CN103558529B - A kind of mode identification method of three-phase cartridge type supertension GIS partial discharge altogether - Google Patents

Abstract

The invention discloses the mode identification method of a kind of three-phase cartridge type supertension GIS partial discharge altogether, including step: use ultra-high-frequency detection three-phase cartridge type GIS partial discharge altogether, utilize type UHF sensor that local discharge signal is sampled；Utilize the small echo threshold values filtering method improved that the local discharge signal collected is carried out denoising Processing, obtain real local discharge signal；By extracting the characteristic parameter of sampled signal based on phase analysis pattern algorithm；The feature space utilizing the core principle component analysis method improved to form characteristic parameter carries out dimension-reduction treatment, obtains the characteristic parameter matrix after dimensionality reduction；Utilize k nearest neighbor classification method based on bunch thought that GIS insulation defect type is carried out pattern recognition.The mode identification method of three-phase of the present invention altogether cartridge type supertension GIS partial discharge, can overcome in prior art that function is few, the scope of application is little and the defect such as poor accuracy, to realize the advantage that function is many, applied widely and accuracy is good.

Description

A kind of mode identification method of three-phase cartridge type supertension GIS partial discharge altogether

Technical field

The present invention relates to high-voltage electrical discharges identification technical field, in particular it relates to a kind of three-phase cartridge type supertension GIS office altogether The mode identification method of portion's electric discharge.

Background technology

Gas-insulated switchgear (Gas lnsulated Switchgear is called for short GIS) is the weight in extra-high voltage grid Want one of component devices, it by the chopper in a transformer station, current transformer, voltage transformer, spark gap, keep apart Pass, earthed switch, bus, cable termination, inlet-outlet sleeve etc. are finally concentrated in being attached separately between each self sealss after optimizing design It is assembled in one and is filled with SF6 as in the monolithic case of dielectric.

The defect of GIS internal influence dielectric performance mainly has: loose contact between serious setup error, conductor, High-pressure conductor outthrust, fixing microgranule, defects of insulator, steam etc..

The development of GIS tends to three-phase cylinder, Composite and intellectuality altogether, owing to achieving miniaturization, can enter in factory Transport to scene with the form at interval after row final assembly and pass the test, therefore can shorten in-site installation duration, simultaneously reliability Get back raising.

Three-phase altogether cartridge type GIS is in internal structure, and the aspect such as Electric Field Distribution has the most different from close coupled type GIS, existing Technical research mainly collects at close coupled type GIS, but for the research of three-phase cartridge type supertension GIS partial discharge detection pattern recognition altogether Less.

During realizing the present invention, inventor find prior art at least exists function is few, the scope of application is little and The defects such as poor accuracy.

Summary of the invention

It is an object of the invention to, for the problems referred to above, propose the mould of a kind of three-phase cartridge type supertension GIS partial discharge altogether Formula recognition methods, to realize the advantage that function is many, applied widely and accuracy is good.

For achieving the above object, the technical solution used in the present invention is: a kind of three-phase cartridge type supertension GIS partial discharge altogether Mode identification method, comprise the following steps:

Step 1: use ultra-high-frequency detection three-phase cartridge type GIS partial discharge altogether, utilize type UHF sensor to local discharge signal Sampling；

Step 2: utilize the small echo threshold values filtering method improved that the local discharge signal collected is carried out denoising Processing, obtain Real local discharge signal；

Step 3: by extracting the characteristic parameter of sampled signal based on phase analysis pattern algorithm；

Step 4: the feature space utilizing the core principle component analysis method improved to form characteristic parameter carries out dimension-reduction treatment, Characteristic parameter matrix after dimensionality reduction；

Step 5: utilize k nearest neighbor classification method based on bunch thought that GIS insulation defect type is carried out pattern recognition.

Further, in step 2, the small echo threshold values filtering method that described utilization the improves local discharge signal to collecting Carry out the operation of denoising Processing, concrete employing adaptive thresholding value calculating method；This adaptive thresholding value calculating method is as follows:

$T_j=\frac{\mathrm{Median}\left(\right|C_{j,k}\left|\right)\sqrt{2\ln \left(N_j\right)}}{0.6745}\mathrm{\αexp}(-\frac{1}{{\mathrm{\β}}_j});$

Wherein, j is yardstick, N_jFor the number of wavelet coefficient on this yardstick, Median (| C_j,k|) it is all little on this yardstick The median of wave system number, α is referred to as the signal to noise ratio factor, be signal signal to noise ratio threshold values calculate in embodiment, β_jBe referred to as yardstick because of Son, is that the maximum of wavelet coefficient on yardstick corrects the estimation difference that sample sequence length difference causes, T_jFor the threshold values calculated.

Further, in step 3, described characteristic parameter includes that degree of skewness Sk, steepness Ku, local peaks are counted Pe, mutually Correlation coefficient Cc and discharge factor Q.

Further, described degree of skewness Sk particularly as follows:

$\mathrm{Sk}=\underset{i=1}{\overset{W}{\mathrm{\Σ}}}{(x_i-\mathrm{\μ})}^3\·p_i\mathrm{\Δx}/{\mathrm{\σ}}^3;$

In above formula, w is the phase window number in the half period；Xi is the phase place of i-th phase window；

$p_i=y_i/\underset{i=1}{\overset{W}{\mathrm{\Σ}}}{y}_i,$

Wherein, y_iIt is the vertical coordinate of spectrogram, represents Apparent discharge magnitude q or discharge time n；Parameter μ represents the office collected The position of electric discharge collection of illustrative plates center, portion, σ represents the precipitous situation that the center axis of symmetry of collection of illustrative plates is embodied, and Δ x is then about office Portion's electric discharge one of collection of illustrative plates to be evenly distributed relevant parameter,Certain phase place corresponding to point in collection of illustrative plates；

Degree of skewness Sk reflection spectral shape is relative to the left and right deflection situation of normal distribution: Sk=0 illustrates that this spectral shape is left Right symmetry；Sk > 0 illustrate this spectrogram relative to normal distribution shape to left avertence；Sk < 0 illustrates that this spectrogram is relative to normal distribution shape Shape is to right avertence.

Further, described steepness Ku particularly as follows:

$\mathrm{Ku}=[\underset{i=1}{\overset{W}{\mathrm{\&Sigma;}}}{(x_i-\mathrm{\&mu;})}^4{p}_i\mathrm{\&Delta;x}/{\mathrm{\&sigma;}}^4]-3;$

In above formula, w is the phase window number in the half period；Xi is the phase place of i-th phase window；

$p_i=y_i/\underset{i=1}{\overset{W}{\mathrm{\&Sigma;}}}{y}_i,$

Wherein, y_iIt is the vertical coordinate of spectrogram, represents Apparent discharge magnitude q or discharge time n；Parameter μ represents the office collected The position of electric discharge collection of illustrative plates center, portion, σ represents the precipitous situation that the center axis of symmetry of collection of illustrative plates is embodied, and Δ x is then about office Portion's electric discharge one of collection of illustrative plates to be evenly distributed relevant parameter,Certain phase place corresponding to point in collection of illustrative plates；

Steepness Ku is for describing the profiles versus of certain shape in the projection degree of normal distribution shape: normal distribution Steepness Ku is 0；If Ku > 0, then illustrate that this spectrogram profile is more precipitous than normal distribution profile；If Ku < 0, then explanation should Spectrogram profile is more smooth than normal distribution profile.

Further, described local peaks is counted Pe, and local peaks is counted for describing the number of local peaks on spectrogram profile；? Profile pointWhether place has local peaks, needs to judge with following difference equation:

(y_i-y_i-1) > 0, (y_i+1-y_i) < 0；

The halved phase window of phase shaft is the most, and local peaks is counted the biggest.

Further, described cross-correlation coefficient Cc particularly as follows:

$\mathrm{Cc}=\frac{\underset{i=1}{\overset{W}{\mathrm{\&Sigma;}}}{q}_i^{+}{q}_i^{-}-\left(\underset{i=1}{\overset{W}{\mathrm{\&Sigma;}}}{q}_i^{+}\underset{i=1}{\overset{W}{\mathrm{\&Sigma;}}}{q}_i^{-}\right)/W}{\sqrt{[\underset{i=1}{\overset{W}{\mathrm{\&Sigma;}}}{\left(q_i^{+}\right)}^2-{\left(\underset{i=1}{\overset{W}{\mathrm{\&Sigma;}}}{q}_i^{-}\right)}^2/W][\underset{i=1}{\overset{W}{\mathrm{\&Sigma;}}}{\left(q_i^{-}\right)}^2-{\left(\underset{i=1}{\overset{W}{\mathrm{\&Sigma;}}}{q}_i^{-}\right)}^2]/W}};$

In formula,The discharge capacity in phase window i, subscript "+", "-" is corresponding to the positive and negative semiaxis of spectrogram；C reacts The strong and weak dependency with PHASE DISTRIBUTION of electric discharge in positive and negative half period, cross-correlation coefficient Cc means close to 1Spectrogram is just The profile of negative half period is quite similar；Cc close to 0,Spectrogram profile difference is huge.

Further, described discharge factor Q particularly as follows:

$Q=\frac{\underset{i=1}{\overset{w}{\mathrm{\&Sigma;}}}{n}_i^{-}{q}_i^{-}}{\underset{i=1}{\overset{w}{\mathrm{\&Sigma;}}}{n}_i^{-}}/\frac{\underset{i=1}{\overset{w}{\mathrm{\&Sigma;}}}{n}_i^{+}{q}_i^{+}}{\underset{i=1}{\overset{w}{\mathrm{\&Sigma;}}}{n}_i^{+}};$

In formula,The electric discharge repetitive rate in phase window i, subscript "+", "-" corresponds toSpectrogram positive and negative Half cycle.

Further, in step 4, the core principle component analysis method that the core principle component analysis method of described improvement is i.e. improved, institute The kernel function of sampling is:

$k(x_i,x_j)={(<x_i,x_j>+a)}^b\exp \left(\frac{-{\left|\right|x_i-x_j\left|\right|}^2}{2{\mathrm{\&sigma;}}^2}\right);$

Wherein, (a ∈ R, b ∈ N, σ > 0), parameter a, the selection of b with σ are true according to the numerical values recited of element in eigenmatrix Fixed, parameter σ is for controlling the radial effect scope of kernel function；x_iAnd x_jRepresent different sample vectors, x_i, x_jRepresentative sample The vector product of this vector, the span of R representation vector is at set of real numbers, and N represents set of integers, k (x_i,x_j) represent combine multinomial The advantage of kernel function and gaussian kernel function and the new kernel function that obtains.

Further, in steps of 5, the algorithm of described k nearest neighbor classification method specifically includes:

First all Partial Discharge Datas are carried out pretreatment and become space vector by Step1: in training set；

The all signal datas belonging to this classification are carried out Similarity Measure two-by-two by Step2: from the beginning of first class, Set a minimum threshold, according to statistics obtain similarity close one by one bunch；

Step3: for each bunch, all signal datas therein are merged, then calculates its center vector；This Outward, calculating bunch number/classification sum, this value represents this bunch contribution coefficient to this class；

Step4: after new text arrives, carries out pretreatment and obtains its vector space；

Step5: by the center vector computed range of every cluster that the space vector of new text is generated with Step3, by this A little distance contribution coefficients with corresponding bunch are multiplied, and belong to the same category of bunch of results added calculated, and compare and obtain maximum that One classification is exactly typical defect shelf depreciation generic to be sorted.

The mode identification method of the three-phase of various embodiments of the present invention cartridge type supertension GIS partial discharge altogether, owing to including step Rapid: to use ultra-high-frequency detection three-phase cartridge type GIS partial discharge altogether, utilize type UHF sensor that local discharge signal is sampled；Utilization changes The small echo threshold values filtering method entered carries out denoising Processing to the local discharge signal collected, and obtains real local discharge signal； By extracting the characteristic parameter of sampled signal based on phase analysis pattern algorithm；Utilize the core principle component analysis method improved to feature The feature space of parameter composition carries out dimension-reduction treatment, obtains the characteristic parameter matrix after dimensionality reduction；Utilize k nearest neighbor based on bunch thought Classification method carries out pattern recognition to GIS insulation defect type；The defect of prior art can be overcome, improve three-phase cartridge type superelevation altogether The accuracy of pressure GIS partial discharge detection pattern recognition；Such that it is able to overcome that function in prior art is few, the scope of application is little and accurate The really defect of property difference, to realize the advantage that function is many, applied widely and accuracy is good.

Other features and advantages of the present invention will illustrate in the following description, and, partly become from description Obtain it is clear that or understand by implementing the present invention.

Below by drawings and Examples, technical scheme is described in further detail.

Accompanying drawing explanation

Accompanying drawing is for providing a further understanding of the present invention, and constitutes a part for description, with the reality of the present invention Execute example together for explaining the present invention, be not intended that limitation of the present invention.In the accompanying drawings:

Fig. 1 is that three-phase of the present invention is total to gas-insulated combination electricity in the mode identification method of cartridge type supertension GIS partial discharge The structural representation of device local discharge test device；

Fig. 2 is the schematic flow sheet of the three-phase of the present invention mode identification method of cartridge type supertension GIS partial discharge altogether；

Fig. 3 is that three-phase of the present invention is total to statistical distribution and Sk, Ku in the mode identification method of cartridge type supertension GIS partial discharge Relation schematic diagram；

Fig. 4 is the effect effect of the three-phase of the present invention mode identification method Kernel Function of cartridge type supertension GIS partial discharge altogether Really schematic diagram；

Fig. 5 be three-phase of the present invention altogether cartridge type supertension GIS partial discharge mode identification method in waveform pair before and after filtering Than figure.

In conjunction with accompanying drawing, in the embodiment of the present invention, reference is as follows:

1-water resistance；2-HT testing transformer；3-transformator；4-bushing；5-disk insulator.

Detailed description of the invention

Below in conjunction with accompanying drawing, the preferred embodiments of the present invention are illustrated, it will be appreciated that preferred reality described herein Execute example be merely to illustrate and explain the present invention, be not intended to limit the present invention.

For defect present in prior art, according to embodiments of the present invention, as Figure 1-Figure 5, it is provided that Yi Zhongsan It is total to the mode identification method of cartridge type supertension GIS partial discharge mutually.

As it is shown in figure 1, the gas that the mode identification method of the three-phase of the present embodiment cartridge type supertension GIS partial discharge altogether uses Body insulation in combined electric appliance local discharge test device, including disk insulator 5, is arranged on the bushing on disk insulator 5 4, water resistance 1, HT testing transformer 2, transformator 3 and the local discharge signal detector being connected with bushing 4 successively (PDSG)；Local discharge signal detector (PDSG) is also connected with disk insulator 5, water resistance 1 and the common port of bushing 4 By ground connection after the potentiometer that is made up of electric capacity.

This local discharge of gas-insulator switchgear assay device, mainly includes transformator 3, the dividing potential drop being made up of electric capacity Device, oscillograph, three-phase cartridge-type gas insulation in combined electric appliance (GIS), sensor and local discharge signal detector (PDSG) altogether；Logical Cross and be respectively provided with high-pressure conductor metallic projections, free metal microgranule in GIS, insulator surface fixes metal, insulate edema of the legs during pregnancy The defects such as gap, detect corresponding local discharge signal, carry out pattern recognition.

The mode identification method of the three-phase of the present embodiment cartridge type supertension GIS partial discharge altogether, comprises the following steps:

1) use hyperfrequency (UHF) detection three-phase cartridge type GIS partial discharge altogether, utilize type UHF sensor to shelf depreciation (PD) signal sampling；

In step 1), use hyperfrequency method (UHF) that three-phase cartridge type GIS partial discharge altogether is detected.Examine by UHF method When surveying GIS partial discharge, the position on power frequency voltage waveform can be occurred according to the spectral characteristic and electric discharge that record signal Identify different faults type.

In step 1), use actual three-phase cartridge type GIS altogether, shelf depreciation collection of illustrative plates under the conditions of typical defect is surveyed Amount.Wherein three-phase cartridge type GIS altogether should have professional high-tension switch gear enterprise to provide.The pressuring method of three-phase conductor is two to connect Ground, one connects high pressure, is provided only with a bushing in model.According to coaxial type GIS test model test result, setting has The Exemplary insulative defect of comparative.There is free metal microgranule in three-phase cartridge type GIS cavity altogether, there is fixing gold in insulator surface Metal particles, insulator exists void defects, and is placed in GIS analog by shelf depreciation physical model, carries out local and puts The measurement of electricity.

2) utilize the small echo threshold values filtering method improved that the local discharge signal collected is carried out denoising Processing, obtain true Local discharge signal；

In step 2) in, the small echo threshold values filtering algorithm of improvement, by median Median(Cj, computational methods k) are visible, The selection of threshold value is the closest with the length relation of analyzed signal.In the middle of reality application, do not ensure that useful signal exists Ratio and position that whole section of sampling number is shared in whole section of sample sequence are constant.Because the length of signal determines small echo After conversion on each yardstick number Nj of wavelet coefficient, it affects median, Median(Cj, value k), thus have impact on The size of threshold value.This impact can cause such result: same useful signal sequence u is comprised in one section of sample sequence s In the middle of, if s length different (i.e. the width of time observation window is different), the result after wavelet filter has larger difference. For comprising the s of the different length of same UHF PD signal, the result of application soft-threshold de-noising algorithm.Original signal samples frequency Rate is 20GHz, and the wider long sample sequence of time window is 50000 points, and the narrower short sample sequence of time window is 16000 points, two What individual sequence was the most complete contains useful UHF PD signal.It is clear that application soft-threshold algorithm filtered UHF PD signal, , there is obvious difference in the part that especially oscillatory extinction will terminate, both correlation coefficienies are 0.6677.

Produce the basic reason of this consequence be in the computing formula of threshold value to have ignored completely the amplitude of useful signal with The factor of signal to noise ratio.In consideration of it, the present embodiment is through repeatedly analyzing, it is contemplated that useful signal amplitude and the factor of signal to noise ratio, carry Having gone out a kind of new adaptive thresholding value calculating method, this adaptive thresholding value calculating method greatly reduces wavelet filter result Sensitivity to sampled point, considers the graphic correlation after wavefront as it is shown in figure 5, (a) is figure before filtering in Fig. 5, and (b) is filtering Rear figure:

$T_j=\frac{\mathrm{Median}\left(\right|C_{j,k}\left|\right)\sqrt{2\ln \left(N_j\right)}}{0.6745}\mathrm{\&alpha;exp}(-\frac{1}{{\mathrm{\&beta;}}_j});$

Wherein, j is yardstick, N_jFor the number of wavelet coefficient on this yardstick, Median (| C_j,k|) it is all little on this yardstick The median of wave system number, α is referred to as the signal to noise ratio factor, be signal signal to noise ratio threshold values calculate in embodiment, β_jBe referred to as yardstick because of Son, is that the maximum of wavelet coefficient on yardstick corrects the estimation difference that sample sequence length difference causes, T_jFor the threshold values calculated.

The effectiveness of Methods for Wavelet Denoising Used depends primarily on wavelet basis function, wavelet decomposition scales, threshold values function, threshold values choosing Several aspects such as take.With substantial amounts of emulation experiment, laboratory simulation and the analysis verification of field measurement data in the present embodiment The effectiveness of method therefor.Result shows, this Methods for Wavelet Denoising Used is compared with the denoising algorithm of other threshold values rule, hence it is evident that improve De-noising ability in signal processing of partial discharge, but also after having process, signal waveform distortion is little, it is more accurate, suffered to extract The advantages such as influence factor is few.

3) by extracting the characteristic parameter of sampled signal based on phase analysis pattern algorithm, it is preferred that characteristic parameter includes: Degree of skewness Sk, steepness Ku, local peaks are counted Pe, cross-correlation coefficient Cc and discharge factor Q；

In step 3), for the signal after sampling, use extraction characteristic parameter based on phase analysis pattern (PRPD), Wherein: degree of skewness Sk, the definition of Sk:

$\mathrm{Sk}=\underset{i=1}{\overset{W}{\mathrm{\&Sigma;}}}{(x_i-\mathrm{\&mu;})}^3\&CenterDot;p_i\mathrm{\&Delta;x}/{\mathrm{\&sigma;}}^3;$

In above formula, w is the phase window number in the half period；Xi is the phase place of i-th phase window；

$p_i=y_i/\underset{i=1}{\overset{W}{\mathrm{\&Sigma;}}}{y}_i,$

Wherein, y_iIt is the vertical coordinate of spectrogram, represents Apparent discharge magnitude q or discharge time n；Parameter μ represents the office collected The position of electric discharge collection of illustrative plates center, portion, σ represents the precipitous situation that the center axis of symmetry of collection of illustrative plates is embodied, and Δ x is then about office Portion's electric discharge one of collection of illustrative plates to be evenly distributed relevant parameter,Certain phase place corresponding to point in collection of illustrative plates.

Degree of skewness Sk reflection spectral shape is relative to the left and right deflection situation of normal distribution: Sk=0 illustrates that this spectral shape is left Right symmetry；Sk > 0 illustrate this spectrogram relative to normal distribution shape to left avertence；Sk < 0 illustrates that this spectrogram is relative to normal distribution shape Shape is to right avertence.

The definition of phase window: the building method of Φ-q-n space curved surface: by operating frequency phase according to.0-360 ° is divided into 256 Minizone, is divided into 128 minizones by discharge pulse amplitude q by maximum amplitude, thus Φ-q plane be divided into 128 × 256 minizones；In statistics Φ-q plane, the discharge time in each minizone, i.e. obtains middle Φ-q-n space curved surface.

The definition of steepness Ku, Ku:

$\mathrm{Ku}=[\underset{i=1}{\overset{W}{\mathrm{\&Sigma;}}}{(x_i-\mathrm{\&mu;})}^4{p}_i\mathrm{\&Delta;x}/{\mathrm{\&sigma;}}^4]-3;$

The definition of wherein each variable is identical with variable-definition in degree of skewness.Steepness Ku is for describing dividing of certain shape Cloth in contrast to the projection degree of normal distribution shape: the steepness Ku of normal distribution is 0；If Ku > 0, then illustrate that this spectrogram is taken turns Wide more precipitous than normal distribution profile；If Ku < 0, then illustrate that this spectrogram profile is more smooth than normal distribution profile.

Local peaks is counted Pe, and local peaks is counted for describing the number of local peaks on spectrogram profile.In profile point Whether place has local peaks, and available following formula judges:

And

Above formula becomes difference equation i.e.:

Difference equation can be reduced to:

(y_i-y_i-1) > 0, (y_i+1-y_i) < 0；

In Practical Calculation, local peaks is counted closely related with the phase window number of spectrogram.It is said that in general, phase shaft is halved Phase window the most, local peaks is counted the biggest.

Cross-correlation coefficient Cc, the definition of cross-correlation coefficient Cc:

$\mathrm{Cc}=\frac{\underset{i=1}{\overset{W}{\mathrm{\&Sigma;}}}{q}_i^{+}{q}_i^{-}-\left(\underset{i=1}{\overset{W}{\mathrm{\&Sigma;}}}{q}_i^{+}\underset{i=1}{\overset{W}{\mathrm{\&Sigma;}}}{q}_i^{-}\right)/W}{\sqrt{[\underset{i=1}{\overset{W}{\mathrm{\&Sigma;}}}{\left(q_i^{+}\right)}^2-{\left(\underset{i=1}{\overset{W}{\mathrm{\&Sigma;}}}{q}_i^{-}\right)}^2/W][\underset{i=1}{\overset{W}{\mathrm{\&Sigma;}}}{\left(q_i^{-}\right)}^2-{\left(\underset{i=1}{\overset{W}{\mathrm{\&Sigma;}}}{q}_i^{-}\right)}^2]/W}};$

In formula,The discharge capacity in phase window i, subscript "+", "-" is corresponding to the positive and negative semiaxis of spectrogram.Cc reacts The strong and weak dependency with PHASE DISTRIBUTION of electric discharge in positive and negative half period, cross-correlation coefficient Cc means close to 1Spectrogram is just The profile of negative half period is quite similar；Cc close to 0,Spectrogram profile difference is huge.

Discharge factor Q,

$Q=\frac{\underset{i=1}{\overset{w}{\mathrm{\&Sigma;}}}{n}_i^{-}{q}_i^{-}}{\underset{i=1}{\overset{w}{\mathrm{\&Sigma;}}}{n}_i^{-}}/\frac{\underset{i=1}{\overset{w}{\mathrm{\&Sigma;}}}{n}_i^{+}{q}_i^{+}}{\underset{i=1}{\overset{w}{\mathrm{\&Sigma;}}}{n}_i^{+}};$

In formula,The electric discharge repetitive rate in phase window i, subscript "+", "-" corresponds toSpectrogram positive and negative Half cycle.Discharge capacity factor Q reactsThe difference of mean discharge magnitude in spectrogram positive-negative half-cycle.

According to the formula of each Statistical Operator above, calculate Statistical Operator by analysis of spectra and algorithm, extract feature Parameter shift degree (SK), steepness (Ku), local peaks are counted (Pe), discharge factor (Q), cross-correlation coefficient (Cc) carry out pattern Identify.

4) feature space utilizing the core principle component analysis method improved to form characteristic parameter carries out dimension-reduction treatment, is dropped Characteristic parameter matrix after dimension；

In step 4), owing to cannot know which characteristic parameter can construct the simplest feature sky of UHF PD signal in advance Between, the most break-even characteristic parameter matrix of being completely lost, the feature space dimension of structure is higher, and there may be dimension redundancy, right Run and recognition result is unfavorable.Therefore, there is a need to carry out the dimension-reduction treatment of feature space.

In the middle of the application of KPCA, the selection of nonlinear transformation (i.e. kernel function) is extremely important.Conventional kernel function has many Item formula kernel function (Polynomial), gaussian kernel function (Gauss) and Sigmoid kernel function are as follows:

k(x_i,x_j)=(<x_i,x_j>+a)^b, (a ∈ R, b ∈ N)；

$k(x_i,x_j)=\exp \left(\frac{-{\left|\right|x_i-x_j\left|\right|}^2}{2{\mathrm{\&sigma;}}^2}\right);$

k(x_i,x_j)=tanh(<x_i,x_j>+a), (a ∈ R)；

Wherein, < x_i,x_j> it is sample vector, x_iAnd x_jVector product, | | x_i-x_j| | for both Euclidean Norms.Multinomial Kernel function is distance (< x_i,x_j>+a) and b power, be<x_i,x_j> monotonic increasing function.Through conversion, if (< x_i,x_j>+a)> 1, initial range < x_i,x_j> can be exaggerated；On the contrary, if (< x_i,x_j>+a)<1, then initial range<x_i,x_j> can be compressed.Visible The effect of Polynomial kernel function is that small distance compression is expanded big distance further.Gaussian kernel function the most radially base core Function, is generally also defined as the index monotonically decreasing function of two vectorial Euclidean distances, and it is the scalar of a kind of radial symmetric Function.Wherein σ is referred to as width parameter, for the radial effect scope of control function, the i.e. width of Gaussian pulse.But generally Gauss The sphere of action of kernel function is less, and action effect is just the most contrary with Polynomial kernel function, i.e. small distance is expanded and greatly away from Tripping contracts.

The effect of kernel function should be to be expanded further by initial range in fact, or by the Range compress between similar sample And the distance between non-similar sample is expanded, in order to carry out Classification and Identification.In consideration of it, the present embodiment combines Polynomial kernel function Pluses and minuses with gaussian kernel function, it is proposed that a kind of new kernel function, its action effect is as shown in Figure 4.

In step 4), the kernel function that the core principle component analysis method of improvement is sampled is:

$k(x_i,x_j)={(<x_i,x_j>+a)}^b\exp \left(\frac{-{\left|\right|x_i-x_j\left|\right|}^2}{2{\mathrm{\&sigma;}}^2}\right);$

Wherein, (a ∈ R, b ∈ N, σ > 0), parameter a, the selection of b with σ are true according to the numerical values recited of element in eigenmatrix Fixed, parameter σ is for controlling the radial effect scope of kernel function；x_iAnd x_jRepresent different sample vectors, x_i, x_jRepresentative sample The vector product of this vector, the span of R representation vector is at set of real numbers, and N represents set of integers, k (x_i,x_j) represent combine multinomial The advantage of kernel function and gaussian kernel function and the new kernel function that obtains.Take a=5, b=1 is to make the distance after conversion herein A certain proportion of change is made with former distance.Parameter σ is for controlling the radial effect scope of kernel function, due in primitive character matrix Two vectorial distances are usually no more than 7, so taking σ=7.From fig. 4, it can be seen that nucleus function proposed by the invention is by original Small distance become big, big distance is suitably reduced.

5) utilize k nearest neighbor classification method based on bunch thought that GIS insulation defect type is carried out pattern recognition；

In step 5), K its basic thought of arest neighbors method is: providing test document, system is in the most categorized good instruction Practice to concentrate and search K the neighbours nearest with it, obtain the classification of test document according to the categorical distribution situation of these neighbours.Wherein Can be weighted by the similarity of these neighbours with test document, thus obtain preferable classifying quality.So-called bunch, the meaning is just Being the class set with the text of similar quality, the present invention is maximum for the spacing belonging to same category of text in training set Those local discharge signal data subsets close be considered one bunch, therefore, the algorithm of k nearest neighbor classification method can be described as follows:

First all Partial Discharge Datas are carried out pretreatment and become space vector by Step1: in training set；

The all signal datas belonging to this classification are carried out Similarity Measure two-by-two by Step2: from the beginning of first class, Set a minimum threshold, according to statistics obtain similarity close one by one bunch；

Step3: for each bunch, all signal datas therein are merged, then calculates its center vector, this Outward, calculating bunch number/classification sum, this value represents this bunch contribution coefficient to this class, is denoted as C；

Step4: after new text arrives, carries out pretreatment and obtains its vector space；

Step5: by the center vector computed range of every cluster that the space vector of new text is generated with Step3, by this A little distance contribution coefficients with corresponding bunch are multiplied, and belong to the same category of bunch of results added calculated, and compare and obtain maximum that One classification is exactly typical defect shelf depreciation generic to be sorted.

The basis of this algorithm is how to find out which text in same category to belong to same cluster, given below find out with The generation bunch algorithm idea of one classification bunch: assume classification:

C={d1, d2 ... ... ...., dm}

Step1: set threshold value a of a similarity；

Step2: first create one bunch, be denoted as T0, by the number of documents that comprised in Ki record bunch, total records wound The number of clusters amount built, initializes processed document i=2；

Step3: from the beginning of di；

Step4: carry out Similarity Measure with first text in Tn and obtain value s；

Step5: if also having the sample not compared with this sample in s >=a, and Tn, then proceed similarity Calculate and update s；Without non-comparative sample, then these data are joined in bunch Tn；If s < a, if there being other Do not compare bunch, then n++, return step4；Without do not compare bunch, then create new bunch, be designated as T++total；By this Document is classified as in T++total bunch；

Step6: if i！=m, then i++；Return Step3；Otherwise, terminate.

In order to overcome the defect that nearest neighbor method's false determination ratio is higher, arest neighbors being generalized to k nearest neighbor, k-nearest neighbor is not chosen One arest neighbors is classified, but chooses and represent a little from the nearest K of text to be sorted, then represents according to this K and to put Classification information determines the classification of text to be sorted.

For the characteristic parameter matrix after dimensionality reduction, a half-sample is used for training k nearest neighbor grader, and second half is for test point The performance of class device.For applying the characteristic parameter matrix after KPCA, RST and CCMDR algorithm dimensionality reduction respectively, application k nearest neighbor divides GIS insulation defect type is identified by class device.The present embodiment has write program file under C language software environment, it is achieved point The design of class device, training and Classification and Identification test.Due to the present embodiment design grader output and not as BP neutral net Equally being distributed in dispersed centered by certain is put, and correspond to 4 class GIS defect types, output value only includes 4 kinds of results [1,2,3,4], so pattern recognition result only represents with recognition correct rate, as shown in table 1.

Table 1:K nearest neighbor algorithm pattern recognition accuracy

Defect type	K-nearest neighbor recognition correct rate
		High-pressure conductor metallic projections	92%
Free metal microgranule	91.5%
		Insulator surface fixes metal	88%
Insulator void defects	90%

In sum, the pattern recognition side of the three-phase of the various embodiments described above of the present invention cartridge type supertension GIS partial discharge altogether Method, including step: use hyperfrequency (UHF) detection three-phase cartridge type GIS partial discharge altogether, utilize type UHF sensor to shelf depreciation (PD) signal sampling；Utilize the small echo threshold values filtering method improved that local discharge signal is carried out denoising Processing；Based on phase analysis The characteristic parameter of schema extraction sampled signal, characteristic parameter includes: degree of skewness Sk, steepness Ku, local peaks are counted Pe, cross-correlation Coefficient Cc and discharge factor Q；The feature space utilizing the core principle component analysis method improved to form characteristic parameter is carried out at dimensionality reduction Reason, obtains the characteristic parameter matrix after dimensionality reduction；Utilize k nearest neighbor classification method that GIS insulation defect type is carried out pattern recognition.These are three years old The beneficial effect that the mode identification method of cartridge type supertension GIS partial discharge at least can reach altogether mutually includes: overcome existing The defect of technology, improves the accuracy of three-phase cartridge type supertension GIS partial discharge detection pattern recognition altogether.

Finally it is noted that the foregoing is only the preferred embodiments of the present invention, it is not limited to the present invention, Although being described in detail the present invention with reference to previous embodiment, for a person skilled in the art, it still may be used So that the technical scheme described in foregoing embodiments to be modified, or wherein portion of techniques feature is carried out equivalent. All within the spirit and principles in the present invention, any modification, equivalent substitution and improvement etc. made, should be included in the present invention's Within protection domain.

Claims (9)

1. the mode identification method of a three-phase cartridge type supertension GIS partial discharge altogether, it is characterised in that comprise the following steps:

Step 1: use ultra-high-frequency detection three-phase cartridge type supertension GIS partial discharge altogether, utilize type UHF sensor that shelf depreciation is believed Number sampling；

Step 2: utilize the improved wavelet threshold filtering method local discharge signal to collecting to carry out denoising Processing, obtains true Real local discharge signal；

Step 3: by extracting the characteristic parameter of sampled signal based on phase analysis pattern algorithm；

Step 4: the feature space utilizing the core principle component analysis method improved to form characteristic parameter carries out dimension-reduction treatment, is dropped Characteristic parameter matrix after dimension；

Step 5: utilize k nearest neighbor classification method based on bunch thought that GIS insulation defect type carries out pattern recognition, in steps of 5, The algorithm of described k nearest neighbor classification method specifically includes:

First all Partial Discharge Datas are carried out pretreatment and become space vector by Step1: in training set；

The all signal datas belonging to this classification are carried out Similarity Measure two-by-two by Step2: from the beginning of first class, set One minimum threshold, according to statistics obtain similarity close one by one bunch；

Step3: for each bunch, all signal datas therein are merged, then calculates its center vector；Additionally, meter Calculating bunch number/classification sum, this value represents this bunch contribution coefficient to this class；

Step4: after new text arrives, carries out pretreatment and obtains its space vector；

Step5: the center vector computed range of every cluster that the space vector of new text and Step3 are generated, by these away from It is multiplied from the contribution coefficient with corresponding bunch, belongs to the same category of bunch of results added calculated, compare that class obtaining maximum It it is not exactly typical defect shelf depreciation generic to be sorted.

The mode identification method of three-phase the most according to claim 1 cartridge type supertension GIS partial discharge altogether, its feature exists In, in step 2, described utilize the improved wavelet threshold filtering method local discharge signal to collecting to carry out denoising Processing Operation, concrete use adaptive thresholding value calculating method；This adaptive thresholding value calculating method is as follows:

Wherein, j is yardstick, N_jFor the number of wavelet coefficient on this yardstick, Median (| C_j,k|) it is all wavelet systems on this yardstick The median of number, α is referred to as the signal to noise ratio factor, is the embodiment in threshold calculations of the signal to noise ratio of signal, β_jIt is referred to as scale factor, is On yardstick, the maximum of wavelet coefficient corrects the estimation difference that sample sequence length difference causes, T_jFor the threshold value calculated.

The mode identification method of three-phase the most according to claim 1 and 2 cartridge type supertension GIS partial discharge altogether, its feature Be, in step 3, described characteristic parameter include degree of skewness Sk, steepness Ku, local peaks count Pe, cross-correlation coefficient Cc and Discharge factor Q.

The mode identification method of three-phase the most according to claim 3 cartridge type supertension GIS partial discharge altogether, its feature exists In, described degree of skewness Sk particularly as follows:

In above formula, w is the phase window number in the half period；x_iIt it is the phase place of i-th phase window；

Wherein, y_iIt is the vertical coordinate of spectrogram, y_iThat represent is the discharge time n that shelf depreciation record occurs in i-th phase window； Parameter μ represents the position of shelf depreciation spectrogram center collected, and it is precipitous that σ represents that the center axis of symmetry of spectrogram embodied Situation, Δ x be then one about shelf depreciation spectrogram to be evenly distributed relevant parameter,Corresponding to certain point in spectrogram Phase place；

Degree of skewness Sk reflection spectral shape is relative to the left and right deflection situation of normal distribution: Sk=0 illustrates about this spectral shape Symmetrical；Sk > 0 illustrate this spectrogram relative to normal distribution shape to left avertence；Sk < 0 illustrates that this spectrogram is relative to normal distribution shape To right avertence.

The mode identification method of three-phase the most according to claim 3 cartridge type supertension GIS partial discharge altogether, its feature exists In, described steepness Ku particularly as follows:

In above formula, w is the phase window number in the half period；x_iIt it is the phase place of i-th phase window；

Wherein, y_iIt is the vertical coordinate of spectrogram, y_iThat represent is the discharge time n that shelf depreciation record occurs in i-th phase window； Parameter μ represents the position of shelf depreciation spectrogram center collected, and it is precipitous that σ represents that the center axis of symmetry of spectrogram embodied Situation, Δ x be then one about shelf depreciation spectrogram to be evenly distributed relevant parameter,Corresponding to certain point in spectrogram Phase place；

Steepness Ku is for describing the profiles versus of certain shape in the projection degree of normal distribution shape: normal distribution precipitous Degree Ku is 0；If Ku > 0, then illustrate that this spectrogram profile is more precipitous than normal distribution profile；If Ku < 0, then this spectrogram is described Profile is more smooth than normal distribution profile.

The mode identification method of three-phase the most according to claim 3 cartridge type supertension GIS partial discharge altogether, its feature exists Counting Pe in, described local peaks, local peaks is counted for describing the number of local peaks on spectrogram profile；Profile point (y_i) Whether place has local peaks, needs to judge with following difference equation:

(y_i-y_i-1) > 0, (y_i+1-y_i)<0；

The halved phase window of phase shaft is the most, and local peaks is counted the biggest, i represent phase window belong to which.

The mode identification method of three-phase the most according to claim 3 cartridge type supertension GIS partial discharge altogether, its feature exists In, described cross-correlation coefficient Cc particularly as follows:

In formula, W represents the phase window sum in positive and negative half period；The discharge capacity in phase window i, subscript "+", "-" corresponding In the positive and negative semiaxis of spectrogram；C has reacted the strong and weak dependency with PHASE DISTRIBUTION of the electric discharge in positive and negative half period, and cross-correlation coefficient Cc connects It is bordering on 1 to meanThe profile similarity of spectrogram positive-negative half-cycle；Cc close to 0,The spectrogram profile differences of positive-negative half-cycle Different greatly.

The mode identification method of three-phase the most according to claim 3 cartridge type supertension GIS partial discharge altogether, its feature exists In, described discharge factor Q particularly as follows:

In formula, W represents the phase window sum in positive and negative half period；The discharge capacity in phase window i, subscript "+", "-" pair Should be in the positive and negative semiaxis of spectrogram；The electric discharge repetitive rate in phase window i, subscript "+", "-" corresponds toSpectrogram Positive-negative half-cycle.

The mode identification method of three-phase the most according to claim 1 and 2 cartridge type supertension GIS partial discharge altogether, its feature It is, in step 4, the core principle component analysis method of described improvement, the kernel function sampled is:

Wherein, (a ∈ R, b ∈ N, σ ＞ 0), parameter a, the selection of b with σ are to determine according to the numerical values recited of element in eigenmatrix , parameter σ is for controlling the radial effect scope of kernel function；x_iAnd x_jRepresent different sample vectors, x_i, x_jRepresentative sample The vector product of vector, the span of R representation vector is at set of real numbers, and N represents set of integers, k (x_i,x_j) represent combine polynomial kernel The advantage of function and gaussian kernel function and the new kernel function that obtains.