CN103198175A - Transformer fault diagnosis method based on fuzzy cluster - Google Patents

Abstract

The invention discloses a transformer fault diagnosis method based on a fuzzy cluster. The method is implemented by the steps as follows: 1), an initial data matrix is built according to on-line monitoring data; 2), data in the initial data matrix are standardized to acquire a fuzzy matrix; 3), a fuzzy similar matrix is built; 4), a transitive closure matrix R' of the fuzzy similar matrix R is calculated with a square method, that is, a fuzzy equivalent matrix t(R)=R'; 5), data in the t(R) are arranged from high to low according to a lower triangle of the t(R); 6), classified threshold values lambda are extracted from high to low; when data in the fuzzy equivalent matrix t(R) are higher than the extracted lambda, corresponding data in the fuzzy equivalent matrix t(R) are replaced with 1, otherwise, with 0, and rows of similar data elements of the fuzzy equivalent matrix t(R) are classified into one classification; and 7), a dynamic cluster graph is acquired according to the lambda, and a DGA (dissolved gas analysis) technology is combined, so that the fault type of a transformer is determined. According to the transformer fault test method, a transformer fault diagnosis model based on the fuzzy cluster is built, so that the fault diagnosis accuracy of a power transformer can be improved.

Description

transformer fault diagnosis method based on fuzzy clustering

Technical Field

The invention belongs to the technical field of transformer fault online monitoring, and particularly relates to a transformer fault diagnosis method based on fuzzy clustering.

Background

At the present stage of China, along with the rapid development of national economy, the demand for electric power also shows a rapid growth trend. Due to the continuous expansion of installed capacity and power grid scale of electric power, the electric power industry in China has entered the development period of large power grid, large unit, high voltage and high automation. Six links of constructing a smart grid taking extra-high voltage as a backbone grid frame, coordinately developing all levels of grids and a smart grid development strategy frame are proposed by the national grid company in 2009, so that the importance of power transmission and transformation is further highlighted. The method has the advantages that the safe and reliable operation of the intelligent transformer substation is one of the main conditions for realizing the stable operation of the whole intelligent power grid, and the intelligent power transformer is an important component of the intelligent transformer substation, so that the potential faults of the intelligent power transformer can be diagnosed timely and reliably, and the method has profound significance for guaranteeing the operation of the intelligent power grid.

The dissolved gas analysis technology is a main means for diagnosing the internal fault of the transformer and provides a basis for indirectly knowing general hidden dangers in the transformer; however, the dissolved gas in oil is a result of the combined action of many factors, so that the relationship between the type of insulation fault of the power transformer and the content of the dissolved gas component in oil has certain ambiguity and uncertainty characteristics. The introduction of the fuzzy clustering analysis method into the fault diagnosis of the transformer is a relatively new research direction at the present stage.

Disclosure of Invention

The invention aims to provide a transformer fault diagnosis method based on fuzzy clustering, which establishes a transformer fault diagnosis model based on fuzzy clustering and can further improve the accuracy of power transformer fault diagnosis.

The technical scheme adopted by the invention is that the transformer fault diagnosis method based on fuzzy clustering is implemented according to the following steps:

step 1, inputting the number n and the number m of characteristic variables of objects to be classified according to online monitoring data, and establishing an original data matrix Y (Y ═ Y)_ij)_n×m：

Discourse domain U is { y ═ y according to online monitoring data₁,y₂,y₃…,y_nAs objects to be classified, where each object has its performance represented by m indices, i.e. Y_i＝{y_i1,y_i2…,y_imAnd (i ═ 1,2, … n), establishing an original data matrix, wherein the original data matrix is Y ═ Y_ij)_n×m；

Step 2, standardizing the data in the original data matrix to be between [0, 1] to obtain a fuzzy matrix:

for the original data matrix Y established in step 1 ═ Y_ij)_n×mIs normalized by normalizing y_ijConversion to y 'by algorithm'_ijSpecifically, the conversion is performed according to the following algorithm:

$y_{\mathrm{ij}}^{\′}=\frac{y_{\mathrm{ij}}-\stackrel{\‾}{y_i}}{T_i},(1\≤i\≤n,1\≤j\≤m);$

wherein, $\stackrel{\‾}{y_i}=\frac{1}{m}{y}_{\mathrm{ij}},$ $T_i=\sqrt{\frac{\underset{j=1}{\overset{m}{\mathrm{\Σ}}}{(y_{\mathrm{ij}}-\stackrel{\‾}{y_i})}^2}{m-1}};$

the original data matrix Y is equal to (Y)_ij)_n×mAfter standard deviation normalization transformation of the data in (1), if any, the data in (1)Then the data is subjected to a range normalization process, which is implemented according to the following algorithm:

$y_{\mathrm{ij}}^{\′\′}=\frac{y_{\mathrm{ij}}-\min \left\{y_{\mathrm{ij}}^{\′}\right\}}{\max \left\{y_{\mathrm{ij}}^{\′}\right\}-\min \left\{y_{\mathrm{ij}}^{\′}\right\}};$

all y' obtained after range normalization_ij∈[0,1]And the fuzzy matrix Y ═ (Y ″) can be obtained without the influence of the dimensional matrix_ij)_n×m；

Step 3, determining a similarity coefficient by adopting one of a similarity coefficient method, a distance method or a proximity method, and establishing a fuzzy similarity matrix;

step 4, calculating a transfer closure matrix R 'of the fuzzy similar matrix R obtained in the step 3 by using a flat method, namely the fuzzy equivalent matrix t (R) is R';

step 5, observing the lower triangle of the fuzzy equivalent matrix t (R) obtained in the step 4, and then arranging the data in the fuzzy equivalent matrix t (R) from large to small, wherein the data from large to small are the value-taking points of the classification threshold lambda;

step 6, taking out the classification threshold value lambda in the step 5 from large to small one by one, and when the data in the fuzzy equivalent matrix t (R) is larger than the taken classification threshold value lambda, replacing the data in the corresponding fuzzy equivalent matrix t (R) with 1, otherwise, replacing the data with 0, and finally, only containing two elements of 0 and 1 in the fuzzy equivalent matrix t (R); observing each column of the fuzzy equivalent matrix t (R), classifying the columns with the same data elements into one class, and further classifying the columns;

and 7, according to the different classification threshold values lambda in the step 5, programming and simulating by using MATLAB software to finally obtain a dynamic cluster map of the transformer fault diagnosis model based on fuzzy clustering, and determining the fault type of the transformer by combining a DGA technology.

The present invention is also characterized in that,

the method for normalizing the original data matrix in the step 2 can also adopt range normalization or maximum normalization.

The range standardization processing method in the step 2 is specifically implemented according to the following algorithm:

the method for performing range standardization processing on the original data matrix obtained in the step 1 is implemented according to the following algorithm:

$y_{\mathrm{ij}}^{\′}=\frac{y_{\mathrm{ij}}-\stackrel{\‾}{y_i}}{\max \left\{y_{\mathrm{ij}}\right\}-\min \left\{y_{\mathrm{ij}}\right\}}.$

the processing method for normalizing the maximum value in the step 2 is specifically implemented according to the following algorithm: carrying out maximum value standardization processing on the original data matrix obtained in the step 1, and specifically implementing the maximum value standardization processing according to the following algorithm:

$y_{\mathrm{ij}}^{\′}=\frac{y_{\mathrm{ij}}}{M_j},$

wherein M is_j＝max(y_1j,y_2j…y_nj)。

The similarity coefficient method in step 3 includes the following two algorithms:

the first method, the cosine method of the included angle, is implemented according to the following algorithm:

$r_{\mathrm{ij}}=\frac{|\underset{l=1}{\overset{m}{\mathrm{\Σ}}}{y}_{\mathrm{il}}*y_{\mathrm{jl}}|}{\sqrt{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}{y}_{\mathrm{il}}^2*\underset{l=1}{\overset{m}{\mathrm{\Σ}}}{y}_{\mathrm{jl}}^2}},(i,j=\mathrm{1,2}\·\·\·n);$

the second kind, the correlation coefficient method is implemented according to the following algorithm:

$r_{\mathrm{ij}}=\frac{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}|y_{\mathrm{il}}-\stackrel{\‾}{y_i}|\·|y_{\mathrm{jl}}-\stackrel{\‾}{y_j}|}{\sqrt{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}{(y_{\mathrm{il}}-\stackrel{\‾}{y_i})}^2}\·\sqrt{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}{(y_{\mathrm{jl}}-\stackrel{\‾}{y_j})}^2}},(i,j=\mathrm{1,2}\·\·\·,n).$

the distance method in step 3 comprises the following three algorithms:

first distance method, Hamming distance:

$d(y_i,y_j)=\underset{l=1}{\overset{m}{\mathrm{\Σ}}}|y_{\mathrm{il}}-y_{\mathrm{jl}}|;$

second distance method, euclidd distance:

$d(y_i,y_j)=\sqrt{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}{(y_{\mathrm{il}}-y_{\mathrm{jl}})}^2};$

third distance method, Chebyshev distance:

$d(y_i,y_j)=\underset{1\≤l\≤n}{\max }|y_{\mathrm{il}}-y_{\mathrm{jl}}|.$

the processing method for maximum value normalization is specifically implemented according to the following algorithm: carrying out maximum value standardization processing on the original data matrix obtained in the step 1, and specifically implementing the maximum value standardization processing according to the following algorithm:

$y_{\mathrm{ij}}^{\′}=\frac{y_{\mathrm{ij}}}{M_j},$

wherein M is_j＝max(y_1j,y_2j…y_nj)。

The closeness method in step 3 includes the following three algorithms:

the first closeness method, the maximum-minimum method:

second closeness method, geometric mean minimization method:

$r_{\mathrm{ij}}=\frac{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}(y_{\mathrm{il}}^y_{\mathrm{jl}})}{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}\sqrt{y_{\mathrm{il}}\·y_{\mathrm{jl}}}},(i,j=\mathrm{1,2}\·\·\·,n);$

the third closeness method, the number average minimum method:

$r_{\mathrm{ij}}=\frac{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}(y_{\mathrm{il}}^y_{\mathrm{jl}})}{\frac{1}{2}\underset{l=1}{\overset{m}{\mathrm{\Σ}}}(y_{\mathrm{il}}+y_{\mathrm{jl}})},(i,j=\mathrm{1,2}\·\·\·,n).$

the invention has the beneficial effects that:

(1) the transformer fault diagnosis method based on the fuzzy clustering is a novel transformer fault diagnosis analysis method provided on the basis of a DGA analysis method researched previously, can classify faults according to a large amount of transformer monitoring data, and is favorable for quickly diagnosing the fault type of the transformer.

(2) The data of the transformer fault diagnosis method based on fuzzy clustering is derived from field gas data monitored by a transformer online monitoring system in real time, so that the information of hydrogen, methane, ethane, ethylene, acetylene, carbon monoxide and carbon dioxide generated when the transformer fails can be monitored, and the fault of the transformer can be predicted according to the mutation of certain gases.

(3) The transformer fault diagnosis method based on the fuzzy clustering is characterized in that a fuzzy clustering analysis method based on a transmission closure method is used for transforming data, a transmission closure matrix of a fuzzy similar matrix, namely a fuzzy equivalent matrix, is solved by using a flat method, and a transformer fault diagnosis model based on the fuzzy clustering is established on the basis by combining a DGA technology and clustering analysis.

Drawings

FIG. 1 is a flow chart of the fuzzy clustering based transformer fault diagnosis method of the present invention;

FIG. 2 is a dynamic clustering chart drawn by the fuzzy clustering-based transformer fault diagnosis method of the present invention.

Detailed Description

The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

The invention discloses a transformer fault testing method based on fuzzy clustering, which has the technical process shown in figure 1, and the specific algorithm is implemented according to the following steps:

step 1, inputting the number n and the number m of characteristic variables of objects to be classified according to online monitoring data, and establishing an original data matrix Y (Y ═ Y)_ij)_n×m：

Discourse domain U is { y ═ y according to online monitoring data₁,y₂,y₃…,y_nAs objects to be classified, where each object has its performance represented by m indices, i.e. Y_i＝{y_i1,y_i2…,y_imAnd (i ═ 1,2, … n), establishing an original data matrix, wherein the original data matrix is Y ═ Y_ij)_n×m；

Step 2, standardizing the data in the original data matrix to be between [0, 1] to obtain a fuzzy matrix:

in practical problems, the dimensions of different data are generally non-uniform, and in order to compare the data of different dimensions, it is usually necessary to normalize the data, that is, to make the data in the original data matrix meet the requirement of fuzzy clustering through appropriate data transformation, and specifically, the original data matrix Y created in step 1 is (Y) according to the following algorithms_ij)_n×mThe data in (1) is normalized:

the first method of normalization is standard deviation normalization, i.e. the original data matrix Y created in step 1 is (Y)_ij)_n×mThe standard deviation normalization processing is carried out on the data to obtain a fuzzy matrix, and the specific method is as follows:

for the original data matrix Y established in step 1 ═ Y_ij)_n×mIs normalized by normalizing y_ijConversion to y 'by algorithm'_ijSpecifically, the conversion is performed according to the following algorithm:

$y_{\mathrm{ij}}^{\′}=\frac{y_{\mathrm{ij}}-\stackrel{\‾}{y_i}}{T_i},(1\≤i\≤n,1\≤j\≤m);$

wherein, $\stackrel{\‾}{y_i}=\frac{1}{m}{y}_{\mathrm{ij}},$ $T_i=\sqrt{\frac{\underset{j=1}{\overset{m}{\mathrm{\Σ}}}{(y_{\mathrm{ij}}-\stackrel{\‾}{y_i})}^2}{m-1}};$

the original data matrix Y is equal to (Y)_ij)_n×mAfter standard deviation normalization transformation of the data in (1), if any, the data in (1)

Then the data is subjected to a range normalization process, which is implemented according to the following algorithm:

$y_{\mathrm{ij}}^{\′\′}=\frac{y_{\mathrm{ij}}-\min \left\{y_{\mathrm{ij}}^{\′}\right\}}{\max \left\{y_{\mathrm{ij}}^{\′}\right\}-\min \left\{y_{\mathrm{ij}}^{\′}\right\}};$

all y' obtained after range normalization_ij∈[0,1]And the fuzzy matrix Y ═ (Y ″) can be obtained without the influence of the dimensional matrix_ij)_n×m；

The second method of normalization is a range normalization, in which the original data matrix Y obtained in step 1 is (Y)_ij)_n×mThe data in (1) is subjected to range normalization to obtain a fuzzy matrix Y ═ Y'_ij)_n×mSpecifically, the following algorithm is implemented:

$y_{\mathrm{ij}}^{\′}=\frac{y_{\mathrm{ij}}-\stackrel{\‾}{y_i}}{\max \left\{y_{\mathrm{ij}}\right\}-\min \left\{y_{\mathrm{ij}}\right\}};$

a third method of normalization is maximum value normalization, i.e., normalizing (Y) the original data matrix Y obtained in step 1_ij)_n×mThe data in (1) is subjected to maximum value normalization processing to obtain a fuzzy matrix Y ═ Y'_ij)_n×mSpecifically, the following algorithm is implemented:

$y_{\mathrm{ij}}^{\′}=\frac{y_{\mathrm{ij}}}{M_j},$

wherein M is_j＝max(y_1j,y_2j…y_nj)；

Step 3, determining a similarity coefficient, and establishing a fuzzy similarity matrix:

the original data matrix Y created according to step 1 ═ (Y)_ij)_n×m，y_iAnd y_jHas a degree of similarity of r_ij＝R(y_i,y_j) The method for establishing the fuzzy similarity matrix according to the similarity coefficient comprises the following steps: a similarity coefficient method, a distance method, or a closeness method;

determining a similarity coefficient by adopting a similarity coefficient method, and establishing a fuzzy similarity matrix, wherein the similarity coefficient method mainly comprises the following two algorithms:

the first method, the angle cosine method, is specifically implemented according to the following algorithm:

$r_{\mathrm{ij}}=\frac{|\underset{l=1}{\overset{m}{\mathrm{\Σ}}}{y}_{\mathrm{il}}*y_{\mathrm{jl}}|}{\sqrt{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}{y}_{\mathrm{il}}^2*\underset{l=1}{\overset{m}{\mathrm{\Σ}}}{y}_{\mathrm{jl}}^2}},(i,j=\mathrm{1,2}\·\·\·n);$

the second method, the correlation coefficient method, is specifically implemented according to the following algorithm:

$r_{\mathrm{ij}}=\frac{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}|y_{\mathrm{il}}-\stackrel{\‾}{y_i}|\·|y_{\mathrm{jl}}-\stackrel{\‾}{y_j}|}{\sqrt{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}{(y_{\mathrm{il}}-\stackrel{\‾}{y_i})}^2}\·\sqrt{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}{(y_{\mathrm{jl}}-\stackrel{\‾}{y_j})}^2}},(i,j=\mathrm{1,2}\·\·\·,n);$

determining a similarity coefficient by adopting a distance method, and establishing a fuzzy similarity matrix, wherein the distance method mainly comprises the following three algorithms:

in general terms, the amount of the solvent to be used,get r_ij＝1-α(d(y_i,y_j))^βWherein α, β are parameters selected such that 0 ≦ r_ij1 or less, and there are three main methods used:

the first distance method, Hamming distance, is specifically implemented according to the following algorithm:

$d(y_i,y_j)=\underset{l=1}{\overset{m}{\mathrm{\Σ}}}|y_{\mathrm{il}}-y_{\mathrm{jl}}|;$

the second distance method, euclidd distance, is specifically implemented according to the following algorithm:

$d(y_i,y_j)=\sqrt{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}{(y_{\mathrm{il}}-y_{\mathrm{jl}})}^2};$

the third distance method, the Chebyshev distance, is specifically implemented according to the following algorithm:

$d(y_i,y_j)=\underset{1\≤l\≤n}{\max }|y_{\mathrm{il}}-y_{\mathrm{jl}}|;$

the similarity coefficient is determined by adopting a proximity method, and a fuzzy similarity matrix is established, wherein the method mainly comprises the following three algorithms: :

the first closeness method, the maximum-minimum method:

second closeness method, geometric mean minimization method:

$r_{\mathrm{ij}}=\frac{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}(y_{\mathrm{il}}^y_{\mathrm{jl}})}{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}\sqrt{y_{\mathrm{il}}\·y_{\mathrm{jl}}}},(i,j=\mathrm{1,2}\·\·\·,n);$

the third closeness method, the number average minimum method:

$r_{\mathrm{ij}}=\frac{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}(y_{\mathrm{il}}^y_{\mathrm{jl}})}{\frac{1}{2}\underset{l=1}{\overset{m}{\mathrm{\Σ}}}(y_{\mathrm{il}}+y_{\mathrm{jl}})},(i,j=\mathrm{1,2}\·\·\·,n);$

among the three algorithms, a closeness method is preferably used for establishing a fuzzy similar matrix of the fuzzy matrix;

step 4, calculating a transfer closure matrix R 'of the fuzzy similar matrix R obtained in the step 3 by using a flat method, namely the fuzzy equivalent matrix t (R) is R';

step 5, observing the lower triangle of the fuzzy equivalent matrix t (R) obtained in the step 4, and then arranging the data in the fuzzy equivalent matrix t (R) from large to small, wherein the data from large to small are the value-taking points of the classification threshold lambda;

step 6, taking out the classification threshold value lambda in the step 5 from large to small one by one, and when the data in the fuzzy equivalent matrix t (R) is larger than the taken classification threshold value lambda, replacing the data in the corresponding fuzzy equivalent matrix t (R) with 1, otherwise replacing the data with 0, wherein the final fuzzy equivalent matrix t (R) only contains two elements of 0 and 1; observing each column of the fuzzy equivalence matrix t (R), wherein the columns with the same data elements are classified into one type, and then classifying the columns;

and 7, according to the different classification threshold values lambda in the step 5, programming and simulating by using MATLAB software to finally obtain a dynamic cluster map of the transformer fault diagnosis model based on fuzzy clustering, and determining the fault type of the transformer by combining a DGA technology.

Examples

The raw data matrix is shown in table 1 below:

TABLE 1 raw data

Data normalization was performed on 7 sets of data in table 1 by the range normalization method, and the data with uniform dimension is shown in table 2:

TABLE 2 normalized data

	1	2	3	4	5	6	7
								H2	0	0.0442	0.0284	0.3754	1	0.0315	0.5016
CH4	0.0569	0.0569	0.0772	1	0.1992	0.0072	0
								C2H6	0.1563	0.2031	0.0781	1	0	0.0781	0.0625
C2H4	0.0163	0.2061	0	1	0.1286	0	0.0143
								C2H2	0	0.1077	0	0.0462	1	0	0.8308

In the determination of fuzzy relation matrix r_ijThen, the proximity method is adopted, that is, as shown in table 3:

TABLE 3 fuzzy relation matrix r_ij

The transitive closure matrix t (r) obtained by the flat method is shown in table 4.

TABLE 4 transitive closure matrix t (R)

Arranging the data in t (R) from small to large as follows:

0.7465＜0.8828＜0.9474＜0.9557＜0.9999

when lambda is 0.7465, the product is

$t{\left(R\right)}_1=\left(\begin{array}{ccccccc}1& 0& 1& 0& 0& 1& 0\\ 0& 1& 0& 1& 1& 0& 1\\ 1& 0& 1& 0& 0& 1& 0\\ 0& 1& 0& 1& 1& 0& 1\\ 0& 1& 0& 1& 1& 0& 1\\ 1& 0& 1& 0& 0& 1& 0\\ 0& 1& 0& 1& 1& 0& 1\end{array}\right)$

Y is divided into 2 classes: {1,3,6},{2,4,5,7}

By analogy, when λ is 0.8828, Y is classified into 3 types: {1, 3, 6}, {2}, {4, 5, 7 };

when λ is 0.9474, Y is classified into 4 types: {1, 3, 6}, {2}, {4}, {5, 7 };

when λ is 0.9557, Y is classified into 4 types: {1, 3, 6}, {2}, {4}, {5, 7 };

when λ is 0.9999, Y is classified into 5 types: {1}, {2}, {3, 6}, {4}, {5}, {7 };

the dynamic cluster map is shown in FIG. 2;

the determination of the transformer fault types in conjunction with the DGA technique is shown in table 5 below:

Claims (7)

1. The transformer fault diagnosis method based on fuzzy clustering is characterized by being implemented according to the following steps:

step 1, inputting the number n and the number m of characteristic variables of objects to be classified according to online monitoring data, and establishing an original data matrix Y (Y ═ Y)_ij)_n×m：

Discourse domain U is { y ═ y according to online monitoring data₁,y₂,y₃…,y_nAs objects to be classified, where each object has its performance represented by m indices, i.e. Y_i＝{y_i1,y_i2…,y_imAnd (i ═ 1,2, … n), establishing an original data matrix, wherein the original data matrix is Y ═ Y_ij)_n×m；

Step 2, standardizing the data in the original data matrix to be between [0, 1] to obtain a fuzzy matrix:

for the original data matrix Y established in step 1 ═ Y_ij)_n×mIs normalized by normalizing y_ijConversion to y 'by algorithm'_ijSpecifically, the conversion is performed according to the following algorithm:

$y_{\mathrm{ij}}^{\′}=\frac{y_{\mathrm{ij}}-\stackrel{\‾}{y_i}}{T_i},(1\≤i\≤n,1\≤j\≤m);$

wherein, $\stackrel{\‾}{y_i}=\frac{1}{m}{y}_{\mathrm{ij}},$ $T_i=\sqrt{\frac{\underset{j=1}{\overset{m}{\mathrm{\Σ}}}{(y_{\mathrm{ij}}-\stackrel{\‾}{y_i})}^2}{m-1}};$

the original data matrix Y is equal to (Y)_ij)_n×mAfter standard deviation normalization transformation of the data in (1), if any, the data in (1)

Then the data is subjected to a range normalization process, which is implemented according to the following algorithm:

$y_{\mathrm{ij}}^{\′\′}=\frac{y_{\mathrm{ij}}-\min \left\{y_{\mathrm{ij}}^{\′}\right\}}{\max \left\{y_{\mathrm{ij}}^{\′}\right\}-\min \left\{y_{\mathrm{ij}}^{\′}\right\}};$

all y' obtained after range normalization_ij∈[0,1]And the fuzzy matrix Y ═ (Y ″) can be obtained without the influence of the dimensional matrix_ij)_n×m；

Step 3, determining a similarity coefficient by adopting one of a similarity coefficient method, a distance method or a proximity method, and establishing a fuzzy similarity matrix;

step 4, calculating a transfer closure matrix R 'of the fuzzy similar matrix R obtained in the step 3 by using a flat method, namely the fuzzy equivalent matrix t (R) is R';

step 5, observing the lower triangle of the fuzzy equivalent matrix t (R) obtained in the step 4, and then arranging the data in the fuzzy equivalent matrix t (R) from large to small, wherein the data from large to small are the value-taking points of the classification threshold lambda;

step 6, taking out the classification threshold value lambda in the step 5 from large to small one by one, and when the data in the fuzzy equivalent matrix t (R) is larger than the taken classification threshold value lambda, replacing the data in the corresponding fuzzy equivalent matrix t (R) with 1, otherwise, replacing the data with 0, and finally, only containing two elements of 0 and 1 in the fuzzy equivalent matrix t (R); observing each column of the fuzzy equivalent matrix t (R), classifying the columns with the same data elements into one class, and further classifying the columns;

and 7, according to the different classification threshold values lambda in the step 5, programming and simulating by using MATLAB software to finally obtain a dynamic cluster map of the transformer fault diagnosis model based on fuzzy clustering, and determining the fault type of the transformer by combining a DGA technology.

2. The transformer fault diagnosis method based on fuzzy clustering according to claim 1, wherein the method of normalizing the raw data matrix in step 2 further adopts range normalization or maximum normalization.

3. The transformer fault diagnosis method based on fuzzy clustering according to claim 2, wherein the range standardization processing method is specifically implemented according to the following algorithm:

the method for performing range standardization processing on the original data matrix obtained in the step 1 is implemented according to the following algorithm:

$y_{\mathrm{ij}}^{\′}=\frac{y_{\mathrm{ij}}-\stackrel{\‾}{y_i}}{\max \left\{y_{\mathrm{ij}}\right\}-\min \left\{y_{\mathrm{ij}}\right\}}.$

4. the transformer fault diagnosis method based on fuzzy clustering according to claim 2, characterized in that the maximum value normalization processing method is specifically implemented according to the following algorithm: carrying out maximum value standardization processing on the original data matrix obtained in the step 1, and specifically implementing the maximum value standardization processing according to the following algorithm:

$y_{\mathrm{ij}}^{\′}=\frac{y_{\mathrm{ij}}}{M_j},$

wherein M is_j＝max(y_1j,y_2j…y_nj)。

5. The transformer fault diagnosis method based on fuzzy clustering according to claim 1, wherein the similarity coefficient method in step 3 comprises the following two algorithms:

the first method, the cosine method of the included angle, is implemented according to the following algorithm:

$r_{\mathrm{ij}}=\frac{|\underset{l=1}{\overset{m}{\mathrm{\Σ}}}{y}_{\mathrm{il}}*y_{\mathrm{jl}}|}{\sqrt{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}{y}_{\mathrm{il}}^2*\underset{l=1}{\overset{m}{\mathrm{\Σ}}}{y}_{\mathrm{jl}}^2}},(i,j=\mathrm{1,2}\·\·\·n);$

the second kind, the correlation coefficient method is implemented according to the following algorithm:

$r_{\mathrm{ij}}=\frac{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}|y_{\mathrm{il}}-\stackrel{\‾}{y_i}|\·|y_{\mathrm{jl}}-\stackrel{\‾}{y_j}|}{\sqrt{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}{(y_{\mathrm{il}}-\stackrel{\‾}{y_i})}^2}\·\sqrt{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}{(y_{\mathrm{jl}}-\stackrel{\‾}{y_j})}^2}},(i,j=\mathrm{1,2}\·\·\·,n).$

6. the transformer fault diagnosis method based on fuzzy clustering according to claim 1, wherein the distance method in step 3 comprises the following three algorithms:

first distance method, Hamming distance:

$d(y_i,y_j)=\underset{l=1}{\overset{m}{\mathrm{\Σ}}}|y_{\mathrm{il}}-y_{\mathrm{jl}}|;$

second distance method, euclidd distance:

$d(y_i,y_j)=\sqrt{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}{(y_{\mathrm{il}}-y_{\mathrm{jl}})}^2};$

third distance method, Chebyshev distance:

$d(y_i,y_j)=\underset{1\≤l\≤n}{\max }|y_{\mathrm{il}}-y_{\mathrm{jl}}|.$

7. the transformer fault diagnosis method based on fuzzy clustering according to claim 1, wherein the closeness method in step 3 comprises the following three algorithms:

the first closeness method, the maximum-minimum method:

second closeness method, geometric mean minimization method:

$r_{\mathrm{ij}}=\frac{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}(y_{\mathrm{il}}^y_{\mathrm{jl}})}{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}\sqrt{y_{\mathrm{il}}\·y_{\mathrm{jl}}}},(i,j=\mathrm{1,2}\·\·\·,n);$

the third closeness method, the number average minimum method:

$r_{\mathrm{ij}}=\frac{\underset{l=1}{\overset{m}{\mathrm{\Σ}}}(y_{\mathrm{il}}^y_{\mathrm{jl}})}{\frac{1}{2}\underset{l=1}{\overset{m}{\mathrm{\Σ}}}(y_{\mathrm{il}}+y_{\mathrm{jl}})},(i,j=\mathrm{1,2}\·\·\·,n).$

Images