Transformer fault diagnosis method based on improved Harris eagle optimized multi-core extreme learning machine

Abstract

The invention discloses a transformer fault diagnosis method based on an improved Harris eagle optimization multi-core extreme learning machine, and the method comprises the following steps: 1, carrying out the data enhancement of a minority class of samples in transformer fault samples through employing an SMOTE algorithm, and enabling each class of samples to achieve the class balance; 2, normalizing the balanced sample data, and dividing the balanced sample data into a training set, a verification set and a test set; 3, selecting a plurality of different kernel functions to construct a multi-kernel extreme learning machine classification model, and training the multi-kernel extreme learning machine classification model by using the training set; step 4, optimizing hyper-parameters of the multi-core extreme learning machine by adopting an improved Harris eagle algorithm, and obtaining an optimal multi-core extreme learning machine model by taking an identification error of the verification set as a fitness function; and 5, testing the classification performance of the optimal multi-core extreme learning machine model by using the test set. According to the method, the nonlinear representation of the transformer fault sample is effectively enhanced by using the multi-core extreme learning machine, and the accuracy and reliability of transformer fault diagnosis can be improved; meanwhile, the robustness of the multi-core extreme learning machine is effectively enhanced by using the improved Harris eagle optimization algorithm with higher convergence speed and higher search capability of parameter optimization.

Description

Technical Field

[0001] The invention belongs to the technical field of oil-immersed transformer equipment state detection, and in particular relates to a transformer fault diagnosis method based on an improved Harris Eagle optimized multi-core extreme learning machine. Background Art

[0002] Transformers play a key role in power systems, and failures may lead to power outages, equipment damage, and even accidents. Therefore, predicting and detecting transformer failures through effective fault diagnosis is crucial to ensure the safe and reliable operation of power systems.

[0003] Dissolved Gas Analysis (DGA) is a widely used technology for transformer fault diagnosis. By monitoring the type and content of dissolved gas in transformer oil, important clues about the internal fault conditions of the transformer can be provided. The content and composition of dissolved gas in oil vary with the working state and fault type of the transformer, so the fault type of the transformer can be identified and determined by analyzing the oil sample, such as aging of the insulation material, partial discharge, overheating, etc. However, there are some challenges in the fault diagnosis of oil-immersed transformers. First, due to the long working time of the transformer, the fault samples of a few categories often occupy a small part of the entire sample set, resulting in an imbalanced class of the data set. This will have a certain impact on the accuracy and reliability of fault diagnosis, because the model may be more inclined to learn the features of the dominant category, while it is more difficult for the fault samples of the minority categories. Secondly, the fault samples of oil-immersed transformers are usually composed of multiple heterogeneous data with complex nonlinear relationships between these heterogeneous data. Traditional linear classification algorithms or machine learning methods with a single kernel function often find it difficult to fully characterize the features of such heterogeneous samples, thereby limiting the accuracy and performance of fault diagnosis.

[0004] Therefore, in order to overcome these problems, a method that can effectively handle class imbalance and nonlinear heterogeneous samples is needed. The transformer fault diagnosis method based on the improved Harris Eagle optimized multi-core extreme learning machine is an innovative solution to these challenges. By introducing the multi-core extreme learning machine, the nonlinear representation space of the sample can be expanded, thereby better capturing heterogeneous samples. Summary of the invention

[0005] The main purpose of the present invention is to provide a transformer fault diagnosis method based on an improved Harris Eagle optimized multi-core extreme learning machine, to solve the transformer fault classification problem and to improve the fault diagnosis accuracy.

[0006] The present invention is implemented by the following technical solution: a transformer fault diagnosis method based on an improved Harris Eagle optimized multi-core extreme learning machine, comprising:

[0007] S1, obtaining transformer fault information and related dissolved gas in oil as sample data;

[0008] S2, use the SMOTE algorithm to perform data enhancement on the obtained minority class samples;

[0009] S3, normalize various sample data and divide them into training set, validation set and test set;

[0010] S4, select multiple different kernel functions and build a multi-core extreme learning machine classification model;

[0011] S5, using the improved Harris Eagle algorithm to optimize the hyperparameters of the multi-core extreme learning machine to obtain the optimal multi-core extreme learning machine model;

[0012] S6, uses the test set to test the classification performance of the optimal multi-core extreme learning machine model.

[0013] Furthermore, the transformer fault types in S1 include medium and low temperature overheating, high temperature overheating, low energy discharge, high energy discharge and partial discharge, wherein the samples of each fault are composed of five types of oil dissolved gas data: H2, C2H2, CH4, C2H6 and C2H4.

[0014] Furthermore, the specific expression for performing data enhancement on minority class samples in transformer fault samples using the SMOTE algorithm in S2 is:

[0015]

[0016] Among them, x new is the new fault data sample after oversampling, x is the original fault data sample, is the adjacent sample of the original fault data sample, rand(0,1) is a random value function from 0 to 1;

[0017] Furthermore, the specific steps of S3 include:

[0018] S31, using range standardization to limit the five types of dissolved gas data in each sample to the range of [0,1], eliminating the influence of different gas dimensions on the judgment of transformer fault type;

[0019] S32, one-hot encoding is used to define each transformer fault type, and all samples are divided into training set, validation set and test set according to the random ratio of 6:2:2.

[0020] Furthermore, the specific steps of S4 include:

[0021] S41, constructing a kernel function of a multi-core extreme learning machine, wherein the basic kernel functions constituting the kernel function of the multi-core extreme learning machine include four types: a linear kernel function, a polynomial kernel function, a radial basis kernel function, and a sigmoid kernel function, and their expressions are respectively:

[0022] K lin (x,x i )=x T x i

[0023] K poly (x,x i )=(x T x i +1) γ

[0024]

[0025] K sig (x,x i ) = tanh(x T x j +β)

[0026] In the formula, K lin (x,x i ) is the linear kernel function, K poly (x,x i ) is the polynomial kernel function, K rbf (x,x i ) is the radial basis kernel function, K sig (x,x i ) is the sigmoid kernel function, γ, σ and β are the kernel parameters respectively;

[0027] On this basis, the weighted sum of the four kernel functions is used to obtain the kernel function of the kernel extreme learning machine:

[0028]

[0029] In the formula, K g (x,x i ) represents a kernel function, λ g is the corresponding weight coefficient;

[0030] S42, build a multi-core extreme learning machine model, its expression is:

[0031]

[0032] Where Y is the actual category label of each input sample, is the corresponding predicted category label, C is the regularization parameter, Ω elm is the kernel function matrix, and its expression is:

[0033]

[0034] Furthermore, the specific steps of S5 include:

[0035] S51, improves the Harris Hawk algorithm. First, chaotic Tent mapping is used to improve the population initialization method:

[0036]

[0037] X i =lb+Y i (ub-lb)

[0038] In the formula, i = 1, 2, 3, ..., N-1, X1 is the first randomly generated individual in the population, X i is the i-th individual in the initial population, Y i is the i-th individual in the mapping space;

[0039] Secondly, in order to prevent each iteration from falling into the local optimum, a chi-square distribution probability density function is added after each iteration so that a certain probability of mutation occurs after the individual position is updated. The expression of the chi-square probability density function is:

[0040]

[0041] In the formula, n is the natural degree, r is the number of iterations;

[0042] S52, construct the fitness function for parameter optimization, and the objective function is defined as follows:

[0043] fitness=1-ACC

[0044]

[0045] In the formula, ACC represents the overall recognition accuracy of the validation set, NC i is the number of correct classifications of the i-th category, NE i is the number of misclassifications of the i-th category, and l is the total number of categories;

[0046] S53, applying the improved Harris Eagle optimization algorithm to the above objective function, obtaining the optimal values ​​of the parameters of the multi-core extreme learning machine through the global search phase, the search to the development phase and the development phase, and obtaining the optimal multi-core extreme learning machine model, wherein the parameters to be optimized include the parameters of each basic kernel function, the weight coefficient of each basic kernel function, and the regularization parameter C.

[0047] Beneficial effects of the present invention:

[0048] The present invention proposes a transformer fault diagnosis method based on an improved Harris Hawk optimized multi-core extreme learning machine. This method makes full use of the advantages of the kernel extreme learning machine, which has good generalization and fast training speed; at the same time, through the kernel mapping of multiple kernel functions, the feature space of the sample is greatly expanded, so that the transformer fault samples composed of different dissolved gases can also be well expressed; in addition, by using the Tent chaos mapping and chi-square probability density function to improve the Harris Hawk optimization algorithm, the convergence speed and spatial search ability of the Harris Hawk algorithm are effectively improved, and the robustness of the classification model is further enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is a flow chart of the method of the present invention;

[0050] Figure 2 This is a schematic diagram of the structure of the multi-core extreme learning machine model proposed in the present invention;

[0051] Figure 3 This is the flow chart of the improved Harris Eagle algorithm proposed by the present invention. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0053] refer to Figure 1 , a transformer fault diagnosis method based on improved Harris Eagle optimized multi-core extreme learning machine, comprising the following steps:

[0054] S1, obtaining transformer fault information and related dissolved gas in oil as sample data;

[0055] S2, use the SMOTE algorithm to perform data enhancement on the obtained minority class samples;

[0056] S3, normalize various sample data and divide them into training set, validation set and test set;

[0057] S4, select multiple different kernel functions and build a multi-core extreme learning machine classification model;

[0058] S5, using the improved Harris Eagle algorithm to optimize the hyperparameters of the multi-core extreme learning machine to obtain the optimal multi-core extreme learning machine model;

[0059] S6, uses the test set to test the classification performance of the optimal multi-core extreme learning machine model.

[0060] The transformer fault types in S1 include medium and low temperature overheating, high temperature overheating, low energy discharge, high energy discharge and partial discharge. The samples of each fault are composed of five types of oil dissolved gas data: H2, C2H2, CH4, C2H6 and C2H4.

[0061] In S2, the SMOTE algorithm is used to perform data enhancement on the minority class samples in the transformer fault samples. The specific expression is:

[0062]

[0063] In the formula, x new is the new fault data sample after oversampling, x is the original fault data sample, is the adjacent sample of the original fault data sample, rand(0,1) is a random value function from 0 to 1;

[0064] Specifically, S3 also includes the following steps:

[0065] S31, the five types of dissolved gas data in each sample are limited to the range of [0,1] by using range standardization to eliminate the influence of different gas dimensions on the judgment of transformer fault type. For each transformer fault sample, the specific expression is:

[0066]

[0067] In the formula, a i represents the content of the i-th dissolved gas constituting the sample, a max Indicates the maximum dissolved gas content in the sample, a min Indicates the dissolved gas content with the smallest value in the sample, i=1,2,3,4,5.

[0068] S32, one-hot encoding is used to define each transformer fault type, and all samples are divided into training set, validation set and test set according to the random ratio of 6:2:2.

[0069] Specifically, S4 further includes the following steps:

[0070] S41, select a mixed weight of a radial basis kernel function, a polynomial kernel function, a linear kernel function and a sigmoid kernel function as a kernel function of a kernel extreme learning machine, wherein their expressions are respectively:

[0071] K lin (x,x i )=x T x i

[0072] K poly (x,x i )=(xT x i +1) γ

[0073]

[0074] K sig (x,x i ) = tanh(x T x j +β)

[0075] In the formula, K lin (x,x i ) is the linear kernel function, K poly (x,x i ) is the polynomial kernel function, K rbf (x,x i ) is the radial basis kernel function, K sig (x,x i ) is the sigmoid kernel function, γ, σ and β are the kernel parameters respectively;

[0076] On this basis, the weighted sum of the four kernel functions is used to obtain the kernel function of the kernel extreme learning machine:

[0077]

[0078] In the formula, K g (x,x i ) represents a kernel function, λ g is the corresponding weight coefficient;

[0079] S42, build a multi-core extreme learning machine model, Figure 2 Describes the specific structure of the model, and its expression is:

[0080]

[0081] Where Y is the actual category label of each input sample, is the corresponding predicted category label, C is the regularization parameter, Ω elm is the kernel function matrix, and its expression is:

[0082]

[0083] Specifically, the S5 further includes the following steps:

[0084] S51, improves the Harris Hawk algorithm. First, chaotic Tent mapping is used to improve the population initialization method:

[0085]

[0086] Xi =lb+Y i (ub-lb)

[0087] In the formula, i = 1, 2, 3, ..., N-1, X1 is the first randomly generated individual in the population, X i is the i-th individual in the initial population, Y i is the i-th individual in the mapping space;

[0088] Secondly, in order to prevent each iteration from falling into the local optimum, a chi-square distribution probability density function is added after each iteration so that a certain probability of mutation occurs after the individual position is updated. The expression of the chi-square probability density function is:

[0089]

[0090] In the formula, n is the natural degree, r is the number of iterations;

[0091] S52, constructing a fitness function for parameter optimization and setting a fitness threshold, the function is defined as follows:

[0092] fitness=1-ACC

[0093]

[0094] Among them, ACC represents the recognition accuracy of the validation set, NC i is the number of correct classifications of the i-th category, NE i is the number of misclassifications of the i-th category, and l is the total number of categories;

[0095] S53, uses the improved Harris Eagle optimization algorithm to optimize the hyperparameters of the multi-core extreme learning machine.

[0096] S54, substitute the hyperparameter value obtained in S53 into the multi-core extreme learning machine model, and use the training set to train the parameters in the multi-core extreme learning machine.

[0097] S55, using the validation set to validate the multi-core extreme learning machine trained in S54, and obtain the corresponding fitness value, if the fitness value is not less than the threshold, proceed to the next step, otherwise return to S53 to continue the hyperparameter optimization and parameter training of the multi-core extreme learning machine.

[0098] Specifically, S53 further includes the following steps:

[0099] S531, determine each parameter to be optimized and their upper and lower limits, the size of the initial population and the maximum number of iterations.

[0100] S532, using Tent chaotic mapping to initialize the population position and select the individual with the best position.

[0101] S533, calculate the escape energy of the prey, and judge the stage of the algorithm based on it. According to the energy of the prey, when entering the development stage, the expression is:

[0102]

[0103] Among them, E represents the escape energy of the prey, E0 represents the initial state of energy, T represents the maximum number of iterations, and the algorithm is in the development stage when |E| < 1, otherwise it is in the global search stage.

[0104] S534, iteratively update the position of the Harris Hawk. At this time, if the algorithm is in the global search stage, the value of the parameter vector (Harris Hawk) is updated through two strategies:

[0105]

[0106] Where H(t) is the current position vector of the eagle, H rabbit (t) is the current position of the individual with the best fitness, H rand (t) is the position of the eagle randomly selected in the current population, H(t+1) is the position vector of the eagle in the next iteration, r1, r2, r3, r4 and q are all random numbers between (0,1), LB and UB are the lower and upper limits of the parameters to be optimized, respectively, and H m (t) is the average position of the population;

[0107] If the algorithm enters the development stage, four attack methods are selected according to the energy E of the prey and the escape probability gl of the prey to update the position of the parameter vector, where:

[0108] When |E|≥0.5 and gl≥0.5, a soft encirclement attack is used;

[0109] When |E|≤0.5 and gl≥0.5, a hard encirclement attack is used;

[0110] When |E|≥0.5 and gl<0.5, a soft encirclement attack method of progressive rapid dive is adopted;

[0111] When |E|≤0.5 and gl<0.5, a hard encirclement attack method with progressive rapid dive is adopted.

[0112] S535, using a chi-square probability density function to mutate the updated position vector.

[0113] S536, determine whether the number of iterations has reached the maximum, if not return to S533, otherwise output the current optimal position vector and end the loop.

[0114] Specifically, S534 also includes the following steps:

[0115] S5341, when the soft encirclement attack method is adopted, the position update expression of the parameter vector is:

[0116] H(t+1)=ΔH(t)-E|JH rabbit (t)-H(t)|

[0117] ΔH(t)=H rabbit (t)-H(t)

[0118] J=2(1-r5)

[0119] Among them, ΔH(t) is the difference between the optimal individual and the current individual, r5 is a random number between (0,1), and J is the jumping distance of the rabbit during the escape process:

[0120] S5342, when the hard encirclement attack method is adopted, the position update expression of the parameter vector is:

[0121] H(t+1)=H rabbit (t)-E|ΔH(t)|

[0122] S5343, when the progressive dive soft encirclement attack method is adopted, the position update expression of the parameter vector is:

[0123]

[0124] Y=H rabbit (t)-EJH rabbit (t)-H(t)

[0125] Z=Y+S×LF(D)

[0126] fitness(·)=1-ACC(·)

[0127] Where fitness(·) is the fitness value of the parameter vector, D is the dimension of the problem, S is a 1×D-dimensional vector, and LF is the Levy flight function, which is expressed as:

[0128]

[0129] Among them, μ, υ are random variables between (0,1), and β is a constant set to 1.5;

[0130] S5344, when using the progressive dive hard encirclement attack method, the parameter position update expression is:

[0131]

[0132] Y=H rabbit (t)-E|JH rabbit (t)-Hm (t)|

[0133] Z=Y+S×LF(D).

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, it should be understood by those skilled in the art that the technical solutions described in the aforementioned embodiments can still be modified, or some or all of the technical features therein can be replaced by equivalents. Therefore, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope defined by the claims of the present invention.

Claims

1. A transformer fault diagnosis method based on improved Harris Eagle optimized multi-core extreme learning machine, characterized in that: The following steps are involved: S1, obtaining transformer fault information and related dissolved gas in oil as sample data; S2, use the SMOTE algorithm to perform data enhancement on the obtained minority class samples; S3, normalize various sample data and divide them into training set, validation set and test set; S4, select multiple different kernel functions and build a multi-core extreme learning machine classification model; S5, using the improved Harris Eagle algorithm to optimize the hyperparameters of the multi-core extreme learning machine to obtain the optimal multi-core extreme learning machine model; S6, uses the test set to test the classification performance of the optimal multi-core extreme learning machine model.

2. A transformer fault diagnosis method based on an improved Harris Hawk optimized multi-core extreme learning machine according to claim 1, characterized in that: The transformer fault types in S1 include medium and low temperature overheating, high temperature overheating, low energy discharge, high energy discharge and partial discharge. The samples of each fault are composed of five types of oil dissolved gas data: H2, C2H2, CH4, C2H6 and C2H4.

3. A transformer fault diagnosis method based on an improved Harris Eagle optimized multi-core extreme learning machine according to claim 1, characterized in that: The specific expression of using the SMOTE algorithm in S2 to perform data enhancement on minority class samples in transformer fault samples is: Among them, x new is the new fault data sample after oversampling, x is the original fault data sample, are the adjacent samples of the original fault data sample, and rand(0,1) is a random value function ranging from 0 to 1.

4. The transformer fault diagnosis method based on improved Harris Hawk optimized multi-core extreme learning machine according to claim 1 is characterized in that: The specific steps of S3 include: S31, using range standardization to limit the five types of dissolved gas data in each sample to the range of [0,1], eliminating the influence of different gas dimensions on the judgment of transformer fault type; S32, one-hot encoding is used to define each transformer fault type, and all samples are divided into training set, validation set and test set according to the random ratio of 6:2:

2.

5. A transformer fault diagnosis method based on an improved Harris Eagle optimized multi-core extreme learning machine according to claim 1, characterized in that: The specific steps of S4 include: S41, construct the kernel function of the multi-core extreme learning machine, where the basic kernel functions constituting the kernel function of the multi-core extreme learning machine include four types: linear kernel function, polynomial kernel function, radial basis kernel function and sigmoid kernel function, and their expressions are respectively: K lin (x,x i )=x T x i K poly (x,x i )=(x T x i +1) γ K sig (x,x i )=tanh(x T x j +β) In the formula, K lin (x,x i ) is the linear kernel function, K poly (x,x i ) is the polynomial kernel function, K rbf (x,x i ) is the radial basis kernel function, K sig (x,x i ) is the sigmoid kernel function, γ, σ and β are the kernel parameters respectively; On this basis, the weighted sum of the four kernel functions is used to obtain the kernel function of the kernel extreme learning machine: In the formula, K g (x,x i ) represents a kernel function, λ g is the corresponding weight coefficient; S42, build a multi-core extreme learning machine model, its expression is: Where Y is the actual category label of each input sample, is the corresponding predicted category label, C is the regularization parameter, Ω elm is the kernel function matrix, and its expression is:

6. A transformer fault diagnosis method based on an improved Harris Eagle optimized multi-core extreme learning machine according to claim 1, characterized in that: The specific steps of S5 include: S51, improves the Harris Hawk algorithm. First, the Tent chaotic map is used to improve the population initialization method: X i =lb+Y i (ub-lb) In the formula, i = 1, 2, 3, ..., N-1, X1 is the first randomly generated individual in the population, X i is the i-th individual in the initial population, Y i is the i-th individual in the mapping space; Secondly, in order to prevent each iteration from falling into the local optimum, a chi-square distribution probability density function is added after each iteration so that a certain probability of mutation occurs after the individual position is updated. The expression of the chi-square probability density function is: In the formula, n is the natural degree, r is the number of iterations; S52, construct the fitness function for parameter optimization, and the objective function is defined as follows: fitness=1-ACC In the formula, ACC represents the overall recognition accuracy of the validation set, NC i is the number of correct classifications of the i-th category, NE i is the number of misclassifications of the i-th category, and l is the total number of categories; S53, applying the improved Harris Eagle optimization algorithm to the above objective function, obtaining the optimal values ​​of the parameters of the multi-core extreme learning machine through the global search phase, the search to the development phase and the development phase, and obtaining the optimal multi-core extreme learning machine model, wherein the parameters to be optimized include the parameters of each basic kernel function, the weight coefficient of each basic kernel function, and the regularization parameter C.

Images