A Transformer DGA Fault Diagnosis Method Based on SMOTE-SSA-CNN

Abstract

The present invention discloses a transformer DGA fault diagnosis method based on SMOTE-SSA-CNN, which obtains the original transformer DGA monitoring data, then expands the minority class samples through the SMOTE algorithm, obtains the equalized transformer DGA monitoring data set, and divides the training set and the test set; uses the hyperparameters of the CNN model as the sparrow individuals of the SSA algorithm, optimizes the optimal hyperparameters of the CNN; and builds a CNN model based on the equalized transformer DGA monitoring data set to perform transformer fault diagnosis. The SMOTE algorithm of the present invention reduces the imbalance of the original data set; optimizes the structure and hyperparameters of the CNN classification model through the sparrow search algorithm to avoid the subjectivity of the hyperparameter setting of the CNN model; combines the expanded data set and the optimal hyperparameters to build a CNN model for transformer fault diagnosis, thereby improving the accuracy of fault diagnosis.

Description

Technical Field

[0001] The invention belongs to the technical field of transformer monitoring, and in particular relates to a transformer DGA fault diagnosis method based on SMOTE-SSA-CNN. Background Art

[0002] As the hub equipment of coal mine power grid, the health of transformer is crucial to the safe and stable operation of power system. However, the structure and operating environment of transformer are complex. In the long-term operation, there are problems such as surface dirt, moisture, insulation aging, etc., which are easy to cause partial discharge, overheating and other faults. Therefore, it is necessary to establish an efficient fault diagnosis model for coal mine transformers to timely and accurately grasp the operating status of transformers, and then provide a reference for the formulation of equipment operation and maintenance plan.

[0003] Existing technologies mainly use traditional neural networks, extreme learning machines (ELMs) and other machine learning methods to perform fault diagnosis on the dissolved gas analysis (DGA) data of transformers. Although these methods have achieved certain results, the actual transformer failure is a low-probability event, resulting in a serious imbalance in the ratio of positive and negative samples in the original data set. When processing unbalanced data sets, the accuracy of fault identification results is low, and the setting of hyperparameters depends on manual experience, which affects the reliability and stability of the model.

[0004] A sufficient number of samples with balanced distribution of positive and negative samples is the basic prerequisite for ensuring the accuracy and performance of deep learning classification. However, the actual transformer failure is a low-probability event, coupled with perfect protection measures, resulting in a lack of fault samples and an imbalance of positive and negative samples in the DGA monitoring data set. When the classification model based on the deep learning algorithm is used to analyze the data set with imbalanced positive and negative samples, the fault identification results tend to be majority class, while the recognition accuracy of the minority class is low. The SMOTE algorithm is an improved method of random oversampling. This method is based on the minority class samples in the unbalanced data set, and obtains new samples by linear random interpolation to achieve balanced processing of the original data set, thereby improving the accuracy and generalization ability of the deep learning classification model. However, when the SMOTE algorithm expands the data, it is easily affected by discrete points and noise, resulting in the expansion may not conform to the actual situation. Summary of the invention

[0005] The present invention is based on the data of dissolved gas in transformer oil, and proposes a transformer DGA fault diagnosis method based on SMOTE-SSA-CNN. The structure and hyperparameters of the CNN model are optimized by the sparrow search algorithm SSA to avoid the subjectivity of the hyperparameter setting of the CNN model. The minority class samples in the original DGA data set are expanded by the SMOTE algorithm using synthetic minority class samples, and multiple expansion values ​​are expanded each time. The predicted value of dissolved gas in oil under working conditions is obtained by predicting other monitoring data, and the expanded value with the smallest Euclidean distance to the predicted value of dissolved gas in oil is selected as the final expanded value, so as to reduce the influence of noise points on the SMOTE algorithm and reduce the imbalance of the original DGA data set; so as to better train the transformer fault diagnosis model.

[0006] The present invention is implemented by the following technical scheme: A transformer DGA fault diagnosis method based on SMOTE-SSA-CNN, comprising the following steps:

[0007] Step 1: Obtain the original transformer DGA monitoring data, and then expand the minority class samples through the SMOTE algorithm;

[0008] Step 2: Based on step 1, the transformer DGA monitoring data set after equalization is obtained and divided into a training set and a test set;

[0009] Step 3: Use the hyperparameters of the CNN model as the sparrow individuals of the SSA algorithm, and the classification accuracy as the fitness function of the sparrow individuals. Calculate the fitness value and update the population until the maximum number of iterations is met to obtain the optimal hyperparameters of the CNN model.

[0010] Step 4: Based on the balanced transformer DGA monitoring data set and the optimal hyperparameters obtained by the SSA algorithm, a CNN model is built for transformer fault diagnosis.

[0011] Further preferably, in step one, for each minority class sample, multiple initial expansion values ​​are expanded each time; based on the historical transformer DGA monitoring data and other transformer monitoring data, on a time scale, the historical transformer DGA monitoring data is used as output and other transformer monitoring data is used as input to train a transformer DGA monitoring data prediction model; based on the trained transformer DGA monitoring data prediction model, the dissolved gas in oil is predicted according to the real-time other transformer monitoring data to obtain a predicted value of dissolved gas in oil; the Euclidean distance between each initial expansion value and the predicted value of dissolved gas in oil is calculated, and the initial expansion value with the smallest Euclidean distance or less than the set Euclidean distance threshold is selected as the final expansion value to be saved, and the other initial expansion values ​​are deleted.

[0012] Further preferably, the specific process of step one is as follows:

[0013] 1) Initialize the minority sample set S in the original transformer DGA monitoring data min，,n ={x1,x2,...,x n}; x1, x2, ..., x n They are the 1st, 2nd,…,nth minority class samples respectively;

[0014] 2) Use K nearest neighbor algorithm to get minority class samples x n The K nearest neighbor samples Y={y1,y2,...,y K},y1,y2,...,y K are the 1st, 2nd,…,Kth samples respectively;

[0015] 3) Randomly select m samples y from K nearest neighbor samples Y i (i=1,2,...,m,m<K);

[0016] 4) Use random linear interpolation to calculate the minority class samples x n and i After processing, new samples with the characteristics of adjacent samples are obtained to achieve the equalization of the original data set. The sample synthesis formula based on random linear interpolation is:

[0017] z i =x n +(x n -y i )·rand(0,1)

[0018] In the formula, z i Represents a newly generated minority class sample; rand(0,1) represents a random number in the range of (0,1).

[0019] Further preferably, the specific process of step three is as follows:

[0020] 1) Initialize the sparrow population and calculate the individual fitness; use the convolution kernel size and number, the number of neurons in the fully connected layer, the learning rate, etc. of the CNN classification model as the sparrow individuals, and the accuracy of the CNN model output results as the fitness function;

[0021] 2) Update the sparrow population according to the rules of the SSA algorithm;

[0022] 3) Calculate the fitness value of each sparrow individual according to the fitness function and sort the population;

[0023] 4) Repeat steps 2) and 3) until the maximum number of iterations is reached;

[0024] 5) Output the optimal CNN hyperparameters.

[0025] Further preferably, the CNN model includes an input layer, two convolutional layers, two pooling layers, a fully connected layer and an output layer, and a one-dimensional CNN is selected for processing; the overall mathematical model expression of the one-dimensional CNN is:

[0026] x output =f Softmax (f fc (f pooling (f conv (x input ))))

[0027] In the formula, x input represents the input feature set; x output Represents the output classification result; f conv (·) indicates convolutional layer calculation; f pooling (·) indicates the pooling layer calculation; f fc (·) indicates the calculation of the fully connected layer; f Softmax (·) represents the output result of the CNN model.

[0028] The present invention provides a terminal, which includes a memory, a processor and a computer program stored in the memory. When the computer program is executed by the processor, each step of the above-mentioned transformer DGA fault diagnosis method based on SMOTE-SSA-CNN is implemented.

[0029] The present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the various steps of the above-mentioned transformer DGA fault diagnosis method based on SMOTE-SSA-CNN are implemented.

[0030] The present invention has the following advantages:

[0031] 1) Aiming at the serious imbalance problem of positive and negative samples in the original DGA data set of the transformer, the SMOTE algorithm is used to generate minority class sample data, which solves the imbalance problem of the positive and negative sample ratios in the original data set and makes the transformer DGA fault diagnosis results more accurate.

[0032] 2) In order to solve the problem that it is difficult to construct the optimal network by selecting CNN hyperparameters based on manual experience, the SSA algorithm is used to optimize the hyperparameters of the CNN model. The example results show that the improved method has better convergence, stronger global search capability and higher fault diagnosis accuracy.

[0033] 3) Compared with the traditional fault diagnosis model, the transformer fault diagnosis model based on SMOTE-SSA-CNN proposed in this patent has stronger fault identification performance, can meet the needs of actual operation and maintenance work, and can provide a reference for operation and maintenance personnel to grasp the transformer operation status in a timely and accurate manner when there is a lack of transformer fault samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0035] Figure 1 This is a flow chart of the method of Example 1.

[0036] Figure 2 Augmenting rationale for minority class samples.

[0037] Figure 3 Schematic diagram of the CNN model.

[0038] Figure 4 This is a flow chart of the method of Example 2.

[0039] Figure 5 This is the fault diagnosis result of the SSA-CNN model for coal mine transformers.

[0040] Figure 6 It is a confusion matrix diagram. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and beneficial technical effect of the present invention clearer, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. It should be understood that the specific implementation methods described in this specification are only for explaining the present invention, not for limiting the present invention.

[0042] Example 1

[0043] like Figure 1 As shown, a transformer DGA fault diagnosis method based on SMOTE-SSA-CNN includes the following steps:

[0044] Step 1: Obtain the original transformer DGA monitoring data, and then expand the minority class samples through the SMOTE algorithm;

[0045] The process of minority class sample expansion is as follows Figure 2 As shown, the specific steps are as follows:

[0046] 1) Initialize the minority sample set S in the original transformer DGA monitoring data min，,n ={x1,x2,...,x n}; x1, x2, ..., x n They are the 1st, 2nd,…,nth minority class samples respectively;

[0047] 2) Use K nearest neighbor algorithm to get minority class samples x n The K nearest neighbor samples Y={y1,y2,...,y K},y1,y2,...,y K are the 1st, 2nd,…,Kth samples respectively;

[0048] 3) Randomly select m samples y from K nearest neighbor samples Y i (i=1,2,...,m,m<K);

[0049] 4) Use random linear interpolation to calculate the minority class samples x n and i After processing, new samples with the characteristics of adjacent samples are obtained to achieve the equalization of the original data set. The sample synthesis formula based on random linear interpolation is:

[0050] z i =x n +(x n -y i )·rand(0,1)

[0051] In the formula, z i Represents a newly generated minority class sample; rand(0,1) represents a random number in the range of (0,1).

[0052] Step 2: Based on step 1, the transformer DGA monitoring data set after equalization is obtained and divided into a training set and a test set;

[0053] Step 3: Use the hyperparameters of the CNN model as the sparrow individuals of the SSA algorithm, and the classification accuracy as the fitness function of the sparrow individuals. Calculate the fitness value and update the population until the maximum number of iterations is met to obtain the optimal hyperparameters of the CNN model.

[0054] Transformer DGA data has the characteristics of strong fuzziness and small differences between features. When using neural networks and other traditional classification models to identify transformer faults, it is easy to produce problems such as low fault identification accuracy and weak generalization ability. Therefore, the present invention uses a convolutional neural network classification model with strong feature characterization ability to replace the traditional classification model, fully explores the differences in DGA data under different fault types, establishes a mapping relationship between DGA data and fault types, and thus improves the accuracy and reliability of transformer DGA fault diagnosis.

[0055] Since the hyperparameters of the CNN model, such as the size and number of convolution kernels, the number of neurons in the fully connected layer, and the learning rate, are usually set based on manual experience, the output results of the model are less accurate. In view of the difficulty in constructing the optimal network by selecting CNN hyperparameters based on manual experience, the present invention optimizes the CNN model hyperparameters through SSA, thereby improving the accuracy and reliability of the model.

[0056] The CNN parameter optimization process based on the SSA algorithm is as follows Figure 1 As shown, the specific steps are as follows:

[0057] 1) Initialize the sparrow population and calculate the individual fitness; use the convolution kernel size and number, the number of neurons in the fully connected layer, the learning rate, etc. of the CNN classification model as the sparrow individuals, and the accuracy of the CNN model output results as the fitness function;

[0058] 2) Update the sparrow population according to the rules of the SSA algorithm;

[0059] 3) Calculate the fitness value of each sparrow individual according to the fitness function and sort the population;

[0060] 4) Repeat steps 2) and 3) until the maximum number of iterations is reached;

[0061] 5) Output the optimal CNN hyperparameters. When the following formula is satisfied, the attribute value corresponding to the maximum fitness function is the optimized hyperparameter of the CNN classification model.

[0062]

[0063] Among them, F represents the fitness value, n 诊断标签＝真实标签 Indicates the number of times the diagnostic label is the same as the true label;

[0064] Step 4: Based on the balanced transformer DGA monitoring data set and the optimal hyperparameters obtained by the SSA algorithm, a CNN model is built for transformer fault diagnosis.

[0065] 1) Establish a CNN model based on hyperparameters. Construct a convolutional neural network (CNN) model, such as Figure 3 As shown in Figure 1, the model includes an input layer, two convolutional layers, two pooling layers, a fully connected layer, and an output layer. According to the characteristics of transformer DGA data, one-dimensional CNN can be selected for processing. The overall mathematical model expression of one-dimensional CNN is:

[0066] x output =f Softmax (f fc (f pooling (f conv (x input))))

[0067] In the formula, x input represents the input feature set; x output Represents the output classification result; f conv (·) indicates convolutional layer calculation; f pooling (·) indicates the pooling layer calculation; f fc (·) indicates the calculation of the fully connected layer; f Softmax (·) represents the output result of the CNN model.

[0068] The transformer DGA monitoring data is divided into a training set and a test set, usually in a certain ratio.

[0069] 2) Training CNN model

[0070] The SSA-CNN model is trained using the training set data, and the model parameters are adjusted through the back-propagation algorithm to minimize the loss function.

[0071] 3) Verify the CNN model

[0072] Use the test set data to validate the trained model and evaluate the performance of the model.

[0073] 4) Transformer fault diagnosis

[0074] In practical applications, when a new transformer needs to be diagnosed, its DGA data is input into the trained CNN model. The model calculates the probability distribution of each fault type based on the input data, determines the fault type of the transformer based on the probability distribution, and selects the fault type with the highest probability as the diagnosis result.

[0075] Example 2

[0076] like Figure 2 As shown, a transformer DGA fault diagnosis method based on SMOTE-SSA-CNN includes the following steps:

[0077] Step 1, obtain the original transformer DGA monitoring data, and then expand the minority class samples through the SMOTE algorithm. For each minority class sample, multiple initial expansion values ​​are expanded each time; based on the historical transformer DGA monitoring data and other transformer monitoring data, on the time scale, the historical transformer DGA monitoring data is used as output, and other transformer monitoring data is used as input to train the transformer DGA monitoring data prediction model. Based on the trained transformer DGA monitoring data prediction model, the dissolved gas in oil is predicted according to the real-time monitoring data of other transformers to obtain the predicted value of dissolved gas in oil.

[0078] The transformer DGA monitoring data prediction model can use LSTM network, CTFNet neural network, or other neural networks for prediction. Other transformer monitoring data include oil temperature, oil pressure, micro-water, dielectric loss factor, capacitance, partial discharge, vibration, sound and service life. The LSTM network or CTFNet neural network is trained with historical data to explore the correlation between dissolved gas in oil and these multi-dimensional state data, thereby realizing the prediction of dissolved gas in oil.

[0079] The Euclidean distance between each calculated initial expansion value and the predicted value of dissolved gas in oil is selected, and the initial expansion value with the smallest Euclidean distance or less than the set Euclidean distance threshold is saved as the final expansion value, and other initial expansion values ​​are deleted;

[0080] Step 3: Finally, the equalized transformer DGA monitoring data set is obtained and divided into a training set and a test set;

[0081] Step 4: Use the hyperparameters of the CNN model as the sparrow individuals of the SSA algorithm, and the classification accuracy as the fitness function of the sparrow individuals. Calculate the fitness value and update the population until the maximum number of iterations is met to obtain the optimal hyperparameters of the CNN model.

[0082] Step 5: Based on the balanced transformer DGA monitoring data set and the optimal hyperparameters obtained by the SSA algorithm, a CNN model is built for transformer fault diagnosis.

[0083] Example 3

[0084] The present invention provides a terminal, which includes a memory, a processor, and a computer program stored in the memory. When the computer program is executed by the processor, each step of the transformer DGA fault diagnosis method based on SMOTE-SSA-CNN as described in any one of the above implementation modes is implemented.

[0085] Example 4

[0086] The present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the computer program implements the various steps of the transformer DGA fault diagnosis method based on SMOTE-SSA-CNN as described in any one of the above implementation modes.

[0087] Example verification

[0088] Taking the 10kV oil-immersed coal mine transformer of a certain regional power department as the research object, a total of 405 groups of historical DGA monitoring data were selected as the sample data set, and the original data set was divided into a test set and a training set according to the ratio of 8:2. The test set removed the data labels, and the training set retained the data labels. The specific sample distribution is shown in Table 1.

[0089] Table 1 Sample distribution

[0090]

[0091] Sample equalization

[0092] As shown in Table 2, the number of samples of the four types of faults in coal mine transformers accounts for less than 15% of the total number of samples, and the ratio of positive and negative samples in the original transformer data set is seriously unbalanced, which will affect the fault diagnosis accuracy of the subsequent SSA-CNN model. Therefore, the SMOTE algorithm is used to expand the fault sample data so that the number of samples for each operating state of coal mine transformers reaches 230 groups.

[0093] In order to further illustrate the feasibility of the proposed minority class sample expansion method for improving the accuracy of fault identification of coal mine transformers, the minority class samples of the original data set are synthesized by random sampling method and SMOTE algorithm respectively. The original data set and the data set after sample expansion are used as the input of the SSA-CNN model, and the model output results are compared. The results are shown in Table 2.

[0094] Table 2 Recognition results of different expansion algorithms

[0095]

[0096] As shown in Table 2, when the original sample data is used as the training set, the accuracy of the fault diagnosis model based on SSA-CNN is only 71.5%, and the fault diagnosis accuracy of coal mine transformers is low. After the minority class samples of the original data set are expanded by random sampling method and SMOTE algorithm, the accuracy of the fault diagnosis model of coal mine transformers is improved by 18.9% and 27.2% respectively. The results show that the minority class sample expansion method based on the SMOTE algorithm can effectively solve the problem of serious imbalance of positive and negative samples in the original data set of coal mine transformers, and significantly improve the accuracy of the fault diagnosis model. Introducing Euclidean distance to select the expansion of the SMOTE algorithm can further improve the accuracy.

[0097] Transformer fault diagnosis

[0098] The hyperparameters of the basic CNN model established in the example are as follows: the kernel sizes of the convolutional layers are 3×3 and 2×2, and the numbers are 32 and 64 respectively; a pooling layer and a batch normalization layer are set after each convolutional layer, the size of the pooling kernel is 2×2, and the step size is 1; the ReLu function is used as the activation function; two fully connected layers are used, and the number of neurons is 64 and 32 respectively; after activation by the Softmax function, the output is 5; the optimizer is Adam; and the learning rate is 0.001.

[0099] The parameter settings of the SSA-CNN model: the convolution layer, fully connected layer, optimizer, etc. are the same as the basic CNN model, but the hyperparameters such as the size and number of convolution kernels, the number of neurons in the fully connected layer, and the initial learning rate are optimized using SSA. Figure 5 , the confusion matrix is ​​as follows Figure 6 .

[0100] Combination Figure 5 and Figure 6 It can be seen that the fault diagnosis results of the SSA-CNN classification model for coal mine transformers are shown in Table 3.

[0101] Table 3 Evaluation index results of SSA-CNN model

[0102]

[0103] From the analysis of Table 3, we can see that the precision and recall rates of the fault diagnosis model based on SSA-CNN for the four operating states of the transformer are both greater than 95%, and the overall fault diagnosis accuracy rate reaches 98.7%. The results show that the overall performance of the fault classification model for coal mine transformers based on SSA-CNN is good, with high sensitivity and credibility.

[0104] Comparison of diagnostic results of different models

[0105] In order to verify the superiority of the SSA-CNN classification model, the balanced sample data set is used as input, and classification models such as BP neural network (BPNN), ELM, CNN, PSO-CNN, GA-CNN, and SSA-CNN are used to diagnose transformer faults. The results are shown in Table 4.

[0106] Table 4 Output results of different models

[0107]

[0108] As shown in Table 4, the accuracy of traditional classification models such as BP neural network and support vector machine is less than 82%, while the CNN model with multi-hidden layer structure can adaptively mine the deep features of multi-source monitoring data, and its fault identification accuracy reaches 91.3%. At the same time, for improved CNN models such as SSA-CNN, due to the optimization of hyperparameters, the transformer fault diagnosis accuracy has been improved to varying degrees compared with the basic CNN model.

[0109] A horizontal comparison of PSO-CNN, GA-CNN, and SSA-CNN shows that compared with the other two classification models, the SSA-CNN model has certain advantages in fault diagnosis accuracy and operation speed. From the perspective of fault diagnosis accuracy, the accuracy of the SSA-CNN model is as high as 98.7%, which is significantly higher than the other two models. Example 2 (Euclidean distance-SSA-CNN) introduces the prediction value of dissolved gas in oil on the basis of the SSA-CNN model to expand the SMOTE algorithm and select it, which can further improve the accuracy.

[0110] From the perspective of operation rate, the number of iterations required for the SSA-CNN model to converge to the highest accuracy is 21, which is lower than that of the PSO-CNN model and the GA-CNN model. The example results show that the proposed method can effectively identify partial discharge faults and overheating faults of transformers used in coal mines, and compared with traditional fault diagnosis methods, the proposed method has better convergence and higher accuracy.

[0111] The present invention is not limited to what is described in the specification and implementation modes, and therefore additional advantages and modifications can be easily realized by those skilled in the art. Therefore, without departing from the spirit and scope of the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details, representative devices, and illustrative examples shown and described herein.

Claims

1. A transformer DGA fault diagnosis method based on SMOTE-SSA-CNN, characterized in that: The following steps are involved: Step 1: Obtain the original transformer DGA monitoring data, and then expand the minority class samples through the SMOTE algorithm; Step 2: Based on step 1, the transformer DGA monitoring data set after equalization is obtained and divided into a training set and a test set; Step 3: Use the hyperparameters of the CNN model as the sparrow individuals of the SSA algorithm, and the classification accuracy as the fitness function of the sparrow individuals. Calculate the fitness value and update the population until the maximum number of iterations is met to obtain the optimal hyperparameters of the CNN model. Step 4: Based on the balanced transformer DGA monitoring data set and the optimal hyperparameters obtained by the SSA algorithm, a CNN model is built for transformer fault diagnosis.

2. The transformer DGA fault diagnosis method based on SMOTE-SSA-CNN according to claim 1 is characterized in that: In step one, for each minority class sample, multiple initial expansion values ​​are expanded each time; based on the historical transformer DGA monitoring data and other transformer monitoring data, on a time scale, the historical transformer DGA monitoring data is used as output and other transformer monitoring data is used as input to train the transformer DGA monitoring data prediction model; based on the trained transformer DGA monitoring data prediction model, the dissolved gas in oil is predicted according to the real-time other transformer monitoring data to obtain the predicted value of dissolved gas in oil; the Euclidean distance between each initial expansion value and the predicted value of dissolved gas in oil is calculated, and the initial expansion value with the smallest Euclidean distance or less than the set Euclidean distance threshold is selected as the final expansion value to be saved, and the other initial expansion values ​​are deleted.

3. The transformer DGA fault diagnosis method based on SMOTE-SSA-CNN according to claim 2 is characterized in that: The transformer DGA monitoring data prediction model is an LSTM network.

4. The transformer DGA fault diagnosis method based on SMOTE-SSA-CNN according to claim 2 is characterized in that: The transformer DGA monitoring data prediction model adopts the CTFNet neural network.

5. The transformer DGA fault diagnosis method based on SMOTE-SSA-CNN according to claim 4 is characterized in that: Other transformer monitoring data include oil temperature, oil pressure, trace water, dielectric loss factor, capacitance, partial discharge, vibration, sound and service life.

6. The transformer DGA fault diagnosis method based on SMOTE-SSA-CNN according to claim 1 is characterized in that: The specific process of step one is as follows: Initialize the minority class sample set S in the original transformer DGA monitoring data min，,n ={x1,x2,...,x n }; x1, x2, ..., x n They are the 1st, 2nd, ..., nth minority class samples respectively; Use K nearest neighbor algorithm to get minority class samples x n The K nearest neighbor samples Y = {y1, y2, ..., y K }, y1, y2, ..., y K are the 1st, 2nd, ..., Kth samples respectively; Randomly select m samples y from K nearest neighbor samples Y i (i=1, 2,..., m, m<K); Use random linear interpolation to calculate the minority class samples x n and i Processing is performed to obtain new samples with the characteristics of adjacent samples, thus achieving the equalization of the original data set; the sample synthesis formula based on random linear interpolation is: z i =x n +(x n -y i )·rand(0,1) In the formula, z i Represents a newly generated minority class sample; rand(0, 1) represents a random number in the range of (0, 1).

7. The transformer DGA fault diagnosis method based on SMOTE-SSA-CNN according to claim 1, characterized in that: The specific process of step three is as follows: 1) Initialize the sparrow population and calculate the individual fitness; use the convolution kernel size and number, the number of neurons in the fully connected layer, the learning rate, etc. of the CNN classification model as the sparrow individuals, and the accuracy of the CNN model output results as the fitness function; 2) Update the sparrow population according to the rules of the SSA algorithm; 3) Calculate the fitness value of each sparrow individual according to the fitness function and sort the population; 4) Repeat steps 2) and 3) until the maximum number of iterations is reached; 5) Output the optimal CNN hyperparameters.

8. According to the transformer DGA fault diagnosis method based on SMOTE-SSA-CNN in claim 7, the CNN model includes an input layer, two convolutional layers, two pooling layers, a fully connected layer and an output layer, and a one-dimensional CNN is selected for processing; the overall mathematical model expression of the one-dimensional CNN is: x output =f Softmax (f fc (f pooling (f conv (x input )))) In the formula, x input represents the input feature set; x output Represents the output classification result; f conv (·) indicates convolutional layer calculation; f pooling (·) indicates the pooling layer calculation; f fc (·) indicates the calculation of the fully connected layer; f Softmax (·) represents the output result of the CNN model.

9. A terminal comprising a memory, a processor, and a computer program stored in the memory, characterized in that: When the computer program is executed by the processor, each step of the transformer DGA fault diagnosis method based on SMOTE-SSA-CNN as described in any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, each step of the transformer DGA fault diagnosis method based on SMOTE-SSA-CNN as described in any one of claims 1 to 8 is implemented.

Images