CN110161389A - A kind of Electric Power Equipment Insulation defect identification method and AEVB self-encoding encoder - Google Patents

Abstract

The invention discloses a kind of Electric Power Equipment Insulation defect identification methods comprising step: (1) constructing AEVB self-encoding encoder, AEVB self-encoding encoder includes probability encoding device and probability decoder；(2) AEVB self-encoding encoder is trained using the history Partial Discharge Data of power equipment, so that: characteristic value of the history Partial Discharge Data of input in the output end output Partial Discharge Data of probability encoding device, the characteristic value input probability decoder of the Partial Discharge Data, to export corresponding insulation defect type in the output end of probability decoder；(3) by the local discharge signal input AEVB self-encoding encoder of power equipment to be identified, AEVB self-encoding encoder exports insulation defect type.In addition, the invention also discloses a kind of AEVB self-encoding encoder for Electric Power Equipment Insulation defect recognition, AEVB self-encoding encoder includes probability encoding device and probability decoder.

Description

A kind of Electric Power Equipment Insulation defect identification method and AEVB self-encoding encoder

Technical field

The present invention relates to a kind of recognition methods and its neural network more particularly to a kind of identifications of Electric Power Equipment Insulation defect Method and its neural network.

Background technique

Currently used insulation defect diagnostic method is to be constructed of partial discharge map progress feature to local discharge information to mention It takes and pattern-recognition.Rapidly and accurately carrying out identification to the partial discharge map of acquisition facilitates insulation defect diagnosis, and then grasps electricity The state of insulation of power equipment.It can be repaired at the first time in the presence of having insulation defect.But current partial discharge spectrum recognition Technology still remains the lower problem of accuracy rate, and failure is easy to cause to misjudge.

Based on this, it is expected that obtaining a kind of Electric Power Equipment Insulation defect identification method, partial discharge map can be significantly improved Recognition accuracy, to preferably assess status of electric power, to be conducive to grasp the state of insulation of power equipment.

Summary of the invention

One of the objects of the present invention is to provide a kind of Electric Power Equipment Insulation defect identification method, which is lacked Sunken recognition methods can effectively improve the recognition accuracy of partial discharge map, so that preferably status of electric power is assessed, To be conducive to grasp the state of insulation of power equipment.

Based on above-mentioned purpose, the invention proposes a kind of Electric Power Equipment Insulation defect identification methods comprising step:

(1) AEVB self-encoding encoder is constructed, AEVB self-encoding encoder includes probability encoding device and probability decoder；

(2) AEVB self-encoding encoder is trained using the history Partial Discharge Data of power equipment, so that: input is gone through Characteristic value of the history Partial Discharge Data in the output end output Partial Discharge Data of probability encoding device, the spy of the Partial Discharge Data Value indicative input probability decoder, to export corresponding insulation defect type in the output end of probability decoder；

(3) by the local discharge signal input AEVB self-encoding encoder of power equipment to be identified, AEVB self-encoding encoder is defeated Insulation defect type out.

It is accurate to the identification of partial discharge map in order to improve in Electric Power Equipment Insulation defect identification method of the present invention Rate, inventor propose the Electric Power Equipment Insulation defect identification method of this case according to depth self-encoding encoder matching algorithm, first Variation Bayes self-encoding encoder (hereinafter referred to as AEVB self-encoding encoder) first is constructed, which includes probability encoding device And probability decoder, then AEVB self-encoding encoder is trained using the history Partial Discharge Data of power equipment, so that: it is defeated Characteristic value of the history Partial Discharge Data entered in the output end output Partial Discharge Data of probability encoding device, the shelf depreciation number According to characteristic value input probability decoder, to export corresponding insulation defect type, the history in the output end of probability decoder Partial Discharge Data can be acquired from case information library, finally, by the local discharge signal of power equipment to be identified AEVB self-encoding encoder is inputted, insulation defect type is exported by AEVB self-encoding encoder.

As a result, the Electric Power Equipment Insulation defect identification method by this case can be accurately and effectively to power equipment Operating condition carries out assessment understanding, the recognition accuracy of partial discharge map is improved, to preferably comment status of electric power Estimate, to be conducive to grasp the state of insulation of power equipment.

Further, in Electric Power Equipment Insulation defect identification method of the present invention, AEVB self-encoding encoder has one A input layer, an output layer, a hidden variable layer and four middle layers；One of input layer, two middle layers and one Hidden variable layer constitutes the probability encoding device, and hidden variable layer exports the characteristic value of Partial Discharge Data；Two middle layers and one A output layer constitutes the probability decoder, and the output layer exports insulation defect type.

Further, in Electric Power Equipment Insulation defect identification method of the present invention, using stochastic gradient descent method Optimize the parameter of probability encoding device and probability decoder.

Further, in Electric Power Equipment Insulation defect identification method of the present invention, probability encoding device is using following Formula characterization:

Z indicates the output of hidden variable layer in formula, is the characteristic value of Partial Discharge Data；σ_encIndicate scale parameter；ε table Show the random parameter for meeting N (0,1) distribution；N (0, I) is standardized normal distribution；F indicates activation primitive；Expression will solve Scale parameter；Indicate the translational movement to be solved；Indicate the scale parameter to be solved；Indicate the translation to be solved Amount；Indicate the h to be solved_lScale parameter；Indicate the h to be solved_lTranslational movement；X is the Partial Discharge Data of input.

Further, in Electric Power Equipment Insulation defect identification method of the present invention, probability decoder is using following Formula characterization:

In formula, x indicates the Partial Discharge Data of input；The characteristic value of z expression Partial Discharge Data；P (x | z) indicate condition Probability-distribution function；N indicates normal distyribution function；μ_decIndicate probability likelihood parameter, σ_decIndicate scale parameter；I indicates offset Amount；Indicate the scale parameter to be solved；Indicate the translational movement to be solved；Indicate the scale parameter to be solved；Indicate the translational movement to be solved；Indicate the h to be solved₂Scale parameter；Indicate the h to be solved₂Translational movement.

Correspondingly, another object of the present invention is to provide a kind of AEVB for Electric Power Equipment Insulation defect recognition is self-editing Code device, can effectively improve the recognition accuracy of partial discharge map by the AEVB self-encoding encoder, thus preferably to power equipment State is assessed, to be conducive to grasp the state of insulation of power equipment.

Based on above-mentioned purpose, the invention also provides a kind of AEV B for Electric Power Equipment Insulation defect recognition to encode certainly Device, AEVB self-encoding encoder include probability encoding device and probability decoder；Wherein probability encoding device is configured to: to its input electric power The local discharge signal of equipment, then its export Partial Discharge Data characteristic value；The probability decoder is configured to: it is inputted The output of input probability encoder is held, insulation defect type is exported.

Further, in AEVB self-encoding encoder of the present invention, AEVB self-encoding encoder has an input layer, one Output layer, a hidden variable layer and four middle layers；One of input layer, two middle layers and a hidden variable layer constitute The probability encoding device, the characteristic value of the hidden variable layer output Partial Discharge Data；Two middle layers and an output layer group At the probability decoder, the output layer exports insulation defect type.

Further, in AEVB self-encoding encoder of the present invention, the parameter of probability encoding device and probability decoder is adopted It is optimized with stochastic gradient descent method.

Further, in AEVB self-encoding encoder of the present invention, probability encoding device is characterized using following formula:

Z indicates the output of hidden variable layer in formula, is the characteristic value of Partial Discharge Data；σ_encIndicate scale parameter；ε table Show the random parameter for meeting N (0,1) distribution；N (0, I) is standardized normal distribution；F indicates activation primitive；Expression will solve Scale parameter；Indicate the translational movement to be solved；Indicate the scale parameter to be solved；Indicate the translation to be solved Amount；Indicate the scale parameter to be solved；Indicate the translational movement to be solved；X is the Partial Discharge Data of input.

Further, in AEVB self-encoding encoder of the present invention, probability decoder is characterized using following formula:

In formula, x indicates the Partial Discharge Data of input；The characteristic value of z expression Partial Discharge Data；P (x | z) indicate condition Probability-distribution function；N indicates normal distyribution function；μ_decIndicate probability likelihood parameter, σ_decIndicate scale parameter；I indicates offset Amount；Indicate the scale parameter to be solved；Indicate the translational movement to be solved；Indicate the scale parameter to be solved； Indicate the translational movement to be solved；Indicate the h to be solved₂Scale parameter；Indicate the h to be solved₂Translational movement.

Electric Power Equipment Insulation defect identification method of the present invention and AEVB self-encoding encoder have the advantages described below And the utility model has the advantages that

Electric Power Equipment Insulation defect identification method of the present invention can effectively improve the recognition accuracy of partial discharge map, To preferably assess status of electric power, to be conducive to grasp the state of insulation of power equipment.

In addition, AEVB self-encoding encoder of the present invention similarly has above advantages and beneficial effect.

Detailed description of the invention

Fig. 1 is the process signal of Electric Power Equipment Insulation defect identification method of the present invention in one embodiment Figure.

Fig. 2 schematically shows the structure of AEVB self-encoding encoder of the present invention in one embodiment.

Fig. 3 to Fig. 6 respectively illustrates Electric Power Equipment Insulation defect identification method of the present invention in different embodiment party Detected detection data in formula.

Specific embodiment

It below will according to specific embodiment and Figure of description is to Electric Power Equipment Insulation defect recognition side of the present invention Method and AEVB self-encoding encoder are described further, but the explanation does not constitute the improper restriction to technical solution of the present invention.

Fig. 1 is the process signal of Electric Power Equipment Insulation defect identification method of the present invention in one embodiment Figure.

As shown in Figure 1, in the present embodiment, Electric Power Equipment Insulation defect identification method comprising steps of

Step 100: building AEVB self-encoding encoder, the AEVB self-encoding encoder includes probability encoding device and probability decoder；

Step 200: AEVB self-encoding encoder is trained using the history Partial Discharge Data of power equipment, so that: it is defeated Characteristic value of the history Partial Discharge Data entered in the output end output Partial Discharge Data of probability encoding device, the shelf depreciation number According to characteristic value input probability decoder, to export corresponding insulation defect type in the output end of probability decoder；

Step 300: by the local discharge signal input AEVB self-encoding encoder of power equipment to be identified, AEVB is encoded certainly Device exports insulation defect type.

It should be noted that AEVB self-encoding encoder is with an input layer, an output layer, a hidden variable layer and four Middle layer；One of input layer, two middle layers and a hidden variable layer constitute the probability encoding device, the hidden variable The characteristic value of layer output Partial Discharge Data；Two middle layers and an output layer constitute the probability decoder, output layer Export insulation defect type.

And Fig. 2 schematically shows the structure of AEVB self-encoding encoder of the present invention in one embodiment.

As shown in Fig. 2, using the parameter of stochastic gradient descent method optimization probability encoding device and probability decoder.Wherein, generally Rate encoder is characterized using following formula:

Z indicates the output of hidden variable layer in formula, is the characteristic value of Partial Discharge Data；σ_encIndicate scale parameter；ε table Show the random parameter for meeting N (0,1) distribution；N (0, I) is standardized normal distribution；F indicates activation primitive；Expression will solve Scale parameter；Indicate the translational movement to be solved；Indicate the scale parameter to be solved；Indicate the translation to be solved Amount；Indicate the scale parameter to be solved；Indicate the translational movement to be solved；X is the Partial Discharge Data of input.

And probability decoder is characterized using following formula:

In formula, x indicates the Partial Discharge Data of input；The characteristic value of z expression Partial Discharge Data；P (x | z) indicate condition Probability-distribution function；N indicates normal distyribution function；μ_decIndicate probability likelihood parameter, σ_decIndicate scale parameter；I indicates offset Amount；Indicate the scale parameter to be solved；Indicate the translational movement to be solved；Indicate the scale parameter to be solved； Indicate the translational movement to be solved；Indicate the h to be solved₂Scale parameter；Indicate the h to be solved₂Translational movement.

In order to be better described this case Electric Power Equipment Insulation defect identification method recognition effect, Fig. 3 to Fig. 5 shows respectively It has anticipated Electric Power Equipment Insulation defect identification method of the present invention detected detection data in various embodiments.

The relationship explanation that table 1 lists the electric discharge type that different local discharge signals detects between position of discharging.

Table 1.

Case No	Electric discharge type	Electric discharge position
			1	Suspended discharge	Isolation switch insulated pull rod and transmission mechanism connecting portion
2	Suspended discharge	Isolation switch insulated pull rod and transmission mechanism connecting portion
			3	Suspended discharge	Built-in sensor joint area
4	The class that insulate electric discharge	The tag of GIS cable termination storehouse is damaged

In addition, it should be noted that, case information is as shown in table 1.Wherein case 1 is identical equipment factory with case 2 Family, same model equipment, the electric discharge case detected in same position.Case 3 is identical as case 1 and 2 electric discharge type of case, but Equipment manufacturer and electric discharge position are different, and case 4 is the comparison case of different electric discharge types.It is detected in above four cases The Partial Discharge Data arrived is as shown in Figures 3 to 6.By the various features for extracting Fig. 3 to Partial Discharge Data shown in fig. 6 Value, mutual matching degree is calculated in conjunction with cosine-algorithm, obtains the flux matched degree result of different characteristic as shown in following table table 2.

Table 2.

Note: statistical characteristics can obtain in the following way in table 2: statistical characteristics be by shelf depreciation amplitude and time Number is in entire power frequency period and the degree of skewness Sk of power frequency positive-negative half-cycle, steepness Ku, degree of asymmetry Q and cross-correlation coefficient Cc etc. 16 A characteristic parameter composition；

DBN characteristic value obtains in the following way: deepness belief network is that Boltzmann machine superposition extension is limited by multilayer Made of network structure model.It is herein 6 layers for the DBN network of comparison, number of nodes 3600,1000,500,100,10, 4.5th layer of output is as the characteristic value extracted；

CNN characteristic value obtains in the following way: depth convolutional network is operated using convolution, the pondization of multilayer to obtain The network structure model of data further feature may be implemented the translation to data, scaling, distort invariance.The CNN applied herein Network input layer is 50 × 72, and two convolutional layers are respectively 63 × 3 convolution kernels and 36 3 × 3 convolution kernels, corresponding pond Changing layer is 1 × 2,1 × 11.Two full articulamentums are 500,10, output layer 4.The output of 2nd full articulamentum is as extraction Characteristic value；

Principal component analysis (principal component analysis, abbreviation PCA)+linear discriminant analysis (linear Discriminant analysis, abbreviation LDA) characteristic value obtains in the following way: sample dropped first with PCA Dimension, eliminates the redundancy of sample, to be ensured of the nonsingularity of scatter matrix.LDA is recycled to solve optimal transformation feature Information.

It should be noted that the number in table 2, which refers to, carries out characteristics extraction for two Case Nos in table 1, and tie The matching degree that cosine-algorithm calculates the two is closed, furthermore case 1-2, which refers to, utilizes a variety of spies for extracting case 1 and case 2 Value indicative, in conjunction with the matching degree that cosine-algorithm both calculates, similarly, case 2-3 refers to using extracting the more of case 2 and case 3 Kind characteristic value, the matching degree of the two is calculated in conjunction with cosine-algorithm.

From Table 2, it can be seen that case 1-2 has compared with other cases in the obtained matching result of AEVB self-encoding encoder Higher matching degree is combined, compared with case 1-3, matching degree is high by 23.09%.Compared with case 1-4, matching degree is high by 89.94%. The counted matching degree of statistical characteristics meter as a comparison then relatively seldom goes out evident regularity, between the data of identical electric discharge type It is closer to degree, the matching degree between the data of different electric discharge types is slightly lower, but the result effect obtained compared with AEVB self-encoding encoder Fruit is poor.DBN model, CNN model and the resulting result of PCA+LDA model can be seen that between case 1-4, case 2-4 It is lower with spending, therefore it can obtain preferable recognition effect to the data of different electric discharge types, but for case 1-2, case 1-3, case 2-3, obtained matching degree are not much different, and cannot distinguish between similar cases, therefore poor as matching application effect.

In order to verify this case AEVB self-encoding encoder recognition effect, based on AEVB self-encoding encoder carry out characteristics extraction, It is then utilized respectively cosine-algorithm, Euclidean distance and best entropy calculate the matching degree between case data, and the results are shown in Table 3.

Table 3.

Case No	Cosine-algorithm	Euclidean distance	Best entropy
				Case 1-2	96.87%	93.03%	98.62%
Case 1-3	73.78%	75.23%	74.44%
				Case 1-4	6.93%	8.31%	5.00%
Case 2-3	61.75%	60.84%	77.43%
				Case 2-4	1.92%	6.80%	3.34%
Case 3-4	9.51%	7.18%	7.15%

Note: cosine-algorithm can be obtained by mode as described below in table 3: the cosine-algorithm of following formula calculates shelf depreciation number The distance between according to, it can be obtained the matching degree MR of Partial Discharge Data.

V in formula_a、V_bRespectively two extracted feature vectors of Partial Discharge Data, | | | | indicate vector field homoemorphism；

The matching degree of Euclidean distance obtains as described below: the Euclidean distance based on two groups of vectors obtains its matching degree.It is based on The matching degree of Euclidean distance is to be difficult to determine suitable measurement standard there are problem, and therefore, it is difficult to normalize.It chooses herein all Maximum distance in sample data calculates its matching degree according to following formula as standard:

Matching degree based on best entropy obtains as described below: the main entropy for calculating signal proposes that calculating parameter is few, can be Error caused by reducing to a certain extent because of time irreversibility.

As can be seen from Table 3, Euclidean distance is utilized respectively for above-mentioned case and best entropy calculates matching degree and cosine is calculated The difference of method is little.

Matching degree calculating further is carried out to the data in 200,000,000 cases, is extracted using AEVB self-encoding encoder special Value indicative is utilized respectively cosine-algorithm, the matching degree of Euclidean distance and best entropy calculating between any two.With similar type case lower It is higher than 80% with degree, matching degree is correct as matching lower than 20% under dissimilar case, calculates matching accuracy and obtains such as table 4 Statistical information.

Table 4.

Electric discharge type	Suspended discharge	Discharge in insulation
			Cosine-algorithm	82.6%	85.7%
Euclidean distance	63.6%	75.3%
			Best entropy	47.5%	72.4%

As can be seen from Table 4, for a large amount of case, the matching degree accuracy based on Euclidean distance and best entropy is lower than Cosine-algorithm, since in the matching degree calculating of Euclidean distance, all sample standard deviations and fixed maximum distance are compared, therefore are easy out Existing singular value, causes overall effect poor, and the matching degree based on best entropy, which calculates, has faster calculating speed, but due to best Entropy is related to the degree of scatter of sample, for the Partial Discharge Data similitude less effective under measurement complex data source.

In summary as can be seen that Electric Power Equipment Insulation defect identification method of the present invention can effectively improve partial discharge The recognition accuracy of map, to preferably assess status of electric power, to be conducive to grasp the insulation of power equipment State.

In addition, AEVB self-encoding encoder of the present invention similarly has above advantages and beneficial effect.

It should be noted that prior art part is not limited to given by present specification in protection scope of the present invention Embodiment, all prior arts not contradicted with the solution of the present invention, including but not limited to first patent document, formerly Public publication, formerly openly use etc., it can all be included in protection scope of the present invention.

In addition, it should also be noted that, institute in the combination of each technical characteristic and unlimited this case claim in this case Combination documented by the combination or specific embodiment of record, all technical characteristics documented by this case can be to appoint Where formula is freely combined or is combined, unless generating contradiction between each other.

The above embodiment is a preferred embodiment of the present invention, but embodiments of the present invention are not by above-described embodiment Limitation, other any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present invention, It should be equivalent substitute mode, be included within the scope of the present invention.

Claims (10)

1. a kind of Electric Power Equipment Insulation defect identification method, which is characterized in that comprising steps of

(1) AEVB self-encoding encoder is constructed, the AEVB self-encoding encoder includes probability encoding device and probability decoder；

(2) AEVB self-encoding encoder is trained using the history Partial Discharge Data of power equipment, so that: the history office of input Characteristic value of portion's discharge data in the output end output Partial Discharge Data of probability encoding device, the characteristic value of the Partial Discharge Data Input probability decoder, to export corresponding insulation defect type in the output end of probability decoder；

(3) by the local discharge signal input AEVB self-encoding encoder of power equipment to be identified, the AEVB self-encoding encoder is defeated Insulation defect type out.

2. Electric Power Equipment Insulation defect identification method as described in claim 1, which is characterized in that the AEVB self-encoding encoder tool There are an input layer, an output layer, a hidden variable layer and four middle layers；One of input layer, two middle layers and One hidden variable layer constitutes the probability encoding device, the characteristic value of the hidden variable layer output Partial Discharge Data；In two Interbed and an output layer constitute the probability decoder, and the output layer exports insulation defect type.

3. Electric Power Equipment Insulation defect identification method as described in claim 1, which is characterized in that use stochastic gradient descent method Optimize the parameter of probability encoding device and probability decoder.

4. Electric Power Equipment Insulation defect identification method as described in claim 1, which is characterized in that the probability encoding device uses Following formula characterizations:

Z indicates the output of hidden variable layer in formula, is the characteristic value of Partial Discharge Data；σ_encIndicate scale parameter；ε indicates full The random parameter of sufficient N (0,1) distribution；N (0, I) is standardized normal distribution；F indicates activation primitive；Indicate the ratio to be solved Example parameter；Indicate the translational movement to be solved；Indicate the scale parameter to be solved；Indicate the translational movement to be solved；Indicate the scale parameter to be solved；Indicate the translational movement to be solved；X is the Partial Discharge Data of input.

5. Electric Power Equipment Insulation defect identification method as claimed in claim 4, which is characterized in that the probability decoder uses Following formula characterizations:

In formula, x indicates the Partial Discharge Data of input；The characteristic value of z expression Partial Discharge Data；P (x | z) indicate conditional probability Distribution function；N indicates normal distyribution function；μ_decIndicate probability likelihood parameter, σ_decIndicate scale parameter；I indicates offset；Indicate the scale parameter to be solved；Indicate the translational movement to be solved；Indicate the scale parameter to be solved；Table Show the translational movement to be solved；Indicate the h to be solved₂Scale parameter；Indicate the h to be solved₂Translational movement.

6. a kind of AEVB self-encoding encoder for Electric Power Equipment Insulation defect recognition, which is characterized in that the AEVB self-encoding encoder Including probability encoding device and probability decoder；Wherein probability encoding device is configured to: to the shelf depreciation of its input electric power equipment Signal, then its export Partial Discharge Data characteristic value；The probability decoder is configured to: its input terminal input probability coding The output of device exports insulation defect type.

7. AEVB self-encoding encoder as claimed in claim 6, which is characterized in that the AEVB self-encoding encoder has an input Layer, an output layer, a hidden variable layer and four middle layers；One of input layer, two middle layers and a hidden variable Layer constitutes the probability encoding device, the characteristic value of the hidden variable layer output Partial Discharge Data；Two middle layers and one Output layer constitutes the probability decoder, and the output layer exports insulation defect type.

8. AEVB self-encoding encoder as claimed in claim 6, which is characterized in that the ginseng of the probability encoding device and probability decoder Number is optimized using stochastic gradient descent method.

9. AEVB self-encoding encoder as claimed in claim 6, which is characterized in that the probability encoding device uses following formula tables Sign:

Z indicates the output of hidden variable layer in formula, is the characteristic value of Partial Discharge Data；σ_encIndicate scale parameter；ε indicates full The random parameter of sufficient N (0,1) distribution；N (0, I) is standardized normal distribution；F indicates activation primitive；Indicate the ratio to be solved Example parameter；Indicate the translational movement to be solved；Indicate the scale parameter to be solved；Indicate the translational movement to be solved；Indicate the scale parameter to be solved；Indicate the translational movement to be solved；X is the Partial Discharge Data of input.

10. AEVB self-encoding encoder as claimed in claim 9, which is characterized in that the probability decoder uses following formula tables Sign:

In formula, x indicates the Partial Discharge Data of input；The characteristic value of z expression Partial Discharge Data；P (x | z) indicate conditional probability Distribution function；N indicates normal distyribution function；μ_decIndicate probability likelihood parameter, σ_decIndicate scale parameter；I indicates offset；Indicate the scale parameter to be solved；Indicate the translational movement to be solved；Indicate the scale parameter to be solved；Table Show the translational movement to be solved；Indicate the h to be solved₂Scale parameter；Indicate the h to be solved₂Translational movement.