CN105223475A - Based on the shelf depreciation chromatogram characteristic algorithm for pattern recognition of Gaussian parameter matching - Google Patents

Abstract

The invention discloses a partial discharge spectrum feature pattern recognition algorithm based on Gaussian parameter fitting, including: a. acquiring the partial discharge spectrum in the process of partial discharge to be measured; b. based on the acquired partial discharge spectrum, using a Gaussian function to obtain the partial discharge spectrum to be measured The distance between the partial discharge to be identified and the standard partial discharge characteristic quantity in the partial discharge process; c. Based on the obtained distance between the partial discharge to be identified and the standard partial discharge characteristic quantity, the result of partial discharge pattern recognition is obtained and output. The partial discharge spectrogram feature pattern recognition algorithm based on Gaussian parameter fitting in the present invention can overcome the defects of poor anti-interference ability, high recognition difficulty and low reliability in the prior art, so as to achieve strong anti-interference ability, low recognition difficulty and The advantage of high reliability.

Description

Based on the shelf depreciation chromatogram characteristic algorithm for pattern recognition of Gaussian parameter matching

Technical field

The present invention relates to electric discharge collection of illustrative plates processing technology field, particularly, relate to the shelf depreciation chromatogram characteristic algorithm for pattern recognition based on Gaussian parameter matching.

Background technology

In order to meet the demand of electric system maintenance, the shelf depreciation online measuring technique of high-voltage electrical equipment, particularly gordian technique---PD Pattern Recognition wherein, obtains significant progress.For describing the characteristic quantity that one group of the partial discharges fault pattern recognition system parameter relevant with classification is exactly this recognition system.The basic task of Characteristic Extraction from all multiple features of system, how to find out those the most effective characteristic quantities and research how high-dimensional feature space is compressed to low dimensional feature space so that design category device effectively.

The nineties in last century, mode identification method starts the identification being applied to shelf depreciation type, judges to replace the range estimation of discharge spectrum.Significantly improve science and the validity of identification.Pattern recognition theory, towards intelligentized future development, namely strengthens the adaptive ability of system, learning ability and fault-tolerant ability etc.But, on the one hand because a lot of Uncertainty affects the collection of local discharge signal; On the other hand in Partial discharge signal comprise the intension of information and rule not yet completely clear, there is no a kind of diagnosis theory or standard of maturation so far; So PD Pattern Recognition technology is still in conceptual phase at present.

Realizing in process of the present invention, inventor finding at least to exist in prior art poor anti jamming capability, identifying the defects such as the large and reliability of difficulty is low.

Summary of the invention

The object of the invention is to, for the problems referred to above, propose the shelf depreciation chromatogram characteristic algorithm for pattern recognition based on Gaussian parameter matching, to realize the advantage that antijamming capability is strong, identification difficulty is little and reliability is high.

For achieving the above object, the technical solution used in the present invention is: based on the shelf depreciation chromatogram characteristic algorithm for pattern recognition of Gaussian parameter matching, comprising:

A, the shelf depreciation collection of illustrative plates obtained in shelf depreciation process to be measured;

B, based on the shelf depreciation collection of illustrative plates obtained, Gaussian function is utilized to obtain distance between shelf depreciation to be identified in shelf depreciation process to be measured and standard local discharge characteristic amount;

C, based on obtain shelf depreciation to be identified and standard local discharge characteristic amount between distance, obtain PD Pattern Recognition result and export.

Further, described step b, specifically comprises:

B1, based on obtain shelf depreciation collection of illustrative plates, to obtain shelf depreciation collection of illustrative plates carry out matching;

B2, based on the fitting result to shelf depreciation collection of illustrative plates, calculate Gaussian function characteristic parameter;

B3, based on calculating gained Gaussian function characteristic parameter, structure binary feature amount;

B4, based on structure gained binary feature amount, calculate the distance between shelf depreciation to be identified in shelf depreciation process to be measured and standard local discharge characteristic amount.

Further, described step b2, comprises further:

In described step b2, described Gaussian function characteristic parameter, comprises and " assembles " center ", " Spread scope " and " partition density ";

In described step b2, described Gaussian function is binary Gaussian function; The prototype function of described binary Gaussian function is:

$f(x,y)=\frac{1}{\sqrt{2\mathrm{\π}\left|\mathrm{\Σ}\right|}}\·\exp \[-\frac{1}{2}{(x-\mathrm{\μ})}^T\·{\mathrm{\Σ}}^{-1}\·(x-\mathrm{\μ})\];$

Wherein, average μ=(μ _x, μ _y), covariance matrix $\mathrm{\Σ}=\left[\begin{array}{cc}{\mathrm{\Σ}}_{11}& {\mathrm{\Σ}}_{12}\\ {\mathrm{\Σ}}_{21}& {\mathrm{\Σ}}_{22}\end{array}\right].$

Further, in described step b3, in the operation of described structure binary feature amount, specifically utilize following formula construction binary feature amount:

$X^{\left(1\right)}=\frac{{\mathrm{\σ}}_{x+}\·{\mathrm{\σ}}_{y-}}{{\mathrm{\μ}}_{y+}\·{\mathrm{\σ}}_{y+}}\·{\mathrm{\η}}^{+},$

$X^{\left(2\right)}=\frac{{\mathrm{\σ}}_{x-}\·{\mathrm{\σ}}_{y+}}{{\mathrm{\μ}}_{y-}\·{\mathrm{\σ}}_{y-}}\·{\mathrm{\η}}^{-};$

Wherein, μ _y+and μ _y-what represent the positive-negative half-cycle of Gaussian function " assembles the Y-axis component of " center ", σ _x+and σ _x-represent the X-axis component of " Spread scope " of the positive-negative half-cycle of Gaussian function, η ⁺and η ^-represent " partition density " of the positive-negative half-cycle of Gaussian function.

Further, in described step b4, in the operation of the shelf depreciation to be identified in described calculating shelf depreciation to be measured process and the distance between standard local discharge characteristic amount, specifically utilize the distance between following formulae discovery shelf depreciation to be identified and standard local discharge characteristic amount:

$D_j^2={(X_a^{\left(1\right)}-X_j^{\left(1\right)})}^2+{(X_a^{\left(2\right)}-X_j^{\left(2\right)})}^2;$

In above formula, D _jfor the distance between shelf depreciation to be identified and standard local discharge characteristic amount, for local discharge characteristic amount to be identified, for standard local discharge characteristic amount.

Further, described step a, specifically comprises:

A1, in the measurement of partial discharge of shelf depreciation process to be measured, obtain shelf depreciation discharge capacity Q and local discharge phase angle φ;

A2, to partial discharge quantity Q at [0, max (Q _i)] interval quantizes, discharge phase angle, local φ is quantized on [0,360] interval, obtains quantizing grid;

A3, the shelf depreciation number of times added up in each quantification grid, calculate shelf depreciation spectrogram.

Further, in described step a3, also comprise the generation step of shelf depreciation spectrogram, that is:

There is the distribution function at phasing degree in the discharge capacity obtaining shelf depreciation in measurement of partial discharge, utilizes distribution function to obtain discharge capacity Q and the local discharge phase angle φ of shelf depreciation to electric discharge.

Further, described step c, specifically comprises:

Structure shelf depreciation probability distribution function, and the result of shelf depreciation probability distribution function as PD Pattern Recognition is exported.

Further, described in described step c, the computing formula of shelf depreciation probability distribution function is as follows:

$P_j=\frac{1/D_j}{{\Σ}_{j\∈\{1,2,\mathrm{...}M\}}1/D_j};$

J ∈ in above formula 1,2 ... M} is the kind of standard partial discharge quantity, P _jfor shelf depreciation probability distribution function.

The shelf depreciation chromatogram characteristic algorithm for pattern recognition based on Gaussian parameter matching of various embodiments of the present invention, owing to comprising: a, the shelf depreciation collection of illustrative plates obtained in shelf depreciation process to be measured; B, based on the shelf depreciation collection of illustrative plates obtained, Gaussian function is utilized to obtain distance between shelf depreciation to be identified in shelf depreciation process to be measured and standard local discharge characteristic amount; C, based on obtain shelf depreciation to be identified and standard local discharge characteristic amount between distance, obtain PD Pattern Recognition result and export; Thus poor anti jamming capability in prior art can be overcome, identify the large and defect that reliability is low of difficulty, with realize antijamming capability strong, identify the advantage that difficulty is little and reliability is high.

Other features and advantages of the present invention will be set forth in the following description, and, partly become apparent from instructions, or understand by implementing the present invention.

Below by drawings and Examples, technical scheme of the present invention is described in further detail.

Accompanying drawing explanation

Accompanying drawing is used to provide a further understanding of the present invention, and forms a part for instructions, together with embodiments of the present invention for explaining the present invention, is not construed as limiting the invention.In the accompanying drawings:

Fig. 1 is the process flow block diagram of in the present invention, local discharge spectrum distribution characteristics being carried out to the algorithm for pattern recognition of Gaussian parameter matching;

Fig. 2 is the surface fitting to the binary Gaussian distribution of electric discharge loose some spectrogram in Q-φ plane in the present invention;

Fig. 3 is the N-Q-φ distributed in three dimensions spectrogram of a corona discharge in the present invention;

Fig. 4 is the binary gauss of distribution function simulated the discharge spectrum in Fig. 3;

Fig. 5 is the characteristic parameter that the discharge spectrum calculated in the present invention distributes at positive and negative half cycle;

Fig. 6 is the X of neotectonics in the present invention ⁽¹⁾and X ⁽²⁾characteristic quantity space to the gathering coordinate of several typical discharges type, wherein, corona discharge in (a) oil, (b) internal air gap discharge, (c) suspended discharge;

Fig. 7 is by discharge probability distribution plan that distance classifier constructs in the present invention;

Fig. 8 is the curve of gaussian curve approximation in the present invention;

Fig. 9 is the process flow diagram of iterative process in the present invention.

Embodiment

Below in conjunction with accompanying drawing, the preferred embodiments of the present invention are described, should be appreciated that preferred embodiment described herein is only for instruction and explanation of the present invention, is not intended to limit the present invention.

For the distribution characteristics of shelf depreciation phase place spectrogram, according to the embodiment of the present invention, as shown in Fig. 1-Fig. 9, provide the shelf depreciation chromatogram characteristic algorithm for pattern recognition based on Gaussian parameter matching, namely local discharge spectrum distribution characteristics is carried out to the algorithm for pattern recognition of Gaussian parameter matching.

Technical scheme of the present invention, adopts the loose distribution in gauss of distribution function matching discharge spectrum, extraction Gaussian function " correlation parameter such as gathering " center ", " Spread scope " and " partition density " etc., then re-construct two new characteristic quantity X ⁽¹⁾and X ⁽²⁾, by calculating the statistical distance D between discharge characteristic amount to be identified and criteria classification discharge characteristic amount _j, by the probability distribution P of electric discharge to be identified on each standard discharge mode _jresult as PD Pattern Recognition algorithm exports.

The shelf depreciation chromatogram characteristic algorithm for pattern recognition based on Gaussian parameter matching of technical solution of the present invention, can the dimension of effective compressive features amount parameter space, is conducive to the design structure of feature classifiers; From the classifying quality to typical discharges, should choose and can reach maximized Classification and Identification to corona discharge type based on the characteristic quantity of the shelf depreciation chromatogram characteristic algorithm for pattern recognition of Gaussian parameter matching.

The technical solution used in the present invention is: a kind of shelf depreciation chromatogram characteristic mode identification method based on Gaussian parameter matching, comprises the following steps:

S100, calculating shelf depreciation spectrogram; Step S100 is specially:

S1001, the discharge capacity Q obtaining shelf depreciation in measurement of partial discharge and local discharge phase angle φ;

S1002, to partial discharge quantity Q at [0, max (Q _i)] interval quantizes, discharge phase angle, local φ is quantized on [0,360] interval, obtains quantizing grid;

S1003, the shelf depreciation number of times added up in each quantification grid, calculate shelf depreciation spectrogram; Step S1003 is specially:

Add up the shelf depreciation number of times in each quantification grid, such as three point of discharges in discharge spectrum are (85.4 respectively, 159.2), (88.7,168.2), (93.1,187.5), horizontal ordinate is discharge phase angle, and ordinate is discharge capacity.If discharge phase quantized interval is [80,90], a quantized interval of discharge capacity is [150,180], so, the first two point of discharge just drops in this electric discharge block, note discharge time is that the 2, three point of discharge does not drop in this block, in next block count.The PRPD calculating shelf depreciation N-Q-φ distributed in three dimensions thus adds up spectrogram.

Step S100 describes the generation step of shelf depreciation spectrogram further, there is the distribution function at phasing degree in the discharge capacity wherein obtaining shelf depreciation in measurement of partial discharge, utilizes described distribution function to obtain discharge capacity Q and the local discharge phase angle φ of shelf depreciation to electric discharge.

S200, Gaussian function is utilized to carry out matching to discharge spectrum;

The Fundamentals of Mathematics of Gaussian function fitting:

By recursive call generalized least square method method and Levenberg-Marquardt method, data fitting is made to be the Gaussian curve that form is described by following equalities:

f＝a*exp-x-μ22σ2+c

X is list entries X, a is amplitude, and μ is center, and σ is standard deviation, and c is side-play amount.

Search a of best-fit observation (X, Y), the value of μ, σ and c, following equalities is for describing the Gaussian curve obtained by Gauss curve fitting algorithm:

yi＝a*exp-xi-μ22σ2+c

Noise as Y is Gaussian distribution, can use least square method.Fig. 8 is the gaussian curve approximation using the method.

If approximating method is least square method, adopt iterative process to obtain the amplitude of exponential model, center, standard deviation and side-play amount, then use the formulae discovery residual error in least square method.Iterative process as shown in Figure 9.

Least residual computing formula is as shown in following:

1N _i＝0N-1w _if _i-y _i2

N is the length of Y, w _ii-th element of weight, f _ii-th element of best Gauss curve fitting, y _ii-th element of Y.

S300, calculating Gaussian function characteristic parameter; The Gaussian function characteristic parameter that step S300 mentions, comprises and " assembles " center ", " Spread scope " and " partition density ".In step S300, binary Gaussian function has following prototype function:

$f(x,y)=\frac{1}{\sqrt{2\mathrm{\π}\left|\mathrm{\Σ}\right|}}\·\exp \[-\frac{1}{2}{(x-\mathrm{\μ})}^T\·{\mathrm{\Σ}}^{-1}\·(x-\mathrm{\μ})\].$

Wherein, average μ=(μ _x, μ _y), covariance matrix $\mathrm{\Σ}=\left[\begin{array}{cc}{\mathrm{\Σ}}_{11}& {\mathrm{\Σ}}_{12}\\ {\mathrm{\Σ}}_{21}& {\mathrm{\Σ}}_{22}\end{array}\right].$ In general, electric discharge convergence point is at positive half cycle phase space [0,180 °] and negative half period phase space [180 °, 360 °] on distribution meet Gauss's (normal state) distribution, discharge phase φ (transverse axis) and discharge capacity Q (longitudinal axis) are not always the case, so utilize binary Gaussian function to carry out Two-dimensional Surfaces matching to the N-Q-φ three-dimensional feature spectrogram distribution scatter diagram mentioned in S100, as shown in Figure 2.

The distribution characteristics of N-Q-φ can represent by following three components cloth parameter:

One is the gathering center of Gaussian distribution: μ=(μ _x, μ _y), there is μ respectively in positive-negative half-cycle ⁺and μ ^-;

Two is Spread scopes of Gaussian distribution: σ=(σ _x, σ _y)=(sqrt (Σ ₁₁), sqrt (Σ ₂₂)), same, there is σ respectively in positive-negative half-cycle ⁺and σ ^-;

Three is Gaussian distribution partition densities in positive-negative half-cycle: η=(η ⁺, η ^-), the density calculation according to loose distribution draws.

S400, based on Gaussian function characteristic parameter structure binary feature amount; Binary feature amount described in formula construction is below utilized in step S400:

$X^{\left(1\right)}=\frac{{\mathrm{\σ}}_{x+}\·{\mathrm{\σ}}_{y-}}{{\mathrm{\μ}}_{y+}\·{\mathrm{\σ}}_{y+}}\·{\mathrm{\η}}^{+};$

$X^{\left(2\right)}=\frac{{\mathrm{\σ}}_{x-}\·{\mathrm{\σ}}_{y+}}{{\mathrm{\μ}}_{y-}\·{\mathrm{\σ}}_{y-}}\·{\mathrm{\η}}^{-}.$

Wherein, μ _y+and μ _y-what represent the positive-negative half-cycle of Gaussian function " assembles the Y-axis component of " center ", σ _x+and σ _x-represent the X-axis component of " Spread scope " of the positive-negative half-cycle of Gaussian function, η ⁺and η ^-represent " partition density " of the positive-negative half-cycle of Gaussian function.

S500, calculate distance between shelf depreciation to be identified and standard local discharge characteristic amount; Following formula is utilized to calculate distance between shelf depreciation to be identified and standard local discharge characteristic amount in step S500:

$D_j^2={(X_a^{\left(1\right)}-X_j^{\left(1\right)})}^2+{(X_a^{\left(2\right)}-X_j^{\left(2\right)})}^2$

In above formula, D _jfor the distance between shelf depreciation to be identified and standard local discharge characteristic amount, for local discharge characteristic amount to be identified, for standard local discharge characteristic amount.

In step S500, each band is identified that local discharge characteristic amount and standard local discharge characteristic amount utilize above-mentioned formula to carry out the calculating of distance, D _jless, then electric discharge to be identified is more close to the standard electric discharge of a jth classification.Following formula is utilized to calculate distance between shelf depreciation to be identified and standard local discharge characteristic amount in step S500:

$D_j^2={(X_a^{\left(1\right)}-X_j^{\left(1\right)})}^2+{(X_a^{\left(2\right)}-X_j^{\left(2\right)})}^2$

In above formula, D _jfor the distance between shelf depreciation to be identified and standard local discharge characteristic amount, for local discharge characteristic amount to be identified, for standard local discharge characteristic amount.

S600, structure shelf depreciation probability distribution function, and the result of shelf depreciation probability distribution function as PD Pattern Recognition is exported; In step S600, the computing formula of shelf depreciation probability distribution function is as follows:

$P_j=\frac{1/D_j}{{\Σ}_{j\∈\{1,2,\mathrm{...}M\}}1/D_j}$

J ∈ in above formula 1,2 ... M} is the kind of standard partial discharge quantity, P _jfor shelf depreciation probability distribution function.

Such as, when specifically implementing, the shelf depreciation chromatogram characteristic mode identification method based on Gaussian parameter matching of technical solution of the present invention, can comprise the following steps:

Q1. the calculating of shelf depreciation phase place spectrogram is resolved

To be classified by standard discharging model the test data of a group " in oil corona discharge " recording, have recorded discharge capacity Q and discharge phase φ.The teachings of discharge phase φ is [0,360 °], and transverse axis quantizes with 400 calibration; Pick and place the maximal value Q in electricity Q array _max, calculate Q _i/ Q _maxobtain relative discharge amount, variation range, at [0,100%], the longitudinal axis quantizes with 100 calibration.Then statistics drops on each and quantizes discharge time in grid as N, so just obtain and represent corona discharge three-dimensional feature distribution function N-Q-φ spectrogram in oil, as shown in Figure 3, wherein transverse axis is discharge phase φ, the longitudinal axis is discharge capacity Q, and discharge time N RGB color range represents.

Q2. the matching of hybrid parameter gauss of distribution function

Statistics software toolkit in MATLAB has Gauss curve fitting function gmdistribution.fit, input parameter is the N-Q-φ spectral data rawdata obtained in Q1, and threshold coefficient h, block count N, the Gauss curved function mgaus that output parameter is matching and goes out, and the principal character parameter info structure of this binary Gaussian function.Wherein, threshold coefficient h, in order to determine the Spread scope on Gauss border, can be described as relaxation factor, using isocontour relative value as value principle.

In shelf depreciation test, distributed by discharge spectrum and seek to discharge whether there is correlativity on operating frequency phase, so block count N is generally taken as 2, as shown in Figure 4, threshold coefficient is wherein h=0.9 to the Gauss curved isogram gone out by " in oil corona discharge " test data rawdata matching of example in Q1.

Q3. the calculating of Gaussian function characteristic parameter

By the gathering center μ=(μ containing Gaussian distribution in the info structure calculated in Q2 _x, μ _y), the ComponentMeans exported by gmdistribu-tion.fit function calculates; The Spread scope of Gaussian distribution: σ=(σ _x, σ _y), the ComponentCovariances exported by gmdistribution.fit function gets diagonal line root and calculates; Gaussian distribution is at the partition density of positive-negative half-cycle: η=(η ⁺, η ^-), the MixtureProportions exported by gmdistribution.fit function draws.

By " in oil corona discharge " test data of example in Q1 gaussian distribution characteristic parameter as shown in Figure 5.

Q4. neotectonics binary feature amount

By the binary feature amount X of neotectonics ⁽¹⁾and X ⁽²⁾represent the N-Q-φ spectrogram distribution characteristics of standard discharging model, obtain in laboratory examination: 1. " in oil corona discharge " test data 2 groups; 2. " bubble-discharge " test data 2 groups; 3. " oil clearance electric discharge " test data 2 groups; 4. " suspended discharge " test data 8 groups; 5. " surface-discharge " test data 1 group.These typical discharges are illustrated respectively in X ⁽¹⁾and X ⁽²⁾feature space, as shown in (a), (b), (c) in Fig. 6.

Q5. the designing and calculating of distance classifier

Newly-increased once new discharge test, electric discharge type is unknown, the characteristic parameter obtained by discharge spectrum: X ⁽¹⁾=8.2, X ⁽²⁾=6.0, calculate " Euclidean distance " D between this discharge characteristic and standard discharge characteristic respectively _j.

Q6. the design structure of discharge probability distribution function

With the probability distribution function P of neotectonics _jrepresent the similarity between unknown discharge characteristic and standard discharge characteristic, corona discharge has the distribution probability of 94.2%, is the maximum probability in various electric discharge classification, therefore, this electric discharge can be determined as " corona discharge " type, as shown in Figure 7.

The beneficial effect of technical solution of the present invention is: the phase place atlas image of this recognizer to shelf depreciation is compressed to the characteristic parameter space only having two dimensions effectively, is conducive to the design structure of feature classifiers.From the classifying quality to typical discharges, the characteristic quantity that this recognizer is chosen can reach maximized discriminator to corona discharge type.

Last it is noted that the foregoing is only the preferred embodiments of the present invention, be not limited to the present invention, although with reference to previous embodiment to invention has been detailed description, for a person skilled in the art, it still can be modified to the technical scheme described in foregoing embodiments, or carries out equivalent replacement to wherein portion of techniques feature.Within the spirit and principles in the present invention all, any amendment done, equivalent replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (9)

1. The partial discharge spectrogram feature pattern recognition algorithm based on Gaussian parameter fitting, is characterized in that, comprises: a. Obtain the partial discharge spectrum in the process of partial discharge to be measured; b. Based on the acquired partial discharge atlas, a Gaussian function is used to obtain the distance between the partial discharge to be identified and the standard partial discharge characteristic quantity in the partial discharge process to be measured; c. Based on the obtained distance between the partial discharge to be identified and the standard partial discharge feature quantity, obtain and output the result of partial discharge pattern recognition. 2. the partial discharge spectrogram feature pattern recognition algorithm based on Gaussian parameter fitting according to claim 1, is characterized in that, described step b, specifically comprises: b1, based on the acquired partial discharge atlas, fitting the acquired partial discharge atlas; b2. Based on the fitting results of the partial discharge spectrum, calculate the characteristic parameters of the Gaussian function; b3. Construct binary feature quantities based on the calculated Gaussian function feature parameters; b4. Calculate the distance between the partial discharge to be identified and the standard partial discharge characteristic quantity in the partial discharge process to be measured based on the obtained binary characteristic quantity. 3. the partial discharge spectrogram feature pattern recognition algorithm based on Gaussian parameter fitting according to claim 2, is characterized in that, described step b2, further comprises: In the step b2, the characteristic parameters of the Gaussian function include "aggregation center", "diffusion range" and "distribution proportion"; In the step b2, the Gaussian function is a binary Gaussian function; the prototype function of the binary Gaussian function is: $\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\sqrt{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; \&pi;</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}<span\; class="notranslate">\; |</span>}}<span\; class="notranslate">\; \&CenterDot;</span>\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; \&lsqb;</span><span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&mu;</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}^{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&mu;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&rsqb;</span><span\; class="notranslate">\; ;</span>$ Among them, mean value μ=(μ _x ,μ _y ), covariance matrix $\mathrm{\&Sigma;}=\left[\begin{array}{cc}{\mathrm{\&Sigma;}}_{11}& {\mathrm{\&Sigma;}}_{12}\\ {\mathrm{\&Sigma;}}_{\mathrm{twenty\; one}}& {\mathrm{\&Sigma;}}_{\mathrm{twenty\; two}}\end{array}\right].$ 4. the partial discharge spectrogram characteristic pattern recognition algorithm based on Gaussian parameter fitting according to claim 3, is characterized in that, in described step b3, in the operation of described construction binary feature quantity, specifically utilize following The formula constructs binary features: ${\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>}<span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; -</span>}}{{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>}<span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>}}<span\; class="notranslate">\; \&CenterDot;</span>{\mathrm{<span\; class="notranslate">\; \&eta;</span>}}^{<span\; class="notranslate">\; +</span>}<span\; class="notranslate">\; ,</span>$ ${\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>}<span\; class="notranslate">\; \&CenterDot;</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>}}{{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; -</span>}<span\; class="notranslate">\; \&CenterDot;</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; -</span>}}<span\; class="notranslate">\; \&CenterDot;</span>{\mathrm{<span\; class="notranslate">\; \&eta;</span>}}^{<span\; class="notranslate">\; -</span>}<span\; class="notranslate">\; ;</span>$ Among them, μ _y+ and μ _y- represent the Y-axis component of the "gathering center" of the positive and negative half-cycle of the Gaussian function, σ _x+ and σ _x- represent the X-axis component of the "diffusion range" of the positive and negative half-cycle of the Gaussian function, η ⁺ and η ^- denote the "distribution weight" of the positive and negative half cycles of the Gaussian function. 5. the partial discharge spectrogram feature pattern recognition algorithm based on Gaussian parameter fitting according to claim 4, is characterized in that, in described step b4, described calculation is to be identified partial discharge and in the partial discharge process to be measured In the operation of the distance between the standard partial discharge characteristic quantities, the following formula is used to calculate the distance between the partial discharge to be identified and the standard partial discharge characteristic quantity: ${\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; a</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; a</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ;</span>$ In the above formula, D _j is the distance between the partial discharge to be identified and the standard partial discharge characteristic quantity, is the characteristic quantity of partial discharge to be identified, is the standard partial discharge characteristic quantity. 6. according to the partial discharge spectrogram characteristic pattern recognition algorithm based on Gaussian parameter fitting according to any one of claim 1-5, it is characterized in that, described step a specifically comprises: a1. In the partial discharge measurement of the partial discharge process to be measured, obtain the partial discharge discharge quantity Q and the partial discharge phase angle φ; a2. Quantify the partial discharge quantity Q in the [0, max(Q _i )] interval, and quantize the partial discharge phase angle φ in the [0, 360] interval to obtain a quantized grid; a3. Count the number of partial discharges in each quantized grid, and calculate the partial discharge spectrum. 7. the partial discharge spectrogram feature pattern recognition algorithm based on Gaussian parameter fitting according to claim 6, is characterized in that, in described step a3, also comprises the generation step of partial discharge spectrogram, namely: In the partial discharge measurement, the distribution function of the discharge amount of partial discharge to the phase angle of discharge occurrence is obtained, and the discharge amount Q of partial discharge and the phase angle φ of partial discharge are obtained by using the distribution function. 8. The partial discharge spectrogram feature pattern recognition algorithm based on Gaussian parameter fitting according to any one of claims 1-5, wherein said step c specifically comprises: The partial discharge probability distribution function is constructed, and the partial discharge probability distribution function is output as the result of partial discharge pattern recognition. 9. the partial discharge spectrogram feature pattern recognition algorithm based on Gaussian parameter fitting according to claim 8, is characterized in that, the computing formula of described partial discharge probability distribution function in described step c is as follows: ${\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; \&Element;</span><span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; ...</span>}\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; \}</span>}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}<span\; class="notranslate">\; ;</span>$ In the above formula, j∈{1,2,...M} is the type of standard partial discharge, and P _j is the probability distribution function of partial discharge.