CN112630564B - Transformer DGA fault diagnosis method based on neighborhood rough set and AMPOS-ELM - Google Patents

Abstract

The transformer DGA fault diagnosis method based on the neighborhood rough set and the AMPOS-ELM comprises the steps of establishing a transformer initial fault characteristic quantity set by combining each sample initial characteristic quantity of a transformer with each DGA fault diagnosis standard, wherein each sample initial characteristic quantity is the content of one most main volatile gas generated due to transformer faults and the volume ratio of two main gases generated due to the same transformer faults; obtaining key characteristic quantity with high attribute importance after analyzing by adopting a neighborhood rough set; and constructing an AMPOS-ELM model, and taking the key characteristic quantity screened out from the neighborhood rough set as the input of the AMPOS-ELM network model to carry out fault diagnosis on the transformer. Aiming at the problems that the accuracy of the existing DGA-based transformer fault diagnosis method is easily influenced by input characteristics and the selection of parameters of an extreme learning machine is difficult, the accuracy of the diagnosis result of the existing intelligent algorithm is improved.

Description

Transformer DGA Fault Diagnosis Method Based on Neighborhood Rough Set and AMPOS-ELM

technical field

The invention relates to the technical field of power system transformer fault state monitoring, in particular to a transformer DGA fault diagnosis method based on neighborhood rough sets and AMPOS-ELM.

Background technique

The power transformer is one of the important and expensive electrical equipment in the power grid system. It plays an indispensable role in the power system and plays a central role in the transformation and transmission of electric energy. Therefore, it is very necessary to ensure the safe operation of power transformers.

After long-term transformer operation and maintenance practice and a large number of fault investigation and analysis, it is found that if there is a potential fault in the transformer or in the initial stage of fault formation, various gases dissolved in the transformer oil will reflect early symptoms. Dissolved Gas Analysis (DGA for short) is to detect the gas components and content of these fault characteristics, so as to analyze and judge the operation status of transformers and hidden faults.

Transformer insulating oil is a kind of mineral oil, and its main components are unsaturated hydrocarbons and other hydrocarbons containing carbon-carbon double or triple bonds. Some carbon-hydrogen bonds or carbon-carbon bonds in some oil molecules will be broken during the internal discharge fault or heating fault of the transformer, thereby generating a small amount of active hydrogen atoms and hydrocarbon free radicals, and these free hydrogen atoms and free radicals will pass through chemical The reaction is combined again, and finally hydrocarbon gas compounds such as H2, CH4, C2H6, C2H4, and C2H2 can be formed.

Power transformers have complex structures and numerous components. Once a fault occurs, it is impossible to find out the type of fault in time, which causes great inconvenience to subsequent maintenance. At present, due to the harsh boundary conditions set by the IEC three-ratio method, it is easy to cause misjudgment, missed judgment, etc., which may cause long-term operation of the transformer with faults, burying potential risks for the normal operation of the power system.

At present, many achievements have been made in the research of intelligent algorithm in the field of transformer fault diagnosis, mainly including artificial neural network method (BPNN), support vector machine method (Support Vector Machine, SVM) (Extreme Learning Machine, ELM), extreme learning machine and other methods, But the shortcomings still exist. First of all, in terms of feature quantity selection, the selection of input features has a great impact on the classification results. The selected input feature dimensions are too high, resulting in the introduction of too many unnecessary variables, which will not only reduce the prediction accuracy, but also make the model participating in the training It is too complicated, and when the amount of selected input features is small, it is difficult to obtain enough information to characterize the output characteristics. Therefore, selecting appropriate input features is crucial to the reliability of subsequent diagnosis. Secondly, in terms of algorithms, BPNN has strong fault tolerance and nonlinear mapping capabilities, but it has problems such as slow convergence and easy overfitting. SVM can better deal with local minima and has strong generalization ability, but the selection of kernel parameters and penalty factors limits the classification performance of SVM, and if the selected parameters are inappropriate, there will be problems such as large errors in diagnostic results. . ELM is widely used in the field of transformer fault diagnosis due to its fast learning speed, excellent generalization ability, and high classification accuracy. , it is easy to lead to the problem of large loss function and poor robustness. Although the particle swarm optimization algorithm has a good diagnostic effect, it has the defect that it is easy to fall into a local optimum, and further optimization is still needed.

Contents of the invention

The present invention aims at the problem that the existing intelligent algorithm for intelligent transformer faults based on DGA data has a low accuracy rate in transformer fault diagnosis, and provides a transformer DGA fault diagnosis method based on neighborhood rough sets and AMPOS-ELM to improve the accuracy of transformer fault diagnosis. Accuracy.

A transformer DGA fault diagnosis method based on neighborhood rough sets and AMPOS-ELM includes the following steps:

Step S001, combine the initial characteristic quantities of each sample of the transformer with each DGA fault diagnosis standard to establish a transformer initial fault characteristic quantity set. The volume ratio of the two main gases produced by the same transformer fault;

Step S002, obtaining key feature quantities with high attribute importance after neighborhood rough set analysis;

Step S003, constructing an AMPOS-ELM model, and using the key feature quantities selected by the neighborhood rough set as the input of the AMPOS-ELM network model to perform transformer fault diagnosis.

Beneficial effect: the transformer DGA fault diagnosis method based on neighborhood rough set and AMPOS-ELM of the present invention is aimed at the problem that the accuracy rate of the transformer fault diagnosis method based on DGA data is easily affected by the input characteristics and the parameter selection of the extreme learning machine is difficult. The transformer fault diagnosis method based on the neighborhood rough set and adaptive mutation particle swarm extreme learning machine algorithm improves the accuracy of the diagnosis results of the current intelligent algorithm.

Detailed ways

Embodiments of the present application are described below through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the content disclosed in this specification. The present application can also be implemented or applied through other different specific implementation modes, and various modifications or changes can be made to the details in this specification based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that, in the case of no conflict, the following embodiments and features in the embodiments can be combined with each other.

A transformer DGA fault diagnosis method based on neighborhood rough sets and AMPOS-ELM includes the following steps:

Step S001, combine the initial characteristic quantities of each sample of the transformer with each DGA fault diagnosis standard to establish a transformer initial fault characteristic quantity set. The volume ratio of the two main gases produced by the same transformer fault;

Step S002, obtaining key feature quantities with high attribute importance after neighborhood rough set analysis;

Step S003, constructing an AMPOS-ELM model, and using the key feature quantities selected by the neighborhood rough set as the input of the AMPOS-ELM network model to perform transformer fault diagnosis.

Further, in step S001, "combining the initial feature quantities of each sample of the transformer with each DGA fault diagnosis standard to establish a transformer initial fault feature set" includes the following steps:

Step S101, based on the initial characteristic quantities of each sample of the transformer, 7 fault types are established according to the DGA fault diagnosis standard, which are normal state group, partial discharge group, low-energy discharge state group, high-energy discharge state group, high-temperature overheating state group, discharge and overheating state group group, medium and low temperature overheating state group, that is, all initial feature quantities of samples can be classified into the seven fault types;

Step S102, according to the rule of constructing fault characteristics by the ratio method, the content of a most important volatile gas produced by a transformer fault and the volume ratio of two main gases produced by the same transformer fault are used as an initial fault feature set.

Further, in step S002, "obtaining key feature quantities with high attribute importance after neighborhood rough set analysis" includes the following steps:

Step S201, taking the transformer fault type as the decision attribute and the initial fault feature quantity as the condition attribute, establishing a decision information table, establishing the lower limit of the attribute importance and the set of neighborhood radii;

Step S202, obtaining the optimized fault feature quantity set through the neighborhood rough set algorithm and obtaining the attribute importance of each optimized fault feature;

Step S203, sort the attribute importance of each optimized fault feature quantity and the attribute importance of the feature quantity selected in the traditional transformer fault diagnosis in descending order, which is more important than the attribute of the feature quantity selected in the traditional transformer fault diagnosis The fault feature quantity corresponding to the attribute importance of the optimized fault feature quantity with large degree is used as the DGA feature quantity, and the DGA feature quantity is the key feature quantity with high attribute importance.

The specific steps of the neighborhood rough set algorithm are as follows:

Step S301, input decision system NDT={U,A∪V,N,f}, where U is the set of fault types and initial fault features, A is the set of initial fault features, V is the set of fault types, N is the number of fault types, f is an information function, and this information function specifies the attribute value of each object in the set U. In order to obtain a more optimized fault characteristic value, the lower limit of attribute importance is set to 0.2;

且子集B独立于A，子集B记为red；Step S302, establishing a subset B of the set A, that is, assuming that A and B are two equivalence relation families of U,

And the subset B is independent of A, and the subset B is recorded as red;

对任意a_n∈(A-red)利用公式

计算正域pos_B(V)，并选择a_n，使正域pos_B(V)值最大，其中Step S303, make

For any a _n ∈ (A-red) use the formula

Calculate the positive field pos _B (V), and choose a _n to maximize the value of the positive field pos _B (V), where

δ _B (X _i )＝{X _i |X _j ∈U,Δ(X _i ,X _j )≤δ},

X _g ∈ U,

δ _B (X _i ) is the neighborhood of X _g , that is, all fault feature quantities in the set δ _B (X _i ) are similar to X _g , and Y _g is an equivalent subset of U;

Step S304, define the dependency of set V on B:

The initial fault feature quantity a∈B, the attribute importance formula of the initial fault feature quantity a to the decision attribute V is: sig(a,B,V)=γ _B (V)-γ _B-{a} (V), if If the calculated result is greater than the lower limit value of attribute importance 0.2, then output red, otherwise output the initial fault feature quantity a, until all the initial fault feature quantities are classified, so as to obtain the optimal fault feature set and each optimized fault feature attribute importance.

The AMPOS-ELM model is constructed, and the key feature quantities selected by the neighborhood rough set are used as the input of the AMPOS-ELM network model, and the method for transformer fault diagnosis is as follows:

Step S401: Determine the ELM topology and training samples, initialize the particle swarm, and select appropriate c _max and c _min , k _max , ω, M;

Step S402: Set the current fitness F _i =σ of the POS algorithm. Compare the fitness value of each particle with the optimal position pbest experienced by the particle, if F _i < pbest, replace F _i with pbest, otherwise maintain the status quo;

Step S403: compare the size of F _i and gbest, if it is better than the former, use it as the current gbest to become the latest global best position;

Step S404: Update particle weight α _i according to formula (1), and judge whether the particle swarm is premature according to formula (2), if it is premature, give the particle an escape operation according to formula (5); 4) Update particle velocity and position. Repeat the above steps, when the running times reach k _max = 160 maximum iterations or gbest reaches stability, exit the program and return the current optimal individual and its fitness;

In the formula: α _i is the weight coefficient of particle i, i=1,2,3...M, M is the population size, pbest _i is the best position of the particle, gbest is the global best position.

In the formula: Δ is the deviation degree of the average fitness value of the particle swarm, f _i is the fitness degree of particle i, f _avg is the average fitness degree of the current group.

分别为第k次迭代时超参数i的第j维变量的速度、位置、个体最优位置和全局最优位置。In the formula, k is the number of iterations; ω is the inertia weight; c ₁ , c ₂ are particle learning factors; r ₁ , r ₂ ∈ rand[0,1];

are the speed, position, individual optimal position and global optimal position of the j-th dimension variable of the hyperparameter i at the k-th iteration, respectively.

In the formula: uniform random number on r ∈ [0,1]; z ^k is the escape control factor.

Step S405: output the input weight and hidden layer threshold corresponding to the optimal fitness, and calculate the optimal output weight matrix;

Step S406: Establish an ELM-based transformer fault diagnosis model according to the output weight matrix;

Step S407: Input the test sample set into the model established in step S406 for transformer fault diagnosis.

The present invention uses a total of 417 groups of confirmed transformer fault type samples provided by an electric power research institute, and can be divided into 7 states according to the diagnostic results of relevant regulations, as follows: (1) normal state (N); (2) partial discharge (PD) )(3) Low energy discharge state (D1); (4) High energy discharge state (D2); (5) High temperature overheat state (T3); (6) Discharge and overheat state (TD); (7) Medium and low temperature overheat state (T12 ). The sample distribution is shown in Table 1. There are 417 groups of samples in total, 317 groups are randomly selected as the training set, and the remaining 100 groups are used as the test set.

The sample distribution is shown in Table 1.

Table 1 Transformer fault samples

Gases generated by transformer faults exist in the following forms:

2. Screen key attribute characteristics

317 pieces of training data are used to establish a decision table, and the attributes are reduced based on the neighborhood rough set algorithm. The final minimum fault feature set and related importance are shown in Table 3. Compared with the feature quantities selected for traditional transformer fault diagnosis, the CH4/H2, C2H4/C2H6, and C2H2/C2H4 fault features ranked first in the importance of attributes in Table 3 are selected as the optimal result of DGA feature quantities.

Table 3 Key attribute feature set and attribute importance

3. ELM parameter optimization and fault diagnosis

Carry out fault diagnosis and recognition for the test set, and use the key characteristic indicators as the input feature quantity of AMPOS-ELM. The diagnosis results of AMPOS-ELM are shown in Table 4.

Table 4 AMPOS-ELM diagnostic results

It can be seen from the table that the test accuracy of the AMPOS-ELM model is up to 89%, so the AMPOS-ELM diagnostic model proposed by this method has high reliability.

What is disclosed above is only a preferred embodiment of the present invention, and certainly cannot limit the scope of rights of the present invention with this. Those of ordinary skill in the art can understand the whole or part of the process of realizing the above-mentioned embodiment, and make according to the claims of the present invention The equivalent changes still belong to the scope covered by the invention.

Claims (3)

1. a transformer DGA fault diagnosis method based on neighborhood rough sets and AMPOS-ELM, is characterized in that: comprise the following steps: Step S001, combine the initial feature quantities of each sample of the transformer with each DGA fault diagnosis standard to establish a transformer initial fault feature set, the initial fault feature set consists of CH ₄ , C ₂ H ₂ , C ₂ H ₄ , C ₂ H ₆ , H ₂ , CH, C ₂ H ₂ /H ₂ , C ₂ H ₂ /C ₂ H ₄ , C ₂ H6/C 2 H ₂ , C ₂ H 2 /CH ₄ , _{C 2 H 4 /} C ₂ H ₆ , C ₂ H ₄ /CH ₄ , C ₂ H ₄ /H _{2 ,} C ₂ H 6 /CH 4 , C ₂ H ₆ /H ₂ , CH ₄ / _{H 2 ,} C ₂ H ₂ /CH, H _2/ CH +H ₂ , C ₂ H ₄ /CH, CH ₄ /CH, C ₂ H ₆ /CH, CH ₄ +C ₂ H ₄ /CH composition; Step S002, using neighborhood rough set analysis to obtain key feature quantities, said key feature quantities are composed of C ₂ H ₄ /C ₂ H ₆ , C ₂ H ₂ /C ₂ H ₄ , (CH ₄ +C ₂ H ₄ ) /(CH), CH ₄ /H ₂ , H ₂ /(H ₂ +CH), C ₂ H ₆ /C ₂ H ₂ , C ₂ H ₄ , C ₂ H ₆ /CH ₄ , H ₂ , C ₂ H ₂ composition; Step S003, constructing an AMPOS-ELM model, using the key feature quantities selected by the neighborhood rough set as the input of the AMPOS-ELM network model, and performing transformer fault diagnosis; Among them, the transformer fault diagnosis method is as follows: Step S401: Determine the ELM topology and training samples, initialize the particle swarm, and select appropriate c ₁ and c ₂ , k _max , ω, M; Step S402: Set the current fitness of the POS algorithm F _i = σ, compare the fitness value of each particle with the best position pbest experienced by the particle, if F _i < pbest, replace F _i with pbest, otherwise maintain the status quo; Step S403: compare the size of F _i and gbest, if it is better than the former, use it as the current gbest to become the latest global best position; Step S404: Update particle weight α _i according to formula (1), and judge whether the particle swarm is premature according to formula (2), if it is premature, give the particle an escape operation according to formula (5); 4) Update the particle velocity and position, repeat the above steps, and when the number of runs reaches k _max = 160 maximum iterations or gbest reaches stability, exit the program and return the current optimal individual and its fitness;

In the formula: α _i is the weight coefficient of particle i, i=1,2,3...M, M is the population size, pbest _i is the best position of the particle, gbest is the global best position;

In the formula: Δ is the deviation of the average fitness value of the particle swarm, f _i is the fitness of particle i, f _avg is the average fitness of the current group;

In the formula, k is the number of iterations; ω is the inertia weight; c ₁ , c ₂ are particle learning factors; r ₁ , r ₂ ∈ rand[0,1];

are the speed, position, individual optimal position and global optimal position of the j-th dimension variable of the hyperparameter i in the k-th iteration, respectively;

In the formula: a uniform random number on r∈[0,1]; z ^k is the escape control factor; Step S405: output the input weight and hidden layer threshold corresponding to the optimal fitness, and calculate the optimal output weight matrix; Step S406: Establish an ELM-based transformer fault diagnosis model according to the output weight matrix; Step S407: Input the test sample set into the model established in step S406 for transformer fault diagnosis. 2. the transformer DGA fault diagnosis method based on neighborhood rough set and AMPOS-ELM as claimed in claim 1, is characterized in that: in step S001, each sample initial feature quantity of transformer is set up 7 according to DGA fault diagnosis standard The fault types are normal state group, partial discharge group, low-energy discharge state group, high-energy discharge state group, high-temperature overheating state group, discharge and overheating state group, and medium-low temperature overheating state group. into the 7 fault types. 3. the transformer DGA fault diagnosis method based on neighborhood rough set and AMPOS-ELM as claimed in claim 1, is characterized in that: the concrete steps of neighborhood rough set algorithm are as follows: Step S301, input decision system NDT={U,A∪V,N,f}, where U is the set of fault types and initial fault features, A is the set of initial fault features, V is the set of fault types, N is the number of fault types, f is an information function, and this information function specifies the attribute value of each object in the set U. In order to obtain a more optimized fault characteristic value, the lower limit of attribute importance is set to 0.2; Step S302, establishing a subset B of the set A, that is, assuming that A and B are two equivalence relation families of U,

And the subset B is independent of A, and the subset B is recorded as red; Step S303, make

For any a _n ∈ (A-red) use the formula

Calculate the positive field pos _B (V), and choose a _n to maximize the value of the positive field pos _B (V), where

δ _B (X _i )＝{X _i |X _j ∈U,Δ(X _i ,X _j )≤δ}, X _g ∈ U, δ _B (X _i ) is the neighborhood of X _g , that is, all fault feature quantities in the set δ _B (X _i ) are similar to X _g , and Y _g is an equivalent subset of U; Step S304, define the dependency of set V on B:

The initial fault feature quantity a∈B, the attribute importance formula of the initial fault feature quantity a to the decision attribute V is: sig(a,B,V)=γ _B (V)-γ _B-{a} (V), if If the calculated result is greater than the lower limit value of attribute importance 0.2, then output red, otherwise output the initial fault feature quantity a, until all the initial fault feature quantities are classified, so as to obtain the optimal fault feature set and each optimized fault feature attribute importance.