Power transformer fault diagnosis method and system for improving long short-term memory network

Abstract

The invention relates to the technical field of fault diagnosis, in particular to a power transformer fault diagnosis method and system for improving a long short-term memory network, and the method comprises the steps: collecting fault data of a power transformer, and carrying out the preprocessing of the fault data; constructing a long short-term memory network structure, and improving the long short-term memory network structure; adopting a non-standard Bayesian algorithm to optimize hyper-parameters in the long short-term memory network structure; constructing a power transformer fault diagnosis model, and training and optimizing the power transformer fault diagnosis model through the power transformer fault data set; and deploying the optimized power transformer fault diagnosis model into an actual power system. Through the power transformer fault diagnosis method, the problem that the application effect of the LSTM network in power transformer fault diagnosis is poor is effectively solved, a new activation function ALEU is provided to enhance the processing capability of the model on complex data, and a dynamic attenuation L1 + L2 regularization method is also provided to improve the generalization capability and diagnosis precision of the model.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a method and system for diagnosing faults of power transformers using an improved long short-term memory network. Background Art

[0002] In the power industry, the stable operation of power transformers is the key to ensuring the safety of the power grid. However, transformer failures occur frequently, and the requirements for fault diagnosis technology are increasing. Traditional fault diagnosis methods mostly rely on expert experience and signal processing technology, which are often difficult to accurately judge under complex and changeable fault modes. With the rapid development of artificial intelligence and deep learning technology, neural network-based fault diagnosis methods have gradually emerged. Among them, long short-term memory networks (LSTMs) have shown great potential in power transformer fault diagnosis because they can effectively process time series data.

[0003] When processing high-dimensional, nonlinear, and complex power data, traditional LSTM networks still face problems such as overfitting, long training time, and poor generalization ability. In particular, in the selection of activation functions, although traditional ReLU and its variants can effectively alleviate the gradient vanishing problem, they have limitations when processing negative inputs, affecting model performance. Therefore, in order to further improve the application effect of LSTM networks in power transformer fault diagnosis, the present invention proposes an improved LSTM network model for power transformer fault diagnosis. Summary of the invention

[0004] The present invention provides a power transformer fault diagnosis method and system with an improved long short-term memory network, which can effectively solve the problems in the background technology.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A method for diagnosing power transformer faults based on an improved long short-term memory network, the method comprising:

[0007] Collecting fault data of the power transformer and preprocessing the fault data to obtain a power transformer fault data set;

[0008] Constructing a long short-term memory network structure, and improving the long short-term memory network structure according to the temporal nature of the fault data;

[0009] A non-standard Bayesian algorithm is used to optimize the hyperparameters in the long short-term memory network structure;

[0010] Based on the improved long short-term memory network structure and the optimized hyperparameters, a power transformer fault diagnosis model is constructed, and the power transformer fault diagnosis model is trained and optimized through the power transformer fault data set;

[0011] The optimized power transformer fault diagnosis model is deployed in an actual power system to perform fault diagnosis on the power transformer.

[0012] Further, obtaining the power transformer fault data set includes:

[0013] Determine a data source, collect fault data of the power transformer according to the data source, and perform data cleaning on the fault data;

[0014] Extracting features from the fault data and normalizing the extracted features to obtain a power transformer fault data set;

[0015] The power transformer fault data set is divided into a training set, a validation set and a test set.

[0016] Furthermore, the constructing of the long short-term memory network structure includes: the long short-term memory network structure is composed of a plurality of long short-term memory units, each of the long short-term memory units is composed of a forget gate, an input gate, an output gate and a unit state.

[0017] Furthermore, according to the temporal nature of the fault data, the improvement of the long short-term memory network structure includes: introducing an improved activation function ALEU into the long short-term memory network structure, combining the negative input non-zero gradient characteristics of Leaky ReLU and the smooth negative value attenuation characteristics of ELU.

[0018] Furthermore, according to the temporal nature of the fault data, the long short-term memory network structure is improved, and further includes: introducing a leakage rate in the improved activation function ALEU, and the leakage rate can be dynamically adjusted, and the adjustment expression is as follows:

[0019]

[0020] Among them, a(x) is an adaptive function about x, and its expression is as follows:

[0021]

[0022] in, is a small positive number, a max is the upper limit of the leakage rate.

[0023] Further, according to the temporal nature of the fault data, the long short-term memory network structure is improved, further comprising:

[0024] Based on L1 and L2 regularization, the importance decay of weights is introduced as a new element to obtain dynamic decay L1+L2 regularization, and the dynamic decay L1+L2 regularization is added to the long short-term memory network structure to obtain a regularization term, which is expressed as:

[0025]

[0026] Among them, λ1 and λ2 are the hyperparameters of L1 and L2 regularization respectively, w is the model weight, t is the number of training steps, and d(t) is the regularization strength decay function, which is expressed as follows:

[0027]

[0028] Here, β is a hyperparameter and T is the total number of training steps.

[0029] Furthermore, according to the temporal nature of the fault data, the long short-term memory network structure is improved, and it also includes: according to the regularization term, generating a loss function of the long short-term memory network, the formula is as follows:

[0030]

[0031] Among them, DataLoss is the data loss item.

[0032] Furthermore, the use of a non-standard Bayesian algorithm to optimize the hyperparameters in the long short-term memory network structure includes: using a non-standard Bayesian algorithm TPE to optimize the hyperparameters.

[0033] A power transformer fault diagnosis system based on an improved long short-term memory network, the system comprising:

[0034] A fault data collection module collects fault data of the power transformer and pre-processes the fault data to obtain a fault data set of the power transformer;

[0035] A network structure optimization module is used to construct a long short-term memory network structure and improve the long short-term memory network structure according to the temporal nature of the fault data;

[0036] A hyperparameter optimization module, which uses a non-standard Bayesian algorithm to optimize the hyperparameters in the long short-term memory network structure;

[0037] A diagnosis model construction module, which constructs a power transformer fault diagnosis model based on the improved long short-term memory network structure and the optimized hyperparameters, and trains and optimizes the power transformer fault diagnosis model through the power transformer fault data set;

[0038] The diagnostic model deployment module deploys the optimized power transformer fault diagnosis model into the actual power system to perform fault diagnosis on the power transformer.

[0039] Furthermore, the fault data collection module includes:

[0040] A data collection unit, which determines a data source, collects fault data of the power transformer according to the data source, and performs data cleaning on the fault data;

[0041] A feature extraction unit extracts features from the fault data and normalizes the extracted features to obtain a power transformer fault data set;

[0042] A data division unit divides the power transformer fault data set into a training set, a verification set and a test set.

[0043] The technical solution of the present invention can achieve the following technical effects:

[0044] The problem of poor application effect of LSTM network in power transformer fault diagnosis is effectively solved. A new activation function ALEU is proposed for LSTM network, which combines the advantages of LeakyReLU and ELU, introduces adaptive leakage rate, and dynamically adjusts it according to the input or network status to enhance the model's ability to process complex data. A dynamic attenuated L1+L2 regularization method is also proposed, which can prevent overfitting in the early stage of training through dynamic adjustment of regularization strength, and allow the model to better fit the data in the later stage, thereby improving the generalization ability and diagnostic accuracy of the model.

[0045] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0047] Figure 1 A flow chart of a method for fault diagnosis of power transformers using improved long short-term memory networks;

[0048] Figure 2 Schematic diagram of the process to obtain the power transformer fault data set;

[0049] Figure 3 Schematic diagram of the structure of the power transformer fault diagnosis system with improved long short-term memory network. DETAILED DESCRIPTION

[0050] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0051] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0052] Embodiment 1:

[0053] like Figure 1 As shown, a power transformer fault diagnosis method based on an improved long short-term memory network includes:

[0054] S1: Collect the fault data of the power transformer and pre-process the fault data to obtain the power transformer fault data set;

[0055] Specifically, the fault data of power transformers can be collected by collecting sensor data, extracting fault records of power transformers, or obtaining fault data of related transformers from public power equipment databases to collect sufficient fault data and ensure the richness and diversity of training data. In order to obtain high-quality training data, the data can be preprocessed, including removing outliers, filling missing values, normalizing or standardizing data, and performing category balancing to eliminate noise, standardize data, and improve the learning effect of the model.

[0056] S2: Build a long short-term memory network structure and improve it according to the temporal nature of fault data;

[0057] S3: A non-standard Bayesian algorithm is used to optimize the hyperparameters in the long short-term memory network structure;

[0058] In this embodiment, according to the characteristics of the power transformer fault data, a long short-term memory network suitable for time series modeling is constructed. A single-layer or multi-layer LSTM can be used to build a model, with the input being a time series data sequence, and the output being a fault category or prediction result, and its structure is optimized in combination with the characteristics of the problem, such as adding a regularization mechanism, optimizing a gating mechanism, using multi-layer stacking, etc. In this step, in order to further improve the performance of the model, non-standard Bayesian algorithms such as TPE and BOHB can be used to optimize the hyperparameters in the long short-term memory network structure, including the number of LSTM layers, the number of hidden units per layer, the learning rate, and the optimizer.

[0059] S4: Based on the improved LSTM network structure and optimized hyperparameters, a power transformer fault diagnosis model is constructed, and the power transformer fault diagnosis model is trained and optimized using the power transformer fault dataset;

[0060] S5: Deploy the optimized power transformer fault diagnosis model to the actual power system to perform fault diagnosis on the power transformer.

[0061] On the basis of the above embodiments, a power transformer fault diagnosis model is constructed according to the improved long short-term memory network structure and the optimized hyperparameters. The improved network structure is optimized for the standard LSTM structure according to the timing and characteristics of the power transformer fault data to ensure that the network is more suitable for the power transformer fault diagnosis task. The hyperparameters play a decisive role in the performance of the model. The hyperparameters can be optimized by a non-standard Bayesian algorithm to ensure that the network structure can be efficiently trained and achieve optimal performance.

[0062] The present invention effectively solves the problem of poor application effect of LSTM network in power transformer fault diagnosis. For LSTM network, a new activation function ALEU is proposed, which combines the advantages of Leaky ReLU and ELU, introduces an adaptive leakage rate, and dynamically adjusts it according to the input or network state to enhance the model's ability to process complex data. A dynamic attenuation L1+L2 regularization method is also proposed, which can prevent overfitting in the early stage of training through dynamic adjustment of the regularization strength, and allow the model to better fit the data in the later stage, thereby improving the generalization ability and diagnostic accuracy of the model.

[0063] In order to obtain high-quality fault data, such as Figure 2 As shown, the power transformer fault data set includes:

[0064] S11: determining a data source, collecting fault data of the power transformer according to the data source, and performing data cleaning on the fault data;

[0065] S12: extracting features from the fault data and normalizing the extracted features to obtain a power transformer fault data set;

[0066] Specifically, in order to comprehensively cover different fault characteristics, the data sources of power transformer fault data can be determined first, such as sensor data, historical records, maintenance logs, etc., covering multiple dimensions of transformer operation to capture various fault characteristics. After collecting data, data quality can be improved through data preprocessing, data cleaning, and removal of duplicate, missing or outlier data points. In order to enhance the model's ability to recognize power transformer fault modes, features that are indicative of the transformer status, such as temperature, current, voltage, vibration frequency, etc., can be extracted from the collected raw data, and the extracted features can be normalized, and the feature values ​​can be scaled to the same scale to improve the efficiency and accuracy of model training.

[0067] S13: Divide the power transformer fault dataset into training set, validation set and test set.

[0068] In order to reasonably allocate data, ensure that the training, verification and testing of the model can fully reflect the data characteristics, and effectively evaluate the performance and generalization ability of the model, the power transformer fault data set can be divided into training set, verification set and test set through the principles of random division and category balance, and ensure that the data between the training set, verification set and test set are completely independent to avoid unreliable results of model evaluation.

[0069] Furthermore, constructing a long short-term memory network structure includes: the long short-term memory network structure is composed of a plurality of long short-term memory units, and each long short-term memory unit is composed of a forget gate, an input gate, an output gate and a unit state.

[0070] Specifically, LSTM is a special recurrent neural network architecture used to handle long-term dependency problems in sequence data. Compared with traditional RNN, LSTM avoids the problem of gradient vanishing or gradient exploding by introducing a "gate" control mechanism, so that it can learn long-term dependencies more effectively. The LSTM network consists of multiple LSTM units, each of which contains three "gate" structures: forget gate, input gate, output gate, and a unit state.

[0071] As a preferred embodiment of this embodiment, according to the temporal nature of fault data, the improvement of the long short-term memory network structure includes: introducing an improved activation function ALEU into the long short-term memory network structure, combining the negative input non-zero gradient characteristics of Leaky ReLU and the smooth negative value attenuation characteristics of ELU.

[0072] In this embodiment, the negative input non-zero gradient feature means that when the input value is negative, the activation function will still output a small negative value instead of completely outputting zero, and its gradient is also non-zero; the negative input non-zero gradient feature can ensure that the gradient in the negative value area can continue to propagate backwards, thereby preventing neurons from being "deactivated" in the negative value area. The smooth negative value attenuation feature means that when the input value is negative, the activation function outputs a negative value that gradually decreases with the input, and the output change is smooth, continuous and differentiable, without mutation points; the smooth negative value attenuation feature can make the activation function perform more smoothly in the negative value area, thereby avoiding the problem of discontinuous output or gradient mutation. ALEU can further improve the performance of the activation function in the negative value area by introducing smooth non-zero gradients and dynamically adjusted attenuation characteristics, which helps to improve the learning ability of LSTM on complex time series data.

[0073] On the basis of the above embodiment, according to the temporal nature of fault data, the long short-term memory network structure is improved, and further includes: introducing a leakage rate in the improved activation function ALEU, and the leakage rate can be dynamically adjusted, and the adjustment expression is as follows:

[0074]

[0075] Among them, a(x) is an adaptive function about x, which is used to dynamically adjust the leakage rate during negative input. Here, a simple adaptive function based on absolute value is defined, and its expression is as follows:

[0076]

[0077] in, is a small positive number used to prevent the denominator from being zero; a max is the upper limit of the leakage rate, which can prevent a(x) from becoming too large when x approaches zero. When the absolute value of x is large, a(x) will approach zero, but due to the existence of , it will never be zero, thus avoiding the "death" problem of ReLU; and when x approaches zero, a(x) will increase, but will not exceed a max , which helps maintain a certain gradient when the input is close to zero.

[0078] This scheme also introduces an adaptive leakage rate, which can be dynamically adjusted according to the input or other parameters of the network, so that the activation function can adaptively adjust its behavior to better adapt to different input data or task requirements, thereby enhancing the LSTM network's modeling ability and learning efficiency for complex data.

[0079] As a preferred embodiment of this invention, the long short-term memory network structure is improved according to the temporal nature of the fault data, and further includes:

[0080] Based on L1 and L2 regularization, the importance decay of weights is introduced as a new element to obtain dynamic decay L1+L2 regularization, and the dynamic decay L1+L2 regularization is added to the long short-term memory network structure to obtain the regularization term, which is expressed as:

[0081]

[0082] Among them, λ1 and λ2 are the hyperparameters of L1 and L2 regularization, which are used to control the strength of regularization. is the regularization strength decay function, whose value gradually decreases with the increase of training steps; w is the model weight, t is the number of training steps, d(t) is the regularization strength decay function, whose value gradually decreases with the increase of training steps. A simple decay function is expressed as follows:

[0083]

[0084] Among them, β is a hyperparameter used to control the decay speed. When β = 0, there is no decay; when β is larger, the decay speed is faster; T is the total number of training steps.

[0085] Specifically, the regularization function plays a vital role in neural networks. It helps the model achieve better performance in practical applications by controlling model complexity, preventing overfitting, improving generalization ability and accelerating convergence. The present invention combines the ideas of L1 and L2 regularization and introduces a new element - weight importance decay, which can gradually reduce the regularization strength as training progresses, to propose a new regularization function, called Dynamic Decay L1+L2 Regularization. By gradually reducing the regularization strength as the training process progresses, it can control the model complexity and prevent overfitting, and can also improve the model fitting ability in the later stage of training, thereby achieving better performance.

[0086] On the basis of the above embodiment, according to the temporal nature of the fault data, the long short-term memory network structure is improved, and further includes: generating a loss function of the long short-term memory network according to the regularization term, and the formula is as follows:

[0087]

[0088] Among them, DataLoss is the data loss term, such as mean square error, cross entropy, etc., which depends on the specific task.

[0089] Specifically, the present invention introduces dynamic attenuation of regularization strength, allowing the model to maintain strong regularization in the early stage of training to prevent overfitting, and gradually weakening regularization in the later stage of training to allow the model to better fit the data.

[0090] Furthermore, a non-standard Bayesian algorithm is used to optimize the hyperparameters in the long short-term memory network structure, including: using a non-standard Bayesian algorithm TPE to optimize the hyperparameters.

[0091] Specifically, hyperparameter optimization has significant benefits for improving the performance of machine learning models, reducing overfitting and underfitting, improving model stability, optimizing the training process, and enhancing the interpretability of models. The present invention uses a non-standard Bayesian algorithm TPE to optimize hyperparameters, divides the hyperparameter space into a good value distribution and a bad value distribution, and optimizes the hyperparameters through a probabilistic model to find a hyperparameter combination that minimizes the verification error.

[0092] Embodiment 2:

[0093] like Figure 3 As shown, the power transformer fault diagnosis system based on the improved long short-term memory network includes:

[0094] A fault data collection module collects fault data of the power transformer and pre-processes the fault data to obtain a fault data set of the power transformer;

[0095] The network structure optimization module builds a long-short-term memory network structure and improves the long-short-term memory network structure according to the temporal nature of fault data;

[0096] The hyperparameter optimization module uses a non-standard Bayesian algorithm to optimize the hyperparameters in the long short-term memory network structure;

[0097] The diagnostic model building module builds a power transformer fault diagnosis model based on the improved long short-term memory network structure and optimized hyperparameters, and trains and optimizes the power transformer fault diagnosis model through the power transformer fault data set;

[0098] The diagnostic model deployment module deploys the optimized power transformer fault diagnosis model into the actual power system to perform fault diagnosis on the power transformer.

[0099] The above-mentioned adjustment system in the present invention can effectively realize the power transformer fault diagnosis method of improving the long short-term memory network, and the technical effects that can be achieved are as described in the above-mentioned embodiments and will not be repeated here.

[0100] Furthermore, the fault data collection module includes:

[0101] A data collection unit determines a data source, collects fault data of the power transformer according to the data source, and performs data cleaning on the fault data;

[0102] A feature extraction unit extracts features from the fault data and normalizes the extracted features to obtain a power transformer fault data set;

[0103] The data partitioning unit divides the power transformer fault data set into a training set, a validation set and a test set.

[0104] Similarly, the above-mentioned optimization schemes for the system can also respectively achieve the optimization effects corresponding to the method in Example 1, which will not be repeated here.

[0105] Although the present application has been described in conjunction with specific features and embodiments thereof, it is apparent that various modifications and combinations may be made thereto without departing from the spirit and scope of the present application. Accordingly, this specification and the accompanying drawings are merely exemplary illustrations of the present application as defined herein, and are deemed to have covered any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application is intended to include these modifications and variations.

Claims

1. A method for diagnosing power transformer faults based on improved long short-term memory network, characterized in that: The method comprises: Collecting fault data of the power transformer and preprocessing the fault data to obtain a power transformer fault data set; Constructing a long short-term memory network structure, and improving the long short-term memory network structure according to the temporal nature of the fault data; A non-standard Bayesian algorithm is used to optimize the hyperparameters in the long short-term memory network structure; Based on the improved long short-term memory network structure and the optimized hyperparameters, a power transformer fault diagnosis model is constructed, and the power transformer fault diagnosis model is trained and optimized through the power transformer fault data set; The optimized power transformer fault diagnosis model is deployed in an actual power system to perform fault diagnosis on the power transformer.

2. The power transformer fault diagnosis method based on improved long short-term memory network according to claim 1 is characterized in that: The obtaining of the power transformer fault data set comprises: Determine a data source, collect fault data of the power transformer according to the data source, and perform data cleaning on the fault data; Extracting features from the fault data and normalizing the extracted features to obtain a power transformer fault data set; The power transformer fault data set is divided into a training set, a validation set and a test set.

3. The power transformer fault diagnosis method based on improved long short-term memory network according to claim 1 is characterized in that: The construction of the long short-term memory network structure includes: the long short-term memory network structure is composed of a plurality of long short-term memory units, and each of the long short-term memory units is composed of a forget gate, an input gate, an output gate and a unit state.

4. The power transformer fault diagnosis method of the improved long short-term memory network according to claim 3 is characterized in that: According to the temporal nature of the fault data, the improvement of the long short-term memory network structure includes: introducing an improved activation function ALEU into the long short-term memory network structure, combining the negative input non-zero gradient characteristics of Leaky ReLU and the smooth negative value attenuation characteristics of ELU.

5. The power transformer fault diagnosis method based on improved long short-term memory network according to claim 4 is characterized in that: According to the temporal nature of the fault data, the long short-term memory network structure is improved, and further includes: introducing a leakage rate in the improved activation function ALEU, and the leakage rate can be dynamically adjusted, and the adjustment expression is as follows: Among them, a(x) is an adaptive function about x, and its expression is as follows: Among them, ζ is a small positive number, a max is the upper limit of the leakage rate.

6. The power transformer fault diagnosis method of improved long short-term memory network according to claim 3 is characterized in that: According to the temporal nature of the fault data, the long short-term memory network structure is improved, further comprising: Based on L1 and L2 regularization, the importance decay of weights is introduced as a new element to obtain dynamic decay L1+L2 regularization, and the dynamic decay L1+L2 regularization is added to the long short-term memory network structure to obtain a regularization term, which is expressed as: Among them, λ1 and λ2 are the hyperparameters of L1 and L2 regularization respectively, w is the model weight, t is the number of training steps, and d(t) is the regularization strength decay function, which is expressed as follows: Here, β is a hyperparameter and T is the total number of training steps.

7. The power transformer fault diagnosis method based on improved long short-term memory network according to claim 6 is characterized in that: According to the temporal nature of the fault data, the long short-term memory network structure is improved, and further includes: according to the regularization term, generating a loss function of the long short-term memory network, the formula is as follows: Among them, DataLoss is the data loss item.

8. The power transformer fault diagnosis method based on the improved long short-term memory network according to claim 1 is characterized in that: The use of a non-standard Bayesian algorithm to optimize the hyperparameters in the long short-term memory network structure includes: using a non-standard Bayesian algorithm TPE to optimize the hyperparameters.

9. Improved long short-term memory network power transformer fault diagnosis system, characterized in that: The system comprises: A fault data collection module collects fault data of the power transformer and pre-processes the fault data to obtain a fault data set of the power transformer; A network structure optimization module is used to construct a long short-term memory network structure and improve the long short-term memory network structure according to the temporal nature of the fault data; A hyperparameter optimization module, which uses a non-standard Bayesian algorithm to optimize the hyperparameters in the long short-term memory network structure; A diagnosis model construction module, which constructs a power transformer fault diagnosis model based on the improved long short-term memory network structure and the optimized hyperparameters, and trains and optimizes the power transformer fault diagnosis model through the power transformer fault data set; The diagnostic model deployment module deploys the optimized power transformer fault diagnosis model into the actual power system to perform fault diagnosis on the power transformer.

10. The power transformer fault diagnosis system of improved long short-term memory network according to claim 9 is characterized in that: The fault data collection module comprises: A data collection unit, which determines a data source, collects fault data of the power transformer according to the data source, and performs data cleaning on the fault data; A feature extraction unit extracts features from the fault data and normalizes the extracted features to obtain a power transformer fault data set; A data division unit divides the power transformer fault data set into a training set, a verification set and a test set.

Images