CN107067041A - A kind of cable insulation defect state appraisal procedure based on high frequency Partial discharge signal figure - Google Patents
A kind of cable insulation defect state appraisal procedure based on high frequency Partial discharge signal figure Download PDFInfo
- Publication number
- CN107067041A CN107067041A CN201710342214.0A CN201710342214A CN107067041A CN 107067041 A CN107067041 A CN 107067041A CN 201710342214 A CN201710342214 A CN 201710342214A CN 107067041 A CN107067041 A CN 107067041A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- gradient
- honeybee
- partial discharge
- 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
Classifications
-
- 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
- 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
- G06F18/211—Selection of the most significant subset of features
- G06F18/2111—Selection of the most significant subset of features by using evolutionary computational techniques, e.g. genetic algorithms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30168—Image quality inspection
Abstract
The invention discloses a kind of cable insulation defect state appraisal procedure based on high frequency Partial discharge signal figure, comprise the following steps, A, the high frequency Partial discharge signal figure for obtaining cable；B, the textural characteristics for extracting high frequency Partial discharge signal figure, gray level co-occurrence matrixes feature, Gray level-gradient co-occurrence matrix feature, as characteristic vector；C, using genetic algorithm to characteristic vector carry out dimensionality reduction；D, the characteristic vector after dimensionality reduction brought into the multi-kernel support vector machine of bee colony optimization and carry out defect recognition；E, progress Classifcation of flaws.Using this method, its accuracy rate recognized is high.
Description
Technical field
The present invention relates to cable insulation defect Fault Identification technical field, and in particular to one kind is based on high frequency Partial discharge signal
The cable insulation defect state appraisal procedure of figure.
Background technology
Power cable primarily serves transmission and the effect of distribution high-power power in power system, is the important group of power network
Into one of part, the proportion increase in power circuit with cable, it is ensured that meaning of the stable operation of power cable to power network
Justice is also more notable.Wherein twisted polyethylene cable be XLPE cable machinery, electrically, thermal stability it is good, laying installation side
Just, therefore it is widely used in power system.In order to ensure the safe and reliable operation of XLPE ac cables, it is necessary to electric power electricity
Cable carries out the preventive trials such as insulation measurement.Meanwhile, in order to grasp the insulation defect of XLPE ac cables accurately and in time
Situation, is being carried out beyond off-line test, in addition it is also necessary to carry out online Insulation test and fault diagnosis.
It is existing that to cable insulation defect state progress evaluation, it uses insulation event of the supporting vector machine model to XLPE cable
Barrier is identified, but its accuracy rate is low.
The content of the invention
In order to solve the above-mentioned technical problem the present invention provides a kind of cable insulation defect shape based on high frequency Partial discharge signal figure
State appraisal procedure.
The present invention is achieved through the following technical solutions：
A kind of cable insulation defect state appraisal procedure based on high frequency Partial discharge signal figure, comprises the following steps,
A, the high frequency Partial discharge signal figure for obtaining cable；
B, the textural characteristics for extracting high frequency Partial discharge signal figure, gray level co-occurrence matrixes feature, Gray Level-Gradient Co-occurrence Matrix are special
Levy, as characteristic vector；
C, using genetic algorithm to characteristic vector carry out dimensionality reduction；
D, the characteristic vector after dimensionality reduction brought into the multi-kernel support vector machine of bee colony optimization and carry out defect recognition；
E, progress Classifcation of flaws.
Because the insulation status paradoxical reaction of XLPE ac cables is in the partial discharge quantity of cable, by XLPE ac cables
Shelf depreciation carries out detection and the insulation defect situation of cable is grasped with pattern-recognition, and insulation defect situation is diagnosed, no
The accuracy rate to Fault Identification can be only improved, and improves the intelligent level of XLPE cable partial discharge detecting system, for protecting
The reliability service for demonstrate,proving cable is significant.By extracting the textural characteristics in signal graph, gray level co-occurrence matrixes feature, ash
Degree-gradient co-occurrence matrix feature fully extracts local feature as characteristic vector, and make this programme can consider Partial discharge signal comprehensively
Each category feature of figure.Dimension-reduction treatment using genetic algorithm to characteristic vector, can effectively solve to be absorbed in locally optimal solution, easily
The defect of the topological structure of data is destroyed, this method can be carried out the optimization selection of the overall situation.Finally optimized using bee colony
Multi-kernel support vector machine defect is identified, can effectively reduce the amount of calculation of this method, lift the speed of service, improve
Recognition accuracy, has stronger adaptability to the partial discharge detection under complex environment.
Preferably, the textural characteristics include roughness, contrast, directionality, line similitude, systematicness and rough
Degree.
Further, the extracting method of the roughness is：
To Image Mean Filtering, average intensity value is asked for, filter window size is 2k×2k, wherein, κ joins for filter window
Number；
For each pixel (i, j), it is calculated respectively in mean intensity difference S (i, j) both horizontally and vertically；
For each pixel (i, j), the filter window yardstick corresponding to maximum mean intensity difference is regard as optimal chi
Spend Sbest；
Calculate the S in entire imagebestAnd ask it to be averagely worth to roughness.
Further, the extracting method of the directionality is：
Calculate the gradient vector of each pixel, including both horizontally and vertically on variable quantity；
One histogram is constructed according to gradient vector, the acuity for calculating peak value in histogram obtains the direction of image
Property.
Preferably, the extracting method of the gray level co-occurrence matrixes feature is：
In entire image, count each (h, k) value appearance number of times, then by they be normalized to occur probability
Phk, and constitute gray level co-occurrence matrixes [Phk]N×N, wherein N is the gray level of image, and k, h are gray scale；
Construct the co-occurrence matrix in 4 directions, including M (0,1), M (- 1,1), M (- 1,0), M (- 1, -1)；
The contrasts of 4 co-occurrence matrixs, the degree of correlation, energy, homogeney, entropy are calculated respectively.
Preferably, the extracting method of the Gray Level-Gradient Co-occurrence Matrix feature is：
The gradient image g (x, y) of original image is extracted with gradient operator；
Gray level sliding-model control is carried out to gradient image and obtains new gradient image G (x, y), x=1,2 ..., M；Y=
1,2 ..., N, its gray level is Lg；
Gray Level-Gradient Co-occurrence Matrix is { Hij, i=0,1 ..., L-1；J=0,1 ..., Lg- 1 }, wherein, HijFor set
The number of element in { (x, y) | f (x, y)=i, G (x, y)=j }；F (x, y) be source images, x=1,2,3...M, y=1,2,
3...N；L is the gray level of source images；
It is rightNormalized is done, is obtainedWherein
Calculate small gradient advantage, big gradient advantage, intensity profile inhomogeneities, gradient nonunf ormity, energy, gray scale
Average, average gradient, gray scale mean square deviation, gradient mean square deviation, correlation, gray level entropy, gradient entropy, the entropy of mixing, inertia, unfavourable balance away from.
Preferably, the step C specific methods are,
C-1, defect characteristic individual UVR exposure：Parameter to characteristic vector is arranged in certain sequence, each base of chromosome
Because of the characteristic item of the corresponding order of correspondence, wherein, chromosome is represented using binary character string, and its allele is by { 0,1 } group
Into；
C-2, the genetic manipulation of defect characteristic optimization：Including Selecting operation, crossing operation, mutation operator, end condition；Its
In,
Selecting operation is specially：
Calculate each chromosome viAdaptive value eval (vi), wherein i=1,2 ... n；
Calculate total adaptive value of colony
Calculate the select probability p of each chromosomei=eval (vi)/F；
Calculate the accumulated probability of each chromosome
Producing one, [[0,1] inner random number r, uses q successively in intervaliCompared with r, first there is qi>=r
Body i is selected as the male parent replicated, repeats the step, untill the individual amount required for individual i number is met；
Crossing operation is specially：Two individuals are selected to make parent from population, the individual that every a pair are mutually paired, at random
It is crosspoint to set the position after a certain locus except first position；
Mutation operator is specially：Using basic bit mutation method, the mutation probability P to individual UVR exposure string to setmAt random
Genic value on a certain position specified or several locus does mutation operator, so as to produce new individual；
End condition is specially the algebraically T=100 of heredity；
C-3, defect characteristic optimum choice：Optimal chromosome binary coding is obtained using genetic algorithm, redundancy is set
Feature delete, as optimize after characteristic parameter.
Preferably, the multi-kernel support vector machine use bee colony optimize method for：
Population is initialized, including determines population number, maximum iteration M, control parameter L and determines search space, and
Random generation initial solution x in search spacei, wherein, i=1,2,3...S, S are food source number, are herein 4；
After initialization, carry out leading honeybee, follow honeybee and investigating the repetitive cycling of honeybee search process in whole population, until
Reach maximum iteration M or error permissible value ε；
In the search procedure incipient stage, honeybee is each led to produce a new explanation by formula (1),
In formula, k ∈ { 1,2, L, S }, j ∈ { 1,2, L, D }, and k ≠ i；For the random number between [- 1,1]；
Calculate the fitness value fit of new explanationiAnd it is evaluated, if the fit of new explanationiBetter than old solution fiti-1, then honeybee is led to remember
Firmly old solution is forgotten about in new explanation；Conversely, it will retain old solution；
It is all lead honeybee to complete search process after, follow honeybee to calculate the select probability that each solves according to formula (2),
A number is randomly generated in [0,1] interval, if the probable value of solution is more than the random number, honeybee is followed by formula
(3) new explanation is produced, and examines the fit of new explanationiIf, the fit of new explanationiCompare fiti-1It is good, then follow honeybee to remember that new explanation is forgotten about
Old solution；Conversely, it will retain old solution,
It is all follow honeybee to complete search process after, if a solution is circulated still not by further more by L time
Newly, then be considered as this solution and be absorbed in local optimum, the food source will be rejected, then this food source is corresponding leads honeybee to change into one
Individual investigation honeybee；Search bee produces a new food source by (4) formula and replaces it,
xij=xminj+rand(0,1)(xmaxj-xminj) (4), wherein j ∈ { 1,2, L, D },
Return leads honeybee search procedure, starts repetitive cycling and eventually finds optimal food source or optimal solution.
Further, greedy selection is carried out as the following formula when evaluating food source,
The present invention compared with prior art, has the following advantages and advantages：
Ground of the invention method is by extracting the textural characteristics in signal graph, gray level co-occurrence matrixes feature, Gray Level-Gradient symbiosis
Matrix character fully extracts local feature as characteristic vector, is utilizing dimension-reduction treatment of the genetic algorithm to characteristic vector, finally
Defect is identified the multi-kernel support vector machine optimized using bee colony, and it can not only improve the accuracy rate to Fault Identification, and
Can effectively solve to be absorbed in locally optimal solution, be easily destroyed data topological structure defect, this method can be carried out entirely
The optimization selection of office, reduces amount of calculation, more adapts to the use demand under complex environment.
Brief description of the drawings
Accompanying drawing described herein is used for providing further understanding the embodiment of the present invention, constitutes one of the application
Point, do not constitute the restriction to the embodiment of the present invention.In the accompanying drawings：
Fig. 1 is the recognition accuracy of each supporting vector machine model.
Embodiment
For the object, technical solutions and advantages of the present invention are more clearly understood, with reference to embodiment and accompanying drawing, to this
Invention is described in further detail, and exemplary embodiment and its explanation of the invention is only used for explaining the present invention, does not make
For limitation of the invention.
Embodiment
A kind of cable insulation defect state appraisal procedure based on high frequency Partial discharge signal figure, comprises the following steps,
A, the high frequency Partial discharge signal figure for obtaining cable；
B, the textural characteristics for extracting high frequency Partial discharge signal figure, gray level co-occurrence matrixes feature, Gray Level-Gradient Co-occurrence Matrix are special
Levy, as characteristic vector；
C, using genetic algorithm to characteristic vector carry out dimensionality reduction；
D, the characteristic vector after dimensionality reduction brought into the multi-kernel support vector machine of bee colony optimization and carry out defect recognition；
E, progress Classifcation of flaws.
Illustrated below for each step.
Textural characteristics are Tamura features, including roughness, contrast, directionality, line similitude, systematicness and rough
Degree.
Wherein, roughness FcrsIt is to describe granule size and the physical quantity of distribution in texture, texture primitive size is bigger, base
Distance is more remote between member, and texture is also more coarse.The extracting method of roughness is：
To Image Mean Filtering, average intensity value is asked for, filter window size is 2k×2k, wherein, κ joins for filter window
Number；
For each pixel (i, j), it is calculated respectively in mean intensity difference S (i, j) both horizontally and vertically；
For each pixel (i, j), the filter window yardstick corresponding to maximum mean intensity difference is regard as optimal chi
Spend Sbest；
Calculate the S in entire imagebestAnd ask it to be averagely worth to roughness, its computational methods is specially：
Wherein, m refers to image length；N refers to picture traverse, and i refers to pixel abscissa
Value is length coordinate value, and J refers to pixel ordinate value i.e. width coordinate value.
Contrast FconIt can be obtained by the statistics to pixel intensity distribution situation, it is right in whole image or region to give
Than the global variable of degree, it can be obtained by following formula：
α4=μ4/σ4, wherein, μ refers to the first moment of pixel distribution, and σ refers to the standard deviation of pixel value, μ4It is four squares, and σ2It is
Variance.
The extracting method of directionality is：
Calculate the gradient vector of each pixel | △ G |, including both horizontally and vertically on variable quantity, image volume can be passed through
Accumulate to calculate, be specially：
| △ G |=(| △H|+|△V|)/2
| Δ G |=(| Δ H |+| Δ V |)/2
Wherein, △HAnd △VRespectively both horizontally and vertically on variable quantity；
After all pixel gradients vectors are calculated, a histogram is constructed according to gradient vector, calculated in histogram
The acuity of peak value obtains the overall directionality of image.
Line similitude FlinFor describing the texel being made up of line, its extracting method is：First, structural grain symbiosis
Matrix, its element PDd(i, j) is defined as the phase for occurring two adjacent cells along edge direction spacing distance Dd on image
To frequency matrix, i and j represent both direction respectively.Using co-occurrence matrix, several important feature features of element can be predicted.
In order to measure line similitude, we define the weighting of equidirectional use+1, and vertical direction use -1 is weighted：
Wherein PDdIt is n × n local direction co-occurrence matrixs.It is n rank matrixes that n, which represents matrix,.
Systematicness FregFor describing the regularity of repeat pattern.Its computational methods is：
Freg=1-r (σcrs+σcon+σdir+σlin), wherein, r is normalization factor, σcrs、σcon、σdir、σlinRepresent one kind
Deviation standard.
The rough degree F of objectrghCan not directly it describe.Through showing that we more emphasize to the research that visual psychology logic is tested
The influence of roughness and contrast, and use it as to measure and carry out approximate calculation and spend roughly, it is therefore intended that it is approximately simplified to
The degree corresponding with human visual perception.It can be characterized with following formula：
Frgh=Fcrs+Fcon。
Gray level co-occurrence matrixes refer to the Combined Frequency point occurred simultaneously at a distance of two gray-scale pixels of (△ x, △ y) in image
Cloth.I.e.：If the gray level of image is set into N grades, then co-occurrence matrix is N N matrix, is represented by M(△x,△y)(h,k).Its
In, positioned at the element value M of (h, k)hkA gray scale is represented to be h and another gray scale is two of k at a distance of the picture for being (△ x, △ y)
Number of times of the plain or pixel to appearance.Its extracting method is：
In entire image, count each (h, k) value appearance number of times, then by they be normalized to occur probability
Phk, obtain square formation [Phk]N×N, square formation [Phk]N×NFor gray scale joint probability matrix, also referred to as gray level co-occurrence matrixes.Gray scale is combined
Probability matrix is actually the joint histogram of two pixels.The co-occurrence matrix in 4 directions of construction is M (0,1), M (- 1,1),
M (- 1,0), M (- 1, -1), are represented at a distance of 0 degree of skew for 1,45 degree of skews, 90 degree of skews and 135 degree of skews respectively.Then divide
5 parametric textures of this 4 co-occurrence matrixs are not calculated：Contrast, the degree of correlation, energy, homogeney, entropy.
Contrast returns to the luminance contrast between pixel and its adjacent pixel in entire image.The consistent image of gray scale, it is right
It is 0 than degree, for open grain, PhkValue set near leading diagonal, now | h-k | value it is smaller, so contrast compared with
It is small；In turn, for close grain, PhkNumeric distribution than more uniform, therefore, contrast is larger.Its computational methods is：
The degree of correlation return pixel in entire image be expected adjacent pixel be how related metric.Describe in matrix and arrange
Or the gray scale similarity between row.It is the measurement of gray scale linear relationship.Its computational methods is：
Wherein, μ refers to pixel average, σhRefer to the standard of longitudinal pixel
Difference, σkRefer to the standard deviation of horizontal pixel.
Energy is the measurement to gradation of image distributing homogeneity, works as PhkWhen distribution is relatively concentrated, energy value is larger；Conversely, then
Energy value is smaller.Its computational methods is：
Homogeney returns to body of the Elemental redistribution to diagonal tightness degree, mainly local homogeneity in gray level co-occurrence matrixes
It is existing.Its computational methods is：
If image does not have any texture, entropy is close to zero.If image is full of close grain, as P in gray level co-occurrence matrixeshkNumber
Value be more or less the same and be distributed than it is more uniform when, then entropy is larger；Conversely, PhkWhen numerical value is relatively concentrated, then entropy is smaller.The meter of entropy
Calculating formula is：
Source images be f (x, y), x=1,2 ..., M；Y=1,2 ..., N, its gray level are L.Its Gray Level-Gradient symbiosis
The extracting method of matrix is：The gradient image g (x, y) of original image is extracted with Sobel operators or other gradient operators, gradient
Image carries out gray level discretization, if number of grayscale levels is Lg, new gray scale is
In formula
After conversion, gradient image is G (x, y), x=1,2 ..., M；Y=1,2 ..., N, its gray level are Lg。
Gray Level-Gradient Co-occurrence Matrix is { Hij, i=0,1 ..., L-1；J=0,1 ..., Lg- 1 }, wherein, HijIt is defined as
The number of element in set { (x, y) | f (x, y)=i, G (x, y)=j }.WillNormalized is done, is obtainedWherein
Using gray scale-gradient co-occurrence matrix, textural characteristics statistic can be calculated：
Small gradient advantage
Big gradient advantage
Intensity profile inhomogeneities
Gradient nonunf ormity
Energy
Gray scale is averaged
Gradient is averaged
Gray scale mean square deviation
Gradient mean square deviation
It is related
Gray level entropy
Gradient entropy
The entropy of mixing
Inertia
Unfavourable balance away from
Finally give, 6 Tamura features, 5 gray level co-occurrence matrixes features and 15 Gray Level-Gradient Co-occurrence Matrix spies
Levy, altogether 26 characteristic parameters.
It is to the specific method of characteristic vector dimensionality reduction,
C-1, defect characteristic individual UVR exposure：Parameter to characteristic vector is arranged in certain sequence, each base of chromosome
Because of the characteristic item of the corresponding order of correspondence, i.e., when some gene in chromosome is " 1 ", represent the corresponding characteristic item quilt of the gene
Choose；Otherwise for " 0 "；Wherein, chromosome is represented using binary character string, and its allele is made up of { 0,1 }, due to altogether
The characteristic parameter of the cable partial discharge signal graph of extraction is 26 dimensions, so the length of chromosome L=32.
C-2, the genetic manipulation of defect characteristic optimization：Including Selecting operation, crossing operation, mutation operator, end condition.
Selecting operation is to sort to select the male parent replicated using individual adaptation degree size, is specially：
Calculate each chromosome viAdaptive value eval (vi), wherein i=1,2 ... n；
Calculate total adaptive value of colony
Calculate the select probability p of each chromosomei=eval (vi)/F；
Calculate the accumulated probability of each chromosome
Producing one, [[0,1] inner random number r, uses q successively in intervaliCompared with r, first there is qi>=r
Body i is selected as the male parent replicated, repeats the step, untill the individual amount required for individual i number is met.
Two homologues are recombinated by mating, new chromosome are formed, so as to produce new individual.Here use
Single-point crossover operator, according to set crossover probability corpse.Crossing operation is specially：Two individuals are selected to make father from population
Generation, the individual that every a pair are mutually paired, it is crosspoint to be randomly provided the position after a certain locus except first position,
Shared L-1=25 possible cross-point locations.
Mutation operator is specially：Using basic bit mutation method, the mutation probability P to individual UVR exposure string to setmAt random
Genic value on a certain position specified or several locus does mutation operator, i.e., the genic value at each specified change point is taken
Inverse operation, gene is become " 1 " by " 0 ", or becomes " 0 " by " 1 ", so as to produce new individual.
End condition is specially the algebraically T=100 of heredity.
C-3, defect characteristic optimum choice：Obtaining optimal chromosome binary coding using genetic algorithm is：
(0110001101011101001110111), from the coded sequence of feature, it is characterized in Isosorbide-5-Nitrae not to be selected, 5,6,
9,11,15,17,18 and 22, the information of this 10 features is redundancy in other words, and the feature that redundancy is set is deleted, selection
The 2nd, 3,7,8,10,12,13,14,16,19,20,21,23,24 and 25, this 15 features for optimization after characteristic parameter.
A sample set is given, is a dimension sample in-real vector, is correspondence output real number, SVM basic thought
It is, input sample DUAL PROBLEMS OF VECTOR MAPPING to N-dimensional feature space, optimal decision function to be constructed in feature space by kernel function,
In formula, ω=(ω1, ω2, ω3...ωN)TFor linear weight value vector, b is threshold value.
The selection of kernel function is the important component in SVMs.Geo-nuclear tracin4 two moulds of the sample input space
The evaluation of the kernel function of formula, to replace the inner product of two points in original higher-dimension Hilbert spaces to calculate.Regression forecasting is carried out in SVM
When, sample vector from former space utilization kernel function can be mapped to high dimension linear space by kernel function first, then in High-dimensional Linear
Best approximation function is constructed in space.Common kernel function has：Linear kernel function：K (x, x ')=xx ', Sigmoid core letters
Number：K (x, x ')=1/1+exp [β (xx ')+b], gaussian kernel function：K (x, x ')=exp [(- | | x-x ' | |2)/(2λ2)]。
Construction is adapted to the multinuclear mixed kernel function of different sample data type inputs：
In formula, M is the number of core, σm>=0,1≤m≤M0 coefficient.
MSVM can be inputted for the sample of different type of action, and different kernel functions are respectively adopted and are mapped, fully
Using the information knowledge of various types of input variable, the precision of prediction of model is improved.
The selection of SVM model parameters has important influence to precision of prediction.Carried out herein using ant colony algorithm in MSVM
Threshold value, the number M0 of core, parameter, optimize.
Artificial bee colony is divided into artificial bee colony algorithm and leads honeybee, follow honeybee and the investigation class of honeybee three, each time search procedure
In, lead honeybee and follow honeybee to be successively to exploit food source, that is, find optimal solution, and it is to see whether to be absorbed in part most to investigate honeybee
It is excellent, other possible food sources are randomly searched for if local optimum is absorbed in.Each food source represents problem one may solution, food
The quality i.e. fitness value fit that the nectar amount correspondence of material resource is accordingly solvedi。
Fitness function is represented with the mean absolute errors returned of SVM after each iteration：
N is sample number in formula,For sample predictions value, YiFor sample actual value.
In artificial bee colony algorithm search procedure, it is necessary first to initialize, including determination population number, maximum iteration
M, control parameter L and determination search space are the scope of solution, the random generation initial solution x in search spacei(i=1,2, L, S),
S is food source number, is herein number M0, parameter lambda, the σ of threshold value b, core in 4, i.e. MSVMm, each solve xiIt is one 4 dimension
Vector.After initialization, whole population will carry out leading honeybee, follow honeybee and investigating the repetitive cycling of honeybee search process, Zhi Daoda
To maximum iteration M or error permissible value ε.
In the search procedure incipient stage, honeybee is each led to produce a new explanation by formula (1),
In formula, k ∈ { 1,2, L, S }, j ∈ { 1,2, L, D }, and k ≠ i；For the random number between [- 1,1]；
Calculate the fitness value fit of new explanationiAnd it is evaluated, if the fit of new explanationiLess than fiti-1, then honeybee is led to remember newly
Solution forgets about old solution；Conversely, it will retain old solution；
It is all lead honeybee to complete search process after, lead honeybee to recruit area's jive the information and fit of solutioni
Information follows honeybee to calculate the select probability each solved according to formula (2) with following honeybee to share,
A number is randomly generated in [0,1] interval, if the probable value of solution is more than the random number, honeybee is followed by formula
(3) new explanation is produced, and examines the fit of new explanationiIf, the fit of new explanationiCompare fiti-1It is small, then follow honeybee to remember that new explanation is forgotten about
Old solution；Conversely, it will retain old solution
It is all follow honeybee to complete search process after, if a solution is circulated still not by further more by L time
Newly, then be considered as this solution and be absorbed in local optimum, the food source will be rejected, then this food source is corresponding leads honeybee to change into one
Individual investigation honeybee；Search bee produces a new food source by (4) formula and replaces it,
xij=xminj+rand(0,1)(xmaxj-xminj) (4), wherein j ∈ { 1,2, L, D },
Return leads honeybee search procedure, starts repetitive cycling.
Artificial bee colony algorithm general greediness that carries out when evaluating food source selects to carry out by formula (5).
Artificial bee colony algorithm is exactly, by cyclic search, to eventually find optimal food source or optimal solution.
By after the Partial discharge signal figure of cable carries out feature extraction and dimensionality reduction, SVM, MSVM and ABC-MSVM are brought into respectively
Model, different defect models are as shown in Figure 1 in the recognition accuracy scope of each model.Wherein strip is average recognition accuracy, I
Molded line describes the scope of each discrimination, from Fig. 1, and we are it can be found that for various defect types, ABC-MSVM average knowledge
Rate is not respectively 85.58%, 89.65%, 88.17%, 93.96%, and only in fault type 1, the average recognition rate than MSVM is low
2.56%, other kinds of average recognition rate all apparently higher than other two kinds of identification types, especially fault types 2, compares MSVM
Average recognition rate it is high by 25.64%.
It is as shown in the table for the overall recognition accuracy and run time of each model.ABC-MSVM is averaged to 4 kinds of defect types
Discrimination reaches 89.34%, and SVM and MSVM are respectively 61.12% and 76.31%.
Above-described embodiment, has been carried out further to the purpose of the present invention, technical scheme and beneficial effect
Describe in detail, should be understood that the embodiment that the foregoing is only the present invention, be not intended to limit the present invention
Protection domain, within the spirit and principles of the invention, any modification, equivalent substitution and improvements done etc. all should be included
Within protection scope of the present invention.
Claims (9)
1. a kind of cable insulation defect state appraisal procedure based on high frequency Partial discharge signal figure, it is characterised in that including following step
Suddenly,
A, the high frequency Partial discharge signal figure for obtaining cable；
B, the textural characteristics for extracting high frequency Partial discharge signal figure, gray level co-occurrence matrixes feature, Gray Level-Gradient Co-occurrence Matrix feature, will
It is used as characteristic vector；
C, using genetic algorithm to characteristic vector carry out dimensionality reduction；
D, the characteristic vector after dimensionality reduction brought into the multi-kernel support vector machine of bee colony optimization and carry out defect recognition；
E, progress Classifcation of flaws.
2. a kind of cable insulation defect state appraisal procedure based on high frequency Partial discharge signal figure according to claim 1, its
It is characterised by, the textural characteristics include roughness, contrast, directionality, line similitude, systematicness and rough degree.
3. a kind of cable insulation defect state appraisal procedure based on high frequency Partial discharge signal figure according to claim 2, its
It is characterised by, the extracting method of the roughness is：
To Image Mean Filtering, average intensity value is asked for, filter window size is 2k×2k, wherein, κ is filter window parameter；
For each pixel (i, j), it is calculated respectively in mean intensity difference S (i, j) both horizontally and vertically；
For each pixel (i, j), the filter window yardstick corresponding to maximum mean intensity difference is regard as best scale
Sbest；
Calculate the S in entire imagebestAnd ask it to be averagely worth to roughness.
4. a kind of cable insulation defect state appraisal procedure based on high frequency Partial discharge signal figure according to claim 2, its
It is characterised by, the extracting method of the directionality is：
Calculate the gradient vector of each pixel, including both horizontally and vertically on variable quantity；
One histogram is constructed according to gradient vector, the acuity for calculating peak value in histogram obtains the directionality of image.
5. a kind of cable insulation defect state appraisal procedure based on high frequency Partial discharge signal figure according to claim 1, its
It is characterised by, the extracting method of the gray level co-occurrence matrixes feature is：
In entire image, count each (h, k) value appearance number of times, then by they be normalized to occur probability Phk,
And constitute gray level co-occurrence matrixes [Phk]N×N, wherein N is the gray level of image, and k, h are gray scale,；
Construct the co-occurrence matrix in 4 directions, including M (0,1), M (- 1,1), M (- 1,0), M (- 1, -1)；
The contrasts of 4 co-occurrence matrixs, the degree of correlation, energy, homogeney, entropy are calculated respectively.
6. a kind of cable insulation defect state appraisal procedure based on high frequency Partial discharge signal figure according to claim 1, its
It is characterised by, the extracting method of the Gray Level-Gradient Co-occurrence Matrix feature is：
The gradient image g (x, y) of original image is extracted with gradient operator；
Gray level sliding-model control is carried out to gradient image and obtains new gradient image G (x, y), x=1,2 ..., M；Y=1,
2 ..., N, its gray level is Lg；
Gray Level-Gradient Co-occurrence Matrix is { Hij, i=0,1 ..., L-1；J=0,1 ..., Lg- 1 }, wherein, HijFor set (x,
Y) | f (x, y)=i, G (x, y)=j in element number；F (x, y) is source images, x=1,2,3...M, y=1,2,3...N；
L is the gray level of source images；
It is rightNormalized is done, is obtainedWherein
Small gradient advantage, big gradient advantage, intensity profile inhomogeneities, gradient nonunf ormity, energy, gray scale is calculated to put down
, gradient is average, gray scale mean square deviation, gradient mean square deviation, correlation, gray level entropy, gradient entropy, the entropy of mixing, inertia, unfavourable balance away from.
7. a kind of cable insulation defect state appraisal procedure based on high frequency Partial discharge signal figure according to claim 1, its
It is characterised by, the step C specific methods are,
C-1, defect characteristic individual UVR exposure：Parameter to characteristic vector is arranged in certain sequence, each gene pairs of chromosome
The characteristic item of corresponding order is answered, wherein, chromosome is represented using binary character string, and its allele is made up of { 0,1 }；
C-2, the genetic manipulation of defect characteristic optimization：Including Selecting operation, crossing operation, mutation operator, end condition；Wherein,
Selecting operation is specially：
Calculate each chromosome viAdaptive value eval (vi), wherein i=1,2 ... n；
Calculate total adaptive value of colony
Calculate the select probability p of each chromosomei=eval (vi)/F；
Calculate the accumulated probability of each chromosome
Producing one, [[0,1] inner random number r, uses q successively in intervaliCompared with r, first there is qi>=r individual i quilts
The male parent of duplication is selected as, the step is repeated, untill the individual amount required for individual i number is met；
Crossing operation is specially：Two individuals are selected to make parent from population, the individual being mutually paired to every a pair is randomly provided
Except the position after a certain locus of first position is crosspoint；
Mutation operator is specially：Using basic bit mutation method, the mutation probability P to individual UVR exposure string to setmIt is randomly assigned
Genic value on a certain position or several locus does mutation operator, so as to produce new individual；
End condition is specially the algebraically T=100 of heredity；
C-3, defect characteristic optimum choice：Optimal chromosome binary coding, the spy that redundancy is set are obtained using genetic algorithm
Levy deletion, the characteristic parameter after as optimizing.
8. a kind of cable insulation defect state appraisal procedure based on high frequency Partial discharge signal figure according to claim 1, its
Be characterised by, the multi-kernel support vector machine use bee colony optimize method for：
Population is initialized, including determines population number, maximum iteration M, control parameter L and determines search space, and in search
Random generation initial solution x in spacei, wherein, i=1,2,3...S, S are food source number；
After initialization, carry out leading honeybee, follow honeybee and investigating the repetitive cycling of honeybee search process in whole population, until reaching
Maximum iteration M or error permissible value ε；
In the search procedure incipient stage, honeybee is each led to produce a new explanation by formula (1),
In formula, k ∈ { 1,2, L, S }, j ∈ { 1,2, L, D }, and k ≠ i；For the random number between [- 1,1]；
Calculate the fitness value fit of new explanationiAnd it is evaluated, if the fit of new explanationiBetter than old solution fiti-1, then honeybee is led to remember newly
Solution forgets about old solution；Conversely, it will retain old solution；
It is all lead honeybee to complete search process after, follow honeybee to calculate the select probability that each solves according to formula (2),
<mrow>
<msub>
<mi>p</mi>
<mi>i</mi>
</msub>
<mo>=</mo>
<mi>f</mi>
<mi>i</mi>
<mi>t</mi>
<mo>/</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>k</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>S</mi>
</munderover>
<msub>
<mi>fit</mi>
<mi>k</mi>
</msub>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>2</mn>
<mo>)</mo>
</mrow>
<mo>;</mo>
</mrow>
A number is randomly generated in [0,1] interval, if the probable value of solution is more than the random number, follows honeybee to be produced by formula (3)
A raw new explanation, and examine the fit of new explanationiIf, the fit of new explanationiCompare fiti-1It is good, then follow honeybee to remember that old solution is forgotten about in new explanation；
Conversely, it will retain old solution,
<mrow>
<msub>
<mi>fit</mi>
<mi>i</mi>
</msub>
<mo>=</mo>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<mfrac>
<mn>1</mn>
<mrow>
<mn>1</mn>
<mo>+</mo>
<msub>
<mi>f</mi>
<mi>i</mi>
</msub>
</mrow>
</mfrac>
<mo>,</mo>
</mrow>
</mtd>
<mtd>
<mrow>
<msub>
<mi>f</mi>
<mi>i</mi>
</msub>
<mo>></mo>
<mn>0</mn>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mn>1</mn>
<mo>+</mo>
<mi>a</mi>
<mi>b</mi>
<mi>s</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>f</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>,</mo>
</mrow>
</mtd>
<mtd>
<mrow>
<msub>
<mi>f</mi>
<mi>i</mi>
</msub>
<mo>&le;</mo>
<mn>0</mn>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>3</mn>
<mo>)</mo>
</mrow>
<mo>;</mo>
</mrow>
It is all follow honeybee to complete search process after, if a solution is not still updated further by L circulation, that
It is considered as this solution and is absorbed in local optimum, the food source will be rejected, then this food source is corresponding leads honeybee to change into one to detect
Look into honeybee；Search bee produces a new food source by (4) formula and replaces it,
xij=xmin j+rand(0,1)(xmax j-xmin j) (4), wherein j ∈ { 1,2, L, D },
Return leads honeybee search procedure, starts repetitive cycling and eventually finds optimal food source or optimal solution.
9. a kind of cable insulation defect state appraisal procedure based on high frequency Partial discharge signal figure according to claim 8, its
It is characterised by, greedy selection is carried out as the following formula when evaluating food source,
<mrow>
<msub>
<mi>v</mi>
<mi>i</mi>
</msub>
<mo>=</mo>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<msub>
<mi>v</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
</mrow>
</mtd>
<mtd>
<mrow>
<mi>f</mi>
<mi>i</mi>
<mi>t</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>v</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>></mo>
<mi>f</mi>
<mi>i</mi>
<mi>t</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
</mrow>
</mtd>
<mtd>
<mrow>
<mi>f</mi>
<mi>i</mi>
<mi>t</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>v</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>&le;</mo>
<mi>f</mi>
<mi>i</mi>
<mi>t</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>.</mo>
</mrow>
3
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710342214.0A CN107067041A (en) | 2017-05-16 | 2017-05-16 | A kind of cable insulation defect state appraisal procedure based on high frequency Partial discharge signal figure |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710342214.0A CN107067041A (en) | 2017-05-16 | 2017-05-16 | A kind of cable insulation defect state appraisal procedure based on high frequency Partial discharge signal figure |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107067041A true CN107067041A (en) | 2017-08-18 |
Family
ID=59609410
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710342214.0A Pending CN107067041A (en) | 2017-05-16 | 2017-05-16 | A kind of cable insulation defect state appraisal procedure based on high frequency Partial discharge signal figure |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107067041A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107991581A (en) * | 2017-12-05 | 2018-05-04 | 国网江西省电力有限公司上饶供电分公司 | Distribution cable intelligent fault based on panoramic state information perceives decision-making platform |
| CN109102508A (en) * | 2018-08-31 | 2018-12-28 | 上海交通大学 | The recognition methods of ac cable insulation defect based on shelf depreciation characteristics of image |
| CN114359112A (en) * | 2022-03-11 | 2022-04-15 | 领伟创新智能系统(浙江)有限公司 | Surface roughness evaluation method based on Hoyer coefficient of machined surface image |
| CN116448019A (en) * | 2023-06-14 | 2023-07-18 | 山西首科工程质量检测有限公司 | Intelligent detection device and method for quality flatness of building energy-saving engineering |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106443379A (en) * | 2016-09-22 | 2017-02-22 | 广州供电局有限公司 | Transformer partial discharge fault type identifying method and transformer partial discharge fault type identifying device |
-
2017
- 2017-05-16 CN CN201710342214.0A patent/CN107067041A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106443379A (en) * | 2016-09-22 | 2017-02-22 | 广州供电局有限公司 | Transformer partial discharge fault type identifying method and transformer partial discharge fault type identifying device |
Non-Patent Citations (5)
| Title |
|---|
| 吴一全 等，: ""基于蜂群优化多核支持向量机的淡水鱼种类识别"", 《农业工程学报》 * |
| 汤勃 等，: ""基于遗传算法的带钢表面缺陷特征降维优化选择"", 《钢铁研究学报》 * |
| 王顺杰 等: ""Tamura 纹理特征在水下目标分类中的应用"", 《应用声学》 * |
| 赵磊 等: ""基于GLCM和LBP的局部放电灰度图像特征提取"", 《电测与仪表》 * |
| 邬长安，: "《数字图像处理、分析及应用》", 30 September 2019 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107991581A (en) * | 2017-12-05 | 2018-05-04 | 国网江西省电力有限公司上饶供电分公司 | Distribution cable intelligent fault based on panoramic state information perceives decision-making platform |
| CN109102508A (en) * | 2018-08-31 | 2018-12-28 | 上海交通大学 | The recognition methods of ac cable insulation defect based on shelf depreciation characteristics of image |
| CN109102508B (en) * | 2018-08-31 | 2022-02-25 | 上海交通大学 | Method for identifying insulation defects of alternating current cable based on partial discharge image characteristics |
| CN114359112A (en) * | 2022-03-11 | 2022-04-15 | 领伟创新智能系统(浙江)有限公司 | Surface roughness evaluation method based on Hoyer coefficient of machined surface image |
| CN114359112B (en) * | 2022-03-11 | 2022-06-14 | 领伟创新智能系统(浙江)有限公司 | Surface roughness evaluation method based on Hoyer statistic value of machined surface image |
| CN116448019A (en) * | 2023-06-14 | 2023-07-18 | 山西首科工程质量检测有限公司 | Intelligent detection device and method for quality flatness of building energy-saving engineering |
| CN116448019B (en) * | 2023-06-14 | 2023-08-25 | 山西首科工程质量检测有限公司 | Intelligent detection device and method for quality flatness of building energy-saving engineering |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107067041A (en) | A kind of cable insulation defect state appraisal procedure based on high frequency Partial discharge signal figure | |
| Balakrishnan et al. | Comparative performance of the FSCL neural net and K-means algorithm for market segmentation | |
| Shintemirov et al. | Power transformer fault classification based on dissolved gas analysis by implementing bootstrap and genetic programming | |
| Onnela et al. | Taxonomies of networks from community structure | |
| CN110135319A (en) | A kind of anomaly detection method and its system | |
| Khotanzad et al. | Feature selection for texture recognition based on image synthesis | |
| CN105787492B (en) | Three value mode texture feature extracting methods of part based on mean value sampling | |
| CN110991549A (en) | Countermeasure sample generation method and system for image data | |
| CN108133225A (en) | A kind of icing flashover fault early warning method based on support vector machines | |
| Wang | Fuzzy clustering analysis by using genetic algorithm | |
| CN110287975A (en) | Flotation dosing abnormity detection method based on NSST morphological characteristics and depth KELM | |
| CN111079805A (en) | Abnormal image detection method combining attention mechanism and information entropy minimization | |
| CN110297469A (en) | The production line fault judgment method of Ensemble feature selection algorithm based on resampling | |
| Habib et al. | Automated textile defect classification by Bayesian classifier based on statistical features | |
| CN114580934A (en) | Early warning method for food detection data risk based on unsupervised anomaly detection | |
| CN110349050B (en) | Intelligent electricity stealing criterion method and device based on power grid parameter key feature extraction | |
| CN117036760A (en) | Multi-view clustering model implementation method based on graph comparison learning | |
| Hendrawan et al. | Neural-discrete hungry roach infestation optimization to select informative textural features for determining water content of cultured Sunagoke moss | |
| CN110533341A (en) | A kind of Livable City evaluation method based on BP neural network | |
| Chen | Correlation analysis of financial indicators and stock price fluctuations based on artificial intelligence system | |
| Salem et al. | New assessment criteria for clustering algorithms | |
| Fewster et al. | Similarity indices for spatia I ecological data | |
| WO2021173715A1 (en) | Methods and systems to train neural networks | |
| CN111626329A (en) | Insulation pipe bus fault diagnosis method based on LDA optimization multi-scale texture features | |
| Goulermas et al. | Feature-based stereo matching via coevolution of epipolar subproblems |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170818 |