Transformer fault diagnosis method, system, equipment and medium

Abstract

The invention provides a transformer fault diagnosis method, system and device and a medium, and belongs to the technical field of power equipment fault diagnosis, and the method comprises the following steps: obtaining sample data of dissolved gas in transformer oil; carrying out weight calculation on the features of the sample data by using a Relief-F algorithm, and sorting and screening out an optimal feature set for transformer fault diagnosis according to the weights of the features; inputting features in the optimal feature set into a deep neural network model, training the deep neural network model, and in the training process, optimizing parameters of the deep neural network model by using a Tianying algorithm to obtain a transformer fault diagnosis model; and obtaining a to-be-tested sample of the transformer, and inputting the to-be-tested sample of the transformer into the transformer fault diagnosis model to obtain a transformer fault diagnosis type. According to the invention, the accuracy of fault diagnosis can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power equipment fault diagnosis, and in particular relates to a transformer fault diagnosis method, system, equipment and medium. Background Art

[0002] As one of the core devices for transmitting and distributing electric energy in the power system, the stable operation of the transformer is directly related to the safety and reliability of the entire power system. Once a transformer fails, it will cause power transmission interruption and may also cause serious accidents such as fire and explosion, causing huge losses to personnel and property. Therefore, timely and accurate diagnosis of transformer faults is an important means to ensure the safe operation of the power system.

[0003] At present, dissolved gas analysis (DGA) technology has become one of the most effective methods for oil-immersed transformer fault diagnosis. Transformer fault diagnosis methods based on DGA include traditional methods and artificial intelligence-based methods. Among them, traditional methods include the main gas method, IEC ratio method, Rogers method and David triangle method, etc. These methods have the advantages of simple principle, easy implementation and small calculation amount, and play an important role in transformer fault diagnosis. However, the judgment criteria of these fault diagnosis methods are based on field experience, which may result in large diagnostic errors. The artificial intelligence-based methods mainly include neural networks, random forests, support vector machines, etc., which have the advantages of continuous learning and updating, and the calculation accuracy is improved compared with traditional methods. However, most of the existing artificial intelligence-based methods use the characteristic ratio of the traditional method as the input feature of the diagnostic model to train the model. However, the matching degree between the input feature and the diagnostic model is low, resulting in low model accuracy, which affects the fault diagnosis accuracy. Summary of the invention

[0004] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a transformer fault diagnosis method, comprising the following steps:

[0005] Obtain sample data of dissolved gases in transformer oil;

[0006] The Relief-F algorithm is used to calculate the weights of the features of the sample data, and the optimal feature set for transformer fault diagnosis is screened out according to the weight sorting of the features;

[0007] The features in the optimal feature set are input into the deep neural network model, and the deep neural network model is trained. During the training process, the parameters of the deep neural network model are optimized using the Sky Eagle algorithm to obtain a transformer fault diagnosis model;

[0008] A transformer sample to be tested is obtained, and the transformer sample to be tested is input into a transformer fault diagnosis model to obtain a transformer fault diagnosis type.

[0009] Preferably, the method of using the Relief-F algorithm to calculate the weights of the features of the sample data and sorting the features by weight to select the optimal feature set for transformer fault diagnosis comprises the following steps:

[0010] A sample R is randomly selected from the sample data, and k neighboring samples are searched for in samples with the same attribute as R to form a set H;

[0011] Search for k nearest neighbor samples in R with different attributes to form a set M;

[0012] Calculate the sample distances between R and H and M respectively, and obtain the weights W of all features based on the distances. A ;

[0013] All features are arranged from large to small according to their weights, and the top 12 features with a total weight of 0.8 are selected as the optimal feature set.

[0014] Preferably, the features in the optimal feature set are input into a deep neural network model, and the deep neural network model is trained. During the training process, the parameters of the deep neural network model are optimized using the Sky Eagle algorithm to obtain a transformer fault diagnosis model, specifically:

[0015] Dividing the optimal feature set into a training set and a validation set;

[0016] The features in the training set are input into the deep neural network model, and the deep neural network model is trained. The maximum number of iterations of the deep neural network model is set to 500, the batch size is 10, the learning rate optimization range is set to 0.001-0.1, and the learning rate is optimized using the Eagle algorithm. The optimized learning rate is 0.0138. The deep neural network model is trained using the optimized learning rate to obtain the transformer fault diagnosis model.

[0017] Preferably, before using the Relief-F algorithm to calculate the weights of the features of the sample data, the method further includes normalizing the sample data.

[0018] Preferably, the sample data of dissolved gas in the transformer oil includes content sequences of H2, CH4, C2H6, C2H4 and C2H2 and corresponding state types.

[0019] Preferably, the deep neural network model includes an input layer, 4 hidden layers and an output layer.

[0020] The present invention also includes a transformer fault diagnosis system, characterized in that it includes:

[0021] A data acquisition module, used to acquire sample data of dissolved gas in transformer oil;

[0022] A feature selection module, used to calculate the weights of the features of the sample data using the Relief-F algorithm, and to select the optimal feature set for transformer fault diagnosis according to the weight sorting of the features;

[0023] A model building module is used to input the features in the optimal feature set into the deep neural network model, train the deep neural network model, and during the training process, use the Eagle algorithm to optimize the parameters of the deep neural network model to obtain a transformer fault diagnosis model;

[0024] The fault diagnosis module is used to obtain a transformer sample to be tested, input the transformer sample to be tested into a transformer fault diagnosis model, and obtain a transformer fault diagnosis type.

[0025] The present invention also includes a computer device, including a memory and a processor; the memory stores a computer program, and the processor is used to run the computer program in the memory to execute the transformer fault diagnosis method.

[0026] The present invention also includes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor to execute the transformer fault diagnosis method.

[0027] The transformer fault diagnosis method, system, device and medium provided by the present invention have the following beneficial effects:

[0028] The present invention uses the Relief-F feature selection algorithm to effectively screen the feature space to be selected, thereby eliminating features that do not match the diagnosis model, thereby obtaining an optimal feature set that is more suitable for the diagnosis model; by inputting the optimal feature set into a deep neural network model and using the Sky Eagle optimization algorithm to determine the optimal parameters in the deep neural network model, a transformer fault diagnosis model with better diagnostic performance can be obtained; by inputting the transformer to-be-tested sample into the obtained transformer fault diagnosis model, transformer faults are diagnosed, thereby improving the accuracy of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiment of the present invention and its design scheme, the following briefly introduces the drawings required for this embodiment. The drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0030] Figure 1 It is a flowchart of the present invention;

[0031] Figure 2 is the deep neural network model structure of the present invention;

[0032] Figure 3 is the characteristic contribution rate of the embodiment of the present invention;

[0033] Figure 4 Comparison of diagnosis results of various state types under the Relief-F feature selection space of an embodiment of the present invention;

[0034] Figure 5 This figure compares the diagnosis results of the deep neural network and the traditional probabilistic neural network in different feature spaces according to the embodiment of the present invention. DETAILED DESCRIPTION

[0035] In order to enable those skilled in the art to better understand the technical solution of the present invention and implement it, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the scope of protection of the present invention.

[0036] Example

[0037] The present invention provides a transformer fault diagnosis method, specifically as follows Figure 1 As shown, the following steps are included:

[0038] Step 1: Obtain sample data of dissolved gas in transformer oil.

[0039] In this embodiment, there are 326 groups of sample data of dissolved gas in transformer oil, including 5 characteristic gas content sequences of H2, CH4, C2H6, C2H4, and C2H2 and corresponding state type labels, totaling six state types. The specific distribution of the collected DGA data is shown in Table 1. Since the values ​​of dissolved gas in oil in different states are quite different, they need to be normalized.

[0040] Based on H2, CH4, C2H2, C2H4 and C2H6 gases, the characteristics of dissolved gases in 21-dimensional selected oils are extracted, as shown in Table 2.

[0041] Table 1 DGA sample data distribution of the embodiment of the present invention

[0042] Status Type Number of samples Normal (NS) 39 Medium and low temperature overheating (LT&MT) 65 High temperature overheating (HT) 72 Low Energy Discharge (LD) 50 High Energy Discharge (HD) 40 Partial Discharge (PD) 60 total 326

[0043] Table 2 21-dimensional features to be selected constructed in the embodiment of the present invention

[0044]

[0045] Step 2: Use the Relief-F algorithm to calculate the weights of the features of the sample data, and select the optimal feature set for transformer fault diagnosis based on the weight sorting of the features.

[0046] The Relief-F algorithm is used to calculate the weights of the 21-dimensional features in the original feature space, and the features are sorted according to the proportion of the weights. Figure 3 As shown in Table 3, the horizontal axis is the weight and the vertical axis is the feature code. The top 12 features with a total weight of 0.8 are selected as the optimal feature set for transformer fault diagnosis (as shown in Table 3), which specifically includes the following steps:

[0047] (1) Randomly select a sample R from the target sample set, and simultaneously search for k neighboring samples with the same attributes as R to form a set H;

[0048] (2) Search for k nearest neighbor samples in R with different attributes to form a set M;

[0049] (3) Calculate the distance between R and samples in H and M respectively, and obtain the weight W of all features based on the distance A , the calculation formula is as follows:

[0050]

[0051] Where W 0(A) is the initial weight value of feature A; n is the number of random sampling times of the sample; Diff(A,R i ,H j ) is R i and H j The distance relative to feature A; p(c) is the proportion of the total number of samples of category c in the original data set; Diff(A,R i ,M j (c)) is the sample R i and M j (c) Distance relative to feature A; class (R i ) is R i Category; M j (c) Category The jth nearest neighbor sample in , k is the number of nearest neighbor samples.

[0052] (4) Arrange all features from large to small according to their weights to obtain the optimal feature subset.

[0053] Table 3 Feature numbers and features after feature selection

[0054] serial number feature serial number feature 4 <![CDATA[C2H4]]> 9 <![CDATA[C2H2 / C2H4]]> 1 <![CDATA[H2]]> 18 <![CDATA[(CH4+C2H4) / THC]]> 17 <![CDATA[H2 / THC]]> 16 <![CDATA[C2H2 / THC]]> 2 <![CDATA[CH4]]> 19 <![CDATA[H2 / THC]]> 5 <![CDATA[C2H2]]> 10 <![CDATA[C2H4 / C2H6]]> 21 <![CDATA[H2 / ALL]]> 3 <![CDATA[C2H6]]>

[0055] Step 3: Input the features in the optimal feature set into the deep neural network model and train the deep neural network model. During the training process, use the Sky Eagle algorithm to optimize the parameters of the deep neural network model to obtain the transformer fault diagnosis model.

[0056] (1) Deep Neural Networks

[0057] As one of the representative algorithms of deep learning, the deep neural network (DNN) model has strong generalization ability and has been widely used in the field of fault diagnosis. Therefore, in this embodiment, the DNN model is adopted, and its structure is as follows: Figure 2 As shown in the figure, it consists of an input layer, 4 hidden layers and an output layer. Compared with other neural network models, the biggest feature is its multi-hidden layer structure. The DNN model mainly consists of two parts:

[0058] 1) The forward propagation process from input to output is calculated as follows:

[0059]

[0060] In the formula, is the output result of the jth neuron in the lth layer, f is the nonlinear activation function, is the connection weight of neurons between the front and back layers, is the output result of the i-th neuron in the l-1th layer, is the bias of the jth neuron in the lth layer.

[0061] 2) The reverse adjustment process of the output and expected error from output to input is calculated as follows:

[0062]

[0063] In the formula, G is the error value, y i is the output category code, Y i l To reverse the output value, is the partial derivative of the error with respect to the weight, η is the algorithm learning rate, this parameter has a great influence on the performance of the model, and the optimization strategy of the Eagle algorithm is used to optimize the learning rate in the DNN model.

[0064] (2) Eagle Algorithm

[0065] The Aquila Optimizer (AO) algorithm is a meta-heuristic algorithm based on the predation behavior of the eagle. It mainly relies on four hunting methods to prey on prey, namely vertical dive, short gliding attack, low-altitude slow descent and walking attack. The specific steps of the AO algorithm are:

[0066] 1) Vertical dive. When the eagle determines the approximate area where the prey is located, it will hover at high altitude to select the best hunting area, that is, to determine the space where the optimal solution is located. The mathematical calculation formula can be expressed as:

[0067]

[0068] Where X1(t+1) is the search result obtained by using the vertical dive hunting method X1 in the t+1 generation, X best (t) is the current optimal position of the population, t and T are the current and maximum iteration times, X M (t) is the current average position of the population, rand is a random value in the range of 0 to 1, N is the size of the search population, and D is the number of dimensions of the problem.

[0069] 2) Contour flight under short glide attack. When the eagle finds the target prey while hovering at high altitude, it will continue to hover above the prey, preparing to land and prey, that is, continuously narrowing the space where the optimal solution is located. The mathematical expression of this behavior is:

[0070] X2(t+1)=X best (t)×Levy(D)+X R (t)+(yx)×rand;

[0071] Where X2(t+1) is the search result obtained by the short gliding attack hunting method X2 in the t+1th generation, X R (t) is a random solution between 1 and N generations, D is the dimension of the problem, Levy is the flight distribution function, y and x are the spiral shapes in the search, and rand is a random value in the range of 0 to 1.

[0072] 3) Low-altitude slow-descent attack. When the eagle is ready to catch prey, it will fly slowly at a low altitude to approach the target prey to test the prey's reaction. The mathematical expression of this behavior is:

[0073] X3(t+1)=(X best (t)-X M (t))×α-rand+

[0074] ((UB-LB)×rand+LB)×δ;

[0075] Where X3(t+1) is the search result obtained by the hunting method X3 using low-altitude slow-descent attack at the t+1th generation, α and δ are adjustment parameters that are always 0.1, and UB and LB are the upper and lower bounds of the given problem, respectively.

[0076] 4) Walking attack. When the eagle approaches the prey, it will adopt the walking attack method to follow the random movement of the prey to capture the prey. The mathematical expression is:

[0077]

[0078] Where X4(t+1) is the search result obtained by using the walking attack hunting method X4 in the t+1th generation, QF is the quality function of the equilibrium search strategy, G1 is the various movements of the eagle during the pursuit of the escaping prey, G2 is the flight slope of the eagle during the flight, and X(t) is the current solution of the tth iteration.

[0079] (3) The Eagle algorithm optimizes the deep neural network model.

[0080] The DNN model and AO algorithm are combined, and the DNN learning rate is input into the AO algorithm. Through repeated iterations, the learning rate with the best diagnostic performance is calculated. The specific steps of AO to optimize the DNN learning rate are as follows:

[0081] 1) Initialize AO parameters.

[0082] 2) Assign the position of AO to the initial learning rate of DNN, update the position of the eagle according to the calculation formula of the four hunting methods, and train the DNN model.

[0083] 3) Calculate the corresponding fitness size (i.e. the value of the objective function) for each eagle and sort them.

[0084] 4) Continuously iterate the optimal Sky Eagle position until the termination condition is met (reaching the maximum number of iterations or fitness threshold).

[0085] 5) Output the optimal learning rate of DNN.

[0086] In the present invention, the 12 extracted features are used as input feature vectors to train the DNN model and construct a transformer fault diagnosis model. The maximum number of iterations of the DNN model is set to 500, the batch size is 10, the learning rate optimization range is set to 0.001-0.1, and the Eagle algorithm is used to optimize it. The optimized learning rate is 0.0138, and the optimized parameters are brought into the DNN for training to obtain a transformer fault diagnosis model.

[0087] Step 4: Obtain a transformer sample to be tested, input the transformer sample to be tested into a transformer fault diagnosis model, and obtain a transformer fault diagnosis type.

[0088] The samples to be tested are input into the trained transformer fault diagnosis model, and the state type of the samples to be tested is diagnosed to obtain the fault type of the samples to be tested. To verify the effectiveness of the transformer fault diagnosis model, the transformer fault diagnosis model is compared with the Convolutional Neural Networks (CNN) model and the Deep Belief Network (DBN) model under the Relief-F selection feature set. The training and diagnostic tests are repeated 30 times, and the diagnostic accuracy and comparison results of each state are obtained as shown in the figure. Figure 4 shown.

[0089] In order to further verify the effectiveness of the DNN model in the Relief-F feature selection space, the DNN model and the traditional probabilistic neural network (PNN) were used to diagnose the transformer in the original feature space and the feature space optimized by different methods. The training and diagnosis tests were repeated 30 times respectively, and the diagnosis results and comparisons of DNN and PNN in different feature spaces were obtained. Figure 5 shown.

[0090] The transformer fault diagnosis method based on Relief-F feature selection and optimized deep neural network of the present invention first extracts transformer features to be selected based on dissolved gas in oil; then, the Relief-F feature selection algorithm is used to effectively screen the feature space to be selected; finally, the Sky Eagle optimization algorithm is used to determine the optimal parameters in the DNN model, and the transformer fault diagnosis model is used to diagnose the transformer fault under the optimal parameters. Comparing various deep network methods in the Relief-F feature selection space, as well as the transformer fault diagnosis model and the traditional neural network model diagnosis results generated in the original feature space and the optimal space by different feature selection methods, the diagnosis results show that the DNN model has the highest accuracy in the Relief-F feature selection space, which provides a new idea for transformer fault diagnosis research.

[0091] The present invention also includes a transformer fault diagnosis system, including a data acquisition module, a feature selection module, a model construction module and a fault diagnosis module. The data acquisition module is used to obtain sample data of dissolved gas in transformer oil; the feature selection module is used to use the Relief-F algorithm to calculate the weight of the features of the sample data, and select the optimal feature set for transformer fault diagnosis according to the weight sorting of the features; the model construction module is used to input the features in the optimal feature set into a deep neural network model, train the deep neural network model, and during the training process, use the Sky Eagle algorithm to optimize the parameters of the deep neural network model to obtain a transformer fault diagnosis model;

[0092] The fault diagnosis module is used to obtain a transformer sample to be tested, input the transformer sample to be tested into a transformer fault diagnosis model, and obtain a transformer fault diagnosis type.

[0093] The present invention also includes a computer device, including a memory and a processor; the memory stores a computer program, and the processor is used to run the computer program in the memory to execute the transformer fault diagnosis method.

[0094] The present invention also includes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor to execute the transformer fault diagnosis method.

[0095] The embodiments described above are only preferred specific implementation modes of the present invention, and the protection scope of the present invention is not limited thereto. Any simple changes or equivalent replacements of the technical solutions that can be obviously obtained by any technician familiar with the field within the technical scope disclosed in the present invention belong to the protection scope of the present invention.

Claims

1. A transformer fault diagnosis method, characterized in that: The steps include: Obtain sample data of dissolved gases in transformer oil; The Relief-F algorithm is used to calculate the weights of the features of the sample data, and the optimal feature set for transformer fault diagnosis is screened out according to the weight sorting of the features; The features in the optimal feature set are input into the deep neural network model, and the deep neural network model is trained. During the training process, the parameters of the deep neural network model are optimized using the Sky Eagle algorithm to obtain a transformer fault diagnosis model; A transformer sample to be tested is obtained, and the transformer sample to be tested is input into a transformer fault diagnosis model to obtain a transformer fault diagnosis type.

2. The transformer fault diagnosis method according to claim 1, characterized in that: The method of using the Relief-F algorithm to calculate the weights of the features of the sample data and sorting the features by weight to select the optimal feature set for transformer fault diagnosis includes the following steps: A sample R is randomly selected from the sample data, and k neighboring samples are searched for in samples with the same attribute as R to form a set H; Search for k nearest neighbor samples in R with different attributes to form a set M; Calculate the sample distances between R and H and M respectively, and obtain the weights W of the features of all sample data based on the distances. A ; All features are arranged from large to small according to their weights, and the top 12 features with a total weight of 0.8 are selected as the optimal feature set.

3. The transformer fault diagnosis method according to claim 1, characterized in that: The features in the optimal feature set are input into the deep neural network model, and the deep neural network model is trained. During the training process, the parameters of the deep neural network model are optimized using the Sky Eagle algorithm to obtain a transformer fault diagnosis model, which is specifically: Dividing the optimal feature set into a training set and a validation set; The features in the training set are input into the deep neural network model, and the deep neural network model is trained. The maximum number of iterations of the deep neural network model is set to 500, the batch size is 10, the learning rate optimization range is set to 0.001-0.1, and the learning rate is optimized using the Eagle algorithm. The optimized learning rate is 0.0138. The deep neural network model is trained using the optimized learning rate to obtain the transformer fault diagnosis model.

4. The transformer fault diagnosis method according to claim 1, characterized in that: Before using the Relief-F algorithm to calculate the weights of the features of the sample data, the method further includes normalizing the sample data.

5. The transformer fault diagnosis method according to claim 1, characterized in that: The sample data of dissolved gas in the transformer oil includes content sequences of H2, CH4, C2H6, C2H4 and C2H2 and corresponding state types.

6. The transformer fault diagnosis method according to claim 1, characterized in that: The deep neural network model consists of an input layer, 4 hidden layers, and an output layer.

7. A transformer fault diagnosis system, characterized in that: include: A data acquisition module, used to acquire sample data of dissolved gas in transformer oil; A feature selection module, used to calculate the weights of the features of the sample data using the Relief-F algorithm, and to select the optimal feature set for transformer fault diagnosis according to the weight sorting of the features; A model building module is used to input the features in the optimal feature set into the deep neural network model, train the deep neural network model, and during the training process, use the Eagle algorithm to optimize the parameters of the deep neural network model to obtain a transformer fault diagnosis model; The fault diagnosis module is used to obtain a transformer sample to be tested, input the transformer sample to be tested into a transformer fault diagnosis model, and obtain a transformer fault diagnosis type.

8. A computer device, characterized in that: It comprises a memory and a processor; the memory stores a computer program, and the processor is used to run the computer program in the memory to execute the transformer fault diagnosis method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor to execute the transformer fault diagnosis method according to any one of claims 1 to 6.

Images