CN101408580B - Method for evaluating oil paper insulation ageing state based on local discharge characteristic parameter - Google Patents

Abstract

The invention provides a new method for evaluating the state of paper oil insulation ageing and realizes the on-line evaluation of the paper oil insulation ageing state of a power transformer. The method includes the following steps: 1) a local discharge signal transducer acquires a local discharge signal of paper oil insulation and transmitted the signal to a data processing device; 2) the processing device analyzes the local discharge signal, obtaining the local discharge feature parameter; 3) the processing device extracts main constituent factor from the local discharge feature parameter; 4) the processing device utilizes the main constituent factor obtained in step 3) and constructs a paper oil insulation ageing evaluation model through an artificial neural net to gain the ageing state of paper oil insulation. By implementing data analysis for the local discharge signal in the paper oil insulation ageing process, the invention extracts the local discharge feature parameter capable of reflecting the insulation ageing, conducts further data mining on the parameter and on the basis, constructs a reasonable mathematics model to evaluate the ageing state of paper oil insulation.

Description

Method for evaluating oil paper insulation ageing state based on local discharge characteristic parameter

Technical field

The invention belongs to power equipment insulation ag(e)ing and life prediction field, be specifically related to a kind of appraisal procedure of oil-immersed power transformer oil paper insulation ageing state.

Background technology

Power transformer is the core of electric energy conversion, transmission, is the equipment of most important and most critical in the electrical network.At present, the transformer of power industry has the late period that has reached its design service life greatly all over the world.The aging assessment and the forecasting technique in life span of transformer receive lot of domestic and foreign scholar's concern always.The main composition form of insulation is insulating oil-fibrous paper combined insulation system in the large-capacity power transformer; It is in operation and wears out gradually because of multifactor effects such as electricity, heat, machineries; Suffer the invasion and attack of extraneous factors such as thunder and lightning, system short-circuit in addition; Make degree of aging further deepen, all possible at any time initiating failure causes large area blackout.Be generally used for diagnosing the aging method of paper oil insulation to comprise: dissolved gas analysis (DGA), furans in the oil (Furan) derivative content is measured, the paper oil insulation degree of polymerization (DP) and tensile strength (TS) etc.Mostly above method is to be applicable to the state of off-line judgement transformer oil Aging of Oil-paper Insulation in Oil, and Analysis of Partial Discharge is a kind of harmless on-line monitoring method, through the local discharge characteristic parameter of research transformer oil Aging of Oil-paper Insulation in Oil, thereby its ageing state is assessed.

The prior art of the assessment aspect of present insulation of electrical installation ageing state; And mostly prior art is the aging assessment to generator stator winding insulation; Like publication number is the Chinese invention patent Shen Qing Publication instructions of CN1402015A, does not appear in the newspapers about the patent of method for evaluating oil paper insulation ageing state.Existing oil paper insulation ageing state assessment based on shelf depreciation, no matter from time domain waveform characteristic or spectrum signature angle, all do not obtain can its ageing state of reliable assessment method.

Summary of the invention

In view of this, in order to address the above problem, the present invention can realize the online evaluation of Research on Power Transformer Oil-paper Insulation ageing state for the oil paper insulation ageing state assessment provides a kind of new appraisal procedure.

The objective of the invention is to realize like this, the method for evaluating oil paper insulation ageing state based on local discharge characteristic parameter comprises the steps:

1) local discharge signal of local discharge signal sensor acquisition paper oil insulation and be transferred to data processing equipment;

2) data processing equipment is analyzed local discharge signal, obtains local discharge characteristic parameter;

3) data processing equipment is to step 2) local discharge characteristic parameter that obtains extracts the major component factor;

4) data processing equipment utilizes the major component factor that step 3) obtains, and makes up the aging assessment models of paper oil insulation through artificial neural network, obtains the ageing state of paper oil insulation.

Further, said step 2) specifically comprise the steps:

21) data processing equipment is according to the operating frequency phase of the generation of the shelf depreciation in the local discharge signal Discharge capacity amplitude q and discharge time n obtain following four kinds of discharge collection of illustrative plates: maximum pd quantity PHASE DISTRIBUTION collection of illustrative plates

Average discharge capacity PHASE DISTRIBUTION collection of illustrative plates

Discharge time PHASE DISTRIBUTION collection of illustrative plates

With discharge time discharge capacity collection of illustrative plates H _n(q);

22) data processing equipment extracts maximum pd quantity PHASE DISTRIBUTION collection of illustrative plates

Average discharge capacity PHASE DISTRIBUTION collection of illustrative plates

With discharge time PHASE DISTRIBUTION collection of illustrative plates

In measure of skewness sk, steepness ku, peak value number peak, degree of asymmetry Asy and the related coefficient cc of positive-negative half-cycle, and discharge time discharge capacity collection of illustrative plates H _n(q) measure of skewness sk, steepness ku and peak value number peaks form primitive character parametric data matrix X as local discharge characteristic parameter with local discharge characteristic parameter;

Further, said step 3) specifically comprises the steps:

31) data processing equipment carries out the standardization variation to local discharge characteristic parameter composition primitive character parametric data matrix X;

32) data processing equipment is tested, is screened the correlativity between the local discharge characteristic parameter standardized data;

33) data processing equipment is to step 32) the local discharge characteristic parameter standardized data that filters out finds the solution covariance, obtains covariance matrix;

34) data processing equipment obtains the major component factor to the principal component analysis (PCA) of covariance matrix;

Further, also comprise the steps: step 34)

35) raw data matrix X is converted into major component factor data matrix F;

36) data processing equipment carries out LOAD FOR to the major component factor;

37) data processing equipment is according to step 36) result of calculation the major component factor is rotated analysis, analyze the physical significance that the major component factor is comprised;

38) data processing equipment calculates the major component factor score, according to factor score, the shelf depreciation raw data matrix is converted into new major component factor data matrix.

Further, in the step 4), adopt the sign amount of degree of polymerization DP as oil paper insulation ageing state;

Further, in the step 4), adopt three layers of backpropagation (Back Propagation) neural network to set up the aging assessment models of paper oil insulation;

Further; The shelf depreciation major component factor that is input as extraction of the aging assessment models of paper oil insulation; Be output as oil paper insulation ageing state, at first data carried out the normalization pre-service before the data input, the hidden layer transport function is selected tanh S type function; The output layer activation function is selected logarithm S type function, and input neuron is counted n ₁Be the number of the major component factor extracted, the output layer neuron number is 1, and hidden layer neuron is counted n ₂According to formula n ₂=2n ₁+ 1 confirms.

It is of the present invention through the local discharge signal in the paper oil insulation ageing process is carried out data analysis; Extraction can be reacted the local discharge characteristic parameter of insulation ag(e)ing; And it is carried out further data mining, make up rational mathematical model on this basis the ageing state of paper oil insulation is assessed.The present invention is analysis means with the shelf depreciation; Extract the statistical characteristic value of shelf depreciation; And checked the correlativity between characteristic quantity; On this basis the local discharge characteristic amount is extracted the major component factor, can eliminate the influence that the correlativity between the primitive character amount is assessed ageing state, and simplify assessment models to a great extent.The aging assessment models based on neural network that proposes can be simulated the nonlinear relationship of the complicacy between shelf depreciation and the ageing state, and can assess the ageing state of paper oil insulation exactly.With of the input of the shelf depreciation major component factor as neural network, can overcome the deficiency of traditional off-line assessment oil paper insulation ageing state method, realize the online accurate assessment of oil paper insulation ageing state.

Other advantages of the present invention, target; To in instructions subsequently, set forth to a certain extent with characteristic; And to a certain extent,, perhaps can from practice of the present invention, obtain instruction based on being conspicuous to those skilled in the art to investigating of hereinafter.Target of the present invention and other advantages can be passed through following instructions, claims, and the structure that is particularly pointed out in the accompanying drawing realizes and obtains.

Description of drawings

In order to make the object of the invention, technical scheme and advantage clearer, will combine accompanying drawing that the present invention is made further detailed description below, wherein:

Fig. 1 shows the process flow diagram that obtains the major component factor in the embodiment of the invention in the method for evaluating oil paper insulation ageing state based on the shelf depreciation parameter;

Fig. 2 shows the structural representation of artificial neural network assessment oil paper insulation ageing state assessment models in the embodiment of the invention.

Embodiment

Below will carry out detailed description to the preferred embodiments of the present invention with reference to accompanying drawing.

The input end of local discharge signal sensor is connected with paper oil insulation electrode as the power transformer of aged samples, and output terminal is connected with data processing equipment, can obtain the local discharge signal of paper oil insulation and be transferred to data processing equipment; But the local discharge signal sensor can be selected sieve Paderewski coil for use, adopts traditional pulse current method to gather local discharge signal, and data processing equipment can be selected intelligent oscillograph for use.

Referring to Fig. 1, the method for evaluating oil paper insulation ageing state based on local discharge characteristic parameter of present embodiment comprises the steps:

1) local discharge signal of local discharge signal sensor acquisition paper oil insulation and be transferred to data processing equipment;

2) data processing equipment is analyzed local discharge signal, obtains local discharge characteristic parameter; Specifically comprise:

21) data processing equipment is according to operating frequency phase

the discharge capacity amplitude q and the discharge time n of the generation of the shelf depreciation in the local discharge signal; With the phase place is the discharge spectrogram of horizontal ordinate; Phase window number according to setting is divided window with phase place, and dividing window with phase place is the various shelf depreciation parameters of the resulting a plurality of power frequency periods of unit statistical measurement.Obtain three kinds of important discharge spectrograms that divide window based on phase place: maximum pd quantity PHASE DISTRIBUTION collection of illustrative plates

Average discharge capacity PHASE DISTRIBUTION collection of illustrative plates

Discharge time PHASE DISTRIBUTION collection of illustrative plates

Except three kinds be the collection of illustrative plates of horizontal ordinate with the phase place, also having obtained a kind of is the discharge time discharge capacity collection of illustrative plates H of horizontal ordinate with discharge capacity q _n(q).

22) data processing equipment extracts maximum pd quantity PHASE DISTRIBUTION collection of illustrative plates

Average discharge capacity PHASE DISTRIBUTION collection of illustrative plates

With discharge time PHASE DISTRIBUTION collection of illustrative plates

In measure of skewness sk, steepness ku, peak value number peak, degree of asymmetry Asy and the related coefficient cc of positive-negative half-cycle, discharge time discharge capacity collection of illustrative plates H _n(q) measure of skewness sk, steepness ku and peak value number peaks are as local discharge characteristic parameter; And composition primitive character amount data matrix X, variable X _i(i=1,2 ..., p) order is according to from collection of illustrative plates

The order of the characteristic quantity of collection of illustrative plates is from sk+ → sk-→ ku+ → ku-→ peaks+ → peaks-→ Asy → cc; H _n(q) order of collection of illustrative plates characteristic quantity is from sk → ku → peaks.

3) data processing equipment is to step 2) local discharge characteristic parameter that obtains extracts the major component factor; Specifically may further comprise the steps:

31) data processing equipment carries out the standardization variation to the primitive character amount data matrix of local discharge characteristic parameter composition, and concrete transform is as follows:

$Z_{\mathrm{ij}}=\frac{X_{\mathrm{ij}}-{\stackrel{\‾}{X}}_i}{S_i}$

In the formula: X _IjBe local discharge characteristic amount raw data, Z _IjBe the data after the standardization, X _iAnd S _iMean value and the standard deviation of representing i characteristic quantity respectively.

32) data processing equipment carries out KMO (Kaiser-Meyer-Olkin) check of correlativity to the local discharge characteristic parameter standardized data; If the KMO test value is greater than 0.5; Then be for further processing; Otherwise the original variable before directly this local discharge characteristic parameter standardization being changed as the major component factor directly as the input of step 4), KMO value ε _KMOComputing formula be:

${\mathrm{\ϵ}}_{\mathrm{KMO}}=\frac{\underset{i}{\mathrm{\Σ}}\underset{j(i\≠j)}{\mathrm{\Σ}}{r_{\mathrm{ij}}}^2}{\underset{i}{\mathrm{\Σ}}\underset{j(i\≠j)}{\mathrm{\Σ}}{r_{\mathrm{ij}}}^2+\underset{i}{\mathrm{\Σ}}\underset{j(i\≠j)}{\mathrm{\Σ}}{S_{\mathrm{ij}}}^2}$

R in the formula _IjBe local discharge characteristic amount X _IjRelated coefficient, s _IjBe local discharge characteristic amount X _IjPartial correlation coefficient;

33) data processing equipment is to step 32) garbled local discharge characteristic parameter standardized data finds the solution covariance, obtains covariance matrix; The correlation matrix R of covariance matrix and original variable is equal, promptly

$r_{\mathrm{ij}}=s_{\mathrm{ij}}=\frac{\underset{t=1}{\overset{n}{\mathrm{\Σ}}}(Z_{\mathrm{ti}}-{\stackrel{\‾}{Z}}_i)(Z_{\mathrm{tj}}-{\stackrel{\‾}{Z}}_j)}{\sqrt{\underset{t=1}{\overset{n}{\mathrm{\Σ}}}{(Z_{\mathrm{ti}}-{\stackrel{\‾}{Z}}_i)}^2}\sqrt{\underset{t=1}{\overset{n}{\mathrm{\Σ}}}{(Z_{\mathrm{tj}}-{\stackrel{\‾}{Z}}_j)}^2}}$

In the formula: Z _TiAnd Z _TjBe respectively the data after the standardization, Z _iAnd Z _jBe respectively the mean value of i characteristic quantity after the standardization;

34) data processing equipment carries out principal component analysis (PCA) to raw data correlation matrix correlation matrix R:

The characteristic root of correlation matrix R is λ ₁>=λ ₂>=λ ₃>=...>=λ _p>0, characteristic of correspondence vector matrix U=[u ₁u ₂... u _p], according to Y=U ^T* Z has promptly obtained the data matrix Y that other one group of incoherent variable is formed.Select the characteristic root to analyze, and calculate the contribution rate and the contribution rate of accumulative total of each major component greater than the corresponding variable of 1 characteristic root, the contribution rate of major component according to ${\mathrm{\μ}}_j={\mathrm{\λ}}_j/\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{\mathrm{\λ}}_i$ Calculate.The present invention gets contribution rate of accumulative total and surpasses 90% a pairing m variable as the major component factor.

35) raw data matrix X is converted into major component factor data matrix F;

36) data processing equipment carries out LOAD FOR to the major component factor:

Y carries out the conversion of following formula to the data matrix:

$Y=\left[\begin{array}{c}{Y}_1\\ Y_2\\ \mathrm{\&Lambda;}\\ Y_p\end{array}\right]=\left[\begin{array}{cccc}\sqrt{{\mathrm{\&lambda;}}_1}& 0& \mathrm{\&Lambda;}& 0\\ 0& \sqrt{{\mathrm{\&lambda;}}_2}& O& M\\ \mathrm{<figure-callout\; id="0"\; label="M\; O\; O"\; filenames="H2008102330960E00021.png"\; state="\{\{state\}\}">M</figure-callout>}& O& O& 0\\ 0& \mathrm{\&Lambda;}& 0& \sqrt{{\mathrm{\&lambda;}}_{\mathrm{<figure-callout\; id="1"\; label="p\; Y"\; filenames="H2008102330960E00021.png"\; state="\{\{state\}\}">p</figure-callout>}}}& \\ \end{array}\right]\left[\begin{array}{c}{Y}_{}\end{array}\right]\left[\begin{array}{c}{}_1/\sqrt{{\mathrm{\&lambda;}}_1}\\ Y_2/\sqrt{{\mathrm{\&lambda;}}_2}\\ M\\ Y_p/\sqrt{{\mathrm{\&lambda;}}_p}\end{array}\right]$

Order $F_{}$ ${}_1={\mathrm{<figure-callout\; id="1"\; label="Y"\; filenames="H2008102330960E00021.png"\; state="\{\{state\}\}">Y</figure-callout>}}_1/\sqrt{{\mathrm{\&lambda;}}_1},$ $F_2=Y_2/\sqrt{{\mathrm{\&lambda;}}_2},$ Λ, $F_p=Y_p/\sqrt{{\mathrm{\&lambda;}}_p},$ So obtain:

$Z=\mathrm{UY}={\left[u_1{u}_2\mathrm{\&Lambda;}{u}_p\right]}_{p\&times;p}\&times;\left[\begin{array}{cccc}\sqrt{{\mathrm{\&lambda;}}_1}& 0& \mathrm{\&Lambda;}& 0\\ 0& \sqrt{{\mathrm{\&lambda;}}_2}& O& M\\ \mathrm{<figure-callout\; id="0"\; label="M\; O\; O"\; filenames="H2008102330960E00021.png"\; state="\{\{state\}\}">M</figure-callout>}& O& O& 0\\ 0& \mathrm{\&Lambda;}& 0& \sqrt{{\mathrm{\&lambda;}}_p}\end{array}\right]\left[\begin{array}{c}{F}_1\\ F_2\\ M\\ F_p\end{array}\right]$

$=\left[\begin{array}{cccc}{u}_{11}\sqrt{{\mathrm{\&lambda;}}_1}& u_{12}\sqrt{{\mathrm{\&lambda;}}_2}& \mathrm{\&Lambda;}& u_{1p}\sqrt{{\mathrm{\&lambda;}}_p}\\ u_{21}\sqrt{{\mathrm{\&lambda;}}_1}& u_{22}\sqrt{{\mathrm{\&lambda;}}_2}& \mathrm{\&Lambda;}& u_{2p}\sqrt{{\mathrm{\&lambda;}}_p}\\ M& M& & M\\ u_{\mathrm{<figure-callout\; id="1"\; label="p"\; filenames="H2008102330960E00021.png"\; state="\{\{state\}\}">p</figure-callout>}1}\sqrt{{\mathrm{\&lambda;}}_1}& u_{p2}\sqrt{{\mathrm{\&lambda;}}_2}& \mathrm{\&Lambda;}& u_{\mathrm{pp}}\sqrt{{\mathrm{\&lambda;}}_p}\end{array}\right]\left[\begin{array}{c}{F}_1\\ F_2\\ M\\ F_p\end{array}\right]=A\&times;\left[\begin{array}{c}{F}_1\\ F_2\\ M\\ F_p\end{array}\right]$

Matrix A is the factor loading matrix of p major component vector, and the pairing main gene loading matrix of preceding m the major component factor of extraction is:

$A_m=\left[\begin{array}{cccc}{u}_{11}\sqrt{{\mathrm{\&lambda;}}_1}& u_{12}\sqrt{{\mathrm{\&lambda;}}_2}& \mathrm{\&Lambda;}& u_{1m}\sqrt{{\mathrm{\&lambda;}}_m}\\ u_{21}\sqrt{{\mathrm{\&lambda;}}_1}& u_{22}\sqrt{{\mathrm{\&lambda;}}_2}& \mathrm{\&Lambda;}& u_{2m}\sqrt{{\mathrm{\&lambda;}}_m}\\ M& M& & M\\ u_{\mathrm{<figure-callout\; id="1"\; label="p"\; filenames="H2008102330960E00021.png"\; state="\{\{state\}\}">p</figure-callout>}1}\sqrt{{\mathrm{\&lambda;}}_1}& u_{p2}\sqrt{{\mathrm{\&lambda;}}_2}& \mathrm{\&Lambda;}& u_{\mathrm{pm}}\sqrt{{\mathrm{\&lambda;}}_m}\end{array}\right]$

37) data processing equipment is rotated analysis to the major component factor:

Select orthogonal matrix T as follows _KjPer two factors to m the major component factor are rotated angle

Make the population variance of postrotational factor loading arrive maximum.

That is: through Orthogonal transformation makes $V=\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{p}{\mathrm{\&Sigma;}}}{({b_{\mathrm{Ij}}}^2-\frac{1}{p}\underset{i=1}{\overset{p}{\mathrm{\&Sigma;}}}\left({b_{\mathrm{Ij}}}^2\right))}^2$ Reach maximum.After factor rotation, the load on each factor is drawn back as far as possible, a part of load is tending towards 1, and another part is tending towards 0.

38) data processing equipment calculates the major component factor score:

According to the factor loading matrix A _mEach factor table is shown as the linear forms of variable X and factor score matrix of coefficients γ:

$\left\{\begin{array}{c}{\mathrm{<figure-callout\; id="1"\; label="F"\; filenames="H2008102330960E00021.png"\; state="\{\{state\}\}">F</figure-callout>}}_1={\mathrm{\&gamma;}}_{11}{X}_{11}+{\mathrm{\&gamma;}}_{12}{X}_{12}+\mathrm{\&Lambda;}+{\mathrm{\&gamma;}}_{1p}{X}_{1p}\\ F_2={\mathrm{\&gamma;}}_{21}{X}_{21}+{\mathrm{\&gamma;}}_{22}{X}_{22}+\mathrm{\&Lambda;}+{\mathrm{\&gamma;}}_{2p}{X}_{2p}\\ \mathrm{\&Lambda;}\\ F_m={\mathrm{\&gamma;}}_{m1}{\mathrm{<figure-callout\; id="1"\; label="X\; m"\; filenames="H2008102330960E00021.png"\; state="\{\{state\}\}">X</figure-callout>}}_m+{\mathrm{\&gamma;}}_{m2}{X}_{m2}+\mathrm{\&Lambda;}+{\mathrm{\&gamma;}}_{\mathrm{mp}}{X}_{\mathrm{mp}}\end{array}\right.$

4) data processing equipment utilizes the major component factor that step 3) obtains, and makes up the aging assessment models of paper oil insulation, the ageing state of assessment paper oil insulation through three layers of backpropagation (BackPropagation) neural network.

With the ageing state of the degree of polymerization as output expression paper oil insulation, consider the numerical value that the degree of polymerization is bigger, therefore at first the degree of polymerization is done following linear transformation:

$\mathrm{Ag}=\frac{1200}{950}-\frac{1}{950}\mathrm{DP}$

Ag=0 representes unagedly so, and Ag=1 then representes aging fully.According to the degree of polymerization value ageing state is divided into five stages of A～E, the degree of polymerization interval of each ageing step is respectively DP _A～DP _E, the Ag interval of corresponding different ageing steps is respectively Ag _A～Ag _EThe Ag value of the different ageing step samples that test obtains is respectively Ag ₁～Ag ₅Select backpropagation (Back Propagation) network in the artificial neural network to set up model, the aging shelf depreciation sample that test is obtained is divided into training sample S-train and test sample book S-test, chooses S-train and Ag ₁～Ag ₅As output come neural network training.According to the interval Ag of the differentiation of different ageing steps _A～Ag _E, the BP network that uses training to accomplish is assessed test sample book, and the result can discern the ageing state of paper oil insulation well.Concrete assessment models structure is as shown in Figure 2.

Assessment models adopts three layers of backpropagation (Back Propagation) neural network among Fig. 2; Be input as the shelf depreciation major component factor of extraction, be output as the Ag of expression oil paper insulation ageing state, network training adopts L-M (Levenberg-Marquardt) algorithm; At first data are carried out the normalization pre-service before the input data; The hidden layer transport function is selected tanh S type function, guarantees the nonlinear characteristic of network, and the output layer activation function is selected logarithm S type function; Guarantee that network output is between [0,1].Input neuron is counted n ₁Be the number of the major component factor extracted, the output layer neuron number is 1, and hidden layer neuron is counted n ₂According to formula n ₂=2n ₁+ 1 confirms.

More than be merely the preferred embodiments of the present invention, be not limited to the present invention, obviously, those skilled in the art can carry out various changes and modification and not break away from the spirit and scope of the present invention the present invention.Like this, belong within the scope of claim of the present invention and equivalent technologies thereof if of the present invention these are revised with modification, then the present invention also is intended to comprise these changes and modification interior.

Claims (4)

1. based on the method for evaluating oil paper insulation ageing state of local discharge characteristic parameter, it is characterized in that: comprise the steps:

1) local discharge signal of local discharge signal sensor acquisition paper oil insulation and be transferred to data processing equipment;

2) data processing equipment is analyzed local discharge signal, obtains local discharge characteristic parameter; Specifically comprise the steps

21) data processing equipment is according to the operating frequency phase of the generation of the shelf depreciation in the local discharge signal

Discharge capacity amplitude q and discharge time n obtain following four kinds of discharge collection of illustrative plates: maximum pd quantity PHASE DISTRIBUTION collection of illustrative plates Average discharge capacity PHASE DISTRIBUTION collection of illustrative plates

Discharge time PHASE DISTRIBUTION collection of illustrative plates

With discharge time discharge capacity collection of illustrative plates H _n(q);

22) data processing equipment extracts maximum pd quantity PHASE DISTRIBUTION collection of illustrative plates Average discharge capacity PHASE DISTRIBUTION collection of illustrative plates

With discharge time PHASE DISTRIBUTION collection of illustrative plates

In measure of skewness sk, steepness ku, peak value number peak, degree of asymmetry Asy and the related coefficient cc of positive-negative half-cycle, and discharge time discharge capacity collection of illustrative plates H _n(q) measure of skewness sk, steepness ku and peak value number peaks form primitive character parametric data matrix X as local discharge characteristic parameter with local discharge characteristic parameter;

3) data processing equipment is to step 2) local discharge characteristic parameter that obtains extracts the major component factor; Concrete steps are following:

31) data processing equipment carries out the standardization variation to the primitive character amount data matrix X of local discharge characteristic parameter composition, and concrete transform is as follows:

In the formula: X _IjBe local discharge characteristic amount raw data, Z _IjBe the data after the standardization,

And S _iMean value and the standard deviation of representing i characteristic quantity respectively;

32) data processing equipment carries out the KMO check of correlativity to the local discharge characteristic parameter standardized data; If the KMO test value is greater than 0.5; Then be for further processing; Otherwise the original variable before directly this local discharge characteristic parameter standardization being changed as the major component factor directly as the input of step 4), KMO value ε _KMOComputing formula be:

R in the formula _IjBe local discharge characteristic amount X _IjRelated coefficient, s _IjBe local discharge characteristic amount X _IjPartial correlation coefficient;

33) data processing equipment is to step 32) garbled local discharge characteristic parameter standardized data finds the solution covariance, obtains covariance matrix; The correlation matrix R of covariance matrix and original variable is equal, that is:

In the formula: Z _TiAnd Z _TjBe respectively the data after the standardization,

With

Be respectively the mean value of i characteristic quantity after the standardization;

34) data processing equipment carries out principal component analysis (PCA) to raw data correlation matrix correlation matrix R:

The characteristic root of correlation matrix R is λ ₁>=λ ₂>=λ ₃>=...>=λ _p＞0, characteristic of correspondence vector matrix U=[u ₁u ₂... u _p], according to Y=U ^T* Z has promptly obtained the data matrix Y that other one group of incoherent variable is formed; Select the characteristic root to analyze greater than the corresponding variable of 1 characteristic root; And calculate the contribution rate and the contribution rate of accumulative total of each major component, get contribution rate of accumulative total and surpass 90% a pairing m variable as the major component factor;

35) raw data matrix X is converted into major component factor data matrix F;

36) data processing equipment carries out LOAD FOR to the major component factor:

Matrix A is the factor loading matrix of p major component vector, and the pairing main gene loading matrix of preceding m the major component factor of extraction is:

37) data processing equipment is rotated analysis to the major component factor:

Select orthogonal matrix T as follows _KjPer two factors to m the major component factor are rotated angle

Make the population variance of postrotational factor loading arrive maximum:

38) data processing equipment calculates the major component factor score, according to the factor loading matrix A _mEach factor table is shown as the linear forms of variable X and factor score matrix of coefficients γ:

4) data processing equipment utilizes the major component factor that step 3) obtains, and makes up the aging assessment models of paper oil insulation through artificial neural network, obtains the ageing state of paper oil insulation.

2. the method for evaluating oil paper insulation ageing state based on local discharge characteristic parameter according to claim 1 is characterized in that: in the step 4), adopt the sign amount of degree of polymerization DP as oil paper insulation ageing state.

3. the method for evaluating oil paper insulation ageing state based on local discharge characteristic parameter according to claim 1 and 2 is characterized in that: in the step 4), adopt three layers of reverse transmittance nerve network to set up the aging assessment models of paper oil insulation.

4. the method for evaluating oil paper insulation ageing state based on local discharge characteristic parameter according to claim 3; It is characterized in that: the shelf depreciation major component factor that is input as extraction of the aging assessment models of paper oil insulation; Be output as oil paper insulation ageing state, at first data carried out the normalization pre-service before the data input, the hidden layer transport function is selected tanh S type function; The output layer activation function is selected logarithm S type function, and input neuron is counted n ₁Be the number of the major component factor extracted, the output layer neuron number is 1, and hidden layer neuron is counted n ₂According to formula n ₂=2n ₁+ 1 confirms.

Images