CN111144431B - Transformer fault diagnosis method based on CBBO-SVM - Google Patents

Abstract

The invention discloses a CBBO-SVM-based transformer fault diagnosis method, which comprises the following specific steps of step 1, collecting a sample set S = { (x) of an oil-immersed transformer with class labels ₁ ,x ₂ ,x ₃ ,...x _n ),(y ₁ ,y ₂ ,y ₃ ,...,x _m ) }; step 2, preprocessing and normalizing the acquired data: normalizing both input and output quantities to [0,1]To achieve the purpose of convenient calculation; step 3, utilizing parameter c of CBBO to support vector machine _i And σ _i Optimizing; step 4, utilizing the parameter c obtained after the optimization in the step 3 _i And σ _i And establishing a multi-stage SVM model, and classifying by using the sample set data so as to achieve the effect of fault diagnosis of the oil-immersed transformer. The method can optimize parameters of the SVM (support vector machine) algorithm by using the CBBO (chaotic biophysics) algorithm, and effectively improve the accuracy of classification.

Description

Transformer fault diagnosis method based on CBBO-SVM

Technical Field

The invention belongs to the technical field of transformer fault online monitoring, and particularly relates to a CBBO-SVM-based transformer fault diagnosis method.

Background

Nowadays, the demand for electricity in the whole country is rapidly increased, the power system is gradually advanced to the linkage direction of a large power grid with high voltage grade and high automation, and the stable operation of the power transformer is not only beneficial to promoting the extra-high voltage construction, but also is closely related to the life of the power system, medical treatment, industry and even consumers. Due to the complex structure of the transformer and the interactive influence of various links such as manufacturing, installation, operation and maintenance, maintenance and the like, the transformer is easy to fail in the operation process. Oil-immersed transformers represent a significant proportion of all transformer classes. Therefore, it is necessary to diagnose a fault in the oil-filled transformer.

The traditional diagnosis method of the oil-immersed transformer mainly relates to the following steps: the three-ratio method, the great defense triangle, the gas graph method, the characteristic gas method, etc., which are well established, but are continuously developed today, the diagnostic methods gradually expose many disadvantages, such as: low accuracy, failure to judge the coexistence of multiple faults, incomplete ratio code combination and the like. People add intelligent algorithms such as BP neural network, bayesian algorithm, firefly algorithm and the like to the traditional diagnosis method, and although the accuracy is improved to a certain extent, new problems occur, such as: the required fault data is more, the iteration times are high, the calculation time is long, and the like.

Therefore, the invention provides a transformer fault diagnosis method based on CBBO-SVM (chaotic biophysical optimization support vector machine), which can obtain optimal parameters on the premise of lower iteration number and greatly improve the diagnosis accuracy.

Disclosure of Invention

The invention aims to provide a CBBO-SVM-based transformer fault diagnosis method, which can optimize parameters of an SVM (support vector machine) algorithm by using a CBBO (chaotic biophysics) algorithm and effectively improve the classification accuracy.

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

step 1, collected sample set S = { (x) of oil-immersed transformer with class label ₁ ，x ₂ ，x ₃ ，...x _n )，(y ₁ ，y ₂ ，y ₃ ，...，x _m ) Of which 50% are training samples, 50% are testing samples, x _i Representing the properties of the sample, including 7 properties of hydrogen, methane, ethane, ethylene, acetylene, carbon monoxide and carbon dioxide, y _i The category labels 1, 2, 3, 4, 5 and 6 are represented and respectively correspond to 6 states of normal state, medium and low temperature overheating, high energy discharge, low energy discharge and arc discharge;

step 2, carrying out pretreatment and normalization treatment on the collected data: normalizing the input quantity and the output quantity to [0,1] by the following formula (1) so as to achieve the purpose of convenient calculation;

in the above formula, X _max 、X _min Respectively is collected H ₂ 、CH ₄ 、C ₂ H ₂ 、C ₂ H ₆ 、C ₂ H ₄ CO and CO ₂ Maximum and minimum values of the concentration of the gas, X _i Is represented by H ₂ 、CH ₄ 、C ₂ H ₂ 、C ₂ H ₆ 、C ₂ H ₄ CO and CO ₂ The concentration of the gas;

step 3, utilizing parameter c of CBBO to support vector machine _i And σ _i Optimizing;

step 4, utilizing the parameter c obtained after the optimization in the step 3 _i And σ _i And establishing a multi-stage SVM model, and classifying by using the sample set data so as to achieve the effect of fault diagnosis of the oil-immersed transformer.

The present invention is also characterized in that,

step 3 is specifically implemented according to the following steps:

step 3.1, SVM parameter c _i And σ _i After the initialization operation is performed, c is _i And σ _i As each habitat P _i Is vector of fitness X _i Then generating an initial population W _i ¹ ；

Step 3.2, calculating initial population W _i ¹ Population habitat suitability index f _i ¹ Obtaining f from the formula (2) _i ¹ ；

In the above formula, i is the number of the habitat variable; n is the total number of habitat variables; j represents each habitat variable P _i A level of (d); c. C _j As a habitat variable P _i The total number of stages; beta is a _ij Is a multiple regression coefficient;

ω _ij is a weighting factor derived from the frequency distribution of habitat variables; a is a constant coefficient;

step 3.3, initial population W _i ¹ Carrying out migration operation, and calculating the migration rate lambda and the migration rate mu according to the formula (3);

in the above formula, I is the maximum mobility; e is the maximum migration rate; lambda is the mobility rate; mu is an migration rate; s. the _max The maximum number of species that can be accommodated by the habitat; s is the number of species in the habitat;

after the migration operation, a new species W is formed according to the migration in and out _i ² Then calculating the suitability index f of the population habitat _i ² ；

Step 3.4, performing variation operation on the population after the migration operation, wherein the variation rate m (x) of the migration habitat is along with the variation rate P of the species _s Is increased, i.e.:

in the above formula: m (x) is habitat variation rate; m is _max The maximum rate of variation; p _s The probability of the species corresponding to the species number S in the habitat is shown; p _max Is P _s Maximum value of (d);

P _s relationship to the mobility λ and the mobility μ:

randomly generating new species W in random habitats based on mutation operations _i ³ Instead of the original species W _i ² Calculating the suitability index f of the population habitat _i ³ ；

Step 3.5, adding the C chaos theory and utilizing the subsection LThe g-istic chaotic mapping equation (6) updates the population to generate a new initialization fitness vector f _i ⁴ ：

Wherein x is _i Is a chaotic variable; k is the number of chaotic iterations; in step 3.5, when μ =4, the mapping equation (6) is in a chaotic state.

In order to ensure the diversity of species, chaotic search is carried out on the basis of Logistic chaotic mapping, and the optimal solution in the new fitness solution is recorded as J _best To J with _best Chaotic search is performed on the basis of:

J _n ＝J _best +ωx _i (7)

wherein x is _i Is the solution of the chaotic mapping equation; omega is an adjustment coefficient;

for the obtained optimal solution J _n And J _best Comparing, taking the optimum of the two, repeating step 3.2, and updating the output parameter c according to the requirement _i And σ _i 。

Step 4 is specifically implemented according to the following steps:

step 4.1, firstly, kernel function selection is carried out;

step 4.2, taking training samples accounting for 50% of the total samples as model input for verification;

step 4.3, using c obtained in step 3.5 _i And σ _i Introducing and obtaining a new objective function f (omega, sigma) _i )：

Wherein c is a penalty factor; sigma _i Is a relaxation factor;

step 4.4, solving by utilizing a Lagrange multiplier method, and combining an objective function and a constraint condition formula (10)

y _i (ωx _i +b)≥1-σ _i (10)

Wherein,

ω and b are weight vectors, σ _i Is a relaxation factor;

solving omega and b to obtain a final decision function;

and 4.5, obtaining 6 decision functions through 6 layers of training, and establishing a 6-level support vector machine transformer fault diagnosis model.

The kernel function in step 4.4 is a Gaussian kernel function K (X, X) _i )。

Gaussian kernel function K (X, X) _i ) The method specifically comprises the following steps:

where ξ is a parameter that controls the height of the gaussian kernel function.

The invention has the beneficial effects that:

(1) According to the transformer fault diagnosis method based on the CBBO-SVM, the BBO algorithm is optimized by using the C chaos theory, and the formed CBBO algorithm greatly improves the parameter optimization capability;

(2) According to the transformer fault diagnosis method based on CBBO-SVM, CBBO is used for optimizing SVM parameters, the principle is simple, and classification performance of SVM can be improved on the premise of ensuring less iteration times;

(3) According to the transformer fault diagnosis method based on the CBBO-SVM, the CBBO-SVM algorithm is combined with gas detection in transformer oil, so that the accuracy of fault diagnosis of the oil-immersed transformer is greatly improved.

Drawings

FIG. 1 is a schematic diagram of a cosine mobility model involved in a CBBO-SVM-based transformer fault diagnosis method of the invention;

FIG. 2 is a schematic diagram of a multi-stage support vector machine involved in the CBBO-SVM based transformer fault diagnosis method of the invention;

FIG. 3 is a flow chart of the transformer fault diagnosis method based on CBBO-SVM of the invention.

Detailed Description

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

The invention relates to a CBBO-SVM-based transformer fault diagnosis method, the flow of which is shown in figure 1 and is implemented according to the following steps:

step 1, collected sample set S = { (x) with class label of oil-immersed transformer ₁ ，x ₂ ，x ₃ ，...x _n )，(y ₁ ，y ₂ ，y ₃ ，...，x _m ) Wherein, the training sample accounts for 50%, the testing sample accounts for 50%, x _i Representing the properties of the sample, including 7 properties of hydrogen, methane, ethane, ethylene, acetylene, carbon monoxide and carbon dioxide, y _i The category labels 1, 2, 3, 4, 5 and 6 are represented and respectively correspond to 6 states of normal state, medium and low temperature overheating, high energy discharge, low energy discharge and arc discharge;

step 2, preprocessing and normalizing the acquired data: normalizing both the input quantity and the output quantity to [0,1] by the following formula (1) so as to achieve the purpose of convenient calculation;

in the above formula, X _max 、X _min Respectively collected H ₂ 、CH ₄ 、C ₂ H ₂ 、C ₂ H ₆ 、C ₂ H ₄ CO and CO ₂ Maximum and minimum values of the concentration of the gas, X _i Represents H ₂ 、CH ₄ 、C ₂ H ₂ 、C ₂ H ₆ 、C ₂ H ₄ CO and CO ₂ The gas concentration;

step 3, utilizing parameter c of CBBO to support vector machine _i And σ _i Optimizing;

step 3 is specifically implemented according to the following steps:

step 3.1, SVM parameter c _i And σ _i After the initialization operation is performed, c is _i And σ _i As each habitat P _i Is (SIV) X _i Then generating an initial population W _i ¹ ；

Step 3.2, calculating initial population W _i ¹ Population Habitat Suitability Index (HSI) f _i ¹ Obtaining f from the formula (2) _i ¹ ；

In the above formula, i is the number of the habitat variable; n is the total number of habitat variables; j represents each habitat variable P _i A level of (d); c. C _j As a habitat variable P _i The total number of stages; beta is a _ij Is a multiple regression coefficient;

ω _ij is a weighting factor derived from the frequency distribution of habitat variables; a is a constant coefficient;

as shown in fig. 2, the cosine mobility model is specifically established as follows:

step 3.3, for the initial population W _i ¹ Carrying out migration operation, wherein in four common mobility models, namely a linear mobility model, a secondary mobility model, an index mobility model and a cosine mobility model, in order to make result parameters more optimal, a cosine mobility model which more accords with a natural rule is selected and established, and the migration rate lambda and the migration rate mu are calculated according to a formula (3);

in the above formula, I is the maximum mobility; e is the maximum migration rate; lambda is the mobility; mu is the migration rate; s _max The maximum number of species that can be accommodated by the habitat; s is the number of species in the habitat;

after the migration operation, according to the migrationThe new species W is formed by emigration _i ² Then calculating the suitability index f of the population habitat _i ² ；

Step 3.4, performing variation operation on the population after the migration operation, wherein the variation rate m (x) of the migration habitat is along with the variation rate P of the species _s Is increased, i.e.:

in the above formula: m (x) is habitat variation rate; m is _max The maximum rate of variation; p _s The probability of the species corresponding to the species number S in the habitat is shown; p _max Is P _s Maximum value of (d);

P _s relationship to the mobility λ and the mobility μ:

randomly generating new species W in random habitats based on mutation operations _i ³ Instead of the original species W _i ² Calculating the suitability index f of the population habitat _i ³ ；

Step 3.5, adding a C chaos theory, updating the population by using a segmented Logistic chaos mapping equation (6) and generating a new initialized fitness vector f _i ⁴ ：

Wherein x is _i Is a chaotic variable; k is the number of chaotic iterations; in step 3.5, when μ =4, the mapping equation (6) is in a chaotic state.

In order to ensure the diversity of species, chaotic search is carried out on the basis of Logistic chaotic mapping, and the optimal solution in the new fitness solution is recorded as J _best In addition to J _best Chaotic search basedNamely:

J _n ＝J _best +ωx _i (7)

wherein x is _i Is the solution of the chaotic mapping equation; omega is an adjustment coefficient;

for the obtained optimal solution J _n And J _best Comparing, taking the optimum of the two, repeating step 3.2, and updating the output parameter c according to the requirement _i And σ _i ；

Step 4, utilizing the parameter c obtained after the optimization in the step 3 _i And σ _i Establishing a multi-stage SVM model, and classifying by using sample set data so as to achieve the effect of fault diagnosis of the oil-immersed transformer;

as shown in fig. 3, the specific establishment method of the multi-stage support vector machine fault diagnosis model is as follows:

step 4 is specifically implemented according to the following steps:

step 4.1, firstly, kernel function selection is carried out;

selecting the most common Gaussian (RBF) kernel function from the linear kernel function, the polynomial kernel function, the sigmoid kernel function and the Gaussian (RBF) kernel function:

where ξ is a parameter controlling the height of a Gaussian (RBF) kernel;

step 4.2, taking training samples accounting for 50% of the total samples as model input for verification;

step 4.3, using c obtained in step 3.5 _i And σ _i Introducing and obtaining a new objective function f (omega, sigma) _i )：

Where C is a penalty factor (meaning that the classification is strict when C tends to be large; meaning that greater error tolerance is possible when C tends to be small), a constant that needs to be specified,

parameters that need to be optimized; sigma _i Is a relaxation factor.

Is also the invention

Step 4.4, solving by utilizing a Lagrange multiplier method, and combining an objective function and a constraint condition formula (10)

y _i (ωx _i +b)≥1-σ _i (10)

Wherein,

ω and b are weight vectors, σ _i Is a relaxation factor;

solving omega and b to obtain a final decision function;

and 4.5, obtaining 6 decision functions through 6 layers of training, and establishing a 6-level support vector machine transformer fault diagnosis model.

Claims (5)

1. A transformer fault diagnosis method based on CBBO-SVM is characterized by comprising the following steps:

step 1, collected sample set S = { (x) with class label of oil-immersed transformer ₁ ，x ₂ ，x ₃ ，...x _n )，(y ₁ ，y ₂ ，y ₃ ，...，x _m ) Wherein, the training sample accounts for 50%, the testing sample accounts for 50%, x _i Representing the properties of the sample, including 7 properties of hydrogen, methane, ethane, ethylene, acetylene, carbon monoxide and carbon dioxide, y _i The category labels 1, 2, 3, 4, 5 and 6 are represented and respectively correspond to 6 states of normal state, medium and low temperature overheating, high energy discharge, low energy discharge and arc discharge;

step 2, preprocessing and normalizing the acquired data: normalizing both the input quantity and the output quantity to [0,1] by the following formula (1) so as to achieve the purpose of convenient calculation;

in the above formula, X _max 、X _min Respectively collected H ₂ 、CH ₄ 、C ₂ H ₂ 、C ₂ H ₆ 、C ₂ H ₄ CO and CO ₂ Maximum and minimum values of the concentration of the gas, X _i Represents H ₂ 、CH ₄ 、C ₂ H ₂ 、C ₂ H ₆ 、C ₂ H ₄ CO and CO ₂ The gas concentration;

step 3, utilizing parameter c of CBBO to support vector machine _i And σ _i Optimizing;

step 4, utilizing the parameter c obtained after the optimization in the step 3 _i And σ _i And establishing a multi-stage SVM model, and classifying by using the sample set data so as to achieve the effect of fault diagnosis of the oil-immersed transformer.

2. The CBBO-SVM based transformer fault diagnosis method as claimed in claim 1, wherein the step 3 is implemented by the following steps:

step 3.1, SVM parameter c _i And σ _i After the initialization operation is performed, c is _i And σ _i As each habitat P _i Is vector of fitness X _i Then generating an initial population W _i ¹ ；

Step 3.2, calculating initial population W _i ¹ Population habitat suitability index f _i ¹ Obtaining f from the formula (2) _i ¹ ；

In the above formula, i is the number of the habitat variable; n is the total number of habitat variables; j represents each habitat variable P _i A level of (d); c. C _j As a habitat variable P _i The total number of stages; beta is a _ij Is a multiple regression coefficient; omega _ij Is a weighting factor derived from the frequency distribution of habitat variables; a is a constant systemCounting;

step 3.3, for the initial population W _i ¹ Carrying out migration operation, and calculating the migration rate lambda and the migration rate mu according to the formula (3);

in the above formula, I is the maximum mobility; e is the maximum migration rate; lambda is the mobility; mu is the migration rate; s _max The maximum number of species that can be accommodated by the habitat; s is the number of species in the habitat;

after the migration operation, a new species W is formed according to the migration in and out _i ² Then calculating the suitability index f of the population habitat _i ² ；

Step 3.4, carrying out variation operation on the population after the migration operation, wherein the variation rate m (x) of the migration habitat is along with the variation rate P of the species _s Is increased, i.e.:

in the above formula: m (x) is habitat variability; m is a unit of _max The maximum rate of variation; p _s The probability of the species corresponding to the species with the number of S in the habitat is set; p _max Is P _s Maximum value of (d);

P _s relationship to the mobility λ and the mobility μ:

randomly generating new species W in random habitats based on mutation operations _i ³ Replacing the original species W _i ² Calculating the suitability index f of the population habitat _i ³ ；

Step 3.5, adding a C chaos theory, and utilizing a segmented Logistic chaos mapping equation (6) to perform population matchingUpdating to generate new initialized fitness vector f _i ⁴ ：

Wherein x is _i Is a chaotic variable; k is the number of chaotic iterations; in the step 3.5, when μ =4, the mapping equation (6) is in a chaotic state;

in order to ensure the diversity of species, chaotic search is carried out on the basis of Logistic chaotic mapping, and the optimal solution in the new fitness solution is recorded as J _best In addition to J _best Chaotic search is carried out on the basis of the chaotic search, namely:

J _n ＝J _best +ωx _i (7)

wherein x is _i Is the solution of the chaotic mapping equation; omega is an adjustment coefficient;

for the obtained optimal solution J _n And J _best Comparing, selecting the optimal one, repeating step 3.2, and updating output parameter c according to requirement _i And σ _i 。

3. The CBBO-SVM based transformer fault diagnosis method as claimed in claim 1, wherein the step 4 is implemented by the following steps:

step 4.1, firstly, kernel function selection is carried out;

step 4.2, taking training samples accounting for 50% of the total samples as model input for verification;

step 4.3, c obtained in step 3.5 _i And σ _i Introducing and obtaining a new objective function f (omega, sigma) _i )：

Wherein c is a penalty factor; sigma _i Is a relaxation factor;

step 4.4, solving by utilizing a Lagrange multiplier method, and combining an objective function and a constraint condition formula (10)

y _i (ωx _i +b)≥1-σ _i (10)

Wherein,

ω and b are weight vectors, σ _i Is a relaxation factor;

solving omega and b to obtain a final decision function;

and 4.5, obtaining 6 decision functions through 6 layers of training, and establishing a 6-level support vector machine transformer fault diagnosis model.

4. The CBBO-SVM based transformer fault diagnosis method as claimed in claim 3, wherein the kernel function in step 4.4 is a Gaussian kernel function K (X, X) _i )。

5. The CBBO-SVM based transformer fault diagnosis method as claimed in claim 4, wherein the Gaussian kernel function K (X, X) _i ) The method specifically comprises the following steps:

where ξ is a parameter that controls the height of the gaussian kernel function.

Images