CN111638428B - GIS-based ultrahigh frequency partial discharge data processing method and system - Google Patents

Abstract

The invention provides a method and system for processing UHF partial discharge data based on GIS. The processing method includes data enhancement and cleaning, and adopts fault feature extraction, fault feature data correlation analysis, data noise addition and data weighting to achieve data enhancement; And according to different labels to obtain the corresponding cleaning factors to achieve data cleaning, thereby ensuring the effectiveness of enhanced data, to achieve GIS UHF partial discharge type diagnosis. The invention realizes data processing through the method of correlation coefficient analysis, enhances the data by adding noise and weighting processing, and uses the cleaning characteristic factor to clean the enhanced data, which can improve the accuracy and generalization when the GIS partial discharge data samples are insufficient. transformation ability.

Description

A method and system for processing UHF partial discharge data based on GIS

technical field

The invention relates to the technical field of partial discharge detection, in particular to a method and system for processing UHF partial discharge data based on GIS.

Background technique

With the introduction of "three types and two networks", the smart grid has developed rapidly. As the basis for the realization of intelligent operation and maintenance, the intelligent detection terminal is also the guarantee for the reliable and safe operation of power equipment. GIS is composed of circuit breakers, isolating switches, grounding switches, transformers, arresters, busbars, connectors and outgoing terminals, etc. These devices or components are all enclosed in a metal grounded shell, which is filled with a certain pressure of SF6 insulation. gas. The advantages of GIS lie in its compact structure, small footprint, high reliability, flexible configuration, convenient installation, strong safety, and strong environmental adaptability.

The fully-sealed structure of GIS also makes the localization and maintenance of partial discharge in GIS difficult, and the maintenance work is complicated. Therefore, power companies often carry out GIS partial discharge detection work. In order to realize GIS partial discharge detection can accurately identify fault types, it is necessary to study GIS UHF partial discharge data. However, at present, there are very few data samples for GIS UHF partial discharge detection, and there is a lack of sufficient data sets to study the characteristics of GIS, resulting in the inability to accurately identify GIS fault types and arrange maintenance work.

At present, the existing data enhancement methods for GIS UHF data are relatively rare, such as transformation and clipping in different ways, which often cause damage to the original feature data, making the identification accuracy of fault types relatively low. This leads to miscalculation of equipment operation and maintenance strategies.

SUMMARY OF THE INVENTION

The invention provides a processing method and system based on GIS ultra-high frequency partial discharge data, which is used to solve the problems that the existing GIS ultra-high frequency partial discharge detection has few data samples and the data enhancement method is unreasonable.

To achieve the above object, the present invention adopts the following technical solutions:

A first aspect of the present invention provides a method for processing UHF partial discharge data based on GIS, the method comprising the following steps:

Obtain GIS UHF partial discharge data under different labels;

Extracting feature values from the partial discharge data to form two-dimensional time series data A;

Correlation analysis is performed on the two-dimensional time series data A and the corresponding label to obtain a first correlation coefficient;

According to the correlation threshold, the two-dimensional time series data A is divided into A1 and A2, wherein A2 is the two-dimensional time series data whose first correlation coefficient is less than the correlation threshold;

Add noise processing to A2 to generate two-dimensional time series data B;

Carrying out a correlation analysis between the two-dimensional time series data B and the corresponding label to obtain a second correlation coefficient;

According to the correlation threshold, the two-dimensional time series data B is divided into B1 and B2, wherein B2 is the two-dimensional time series data whose second correlation coefficient is less than the correlation threshold;

B2 is weighted and fused, combined with B1, and then combined with A1 to obtain two-dimensional time series data C;

Correlation analysis is performed between the two-dimensional time series data C and the corresponding label, a third correlation coefficient is obtained, the cleaning characteristic factor of the corresponding label is obtained, C is cleaned based on the cleaning characteristic factor, and a data sample D is obtained.

Further, the label includes air gap, suspension, corona, creepage and particle.

Further, the eigenvalues are obtained by the regional mean decomposition method.

Further, the specific process of adding noise to A2 is as follows:

A Gaussian random variable is added to the two-dimensional time series data A2 and data correction is performed.

Further, the calculation method of the Gaussian random variable rv is as follows:

rv=sqrt(-2.0*log(U ₁ ))*cos(2*π*U ₂ )

where the random variables U1 and U2 are expressed as:

In the formula, the random variables U ₁ and U ₂ are independent of each other, and both obey the uniform distribution between (0, 1); the random variables Z ₀ , Z ₁ obey the standard Gaussian distribution and satisfy the normal distribution, the mean is 0, and the variance is 1.

Further, the specific process of performing data correction is:

Define the sample data A2 to be enhanced as dstImage[x][y], define the enhanced sample data B as EnhDstImage[x][y], define the value of val after adding the Gaussian random variable rv, and calculate val by the following formula:

val=dstImage[x][y]+rv

Correct the range of val:

If val<0, then val=0;

If val>255, then val=255;

The enhanced sample data B is redefined:

EnhDstImage[x][y]=val

After all the data in the dstImage array is processed above, the enhanced sample data EnhDstImage is formed.

Further, the weighted fusion of B2 is specifically:

B2[x][y]=a1*W1+a2*W2+…+an*Wn

Among them, W1+W2+...+Wn=1; a1, a2,...an are the feature data of the xth row.

Further, the specific process of obtaining the cleaning characteristic factor of the corresponding label is:

Sort the obtained third correlation coefficient according to the numerical value;

According to the sorting result, the third correlation coefficient is segmented, and the average value of each segment is calculated respectively;

The average value of each segment is then averaged to obtain the cleaning characteristic factor.

Further, the specific process of cleaning C based on the cleaning characteristic factor is:

comparing the cleaning characteristic factor with the third correlation coefficient matrix;

If the cleaning characteristic factor is greater than the corresponding third correlation coefficient, the data corresponding to the current third correlation coefficient is discarded, and the data values before and after the modified data are averaged to replace the currently discarded data;

Otherwise, the data corresponding to the current third correlation coefficient is retained.

A second aspect of the present invention provides a GIS-based UHF partial discharge data processing system, the system comprising:

The data acquisition unit is used to obtain GIS UHF partial discharge data under different labels;

a first data processing unit, configured to perform feature value extraction on the partial discharge data to form two-dimensional time series data A;

The second data processing unit performs a correlation analysis on the two-dimensional time series data A and the corresponding label to obtain a first correlation coefficient;

The first comparison unit, according to the correlation threshold, divides the two-dimensional time series data A into A1 and A2, wherein A2 is the two-dimensional time series data whose first correlation coefficient is less than the correlation threshold;

The third data processing unit adds noise processing to A2 to generate two-dimensional time series data B;

the fourth data processing unit, performing correlation analysis on the two-dimensional time series data B and the corresponding label to obtain a second correlation coefficient;

The second comparison unit divides the two-dimensional time series data B into B1 and B2 according to the correlation threshold, wherein B2 is the two-dimensional time series data whose second correlation coefficient is less than the correlation threshold;

The fifth data processing unit performs weighted fusion of B2, splices with B1, and then synthesizes with A1 to obtain two-dimensional time series data C;

The sixth data processing unit performs correlation analysis on the two-dimensional time series data C and the corresponding label, obtains a third correlation coefficient, obtains the cleaning characteristic factor of the corresponding label, cleans C based on the cleaning characteristic factor, and obtains data sample D.

The processing system of the second aspect of the present invention can implement the methods in the first aspect and each implementation manner of the first aspect, and achieve the same effect.

The effects provided in the summary of the invention are only the effects of the embodiments, rather than all the effects of the invention. One of the above technical solutions has the following advantages or beneficial effects:

The invention realizes data processing through the method of correlation coefficient analysis, enhances the data by adding noise and weighting processing, and uses the cleaning characteristic factor to clean the enhanced data, which can improve the accuracy and generalization when the GIS partial discharge data samples are insufficient. transformation ability.

Description of drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments or the prior art. In other words, other drawings can also be obtained based on these drawings without creative labor.

Fig. 1 is the schematic flow chart of the method of the present invention;

FIG. 2 is a schematic structural diagram of the system according to the present invention.

Detailed ways

In order to clearly illustrate the technical features of the solution, the present invention will be described in detail below through specific embodiments and in conjunction with the accompanying drawings. The following disclosure provides many different embodiments or examples for implementing different structures of the invention. In order to simplify the disclosure of the present invention, the components and arrangements of specific examples are described below. Furthermore, the present invention may repeat reference numerals and/or letters in different instances. This repetition is for the purpose of simplicity and clarity and does not in itself indicate a relationship between the various embodiments and/or arrangements discussed. It should be noted that the components illustrated in the figures are not necessarily drawn to scale. Descriptions of well-known components and processing techniques and processes are omitted from the present invention to avoid unnecessarily limiting the present invention.

A method for processing GIS UHF partial discharge data proposed by the present invention includes data enhancement and cleaning, and adopts fault feature extraction, fault feature data correlation analysis, data noise addition and data weighting to achieve data enhancement; and obtains corresponding data according to different labels. The cleaning factor is used to clean the data, thereby ensuring the validity of the enhanced data, and realizing the type diagnosis of GIS UHF partial discharge.

As shown in Figure 1, a method for processing UHF partial discharge data based on GIS of the present invention specifically includes the following steps:

S1, obtain GIS UHF partial discharge data under different labels;

S2, carrying out feature value extraction on the partial discharge data to form two-dimensional time series data A;

S3, the two-dimensional time series data A and the corresponding label are subjected to correlation analysis to obtain the first correlation coefficient;

S4, according to the correlation threshold, the two-dimensional time series data A is divided into A1 and A2, wherein A2 is the two-dimensional time series data whose first correlation coefficient is less than the correlation threshold;

S5, adding noise processing to A2 to generate two-dimensional time series data B;

S6, the two-dimensional time series data B and the corresponding label are subjected to correlation analysis to obtain the second correlation coefficient;

S7, according to the correlation threshold, the two-dimensional time series data B is divided into B1 and B2, wherein B2 is the two-dimensional time series data whose second correlation coefficient is less than the correlation threshold;

S8, weighted fusion of B2 is performed, and combined with B1, and then combined with A1 to obtain two-dimensional time series data C;

S9, perform correlation analysis on the two-dimensional time series data C and the corresponding label, obtain a third correlation coefficient, obtain the cleaning characteristic factor of the corresponding label, and clean C based on the cleaning characteristic factor to obtain a data sample D.

The labels in step S1 include air gap, suspension, corona, creepage and particle. In the following embodiments, an air gap label is used as an example for description.

The formation of the two-dimensional time series array A in step S2 is specifically:

S21, UHF detection collects 1 second of data at a time, each cycle is 20ms, divides 20ms into 60 time slices, and obtains 50×60 two-dimensional time series data;

S22, using the regional mean decomposition method to obtain the eigenvalues of the two-dimensional time series data, and form new two-dimensional time series data A;

The regional mean value decomposition method in step S22 obtains the instantaneous value of the envelope function in the envelope map, and the envelope function is as follows:

In the formula, n _i and n _i+1 are the adjacent extreme points respectively, a _i is the average value of the adjacent extreme points, that is, the eigenvalue, and a(t) is the instantaneous value of the envelope function. Two-dimensional time series data A is constructed according to the eigenvalues obtained from the set of data.

Set the correlation coefficient threshold in step S4, where the threshold is set to 0.85, according to the set correlation coefficient threshold, separate the two-dimensional time series data A, and obtain A1 and A2 respectively, specifically:

A1 is two-dimensional time series data whose first correlation coefficient is greater than or equal to 0.85;

A2 is two-dimensional time series data whose first correlation coefficient is less than 0.85.

The specific process of adding noise to A2 in step S5 is as follows:

A Gaussian random variable is added to the two-dimensional time series data A2 and data correction is performed.

The Gaussian random variable rv is calculated as follows:

rv=sqrt(-2.0*log(U ₁ ))*cos(2*π*U ₂ )

where the random variables U1 and U2 are expressed as:

In the formula, the random variables U ₁ and U ₂ are independent of each other, and both obey the uniform distribution between (0, 1); the random variables Z ₀ , Z ₁ obey the standard Gaussian distribution and satisfy the normal distribution, the mean is 0, and the variance is 1.

The specific process of data correction is as follows:

Define the sample data A2 to be enhanced as dstImage[x][y], define the enhanced sample data B as EnhDstImage[x][y], define the value of val after adding the Gaussian random variable rv, and calculate val by the following formula:

val=dstImage[x][y]+rv

Correct the range of val:

If val<0, then val=0;

If val>255, then val=255;

The enhanced sample data B is redefined:

EnhDstImage[x][y]=val

After all the data in the dstImage array is processed above, the enhanced sample data EnhDstImage is formed.

In step S7, the threshold value of the correlation coefficient is set to 0.8 according to the air gap label, and compared with the second correlation coefficient, the characteristic data of the correlation coefficient greater than or equal to 0.8 is denoted as B1; the characteristic data of the correlation coefficient less than 0.8 is denoted as B1. for B2.

In step S8, the weighted fusion of B2 is specifically:

B2[x][y]=a1*W1+a2*W2+…+an*Wn

Among them, W1+W2+...+Wn=1; a1, a2,...an are the feature data of the xth row.

In step S9, the specific process of obtaining the cleaning characteristic factor of the corresponding label is as follows:

Sort the obtained third correlation coefficient according to the numerical value; divide the third correlation coefficient into segments according to the sorting result, and calculate the average value of each subsection respectively; and then remove the mean value from the average value of each subsection to obtain Clean the characteristic factor. In this embodiment, 20% is used as a subsection, that is, the average value of the 20% correlation coefficient is taken, the average value of the 20% correlation coefficient is taken in turn, and the average value of the last 20% correlation coefficient is taken. Perform de-averaging to obtain the cleaning characteristic factor a.

Compare the cleaning characteristic factor a with the correlation coefficient matrix, if the cleaning characteristic factor is less than or equal to the corresponding third correlation coefficient, the data corresponding to the coefficient is retained; if the cleaning characteristic factor is greater than the corresponding third correlation coefficient, then The data is discarded, and the values before and after the data are averaged to replace the discarded values.

For example: Suppose the data enhancement sample is Enhance[x,y], the correlation coefficient matrix is Correlation[m,n], and the cleaning feature factor is a.

If a≤Correlation[3,2], Enhance[3,2] is reserved;

If a>Correlation[5,6], then Enhance[5,6] is discarded, and (Enhance[5,5]+Enhance[5,7])/2 is filled in this position;

If several consecutive values are less than a, take the mean of the row to fill. After cleaning the enhanced data, the final GIS air-gap partial discharge data enhanced sample D is obtained, which is used for training the deep learning network model.

The air-gap tag is updated to dynamically adjust the data cleaning factor, thereby improving the effect and quality of data cleaning and ensuring the accuracy and integrity of data cleaning.

Using the above method, the correlation analysis between different partial discharge types and corresponding tags is carried out. Among them, the processed samples are the data samples processed by the above-mentioned methods, and the obtained correlation coefficient mean values are compared as shown in the following table.

air gap Suspended corona along the surface particles original sample 0.88 0.89 0.82 0.8 0.81 Process samples 0.95 0.97 0.91 0.87 0.89

It can be seen that compared with directly using samples as training data, the present invention can generate samples with more characteristic parameters with less calculation amount, can solve the problem of insufficient GIS partial discharge data sets, increase the number of samples, and avoid overfitting, Improve the recognition accuracy of GIS UHF partial discharge fault types.

As shown in FIG. 2, the processing system based on GIS UHF partial discharge data of the present invention includes a data acquisition unit, a first data processing unit, a second data processing unit, a first comparison unit, a third data processing unit, A fourth data processing unit, a second comparison unit, a fifth data processing unit and a sixth data processing unit.

The data acquisition unit is used to obtain the GIS UHF partial discharge data under different labels; the first data processing unit is used to extract the characteristic value of the partial discharge data to form the two-dimensional time series data A; the second data processing unit converts the two-dimensional time series Perform correlation analysis on the data A and the corresponding label to obtain a first correlation coefficient; the first comparison unit divides the two-dimensional time series data A into A1 and A2 according to the correlation threshold, where A2 is that the first correlation coefficient is less than the correlation Threshold two-dimensional time series data; the third data processing unit adds noise processing to A2 to generate two-dimensional time series data B; the fourth data processing unit performs correlation analysis on the two-dimensional time series data B and the corresponding label, and obtains a second correlation coefficient The second comparison unit divides the two-dimensional time series data B into B1 and B2 according to the correlation threshold value, wherein B2 is the two-dimensional time series data whose second correlation coefficient is less than the correlation threshold value; The fifth data processing unit carries out weighted fusion by B2 , and combined with B1, and then combined with A1 to obtain two-dimensional time series data C; the sixth data processing unit performs correlation analysis on the two-dimensional time series data C and the corresponding label, obtains the third correlation coefficient, and obtains the cleaning of the corresponding label. Characteristic factor. Based on the cleaning characteristic factor, C is cleaned to obtain a data sample D.

Although the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, they do not limit the scope of protection of the present invention. Those skilled in the art should understand that on the basis of the technical solutions of the present invention, those skilled in the art do not need to pay creative efforts. Various modifications or deformations that can be made are still within the protection scope of the present invention.

Claims (6)

1. A GIS-based ultrahigh frequency partial discharge data processing method is characterized by comprising the following steps:

acquiring GIS ultrahigh frequency partial discharge data under different labels, wherein the labels comprise air gaps, suspension, corona, surfaces and particles;

extracting characteristic values of the partial discharge data to form two-dimensional time sequence data A;

performing correlation analysis on the two-dimensional time sequence data A and the corresponding label to obtain a first correlation coefficient;

dividing the two-dimensional time series data A into A1 and A2 according to a correlation threshold, wherein A2 is the two-dimensional time series data with a first correlation coefficient smaller than the correlation threshold;

adding noise processing to A2 to generate two-dimensional time series data B;

performing correlation analysis on the two-dimensional time sequence data B and the corresponding label to obtain a second correlation coefficient;

dividing the two-dimensional time series data B into B1 and B2 according to a correlation threshold, wherein B2 is the two-dimensional time series data of which the second correlation number is less than the correlation threshold;

b2 is subjected to weighted fusion, is spliced with B1 and is synthesized with A1 to obtain two-dimensional time sequence data C;

performing correlation analysis on the two-dimensional time sequence data C and the corresponding label to obtain a third correlation coefficient, obtaining a cleaning characteristic factor of the corresponding label, and cleaning the C based on the cleaning characteristic factor to obtain a data sample D;

the specific process of extracting the characteristic value of the partial discharge data to form the two-dimensional time sequence data A is as follows:

s21, collecting data of 1 second at a time by ultrahigh frequency detection, dividing 20ms into 60 time slices in each period of 20ms, and obtaining two-dimensional time sequence data of 50 x 60;

s22, obtaining the characteristic value of the two-dimensional time sequence data by using a region mean decomposition method, and forming new two-dimensional time sequence data A;

in step S22, the region mean decomposition method obtains an instantaneous value of the envelope function in the envelope map, where the envelope function is as follows:

in the formula,

are respectively the adjacent extreme points, and are respectively the adjacent extreme points,

the average value of the adjacent extreme points, i.e. the characteristic value,

i.e. the instantaneous value of the envelope function; constructing two-dimensional time sequence data A according to the characteristic values of the group of data;

adding noise to the A2, specifically, adding a Gaussian random variable in the two-dimensional time sequence data A2 and performing data correction;

the specific process of acquiring the cleaning characteristic factor of the corresponding label comprises the following steps:

sequencing the obtained third phase relation number according to the numerical value; segmenting the third phase relation number according to the sequencing result, and respectively calculating the average value of each segment; and averaging the average values of the segments to obtain the cleaning characteristic factors.

2. The method for processing GIS-based ultrahigh frequency partial discharge data according to claim 1, wherein the Gaussian random variable rv is calculated as follows:

wherein the random variables U1 and U2 are represented as:

in the formula, a random variable U ₁ 、U ₂ Independent of each other and all obey uniform distribution among (0, 1); random variable Z ₀ ，Z ₁ Obey a standard gaussian distribution and satisfy a normal distribution with a mean of 0 and a variance of 1.

3. The method for processing the GIS-based ultrahigh frequency partial discharge data according to claim 2, wherein the specific process of data correction is as follows:

defining sample data A2 to be enhanced as dstImage [ x ] [ y ], defining sample data B after enhancement as EnhDstImage [ x ] [ y ], defining the value after adding a Gaussian random variable rv as val, and calculating val according to the following formula:

correcting the range of val:

redefining the enhanced sample data B:

and after all the data in the dstImage array are processed, the enhanced sample data EnhDstimage is formed.

4. The method for processing the GIS ultrahigh frequency partial discharge data according to claim 1, wherein the weighted fusion of B2 is specifically as follows:

wherein,

is the x-th row characteristic data.

5. The GIS-based ultrahigh frequency partial discharge data processing method according to claim 4, wherein the specific process of cleaning C based on the cleaning characteristic factors is as follows:

comparing the cleaning characteristic factor with a third phase relation number matrix;

if the cleaning characteristic factor is larger than the corresponding third correlation coefficient, discarding data corresponding to the current third phase relation number, averaging previous and next data values corresponding to the discarded data, and replacing the currently discarded data;

otherwise, retaining the data corresponding to the current third phase relation number.

6. A GIS ultrahigh frequency partial discharge data-based processing system is characterized by comprising:

the system comprises a data acquisition unit, a data acquisition unit and a data processing unit, wherein the data acquisition unit is used for acquiring GIS ultrahigh frequency partial discharge data under different labels, and the labels comprise air gaps, suspension, corona, surfaces and particles;

the first data processing unit is used for extracting a characteristic value of the partial discharge data to form two-dimensional time sequence data A;

the second data processing unit is used for carrying out correlation analysis on the two-dimensional time sequence data A and the corresponding label to obtain a first correlation coefficient;

a first comparing unit that divides the two-dimensional time series data a into a1 and a2 according to a correlation threshold, wherein a2 is the two-dimensional time series data whose first correlation coefficient is smaller than the correlation threshold;

a third data processing unit which adds noise processing to A2 and generates two-dimensional time series data B;

the fourth data processing unit is used for carrying out correlation analysis on the two-dimensional time sequence data B and the corresponding label to obtain a second correlation coefficient;

a second comparing unit which divides the two-dimensional time series data B into B1 and B2 according to a correlation threshold, wherein B2 is the two-dimensional time series data of which the second correlation number is less than the correlation threshold;

the fifth data processing unit is used for performing weighted fusion on the B2, splicing the B2 with the B1, and then synthesizing the B1 with the A1 to obtain two-dimensional time sequence data C;

the sixth data processing unit is used for carrying out correlation analysis on the two-dimensional time sequence data C and the corresponding label to obtain a third correlation coefficient, obtaining a cleaning characteristic factor of the corresponding label, and cleaning the C based on the cleaning characteristic factor to obtain a data sample D;

the specific process of extracting the characteristic value of the partial discharge data to form the two-dimensional time sequence data A is as follows:

s21, the ultrahigh frequency detection collects data for 1 second at a time, each period is 20ms, 20ms is divided into 60 time slices, and the data are obtained

Two-dimensional time series data of (a);

s22, obtaining the characteristic value of the two-dimensional time sequence data by using a region mean decomposition method, and forming new two-dimensional time sequence data A;

in step S22, the region mean decomposition method obtains an instantaneous value of the envelope function in the envelope map, where the envelope function is as follows:

in the formula,

are respectively the adjacent extreme points, and are respectively the adjacent extreme points,

the average value of the adjacent extreme points, i.e. the feature value,

i.e. the instantaneous value of the envelope function; constructing two-dimensional time sequence data A according to the characteristic values of the group of data;

adding noise to the A2, specifically, adding a Gaussian random variable in the two-dimensional time sequence data A2 and performing data correction;

the specific process of acquiring the cleaning characteristic factor of the corresponding label comprises the following steps:

sequencing the obtained third phase relation number according to the numerical value; segmenting the third phase relation number according to the sequencing result, and respectively calculating the average value of each segment; and averaging the average values of the segments to obtain the cleaning characteristic factors.

Images