CN113343550A - Partial discharge fault diagnosis method based on local image characteristics - Google Patents
Partial discharge fault diagnosis method based on local image characteristics Download PDFInfo
- Publication number
- CN113343550A CN113343550A CN202110641395.3A CN202110641395A CN113343550A CN 113343550 A CN113343550 A CN 113343550A CN 202110641395 A CN202110641395 A CN 202110641395A CN 113343550 A CN113343550 A CN 113343550A
- Authority
- CN
- China
- Prior art keywords
- partial discharge
- image
- longicorn
- iteration
- feature
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000003745 diagnosis Methods 0.000 title claims abstract description 22
- 238000012706 support-vector machine Methods 0.000 claims abstract description 51
- 241001481710 Cerambycidae Species 0.000 claims abstract description 43
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 20
- 230000000694 effects Effects 0.000 claims abstract description 5
- 238000007781 pre-processing Methods 0.000 claims abstract 2
- 239000013598 vector Substances 0.000 claims description 38
- 230000007547 defect Effects 0.000 claims description 28
- 230000000007 visual effect Effects 0.000 claims description 27
- 239000011159 matrix material Substances 0.000 claims description 26
- 238000012360 testing method Methods 0.000 claims description 8
- 238000002790 cross-validation Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 230000001629 suppression Effects 0.000 claims description 3
- 238000012935 Averaging Methods 0.000 claims description 2
- 238000003384 imaging method Methods 0.000 claims 1
- 238000005457 optimization Methods 0.000 abstract description 4
- 239000000284 extract Substances 0.000 abstract description 2
- 238000013145 classification model Methods 0.000 abstract 2
- 208000028659 discharge Diseases 0.000 description 57
- 238000001514 detection method Methods 0.000 description 5
- 238000009413 insulation Methods 0.000 description 4
- 238000000605 extraction Methods 0.000 description 3
- 238000003909 pattern recognition Methods 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000015556 catabolic process Effects 0.000 description 2
- 238000006731 degradation reaction Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 239000002923 metal particle Substances 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 208000024891 symptom Diseases 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/25—Design optimisation, verification or simulation using particle-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Medical Informatics (AREA)
- Health & Medical Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Testing Relating To Insulation (AREA)
Abstract
A partial discharge fault diagnosis method based on partial image features comprises the following steps: collecting partial discharge images of different fault types, preprocessing the images and generating a partial discharge image database; extracting local features of the partial discharge image by adopting a U-SURF-BoW algorithm, and constructing a partial discharge image feature space; and then, carrying out dimensionality reduction on the feature space by adopting PCA, then carrying out parameter optimization operation on the support vector machine by using a longicorn algorithm, constructing an optimal partial discharge support vector machine classification model, and finally, identifying and classifying the partial discharge image data features by adopting the classification model to obtain a final fault diagnosis result. The invention can effectively and stably extract the partial discharge characteristics, improve the efficiency and the accuracy of the partial discharge fault diagnosis and simultaneously ensure that the partial discharge fault diagnosis has better identification effect under the condition of high noise.
Description
Technical Field
The invention belongs to the field of partial discharge mode identification, and particularly relates to a partial discharge fault diagnosis method based on partial image characteristics.
Background
Partial Discharge (PD) is an important symptom reflecting insulation degradation of electrical equipment, and different Discharge types differ in insulation degradation mechanism, Discharge development process, and harmfulness. Therefore, the type of the partial discharge is timely and effectively identified, and the method has important significance for equipment fault identification and insulation state evaluation. An ultrahigh frequency (UHF) detection method is widely applied to PD detection and pattern recognition as a detection method which has high sensitivity, can position a discharge source and has strong anti-interference capability. The principle of the uhf detection method is to monitor the occurrence of partial discharge by using high frequency electromagnetic waves excited by a pulse current generated by the partial discharge.
In order to improve the efficiency of partial discharge fault diagnosis, an intelligent algorithm needs to be used instead of manpower, and pattern recognition is a tool for intelligent diagnosis. For GIS partial discharge mode identification, the current common feature extraction method comprises statistical feature, texture feature, shape feature and other methods. However, these algorithms also have certain limitations, are not good for recognition in a noisy field environment, and are easy to extract a large number of redundant features. Currently, the classifiers mainly include back propagation neural networks (called BPNN), KNN algorithms, etc., however, the structure of BPNN is difficult to determine, and the parameters to be adjusted are many. The KNN principle is simple, and the method is well applied to the field of partial discharge mode identification, but the algorithm is low in efficiency when new data are tested.
Disclosure of Invention
The invention aims to overcome the defects in the prior art, and provides a partial discharge fault diagnosis method based on local image characteristics so as to obtain a better partial discharge fault type classification effect and further realize automatic diagnosis and accurate identification of partial discharge faults.
In order to achieve the purpose, the invention adopts the following technical scheme:
a partial discharge fault diagnosis method based on partial image features is characterized by comprising the following steps:
1) respectively acquiring ultrahigh frequency signal images of partial discharge of A fault types by using ultrahigh frequency sensors to serve as a sample set X ═ X1,X2,…,Xi,…,XA},1<i≤A,XiA set of samples representing signals of the i-th class; and isn represents the total number of partial discharge signal samples for the i-th fault,j is more than or equal to 1 and less than or equal to N, and the total number of partial discharge ultrahigh frequency signal image samples is N;
2) using a document [1 ]]The method calculates the BoW feature packet of all partial discharge ultrahigh frequency signal image features (see the literature [1 ]]Sivic, Josef. effective visual search of video cassette as a text retrieval. IEEE Transactions on pattern analysis and machine interpretation, vol.31, No.4.IEEE: 591-605. April 2009) to extract the U-SURF characteristics of all partial discharge ultrahigh frequency signal image sample sets X to obtain all sample sets XSet of features B ═ B1,B2,…,Bi,…,BA}. Wherein, BiA feature set representing the signal of the i-th class; and is And representing the characteristics of the jth sample in the ith fault, wherein the characteristics of each sample are 1 × H-dimensional vectors, and a corresponding fault type label is added to the characteristic set of the A fault types and is marked as Y ═ Y1,Y2,…,Yi,…,YA}; using a document [1 ]]Calculating BoW feature packets of all partial discharge ultrahigh frequency signal image features to obtain a visual word frequency feature set C ═ C1,C2,…,Ci,…,CA},CiA feature set representing the signal of the i-th class; and is Representing the characteristics of the jth sample in the ith type of fault, wherein the characteristics of each sample are vectors with the dimensions of 1 x L;
2.1) converting the sample set X into a gray scale graph G, removing numerical information of coordinates, controlling the pixel size of each picture to be consistent, wherein the pixel size is r, the expression of any pixel point in the picture is D (a, b), and the value ranges of a and b are as follows: a is more than or equal to 1 and less than or equal to r, and b is more than or equal to 1 and less than or equal to r;
2.2) for each gray-scale image GiDetermining the U-SURF characteristic points of the image according to the following method:
extracting a Hessian matrix with the scale of sigma at the point D in the image:
wherein L isxx(D,σ)、Lxy(D,σ)、Lyy(D, σ) represents the convolution of the gaussian second-order partial derivative with the image at D;
the convolution of the image and the second derivative is respectively expressed by Dxx, Dxy and Dyy approximation, and the approximate determinant value of the Hessian matrix is obtained according to the following formula:
det(Happroximate)=DxxDyy-(wDxy)2
then adopting a non-maximum suppression principle in a neighboring area with the size of 3 multiplied by 3, selecting an image where the maximum value of the Hessian matrix determinant is located and the scale space position thereof as characteristic points, and calling the characteristic points as the gray image GiRepeating the step 2.2) to obtain all gray level images GiThe U-SURF feature points of (1);
2.3) calculating each grayscale image GiThe description vector of SURF feature points of (1):
taking the U-SURF feature point as a center, selecting a 20s × 20s square area, where s represents the scale of the feature point, and dividing the area into 4 × 4 sub-areas with the best effect, then calculating a corresponding 4-dimensional feature vector V ═ Σ dx, Σ | dx |, Σ dy, Σ | dy | } in each sub-area, and forming a 4 × 4 × 4 ═ 64-dimensional description vector at each feature point;
2.4) the features obtained in step 2.3) are all mapped to the vocabulary of the visual dictionary by calculating the distance between the features according to the following method:
constructing a word list by using a K-means algorithm, dividing all characteristics into L clusters by taking K as a parameter, enabling the clusters to have higher similarity and lower inter-cluster similarity, regarding each cluster center as a visual vocabulary in a dictionary, obtaining L visual vocabularies in total, and enabling all the visual vocabularies to form a visual dictionary;
2.5) counting the occurrence frequency and the occurrence frequency of each visual word in each picture to obtain the frequency of the visual words with different partial discharge image characteristics;
3) calculating BoW feature packets of all partial discharge ultrahigh frequency signal image features to obtain a visual word frequency feature set C ═ C of all sample sets1,C2,…,Ci,…,CA},CiA feature set representing the signal of the i-th class; and is Representing the characteristics of the jth sample in the ith type of fault, wherein the characteristics of each sample are vectors with the dimensions of 1 x L;
4) after the acquired U-SURF characteristics are subjected to dimensionality reduction by using the following PCA dimensionality reduction method, a dimensionality reduction characteristic matrix P is obtained:
4.1) zero-averaging each column of the U-SURF feature, i.e. subtracting the average of this row;
4.2) solving the covariance matrix and the eigenvalue and eigenvector of the covariance matrix;
4.3) arranging the eigenvectors into a matrix from top to bottom according to the sizes of the corresponding eigenvalues, taking the data of the first k columns to form a dimension-reduced eigenvector matrix P, wherein P is { P ═ P { (P)1,P2,…,Pi,…,PA},PiA feature set representing the signal of the i-th class; and is Features representing the q-th sample in a class i fault, where 1<q≤k<L;
5) Normalizing the dimension-reduced feature matrix P to obtain a defect feature set U ═ U1,U2,…,Ui,…,UZ},UiThe method is characterized in that an ith defect characteristic sample corresponding to a GIS partial discharge ultrahigh frequency signal is represented, and the method comprises the following steps: representing the jth sample in the ith type defect characteristic sample; and the jth sampleK normalized statistical characteristics are included, i is more than or equal to 1 and less than or equal to A, j is more than or equal to 1 and less than or equal to M, and M represents the total number of the ith defect characteristic samples;
6) initializing the longicorn tentacle length, the longicorn motion step length, the longicorn iteration times and the longicorn tentacle position, and constructing an initial support vector machine model:
6.1) initializing the antenna whisker length to be s, the antenna movement step length to be u and the antenna iteration number to be tmaxThe two-dimensional position coordinate vector of the two tentacles of the longicorn is P0={PL,PRIn which P isLRepresenting the coordinates of the longicorn left-tentacle position, PRRepresenting the coordinates of the right tentacle of the longicorn;
using said two-dimensional position coordinate vector P0Initializing parameters of a support vector machine by coordinate values in the x direction and the y direction of the two tentacles, wherein the coordinate value in the x direction represents a parameter c of the support vector machine, and the coordinate value in the y direction represents a parameter g of the support vector machine;
the support vector machine finds a partition hyperplane in a sample space based on a training set, and separates samples of different classes. The support vector machine can be converted to solve the convex quadratic programming problem:
yi(ωTxi+b)≥1-ξi,i=1,2,...,N
ξi≥0,i=1,2,...,N
wherein: omega is hyperplane normal vector, b is offset, C is penalty factor, xiiIs a relaxation factor;
introducing a kernel function k (x)iX), then the final decision equation for the classifier is:
wherein: alpha is alphai *Is the optimal Lagrangian multiplier, b*Is the optimum offset;
6.2) defining the current iteration number as t, and initializing t to be 1;
two-dimensional coordinate vector P of longicorn beard0Two-dimensional coordinate vector P as the t-th iterationt(ii) a Taking the initial support vector machine model as a support vector machine model of the t iteration;
6.3) two-dimensional coordinate vector P with the t-th iterationtConstructing a support vector machine model of the t iteration, performing cross validation on the e defect feature sample subsets by using the support vector machine model of the t iteration to obtain a partial discharge defect error rate, and taking the partial discharge defect error rate as a fitness value of the t iteration in the longicorn algorithm;
6.4) selecting a smaller value in the fitness values of the t iteration corresponding to the left and right longicorn whiskers, taking the smaller value as a local optimal value of the t iteration, and taking the tentacle coordinate of the longicorn corresponding to the local optimal value;
6.5) the longicorn moves to the tentacle side corresponding to the local optimal value of the t iteration according to the motion step u of the longicorn, thereby obtaining a two-dimensional coordinate vector P of the t +1 iterationt+1;
6.6) assigning t +1 to t, and then judging whether t reaches tmaxIf not, returning to step 6.3), if yes, selecting tmaxTaking the minimum value in the local optimal values of the secondary iteration as a global optimal value; constructing a support vector machine model for diagnosing the partial discharge fault by taking the two-dimensional coordinate vector corresponding to the global optimal value as an optimal support vector machine parameter; and go to the next step;
7) and diagnosing the test sample set of the partial discharge ultrahigh frequency signal image by using the support vector machine model for partial discharge fault diagnosis, and obtaining a final classification result so as to output a fault diagnosis result of partial discharge.
Compared with the prior art, the invention has the beneficial effects that:
1. the method for extracting the partial discharge characteristics based on the U-SURF-BoW solves the problems that the traditional method is difficult to apply to a noisy environment and extracts a large number of redundant characteristics, and meanwhile, the PCA method is adopted to perform dimension reduction processing on the extracted characteristics, so that the problems of characteristic redundancy and large calculated amount in the model are solved. The method improves the effectiveness and the recognition efficiency of the feature extraction of the partial discharge image, and simultaneously ensures the accuracy of the partial discharge fault diagnosis.
2. The invention introduces the longicorn whisker algorithm into the support vector machine, and performs characteristic selection and parameter optimization on the support vector machine through the longicorn whisker algorithm. Compared with the traditional method for optimizing the support vector machine by using the intelligent algorithm, the method can more efficiently and accurately finish the diagnosis and identification of the partial discharge fault, and improves the classification precision and efficiency of the support vector machine.
Drawings
FIG. 1 is a schematic flow diagram of the present invention;
Detailed Description
In this embodiment, a partial discharge fault diagnosis method based on local image features, as shown in fig. 1, includes the following steps:
1) respectively acquiring partial discharge ultrahigh frequency signal images of A fault types by using ultrahigh frequency sensors as a sample set X ═ X1,X2,…,Xi,…,XA},1<i≤A,XiA set of samples representing signals of the i-th class; and isn represents the total number of samples of the ith type fault partial discharge signal,j is more than or equal to 1 and less than or equal to N, and the total number of partial discharge ultrahigh frequency signal image samples is N; the types of the partial discharge defects are generally divided into 4 types, namely metal tip defects, suspended electrode defects, free metal particle defects and air gap model defects;
2) extracting U-SURF characteristics of all partial discharge ultrahigh frequency signal image sample sets X to obtain all partial discharge ultrahigh frequency signal image sample sets XCharacteristic set B ═ B of sample set1,B2,…,Bi,…,BA},BiA feature set representing the signal of the i-th class; and is And representing the characteristics of the jth sample in the ith fault, wherein the characteristics of each sample are 1 × H-dimensional vectors, and a corresponding fault type label is added to the characteristic set of the A fault types and is marked as Y ═ Y1,Y2,…,Yi,…,YA}; using a document [1 ]]Calculating BoW feature packets of all partial discharge ultrahigh frequency signal image features to obtain a visual word frequency feature set C ═ C of all sample sets1,C2,…,Ci,…,CA},CiA feature set representing the signal of the i-th class; and is And representing the characteristics of the jth sample in the ith fault, wherein the characteristics of each sample are vectors with the dimensions of 1 x L.
2.1) converting the partial discharge ultrahigh frequency signal image X into a gray level image G, thereby achieving the purpose of improving the calculation efficiency; removing the numerical information of the coordinates to avoid the interference of the numerical information on the identification process; the pixel sizes of all the pictures are controlled to be consistent and are r multiplied by r, wherein D (a, b) is any pixel point in the pictures, a is larger than or equal to 1 and smaller than or equal to r, and b is larger than or equal to 1 and smaller than or equal to r.
2.2) for each gray-scale GiAnd determining U-SURF characteristic points of the image.
Extracting a Hessian matrix with the scale of sigma at the point D in the image:
wherein L isxx(D,σ)、Lxy(D,σ)、Lyy(D, σ) represents the convolution of the gaussian second-order partial derivative with the image at D;
in order to improve the calculation efficiency, the convolution of the image and the second derivative is respectively approximately represented by Dxx, Dxy and Dyy, and the approximate determinant value of the Hessian matrix is obtained:
det(Happroximate)=DxxDyy-(wDxy)2
then adopting a non-maximum suppression principle in a neighboring region with the size of 3 multiplied by 3, and selecting an image where the maximum value of the Hessian matrix determinant is located and the scale space position of the image as feature points, namely SURF feature points;
2.3) calculating each grayscale image GiThe description vector of SURF feature points of (1):
and selecting a 20s × 20s square area by taking the U-SURF feature point as a center, and dividing the area into 4 × 4 sub-areas. Then, the corresponding 4-dimensional feature vector V ═ Σ dx, Σ | dx |, Σ dy, Σ | dy | } in each sub-region is calculated, respectively. A description vector of 4 × 4 × 4 ═ 64 dimensions is formed at each feature point.
And 2.4, mapping all the characteristics obtained in the step 2.3 into the vocabulary of the visual dictionary by calculating the distance between the characteristics. And constructing a word list by using a K-means algorithm. And taking K as a parameter, and dividing all the characteristics into L clusters, so that the clusters have higher similarity and the inter-cluster similarity is lower. Each cluster center is considered to be a visual vocabulary in the dictionary. Obtaining L visual vocabularies in total, wherein all the visual vocabularies form a visual dictionary;
and 2.5, counting whether each visual word appears or not and the appearance frequency of each visual word in each picture to obtain the visual word frequency of different partial discharge image characteristics.
And 3, performing dimensionality reduction on the acquired U-SURF characteristics by using a PCA dimensionality reduction method.
Step 3.1, carrying out zero equalization on each column of the U-SURF characteristics, namely subtracting the average value of the row;
step 3.2, solving the covariance matrix and the eigenvalue and eigenvector of the covariance matrix
Step 3.3, arranging the eigenvectors into a matrix from top to bottom according to the sizes of the corresponding eigenvalues, and taking the data of the first k columns to form a matrix P, wherein P is { P ═ P1,P2,…,Pi,…,PA},PiA feature set representing the signal of the i-th class; and is Features representing the q-th sample in a class i fault, where 1<q≤k<L。
Step 4, the extracted feature matrix is subjected to normalization processing, so that the speed and the efficiency of subsequent pattern recognition are improved conveniently, and a defect feature set U is obtained1,U2,…,Ui,…,UZ},UiThe method is characterized in that a sample of the ith defect characteristics corresponding to the partial discharge ultrahigh frequency signal is represented, and the sample comprises the following components: representing the jth sample in the ith type defect characteristic sample; and the jth sampleK normalized statistical characteristics are included, i is more than or equal to 1 and less than or equal to A, j is more than or equal to 1 and less than or equal to M, and M represents the total number of the ith defect characteristic samples;
step 5, initializing the longicorn tentacle length, the longicorn motion step length, the longicorn iteration times and the longicorn tentacle position, and constructing an initial support vector machine model;
step 5.1, initializing the antenna whisker length to be s, the antenna motion step length to be u and the antenna iteration number to be tmaxThe two-dimensional position coordinate vector of the two tentacles of the longicorn is P0={PL,PRIn which P isLRepresenting the coordinates of the longicorn left-tentacle position, PRRepresenting the coordinates of the right tentacle of the longicorn;
using said two-dimensional position coordinate vector P0Initializing parameters of a support vector machine by coordinate values in the x direction and the y direction of the two tentacles, wherein the coordinate value in the x direction represents a parameter c of the support vector machine, and the coordinate value in the y direction represents a parameter g of the support vector machine;
step 5.2, defining the current iteration number as t, and initializing t to be 1;
two-dimensional coordinate vector P of longicorn beard0Two-dimensional coordinate vector P as the t-th iterationt(ii) a Taking the initial support vector machine model as a support vector machine model of the t iteration;
step 5.3, utilizing the t-th iteration two-dimensional coordinate vector PtConstructing a support vector machine model of the t iteration, performing cross validation on the e defect feature sample subsets by using the support vector machine model of the t iteration to obtain the partial discharge defect error rate of the support vector machine model of the t iteration, and taking the partial discharge defect error rate as the fitness value of the t iteration in the Tianniu whisker algorithm;
step 5.4, selecting a smaller value of the fitness values of the t iteration corresponding to the left and right longicorn whiskers, taking the smaller value as a local optimal value of the t iteration, and obtaining the tentacle coordinates of the longicorn corresponding to the local optimal value;
step 5.5, the longicorn moves to one side of the tentacle corresponding to the local optimal value of the t iteration according to the motion step u of the longicorn, and therefore the two-dimensional coordinate vector P of the t +1 iteration is obtainedt+1;
Step 5.6, after assigning t +1 to t, judging whether t reaches tmaxIf yes, then select tmaxTaking the minimum value in the local optimal values of the secondary iteration as a global optimal value; taking the two-dimensional coordinate vector corresponding to the global optimal value as an optimal support vector machine parameter, thereby constructing a support vector machine model for diagnosing the partial discharge fault; otherwise, returning to the step 6.3;
and 6, diagnosing the test sample set of the partial discharge ultrahigh frequency signal image by using the support vector machine model for partial discharge fault diagnosis, and outputting a fault classification result of partial discharge.
In order to verify the accuracy of the method, a support vector machine (BAS-SVM) model based on the longicorn whisker algorithm and basic SVM and BPNN are simultaneously established for comparison.
The method comprises the steps of acquiring 400 groups of GIS insulation defect partial discharge data in an ultrahigh frequency partial discharge experiment, dividing samples into training samples and testing samples, randomly selecting 300 groups of training samples, and testing the rest testing samples to test a trained classifier.
The partial discharge fault diagnosis is realized by respectively adopting 3 different classifiers of SVM (support vector machine) based on Particle Swarm Optimization (PSO), SVM based on celestial cow whisker optimization (BAS) and BPNN. The U-SURF-BoW features are put into a classifier for learning, then the trained classifier is used for classifying and identifying the test samples, and the obtained identification result is shown in the following table:
TABLE 1 comparison of recognition effects of different algorithms
As is apparent from comparing the values in table 1, the overall recognition accuracy of the BAS-SVM algorithm is the highest among the 3 classifiers, which is 98.9%, and is much higher than the accuracy of 92.6% of the basis SVM model and 89.1% of the RF model. Meanwhile, the recognition results of the BAS-SVM and the basic SVM are compared, so that the recognition accuracy of the BAS-SVM model is improved by 6.2% compared with that of the basic SVM model, and the longicorn algorithm is proved to be feasible and effective for optimizing the support vector machine. Meanwhile, the effectiveness of the U-SURF-BoW feature extraction method can be seen through the identification result. The experimental result proves that the method can accurately diagnose the type of the partial discharge fault, and simultaneously provides detection and judgment basis for operation and maintenance personnel in an actual field, thereby being beneficial to the safe and stable operation of the power grid.
Claims (1)
1. A partial discharge fault diagnosis method based on partial image features is characterized by comprising the following steps:
1) local discharge for collecting different fault typesAnd (3) electrically imaging, preprocessing the image, and generating a local discharge image database: respectively acquiring ultrahigh frequency signal images of partial discharge of A fault types by using ultrahigh frequency sensors to serve as a sample set X ═ X1,X2,…,Xi,…,XA},1<i≤A,XiA set of samples representing signals of the i-th class; and isn represents the total number of partial discharge signal samples for the i-th fault,j is more than or equal to 1 and less than or equal to N, and the total number of partial discharge ultrahigh frequency signal image samples is N;
2) extracting local features of the partial discharge image by adopting a U-SURF-BoW algorithm, and constructing a partial discharge image feature space: comprises the following steps:
2.1) converting the sample set X into a gray scale graph G, removing numerical information of coordinates, controlling the pixel size of each picture to be consistent, wherein the pixel size is r, the expression of any pixel point in the picture is D (a, b), and the value ranges of a and b are as follows: a is more than or equal to 1 and less than or equal to r, and b is more than or equal to 1 and less than or equal to r;
2.2) for each gray-scale image GiDetermining the U-SURF characteristic points of the image according to the following method:
extracting a Hessian matrix with the scale of sigma at the point D in the image:
wherein L isxx(D,σ)、Lxy(D,σ)、Lyy(D, σ) represents the convolution of the gaussian second-order partial derivative with the image at D;
the convolution of the image and the second derivative is approximately represented by Dxx, Dxy and Dyy respectively, and the approximate determinant value of the Hessian matrix is obtained as follows:
det(Happroximate)=DxxDyy-(wDxy)2
then adopting a non-maximum suppression principle in a neighboring area with the size of 3 multiplied by 3, selecting an image where the maximum value of the Hessian matrix determinant is located and the scale space position thereof as characteristic points, and calling the characteristic points as the gray image GiRepeating the step 2.2) to obtain all gray level images GiThe U-SURF feature points of (1);
2.3) calculating each grayscale image GiThe description vector of SURF feature points of (1):
taking the U-SURF feature point as a center, selecting a 20s × 20s square area, where s represents the scale of the feature point, dividing the area into 4 × 4 sub-areas with the best effect, and then calculating a corresponding 4-dimensional feature vector V ═ Σ dx, Σ | dx |, Σ dy, Σ | dy | } in each sub-area, so as to form a 4 × 4 × 4 ═ 64-dimensional description vector at each feature point;
2.4) the features obtained in step 2.3) are all mapped to the vocabulary of the visual dictionary by calculating the distance between the features according to the following method:
constructing a word list by using a K-means algorithm, dividing all characteristics into L clusters by taking K as a parameter, enabling the clusters to have higher similarity and lower inter-cluster similarity, regarding each cluster center as a visual vocabulary in a dictionary, obtaining L visual vocabularies in total, and enabling all the visual vocabularies to form a visual dictionary;
2.5) counting the occurrence frequency and the occurrence frequency of each visual word in each picture to obtain the frequency of the visual words with different partial discharge image characteristics;
extracting U-SURF characteristics of all partial discharge ultrahigh frequency signal image sample sets X to obtain characteristic sets B ═ B of all sample sets1,B2,…,Bi,…,BA},BiA feature set representing the signal of the i-th class; and is Is shown asThe characteristics of the jth sample in the i-type faults are 1 × H-dimensional vectors, corresponding fault type labels are added to the characteristic set of the A-type fault types, and the labels are recorded as Y ═ Y1,Y2,…,Yi,…,YA};
3) Calculating BoW feature packets of all partial discharge ultrahigh frequency signal image features to obtain a visual word frequency feature set C ═ C of all sample sets1,C2,…,Ci,…,CA},CiA feature set representing the signal of the i-th class; and is Representing the characteristics of the jth sample in the ith type of fault, wherein the characteristics of each sample are vectors with the dimensions of 1 x L;
4) after the dimension reduction processing is carried out on the obtained U-SURF-BOW features by using a PCA dimension reduction method, obtaining a dimension reduction feature matrix P comprises the following steps:
4.1) zero-averaging each column of the U-SURF feature, i.e. subtracting the average of this row;
4.2) solving the covariance matrix and the eigenvalue and eigenvector of the covariance matrix;
4.3) arranging the eigenvectors into a matrix from top to bottom according to the sizes of the corresponding eigenvalues, taking the data of the first k columns to form a dimension-reduced eigenvector matrix P, wherein P is { P ═ P { (P)1,P2,…,Pi,…,PA},PiA feature set representing the signal of the i-th class; and P isi={Pi 1,Pi 2,…,Pi q,…,Pi k},Pi qFeatures representing the q-th sample in a class i fault, where 1<q≤k<L;
5) Normalizing the dimension-reduced feature matrix P to obtain a defect feature set U ═ U1,U2,…,Ui,…,UZ},UiIndicating the class i defect characteristics corresponding to GIS partial discharge ultrahigh frequency signalsCharacterizing the sample, and having: representing the jth sample in the ith type defect characteristic sample; and the jth sampleK normalized statistical characteristics are included, i is more than or equal to 1 and less than or equal to A, j is more than or equal to 1 and less than or equal to M, and M represents the total number of the ith defect characteristic samples;
6) initializing the longicorn tentacle length, the longicorn motion step length, the longicorn iteration times and the longicorn tentacle position, and constructing an initial support vector machine model:
6.1) initializing the antenna whisker length to be s, the antenna movement step length to be u and the antenna iteration number to be tmaxThe two-dimensional position coordinate vector of the two tentacles of the longicorn is P0={PL,PRIn which P isLRepresenting the coordinates of the longicorn left-tentacle position, PRRepresenting the coordinates of the right tentacle of the longicorn; using said two-dimensional position coordinate vector P0Initializing parameters of a support vector machine by coordinate values in the x direction and the y direction of the two tentacles, wherein the coordinate value in the x direction represents a parameter c of the support vector machine, and the coordinate value in the y direction represents a parameter g of the support vector machine;
6.2) defining the current iteration number as t, and initializing t to be 1 by using a two-dimensional coordinate vector P of the longicorn beard0Two-dimensional coordinate vector P as the t-th iterationt(ii) a Taking the initial support vector machine model as a support vector machine model of the t iteration;
6.3) two-dimensional coordinate vector P with the t-th iterationtConstructing a support vector machine model of the t iteration, performing cross validation on the e defect feature sample subsets by using the support vector machine model of the t iteration to obtain a partial discharge defect error rate, and taking the partial discharge defect error rate as a fitness value of the t iteration in the longicorn algorithm;
6.4) selecting a smaller value in the fitness values of the t iteration corresponding to the left and right longicorn whiskers, taking the smaller value as a local optimal value of the t iteration, and taking the tentacle coordinate of the longicorn corresponding to the local optimal value;
6.5) the longicorn moves to the tentacle side corresponding to the local optimal value of the t iteration according to the motion step u of the longicorn, thereby obtaining a two-dimensional coordinate vector P of the t +1 iterationt+1;
6.6) assigning t +1 to t, and then judging whether t reaches tmaxIf not, returning to step 6.3), if yes, selecting tmaxTaking the minimum value in the local optimal values of the secondary iteration as a global optimal value; constructing a support vector machine model for diagnosing the partial discharge fault by taking the two-dimensional coordinate vector corresponding to the global optimal value as an optimal support vector machine parameter; and go to the next step;
7) and diagnosing the test sample set of the partial discharge ultrahigh frequency signal image by using the support vector machine model for partial discharge fault diagnosis, and outputting a fault diagnosis result of partial discharge.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110641395.3A CN113343550A (en) | 2021-06-09 | 2021-06-09 | Partial discharge fault diagnosis method based on local image characteristics |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110641395.3A CN113343550A (en) | 2021-06-09 | 2021-06-09 | Partial discharge fault diagnosis method based on local image characteristics |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113343550A true CN113343550A (en) | 2021-09-03 |
Family
ID=77475706
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110641395.3A Pending CN113343550A (en) | 2021-06-09 | 2021-06-09 | Partial discharge fault diagnosis method based on local image characteristics |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113343550A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117609831A (en) * | 2023-11-21 | 2024-02-27 | 国网宁夏电力有限公司电力科学研究院 | Transformer partial discharge detection method, medium and system based on random matrix |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2015070513A1 (en) * | 2013-11-14 | 2015-05-21 | 国家电网公司 | Pattern recognition method for partial discharge of three-phase in one enclosure type ultrahigh voltage gis |
CN105389593A (en) * | 2015-11-16 | 2016-03-09 | 上海交通大学 | Image object recognition method based on SURF |
US20180039865A1 (en) * | 2015-08-07 | 2018-02-08 | Hefei University Of Technology | Analog circuit fault mode classification method |
CN112364756A (en) * | 2020-11-09 | 2021-02-12 | 国网安徽省电力有限公司电力科学研究院 | GIS fault classification method based on longicorn stigma and random forest |
-
2021
- 2021-06-09 CN CN202110641395.3A patent/CN113343550A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2015070513A1 (en) * | 2013-11-14 | 2015-05-21 | 国家电网公司 | Pattern recognition method for partial discharge of three-phase in one enclosure type ultrahigh voltage gis |
US20180039865A1 (en) * | 2015-08-07 | 2018-02-08 | Hefei University Of Technology | Analog circuit fault mode classification method |
CN105389593A (en) * | 2015-11-16 | 2016-03-09 | 上海交通大学 | Image object recognition method based on SURF |
CN112364756A (en) * | 2020-11-09 | 2021-02-12 | 国网安徽省电力有限公司电力科学研究院 | GIS fault classification method based on longicorn stigma and random forest |
Non-Patent Citations (2)
Title |
---|
方涛 等: "基于天牛须搜索优化支持向量机的变压器故障诊断研究", 《电力系统保护与控制》 * |
李泽 等: "基于 SURF的含噪局部放电模式识别", 《电工技术学报》 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117609831A (en) * | 2023-11-21 | 2024-02-27 | 国网宁夏电力有限公司电力科学研究院 | Transformer partial discharge detection method, medium and system based on random matrix |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109582003B (en) | Bearing fault diagnosis method based on pseudo label semi-supervised kernel local Fisher discriminant analysis | |
CN107505133B (en) | The probability intelligent diagnosing method of rolling bearing fault based on adaptive M RVM | |
CN105738109B (en) | Bearing fault classification diagnosis method based on rarefaction representation and integrated study | |
CN106599797B (en) | A kind of infrared face recognition method based on local parallel neural network | |
CN111832608B (en) | Iron spectrum image multi-abrasive particle identification method based on single-stage detection model yolov3 | |
Stuhlsatz et al. | Feature extraction with deep neural networks by a generalized discriminant analysis | |
CN106295124B (en) | The method of a variety of image detecting technique comprehensive analysis gene subgraph likelihood probability amounts | |
CN110363746B (en) | Ultrasonic nondestructive detection signal classification method based on convolutional neural network | |
CN112364756B (en) | GIS fault classification method based on longicorn stigma and random forest | |
CN111882554B (en) | SK-YOLOv 3-based intelligent power line fault detection method | |
CN113673346A (en) | Motor vibration data processing and state recognition method based on multi-scale SE-Resnet | |
CN113447771A (en) | Partial discharge pattern recognition method based on SIFT-LDA characteristics | |
CN111639697B (en) | Hyperspectral image classification method based on non-repeated sampling and prototype network | |
CN110097067B (en) | Weak supervision fine-grained image classification method based on layer-feed feature transformation | |
Chen et al. | Fast and reliable probabilistic face embeddings in the wild | |
CN113343550A (en) | Partial discharge fault diagnosis method based on local image characteristics | |
Artola et al. | Glad: A global-to-local anomaly detector | |
CN111025100A (en) | Transformer ultrahigh frequency partial discharge signal mode identification method and device | |
Wani | Microarray classification using sub-space grids | |
CN117274988A (en) | Wheat stripe rust spore image detection method and system based on Yolov5s | |
CN112990279B (en) | Radar high-resolution range profile library outside target rejection method based on automatic encoder | |
CN115452957A (en) | Small sample metal damage identification method based on attention prototype network | |
CN104077608A (en) | Behavior recognition method based on sparsely coded slow characteristic functions | |
Zhou et al. | JCS: An explainable surface defects detection method for steel sheet by joint classification and segmentation | |
SALIBA et al. | An overview of pattern recognition |
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: 20210903 |
|
RJ01 | Rejection of invention patent application after publication |