Power Supply Bureau of Yunnan Power Grid Co Ltd
Priority to CN202010062693.2A priority Critical patent/CN111103517A/en
Publication of CN111103517A publication Critical patent/CN111103517A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/12—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing
    • G01R31/1227—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Testing Relating To Insulation (AREA)

Abstract

The invention relates to a vacuum degree partial discharge pulse group classification and identification method. The method comprises the steps of extracting characteristic parameters of pulses through an equivalent time-frequency method to form a 2D characteristic parameter space, and classifying a plurality of partial discharge pulse groups by utilizing an improved fuzzy C-means clustering algorithm. In the insulation on-line monitoring of the power equipment with multiple local discharge sources, the method classifies the obtained discharge pulse groups, converts the discharge pulse groups with high similarity in the same class into a pulse peak value-time sequence, and identifies the discharge mode according to the traditional method, so that the discharge mode can be compared with a discharge mode database constructed based on a single artificial defect model, and the discharge pulse groups irrelevant to the database can be judged as invalid signals or noise. While other associated discharge pulse data identify their corresponding discharge type. Therefore, the problem of aliasing of the discharge pulse peak value-time sequence is solved, and a plurality of partial discharge sources or discharge modes containing interference sources can be identified.

Description

Vacuum degree partial discharge pulse group classification and identification method
Technical Field
The invention belongs to the technical field of partial discharge, and particularly relates to a vacuum degree partial discharge pulse group classification and identification method which is applicable to partial discharge ultra-wideband detection technology of vacuum degree multiple partial discharge sources and discharge pulse group classification in a pattern identification system.
Background
At present, when an alternating current partial discharge monitoring system in China is used for online monitoring of insulation conditions of electric equipment such as a motor, a power cable and a Gas Insulated Switchgear (GIS), peak value-time sequences of partial discharge signals are mostly extracted. When a plurality of partial discharge sources (two or more) or interference sources exist, the obtained partial discharge signals are peak-time sequences doped with various discharge signals or abnormal interference signals and various discharge spectrograms correspondingly generated for pattern recognition are also doped with various signals and are subjected to random aliasing. Typically, the system database for the partial discharge pattern recognition is constructed based on a single artificial defect model. Therefore, in the case that a plurality of partial discharge sources or abnormal interference sources exist, the judgment of the discharge mode by the partial discharge identification system based on the pulse peak value-time sequence can be inaccurate, and the method is particularly provided for identifying the partial discharge signal caused by the vacuum degree.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a vacuum degree partial discharge pulse group classification and identification method, which can solve the aliasing problem in the traditional discharge pulse peak value-time sequence and can identify the discharge mode of a plurality of partial discharge sources (or interference sources).
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
a classification and identification method for vacuum degree partial discharge pulse groups is characterized in that characteristic parameters of pulses are extracted through an equivalent time-frequency method to form a 2D characteristic parameter space, partial discharge pulse groups are classified through an improved fuzzy C-means clustering algorithm, and the classification and identification method is specifically divided into two stages:
partial discharge pulse waveform feature extraction stage: extracting characteristic quantity of the pulse waveform by an equivalent time-frequency method, and compressing the characteristic quantity into a vector of a characteristic space; normalizing the characteristic values of the discharge pulses on the equivalent time domain and the equivalent frequency domain;
a classification stage of a fuzzy C-means clustering algorithm: the method is realized by an iterative hill climbing method, namely, the optimal fuzzy classification is obtained by continuously iterating an iteration function, a target function is the square sum of weighted distances from a sample to a clustering center, and classification is carried out according to a final membership matrix.
Further, preferably, the characteristic quantity of the pulse waveform is extracted by an equivalent time-frequency method and is compressed into a vector of a characteristic space, and the specific method is as follows:
the pulse group data is processed as follows:
Figure BDA0002374998840000021
wherein j represents the jth pulse, and n represents that the pulse is composed of n points; diAt point iDomain waveform value, mv; Δ h is the sampling time interval; delta h (i-1) is the time corresponding to the ith point;
performing Fourier transform on the pulse waveform:
Figure BDA0002374998840000022
in the formula, DiThe spectral amplitude of the ith point; Δ g (i-1) is the frequency value of the ith point;
the signal is then processed as follows:
Figure BDA0002374998840000023
Figure BDA0002374998840000024
hitime corresponding to the ith point, giThe frequency value of the ith point;
Figure BDA0002374998840000025
denotes the normalization of the time and frequency domains of 0 to j pulses, Hj、GjRepresenting the equivalent time domain characteristics and the equivalent frequency domain characteristics of the jth pulse;
extracting the characteristics of all discharge pulses in the same class to obtain the characteristic vector (H) of the pulse groupj,Gj) J is 1,2, and N is the number of pulses.
Further, preferably, the characteristic values of the discharge pulse in the equivalent time domain and the equivalent frequency domain are normalized, and the specific method is as follows:
Figure BDA0002374998840000026
in the formula (II), H'j、G'jRespectively, feature vector values after normalization processing.
Further, preferably, the optimal fuzzy classification is obtained by continuously iterating an iteration function, the objective function is the sum of squares of weighted distances from the samples to the clustering center, and the classification is performed according to the final membership matrix, and the specific method is as follows:
sample space X ═ X1,...,xn) N samples in (a) are classified into class c, P ═ Pij) I 1,2, c, j 1,2, n, where P is the membership matrix, i.e. the classification result, PijIs a sample xjMembership to class i;
the clustering criterion of the FCM algorithm is to minimize the sum of the squares of the weighted distances from all samples to the cluster center; the objective function is defined as:
Figure BDA0002374998840000031
in the formula, JFCMThe sum of squares of the weighted distances from all samples to the cluster center; q. q.siThe cluster center of the ith class; m ∈ (1, ∞) is a weighting index; obtaining an optimal classification of the sample by continuously iteratively updating, the iterative function being:
Figure BDA0002374998840000032
in the formula, xjFor the jth sample, pijIs a sample xjDegree of membership to class i, qiThe cluster center of the ith class;
on the basis of determining the clustering number c and the weighting index m, initializing a membership degree matrix P by using random numbers between (0,1), and calculating a clustering center Q(n)，Q(n)Is composed of qiForming a vector, wherein n is the iteration number; when inequality | | Qn-Qn+1Finishing iteration when | | is less than or equal to epsilon, and outputting a membership matrix classification result;
where ε represents an allowable error.
Further, it is preferable that the method of initializing the membership matrix P is: c × n random numbers which are evenly distributed are generated to form a P matrix; m is 2; epsilon is 1 x 10-5。
Further, it is preferable to define compactness and separability of the data set, specifically as follows;
Figure BDA0002374998840000041
in the formula: j. the design is a squaremIs a compactness measure that characterizes the degree of compactness of the sample within the class; l ismIs a separability measure used for characterizing the dispersion degree of the samples among the classes; p is a radical ofijIs a sample xjDegree of membership to class i, qiThe cluster center of the ith class;
Figure BDA0002374998840000042
is the mean of the cluster centers;
the cluster validity function is as follows:
Figure BDA0002374998840000043
in the formula:
Figure BDA0002374998840000044
is the maximum value of the compactness measure;
Figure BDA0002374998840000045
is the maximum value of the separability measure;
GFCMcorresponds to the optimal number of clusters c.
Further, it is preferable to take the maximum value of the clustering number c for n samples
Figure BDA0002374998840000046
Further, it is preferable that the maximum value of c is 6.
The method adopts a pulse waveform-time sequence, extracts characteristic parameters of pulses through an equivalent time-frequency method to form a 2D or high-dimensional characteristic parameter space, and classifies partial discharge pulse groups by utilizing an improved fuzzy C-means clustering algorithm.
The method classifies the obtained discharge pulse groups in the insulation online monitoring of the power equipment with multiple local discharge sources, converts the discharge pulse groups with high similarity in the same class into a pulse peak value-time sequence, and then identifies the discharge mode according to the traditional method, so that the method can be compared with a discharge mode database constructed based on a single artificial defect model, the discharge pulse groups irrelevant to the database can be judged as invalid signals or noise, and other relevant discharge pulse data identify the corresponding discharge types. The partial discharge test result in the GIS shows that: the method can accurately classify the pulse group formed by the multiple partial discharge sources, and provides experimental and theoretical bases for developing the partial discharge ultra-wideband detection technology and the pattern recognition system of the multiple partial discharge sources.
The method is realized in two stages, and the method comprises the following steps of: extracting characteristic quantity of the pulse waveform by an equivalent time-frequency method, and compressing the characteristic quantity into a vector of a characteristic space; classifying the discharge pulses with similar characteristics into one class by using an unsupervised clustering method; converting the waveform-time sequence of all the discharge pulses of each subclass into a peak-time sequence; and identifying a discharge mode. The characteristic values of the discharge pulse on the equivalent time domain and the equivalent frequency domain are normalized to be in the same order of magnitude, so that the classification result is influenced the same. Then, a fuzzy C-means clustering (FCM) algorithm classification stage: the FCM algorithm is realized by an iterative hill-climbing method, namely, the optimal fuzzy classification is obtained by continuously iterating an iteration function. And the target function is the square sum of weighted distances from the samples to the clustering center, and classification is carried out according to the final membership matrix.
In the power equipment insulation structure, although we are not aware of the discharge source inside, in the case where the partial discharge source is stable, the unknown system is determined. Because the insulation aging of the power equipment and the change of the external environment condition do not affect the partial discharge pulse signal in a short time, the discharge pulse generated by the partial discharge source is relatively stable. The key for researching the partial discharge ultra-wideband detection and discharge pattern recognition system based on the single artificial defect model is to research an accurate discharge pulse group classification technology.
Compared with the prior art, the invention has the beneficial effects that:
the method solves the problem of classification technology which is a key technology of a current artificial defect model identification system. Compared with a simple FCM method and a manual classification method, the classification method can automatically output the optimal classification number and classification results of the partial discharge pulse sequence without any manual participation, and even people without any knowledge of classification knowledge can well finish the classification work of the partial discharge pulse by using the algorithm. The test results based on the single partial discharge source and the multiple partial discharge sources show the effectiveness and the practicability of the method, and provide a basis for developing an identification system based on a single artificial defect model.
Drawings
FIG. 1 is a graph of the effect of the clustering number c on the classification results according to the present invention;
FIG. 2 is a schematic diagram of a partial discharge pulse sequence for a single partial discharge source according to the present invention; wherein (a) the peak-phase sequence of the discharge pulses; (b) equivalent time-frequency characteristics of the discharge pulses;
FIG. 3 is a diagram of the classification results, equivalent time-frequency characteristics and peak-phase sequence thereof under the condition of a single partial discharge source according to the present invention; wherein, (a) the equivalent time-frequency characteristic of class 1 and the peak value-phase sequence thereof; (b) class 2 equivalent time-frequency characteristics and their peak-phase sequences; (c) class 3 equivalent time-frequency characteristics and peak-phase sequence thereof
FIG. 4 is a time domain and frequency domain characteristic diagram of partial discharge pulses with a single partial discharge source in accordance with the present invention; wherein, (a) the time domain and the frequency domain of class 1; (b) class 2 time and frequency domains; (c) class 3 time and frequency domain waveforms;
fig. 5 is a sequence diagram of partial discharge pulses in the case of multiple partial discharge sources according to the present invention; wherein (a) the peak-phase sequence of the discharge pulses; (b) equivalent time-frequency characteristics of the discharge pulses;
FIG. 6 is a diagram of the classification results, their equivalent time-frequency characteristics and their peak-phase sequence under the condition of multiple partial discharge sources according to the present invention; wherein, (a) the equivalent time-frequency characteristic of class 1 and the peak value-phase sequence thereof; (b) class 2 equivalent time-frequency characteristics and their peak-phase sequences; (c) class 3 equivalent time-frequency characteristics and their peak-phase sequences; (d) class 4 equivalent time-frequency characteristics and their peak-phase sequences; (e) class 5 equivalent time-frequency characteristics and their peak-phase sequences;
FIG. 7 is a time domain and frequency domain characterization plot of partial discharge pulses under multiple partial discharge sources in accordance with the present invention; wherein, (a) the time domain and the frequency domain of class 1; (b) class 2 time and frequency domains; (c) class 3 time and frequency domains; (d) class 4 time and frequency domains; (e) class 5 time and frequency domains.
Detailed Description
The present invention will be described in further detail with reference to examples.
It will be appreciated by those skilled in the art that the following examples are illustrative of the invention only and should not be taken as limiting the scope of the invention. The examples do not specify particular techniques or conditions, and are performed according to the techniques or conditions described in the literature in the art or according to the product specifications. The materials or equipment used are not indicated by manufacturers, and all are conventional products available by purchase.
In the embodiment, an equivalent time-frequency method is applied to the extraction of the characteristic quantity of the partial discharge pulse waveform, and the characteristic quantity extraction is carried out on the ultra-wideband partial discharge pulse waveform-time sequence of 100MHz, so that the invention mainly discloses the improvement of the fuzzy C algorithm to realize the accurate classification of multiple partial discharge pulse groups.
Extracting characteristic quantity of discharge pulse:
and extracting the characteristic quantity of the pulse waveform by an equivalent time-frequency method, and compressing the characteristic quantity into a vector of a characteristic space. The extraction of the characteristic quantity needs to contain most information on the premise of maximally representing the pulse waveform characteristics, so that the discharge pulses with similar characteristics are gathered together in a characteristic space. And then classifying the discharge pulses with similar characteristics into one class by using an unsupervised clustering method. Thus, all samples in the feature space are divided into several sub-classes, each sub-class corresponding to a sub-pulse group. Then, the waveform-time series of all the discharge pulses of each subclass is converted into a peak-time series, and the discharge pattern is identified.
In order to facilitate the classification of the pulse groups, the data is processed as follows;
Figure BDA0002374998840000061
wherein j represents the jth pulse, and n represents that the pulse is composed of n points; diIs the time domain waveform value of the ith point, mv; Δ h is the sampling time interval; Δ h (i-1) is the time corresponding to the ith point. Similarly, the pulse waveform is subjected to Fourier transform to obtain the pulse waveform
Figure BDA0002374998840000071
In the formula, DiThe spectral amplitude of the ith point; Δ g (i-1) is the frequency value at the ith point.
The simplest way to characterize the time-frequency characteristics of a signal is to use a spread representation of the signal, so that the signal is processed as follows:
Figure BDA0002374998840000072
Figure BDA0002374998840000073
hitime corresponding to the ith point, giThe frequency value of the ith point;
Figure BDA0002374998840000074
denotes the normalization of the time and frequency domains of 0 to j pulses, Hj、GjRepresenting the equivalent time domain characteristics and the equivalent frequency domain characteristics of the jth pulse;
extracting the characteristics of all discharge pulses in the same class to obtain the characteristic vector (H) of the pulse groupj,Gj) J is 1,2, and N is the number of pulses. However, if more information is required to characterize the discharge pulse, the feature vector sequence can be extracted by the following formula;
Figure BDA0002374998840000075
Figure BDA0002374998840000076
a feature vector sequence representing the extracted l to j pulses;
where l is 2, j, equation (5) obeys the requirements of equations (1) through (4) at any time.
Preprocessing of characteristic quantity:
the characteristic values of the discharge pulse obtained according to the formula (4) in the equivalent time domain and the equivalent frequency domain have very large differences in magnitude, which are in different orders of magnitude, and thus, the influence of the characteristic values on the classification result is weighted differently. In order to have the same effect on the classification result, it is normalized, i.e. it has
Figure BDA0002374998840000081
In the formula (II), H'j、G'jRespectively, feature vector values after normalization processing.
Fuzzy C-means clustering (FCM) algorithm principle:
the FCM algorithm is realized by an iterative hill-climbing method, namely, the optimal fuzzy classification is obtained by continuously iterating an iteration function. The target function is the square sum of weighted distances from the samples to the clustering center, and is classified according to the final membership matrix, and the algorithm is as follows;
sample space X ═ X1,...,xn) N samples in (a) are classified into class c, P ═ Pij) I 1,2, c, j 1,2, n, where P is the membership matrix, i.e. the classification result, PijIs a sample xjMembership to class i. X is H or G;
each H 'is calculated by formula (6)'jG'jConstituting a feature vector H, G. H is H'jIs made of (H'0H1'..), G is G'jIs composed of (G'0G1'..) formula (5) extracts a feature vector sequence containing more informationThe characteristics of the discharge pulse can be better characterized.
The clustering criterion of the FCM algorithm is to minimize the sum of the squares of the weighted distances of all samples to the cluster center. The objective function is defined as:
Figure BDA0002374998840000082
in the formula, JFCMThe sum of squares of the weighted distances from all samples to the cluster center; q. q.siThe cluster center of the ith class; m ∈ (1, ∞) is a weighting index. Obtaining an optimal classification of the sample by continuously iteratively updating, the iterative function being
Figure BDA0002374998840000091
In the formula, xjFor the jth sample, pijIs a sample xjDegree of membership to class i, qiThe cluster center of the ith class;
on the basis of determining the clustering number c and the weighting index m, initializing a membership degree matrix U by using numbers between (0,1), and calculating a clustering center Q(n)And n is the number of iterations. When inequality | | Qn-Qn+1And finishing iteration when | | is less than or equal to epsilon, and outputting a membership matrix classification result. Where ε represents an allowable error, 1 × 10-5。
The method for initializing the membership matrix P comprises the following steps: c × n random numbers are generated, and the P matrix is formed.
The weighting index m controls the sharing degree of the model among fuzzy classes, and the proper value of m is selected to smooth the membership function and inhibit noise. However, there is no theoretical guidance on how to select the optimal value of m. Where m is 2.
Determination of the number of clusters c:
the clustering analysis is an important unsupervised clustering pattern classification method, but the FCM algorithm must give a clustering number in advance, and whether the clustering number is selected properly or not will have a great influence on a clustering result. If the predetermined number of clusters is greater than the true number of clusters, the data of the same category is separated; if the predetermined number of clusters is smaller than the true number of clusters, this results in merging of data of different classes. As shown in fig. 1, the same group of data is classified by c-3 and c-4.
As can be seen from fig. 1, if the predetermined number of clusters is not appropriate, the cluster analysis algorithm does not obtain the best clustering result. The optimal classification number problem of the sample is determined according to the following method, and the compactness and the separability of the data set are defined as follows;
Figure BDA0002374998840000092
in the formula: j. the design is a squaremIs a compactness measure that characterizes the degree of compactness of the sample within the class; l ismIs a measure of separability, and is used to characterize the degree of dispersion of samples between classes. p is a radical ofijIs a sample xjDegree of membership to class i, qiThe cluster center of the ith class;
Figure BDA0002374998840000101
is the mean of the cluster centers.
The invention provides a new clustering validity function for partial discharge classification, which has the following form:
Figure BDA0002374998840000102
in the formula:
Figure BDA0002374998840000103
is the maximum value of the compactness measure;
Figure BDA0002374998840000104
is the maximum value of the separability measure. Cluster validity function GFCMIs defined as the difference between the compactness metric and the separability metric normalized (i.e., processed to the same order of magnitude). Since the compactness metric and the separability metric may be on different orders of magnitude, this will result inA term at a lower order of magnitude is meaningless. Therefore, we treated first and then made the difference. The smaller the compactness measure is, the smaller the data difference in the same class is, i.e. the greater the compactness degree of the data in the class is; the greater the separability measure, the greater the difference in data between different classes, i.e., the greater the degree of dispersion of data between classes. Thus, GFCMCorresponds to the optimal number of clusters c.
The maximum value of the cluster number c for n samples may be taken
Figure BDA0002374998840000105
The practical case of partial discharge and the rapidity of the calculation are considered herein to take the maximum value of c to be 6.
The specific implementation results of this embodiment are shown in fig. 2-7.
The partial discharge pulse sequence diagram under the condition of a single partial discharge source of the invention is shown in fig. 2, each point in fig. 2a represents a discharge pulse, the abscissa is the phase corresponding to the pulse occurrence time, and the ordinate is the pulse signal voltage amplitude. Each point in fig. 2b corresponds one-to-one to the point in fig. 2a, with the abscissa being the equivalent time of the pulse and the ordinate being the equivalent frequency. Based on the aggregate condition of the points in fig. 2b, all pulses are divided into 3 classes, each of which is a subset of fig. 2a and 2b, shown in a, b, c, respectively, of fig. 3; the highest amplitude pulse is found in each class, and its corresponding waveform and amplitude-frequency characteristics are shown in a, b, c of fig. 4.
The partial discharge pulse sequence diagram under the condition of multiple partial discharge sources of the invention is shown in fig. 5, each point in fig. 5a represents a discharge pulse, the abscissa is the phase corresponding to the pulse occurrence time, and the ordinate is the pulse signal voltage amplitude. Each point in fig. 5b corresponds one-to-one to the point in fig. 5a, with the abscissa being the equivalent time of the pulse and the ordinate being the equivalent frequency. Based on the aggregate condition of the points in fig. 5b, all pulses are divided into 5 classes, each of which is a subset of fig. 5a and 5b, shown in fig. 6 a, b, c, d, e, respectively; the highest amplitude pulse is found in each class, with the corresponding waveform and amplitude-frequency characteristics shown in a, b, c, d, e of fig. 7.
The foregoing shows and describes the general principles, essential features, and advantages of the invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are described in the specification and illustrated only to illustrate the principle of the present invention, but that various changes and modifications may be made therein without departing from the spirit and scope of the present invention, which fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.

Claims (8)

1. The classification and identification method for the partial discharge pulse group under the vacuum degree is characterized in that characteristic parameters of pulses are extracted through an equivalent time-frequency method to form a 2D characteristic parameter space, the partial discharge pulse group is classified by an improved fuzzy C-means clustering algorithm, and the classification and identification method is specifically divided into two stages:
partial discharge pulse waveform feature extraction stage: extracting characteristic quantity of the pulse waveform by an equivalent time-frequency method, and compressing the characteristic quantity into a vector of a characteristic space; normalizing the characteristic values of the discharge pulses on the equivalent time domain and the equivalent frequency domain;
a classification stage of a fuzzy C-means clustering algorithm: the method is realized by an iterative hill climbing method, namely, the optimal fuzzy classification is obtained by continuously iterating an iteration function, a target function is the square sum of weighted distances from a sample to a clustering center, and classification is carried out according to a final membership matrix.
2. The vacuum degree partial discharge pulse group classification and identification method as claimed in claim 1, wherein the characteristic quantity of the pulse waveform is extracted by an equivalent time-frequency method and compressed into a vector of a characteristic space, and the specific method is as follows:
the pulse group data is processed as follows:
Figure FDA0002374998830000011
wherein j represents the jth pulse, and n represents that the pulse is composed of n points; diIs the time domain waveform value of the ith point, mv; Δ hIs a sampling time interval; delta h (i-1) is the time corresponding to the ith point;
performing Fourier transform on the pulse waveform:
Figure FDA0002374998830000012
in the formula, DiThe spectral amplitude of the ith point; Δ g (i-1) is the frequency value of the ith point;
the signal is then processed as follows:
Figure FDA0002374998830000013
Figure FDA0002374998830000021
hitime corresponding to the ith point, giThe frequency value of the ith point;
Figure FDA0002374998830000024
denotes the normalization of the time and frequency domains of 0 to j pulses, Hj、GjRepresenting the equivalent time domain characteristics and the equivalent frequency domain characteristics of the jth pulse;
extracting the characteristics of all discharge pulses in the same class to obtain the characteristic vector (H) of the pulse groupj,Gj) J is 1,2, and N is the number of pulses.
3. The vacuum degree partial discharge pulse group classification and identification method according to claim 2, wherein the characteristic values of the discharge pulses in the equivalent time domain and the equivalent frequency domain are normalized by:
Figure FDA0002374998830000022
in the formula (II), H'j、G'jRespectively normalized eigenvector value。
4. The vacuum degree partial discharge pulse group classification and identification method according to claim 3, wherein the optimal fuzzy classification is obtained by continuously iterating an iteration function, an objective function is the square sum of weighted distances from a sample to a clustering center, and classification is performed according to a final membership matrix, and the method comprises the following steps:
sample space X ═ X1,...,xn) N samples in (a) are classified into class c, P ═ Pij) I 1,2, c, j 1,2, n, where P is the membership matrix, i.e. the classification result, PijIs a sample xjMembership to class i;
the clustering criterion of the FCM algorithm is to minimize the sum of the squares of the weighted distances from all samples to the cluster center; the objective function is defined as:
Figure FDA0002374998830000023
in the formula, JFCMThe sum of squares of the weighted distances from all samples to the cluster center; q. q.siThe cluster center of the ith class; m ∈ (1, ∞) is a weighting index; obtaining an optimal classification of the sample by continuously iteratively updating, the iterative function being:
Figure FDA0002374998830000031
in the formula, xjFor the jth sample, pijIs a sample xjDegree of membership to class i, qiThe cluster center of the ith class;
on the basis of determining the clustering number c and the weighting index m, initializing a membership degree matrix P by using random numbers between (0,1), and calculating a clustering center Q(n)，Q(n)Is composed of qiForming a vector, wherein n is the iteration number; when inequality | | Qn-Qn+1Finishing iteration when | | is less than or equal to epsilon, and outputting a membership matrix classification result;
where ε represents an allowable error.
5. The vacuum degree partial discharge pulse group classification and identification method as claimed in claim 4, wherein the method for initializing the membership matrix P is as follows: c × n random numbers which are evenly distributed are generated to form a P matrix; m is 2; epsilon is 1 x 10-5。
6. The vacuum degree partial discharge pulse group classification and identification method according to claim 4, characterized in that compactness and separability of a data set are defined, specifically as follows;
Figure FDA0002374998830000032
in the formula: j. the design is a squaremIs a compactness measure that characterizes the degree of compactness of the sample within the class; l ismIs a separability measure used for characterizing the dispersion degree of the samples among the classes; p is a radical ofijIs a sample xjDegree of membership to class i, qiThe cluster center of the ith class;
Figure FDA0002374998830000033
is the mean of the cluster centers;
the cluster validity function is as follows:
Figure FDA0002374998830000041
in the formula:
Figure FDA0002374998830000042
is the maximum value of the compactness measure;
Figure FDA0002374998830000043
is the maximum value of the separability measure;
GFCMcorresponds to the optimal number of clusters c.
7. The method of claim 4The classification and identification method of the vacuum degree partial discharge pulse group is characterized in that the maximum value of n sample clustering numbers c is taken
Figure FDA0002374998830000044
8. The vacuum degree partial discharge pulse group classification recognition method according to claim 4, wherein the maximum value of c is 6.
CN202010062693.2A 2020-01-20 2020-01-20 Vacuum degree partial discharge pulse group classification and identification method Pending CN111103517A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010062693.2A CN111103517A (en) 2020-01-20 2020-01-20 Vacuum degree partial discharge pulse group classification and identification method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010062693.2A CN111103517A (en) 2020-01-20 2020-01-20 Vacuum degree partial discharge pulse group classification and identification method

Publications (1)

Publication Number Publication Date
CN111103517A true CN111103517A (en) 2020-05-05

Family

ID=70427535

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010062693.2A Pending CN111103517A (en) 2020-01-20 2020-01-20 Vacuum degree partial discharge pulse group classification and identification method

Country Status (1)

Country Link
CN (1) CN111103517A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112198399A (en) * 2020-09-07 2021-01-08 红相股份有限公司 Identification method and terminal for multi-source electromagnetic wave signals
CN112305381A (en) * 2020-09-21 2021-02-02 国网山东省电力公司临沂供电公司 Method and system for monitoring and positioning online partial discharge of distribution cable
CN112364704A (en) * 2020-10-16 2021-02-12 康威通信技术股份有限公司 Clustering method and system based on clock synchronization partial discharge
CN112763871A (en) * 2020-12-30 2021-05-07 珠海华网科技有限责任公司 Partial discharge classification identification method
CN113447769A (en) * 2021-05-31 2021-09-28 华东电力试验研究院有限公司 PD pulse group manual rapid extraction system based on two-dimensional characteristic parameter plane

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102645620A (en) * 2012-05-17 2012-08-22 广东电网公司电力科学研究院 Multisource partial discharge detection method and device of transformer substation based on time-frequency characteristic parameters
CN103698669A (en) * 2013-12-12 2014-04-02 广东电网公司东莞供电局 Local discharge pulse separating method for high-voltage power equipment
CN106896306A (en) * 2017-04-26 2017-06-27 国网上海市电力公司 A kind of GIS oscillatory surges pressure test signal antinoise method
CN107656154A (en) * 2017-09-18 2018-02-02 杭州安脉盛智能技术有限公司 Based on the Diagnosis Method of Transformer Faults for improving Fuzzy C-Means Cluster Algorithm
CN108132428A (en) * 2017-11-24 2018-06-08 国网北京市电力公司 The method, apparatus of shelf depreciation multi-source separation
CN109670536A (en) * 2018-11-30 2019-04-23 中国南方电网有限责任公司超高压输电公司检修试验中心 A kind of local discharge signal clustering method in the case of multi-source electric discharge and interference superposition
CN109901031A (en) * 2019-02-27 2019-06-18 西安电子科技大学 Signal De-noising Method, information data processing terminal for local discharge signal

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102645620A (en) * 2012-05-17 2012-08-22 广东电网公司电力科学研究院 Multisource partial discharge detection method and device of transformer substation based on time-frequency characteristic parameters
CN103698669A (en) * 2013-12-12 2014-04-02 广东电网公司东莞供电局 Local discharge pulse separating method for high-voltage power equipment
CN106896306A (en) * 2017-04-26 2017-06-27 国网上海市电力公司 A kind of GIS oscillatory surges pressure test signal antinoise method
CN107656154A (en) * 2017-09-18 2018-02-02 杭州安脉盛智能技术有限公司 Based on the Diagnosis Method of Transformer Faults for improving Fuzzy C-Means Cluster Algorithm
CN108132428A (en) * 2017-11-24 2018-06-08 国网北京市电力公司 The method, apparatus of shelf depreciation multi-source separation
CN109670536A (en) * 2018-11-30 2019-04-23 中国南方电网有限责任公司超高压输电公司检修试验中心 A kind of local discharge signal clustering method in the case of multi-source electric discharge and interference superposition
CN109901031A (en) * 2019-02-27 2019-06-18 西安电子科技大学 Signal De-noising Method, information data processing terminal for local discharge signal

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
鲍永胜: "局部放电脉冲波形特征提取及分类技术", 《中国电机工程学报》 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112198399A (en) * 2020-09-07 2021-01-08 红相股份有限公司 Identification method and terminal for multi-source electromagnetic wave signals
CN112198399B (en) * 2020-09-07 2023-12-19 红相股份有限公司 Multi-source electromagnetic wave signal identification method and terminal
CN112305381A (en) * 2020-09-21 2021-02-02 国网山东省电力公司临沂供电公司 Method and system for monitoring and positioning online partial discharge of distribution cable
CN112364704A (en) * 2020-10-16 2021-02-12 康威通信技术股份有限公司 Clustering method and system based on clock synchronization partial discharge
CN112763871A (en) * 2020-12-30 2021-05-07 珠海华网科技有限责任公司 Partial discharge classification identification method
CN113447769A (en) * 2021-05-31 2021-09-28 华东电力试验研究院有限公司 PD pulse group manual rapid extraction system based on two-dimensional characteristic parameter plane

Similar Documents

Publication Publication Date Title
CN111103517A (en) Vacuum degree partial discharge pulse group classification and identification method
CN106483370B (en) Non-intrusion type household loads real-time identification method and device based on multi-feature fusion
CN104198898B (en) Local discharge development process diagnosis method based on pulse-train analysis
Kranz Diagnosis of partial discharge signals using neural networks and minimum distance classification
CN109061426B (en) Transformer partial discharge fault diagnosis method and online monitoring device
CN111444784B (en) Power equipment discharge signal separation and classification method based on kernel principal component analysis
Chang et al. Application of pulse sequence partial discharge based convolutional neural network in pattern recognition for underground cable joints
CN111538960B (en) Alternating current asynchronous motor fault diagnosis method based on improved fuzzy C-means clustering
CN110701087A (en) Axial flow compressor pneumatic instability detection method based on single-classification overrun learning machine
Liao et al. Recognition of partial discharge patterns
CN113391244A (en) VMD-based transformer switching-on vibration signal characteristic frequency calculation method
CN112636328A (en) Medium-voltage distribution network equipment load state identification method
CN111025100A (en) Transformer ultrahigh frequency partial discharge signal mode identification method and device
Chung et al. Non-intrusive load monitoring using wavelet design and co-testing of machine learning classifiers
Zhao et al. A fault diagnosis approach for rolling element bearing based on S-transform and artificial neural network
CN111539275B (en) Electrical load classification method and system based on load characteristic visualization
CN114492543A (en) Partial discharge classification and identification method based on time-frequency domain characteristics
CN109917245B (en) Ultrasonic detection partial discharge signal mode identification method considering phase difference
CN114021275A (en) Rolling bearing fault diagnosis method based on deep convolution fuzzy system
Wang et al. Time-frequency features extraction and classification of partial discharge UHF signals
CN113408383A (en) Audible noise invalid data judgment method based on bounded beta (g, h) distribution and MWKPCA
Guzmán et al. Classification of High Frequency NILM Transients Based on Convolutional Neural Networks
CN112098744B (en) Method for identifying voltage sag source
CN114861699A (en) Partial discharge diagnosis method based on improved ANFIS
Zeng et al. Type identification of partial discharge signal based on emd decomposition and gwo-svm

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20200505

RJ01 Rejection of invention patent application after publication