Power transformer fault diagnosis method and system based on enhanced time convolution network
Through the power transformer fault diagnosis method based on enhanced time convolution network, the problems of strong subjectivity and low efficiency in the traditional methods are solved, and the accuracy and prediction capabilities of transformer fault recognition are improved, and the ability to achieve higher robustness and generalization are achieved.
Patent Information
- Application Number
- CN202510035194.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-09
- Publication Date
- 2025-04-15
AI Technical Summary
Traditional transformer fault diagnosis methods have problems such as strong subjectivity, low efficiency and high cost, and it is difficult to meet the needs of modern power grids for intelligent and automated operations and maintenance.
The power transformer fault diagnosis method based on the enhanced time convolution network is adopted. By collecting data related to the operating status of the power transformer, feature screening and data preprocessing, the optimized time convolution network is designed and transformed, the power transformer fault diagnosis model is built, and a new activation function ALSU and HER regularization method is introduced into the model.
It effectively improves the accuracy and robustness of transformer fault recognition, prevents neuronal inactivation, improves the training efficiency and stability of the network, and improves the generalization ability of the model.
Smart Images
Figure CN119829950A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of transformer fault detection, and in particular to a power transformer fault diagnosis method and system based on an enhanced temporal convolutional network. Background Art
[0002] In the power system, the transformer is one of the key equipment, and its operating status is directly related to the safety and stability of the power grid. However, in the long-term operation of the transformer, due to the influence of various factors such as environment, load changes, material aging, etc., various faults are prone to occur, such as insulation aging, winding deformation, oil deterioration, etc. If these faults are not discovered and handled in time, they will not only cause equipment damage, but may also cause large-scale power outages, causing huge economic losses and social impacts. Therefore, how to efficiently and accurately diagnose and predict the faults of power transformers has become an important issue that needs to be solved in the power industry.
[0003] Traditional transformer fault diagnosis methods mostly rely on manual inspections, regular tests, and expert experience judgment. These methods have the problems of strong subjectivity, low efficiency, and high cost, and are difficult to meet the needs of modern power grids for intelligent and automated operation and maintenance. With the rapid development of artificial intelligence technology, especially the successful application of deep learning in image recognition, natural language processing, and other fields, new ideas and technical means have been provided for the diagnosis and prediction of power transformer faults. As a deep learning model that specializes in processing time series data, temporal convolutional networks have attracted much attention because of their ability to capture long-term dependencies in data. However, directly applying traditional temporal convolutional networks to the field of power transformer fault diagnosis and prediction still faces many challenges, such as high model complexity, long training time, high risk of overfitting, and insufficient ability to recognize complex fault patterns. In view of this, a method for power transformer fault diagnosis and prediction based on enhanced temporal convolutional networks has injected new vitality into the development of the field of power transformer fault diagnosis and prediction. Summary of the invention
[0004] The present invention provides a power transformer fault diagnosis method and system based on an enhanced temporal convolutional 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 power transformer fault diagnosis method based on an enhanced temporal convolutional network, the method comprising:
[0007] Collect data related to the operating status of power transformers in the power system to obtain a transformer operating status data set;
[0008] Performing feature screening on the transformer operation status data set to obtain a transformer operation status feature data set;
[0009] Designing a temporal convolutional network, and transforming and optimizing the temporal convolutional network;
[0010] Based on the modified temporal convolutional network, a power transformer fault diagnosis model is built, and the power transformer fault diagnosis model is trained and optimized through the transformer operation status feature data set;
[0011] The optimized power transformer fault diagnosis model is deployed in the power system to perform fault diagnosis on the power transformer.
[0012] Furthermore, the transformer operation status data set is subjected to feature screening to obtain a transformer operation status feature data set including:
[0013] Preprocessing the data in the transformer operation status data set;
[0014] The preprocessed data is screened by feature selection algorithm to obtain key features;
[0015] The key features are transformed and encoded to obtain a transformer operation status feature data set.
[0016] Furthermore, the temporal convolutional network adopts a one-dimensional dilated convolutional network and a residual connection.
[0017] Furthermore, the transformation and optimization of the temporal convolutional network includes: introducing a new activation function ALSU into the temporal convolutional network, combining the immortality of Leaky ReLU and the smoothness of Sigmoid function.
[0018] Furthermore, the transformation and optimization of the temporal convolutional network also includes: introducing an adaptive mechanism in the temporal convolutional network to adjust the negative slope parameter of the Leaky ReLU, and the form of the adaptive mechanism is as follows:
[0019]
[0020] Among them, μ is a learnable threshold parameter; α is the scaling factor of the Sigmoid function; β is the negative slope of the LeakyReLU.
[0021] Furthermore, the transformation and optimization of the temporal convolutional network also includes: adding a HER regularization method to the temporal convolutional network and introducing a regularization strength dynamic adjustment mechanism.
[0022] Furthermore, the introduction of the regularization strength dynamic adjustment mechanism includes: adding a dynamic adjustment factor γ on the basis of Elastic Net to dynamically adjust the regularization strength, and the adjustment formula is as follows:
[0023]
[0024] Among them, L0 is the original loss function, λ is the regularization coefficient, and w i is the model weight, γ1 and γ2 are dynamic adjustment factors.
[0025] Furthermore, the construction of the power transformer fault diagnosis model includes:
[0026] Build a power transformer fault diagnosis model architecture, and design the number of layers and parameters of each layer of the power transformer fault diagnosis model;
[0027] Designing an output layer of the power transformer fault diagnosis model, and adding a fully connected layer before the output layer;
[0028] Integrating the HER regularization method into the power transformer fault diagnosis model;
[0029] Dividing the transformer operating state feature data set into a training set, a validation set and a test set, and using the training set and the validation set to train the power transformer fault diagnosis model;
[0030] The trained power transformer fault diagnosis model is evaluated using the test set, and the power transformer fault diagnosis model is optimized according to the evaluation result.
[0031] A power transformer fault diagnosis system based on an enhanced temporal convolutional network, the system comprising:
[0032] A data collection module collects data related to the operating status of power transformers in the power system and obtains a transformer operating status data set;
[0033] A feature screening module performs feature screening on the transformer operation status data set to obtain a transformer operation status feature data set;
[0034] A network design module designs a temporal convolutional network and performs transformation and optimization on the temporal convolutional network;
[0035] A model building module, based on the modified temporal convolutional network, builds a power transformer fault diagnosis model, and trains and optimizes the power transformer fault diagnosis model through the transformer operation status feature data set;
[0036] The model deployment module deploys the optimized power transformer fault diagnosis model into the power system to perform fault diagnosis on the power transformer.
[0037] Furthermore, the model building module includes:
[0038] A model architecture design unit, which builds a power transformer fault diagnosis model architecture and designs the number of layers and parameters of each layer of the power transformer fault diagnosis model;
[0039] An output layer design unit designs the output layer of the power transformer fault diagnosis model and adds a fully connected layer before the output layer;
[0040] A regularization integration unit, integrating the HER regularization method in the power transformer fault diagnosis model;
[0041] A diagnostic model training unit, which divides the transformer operating state feature data set into a training set, a validation set and a test set, and uses the training set and the validation set to train the power transformer fault diagnosis model;
[0042] A model evaluation and optimization unit is used to evaluate the trained power transformer fault diagnosis model using the test set, and optimize the power transformer fault diagnosis model according to the evaluation result.
[0043] The technical solution of the present invention can achieve the following technical effects:
[0044] The accuracy of transformer fault identification and the robustness of prediction ability are effectively improved. Based on the traditional time convolutional network, a new activation function ALSU is introduced. This activation function effectively prevents neuron deactivation while maintaining efficient probability output, thereby improving the training efficiency and stability of the network. In addition, the present invention proposes a HER regularization method, which can dynamically adjust the regularization strength, can more flexibly adapt to the needs of different training stages, and further improve the generalization ability 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 It is a flowchart of a power transformer fault diagnosis method based on enhanced temporal convolutional network;
[0048] Figure 2 Schematic diagram of the process for obtaining transformer operating status characteristic data set;
[0049] Figure 3 Schematic diagram of the structure of the power transformer fault diagnosis system based on enhanced temporal convolutional 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 enhanced temporal convolutional network includes:
[0054] S1: Collect data related to the operating status of power transformers in the power system to obtain a transformer operating status data set;
[0055] S2: Perform feature screening on the transformer operation status data set to obtain a transformer operation status feature data set;
[0056] In this embodiment, in order to improve the diversity and robustness of the model input, multi-source data related to the operating status of the transformer can be collected from the power system. These data may include real-time operating data of the transformer, such as current, voltage, temperature, vibration signals, etc., historical fault records, environmental parameters such as temperature, humidity, wind speed, etc., and regular maintenance records, etc.; after collecting the data, feature selection and standardization operations can be performed on the data to reduce data noise and improve the subsequent model training effect.
[0057] S3: Design a temporal convolutional network and optimize it;
[0058] S4: Based on the modified temporal convolutional network, a power transformer fault diagnosis model is built, and the power transformer fault diagnosis model is trained and optimized through the transformer operation status feature dataset;
[0059] S5: Deploy the optimized power transformer fault diagnosis model into the power system to perform fault diagnosis on the power transformer.
[0060] Specifically, according to the characteristics of the power transformer operating status data, this step can design a time convolution network suitable for capturing time dependencies. The time convolution network is a model specifically used to process time series data. Its core feature is the use of dilated causal convolution to capture long-term dependencies in time series data. In order to enhance its performance to better meet the needs of power transformer fault diagnosis tasks, the traditional time convolution network can be optimized to solve the shortcomings of the traditional time convolution network in capturing long-term dependencies and improve the training efficiency and prediction accuracy of the model. Based on the optimized time convolution network structure, the specific implementation of the fault diagnosis task is completed. The characteristic data of the transformer operating status can be used to train the model and perform hyperparameter adjustment or structural optimization to enable the model to have a strong fault diagnosis capability. Finally, the trained and optimized model is applied to the power system for real-time monitoring and diagnosis of the transformer operating status and prediction of fault risks.
[0061] Through the present invention, the accuracy of transformer fault identification and the robustness of prediction ability are effectively improved. On the basis of the traditional time convolutional network, a new activation function ALSU is introduced. While maintaining efficient probability output, this activation function effectively prevents neuron deactivation and improves the training efficiency and stability of the network. In addition, the present invention proposes a HER regularization method, which can dynamically adjust the regularization strength, can more flexibly adapt to the needs of different training stages, and further improve the generalization ability of the model.
[0062] In order to extract features useful for transformer fault diagnosis and prediction, such as Figure 2 As shown, the transformer operation status data set is feature screened to obtain the transformer operation status feature data set including:
[0063] S21: preprocessing the data in the transformer operation status data set;
[0064] On the basis of the above embodiments, in order to improve data quality, the collected data can be preprocessed and standardized to ensure that the collected data is standardized in terms of format, unit, timestamp, etc., to facilitate subsequent processing and analysis; data cleaning can be performed, and technical means such as data smoothing, interpolation, and denoising can be used to remove noise, outliers, missing values, etc. in the data to ensure data quality.
[0065] S22: feature selection algorithm is used to screen the preprocessed data to obtain key features;
[0066] S23: Perform feature transformation and encoding on key features to obtain a transformer operation status feature data set.
[0067] Specifically, features useful for transformer fault diagnosis and prediction are extracted from the original data. These features may include statistical features, such as mean, variance, kurtosis, skewness, etc., time-frequency domain features, such as features extracted by Fourier transform, wavelet transform, etc., and features based on expert experience, etc. In order to reduce the complexity of the model, improve training efficiency and prediction accuracy, feature selection algorithms such as principal component analysis PCA, mutual information, and feature importance scoring can be used to evaluate and screen the extracted features, remove redundant or irrelevant features, and retain the features that contribute most to model performance; based on the screened features, necessary feature transformation and encoding, such as normalization, standardization, one-hot encoding, etc., can be performed to adapt to the input requirements of the model, and the potential relationship between features may be further explored.
[0068] As a preferred embodiment of this embodiment, the temporal convolutional network adopts a one-dimensional dilated convolutional network and a residual connection.
[0069] Specifically, the enhanced temporal convolutional network model of this step can capture long-term dependencies in time series data by introducing dilated causal convolutions, and solve the gradient vanishing problem in deep networks by combining residual connections. By adding residual connections, the temporal convolutional network can train deep networks more efficiently, solve the gradient vanishing problem, and enhance the modeling ability of long-term dependencies. The temporal convolutional network is a deep learning model specifically used to process sequence data. It uses a one-dimensional fully convolutional network structure to capture the temporal dependencies in time series data, and the temporal convolutional network contains residual connections, which helps to solve the gradient vanishing problem in deep networks, allowing the network to be trained deeper, thereby capturing dependencies over longer time spans. The core of the temporal convolutional network of this embodiment is a one-dimensional dilated causal convolution, which combines the advantages of causal convolution and dilated convolution. It can not only maintain the temporal constraints of time series data, but also capture long-term dependencies in the sequence by expanding the receptive field. The basic formula is as follows:
[0070]
[0071] Where: f is the filter; d is the dilation coefficient; k is the convolution kernel size; s-di ensures that convolution operations can only be performed on past inputs.
[0072] In order to better meet the needs of power transformer fault diagnosis tasks, the temporal convolutional network is transformed and optimized, including: introducing a new activation function ALSU in the temporal convolutional network, combining the immortality of Leaky ReLU and the smoothness of Sigmoid function.
[0073] In a neural network, an activation function is used to introduce nonlinearity so that the network can learn and represent more complex patterns. In order to improve the training efficiency and stability of the model, this embodiment introduces a new activation function ALSU. The ALSU activation function is designed to combine the immortality of Leaky ReLU and the smoothness of the Sigmoid function, which not only avoids neuron death but also achieves accurate probability output. Immortality means avoiding neuron death, so that even in the negative input area, neurons can maintain non-zero gradients and continue to learn; smoothness ensures smooth changes in output and avoids extreme fluctuations in gradients during training, making it easier for the network to converge stably.
[0074] Furthermore, the transformation and optimization of the temporal convolutional network also includes: introducing an adaptive mechanism in the temporal convolutional network to adjust the negative slope parameter of the Leaky ReLU. The form of the adaptive mechanism is as follows:
[0075]
[0076] Among them, μ is a learnable threshold parameter used to decide when to switch from Leaky ReLU to the Sigmoid part; α is the scaling factor of the Sigmoid function, which is used to adjust the sensitivity of the Sigmoid function; β is the negative slope of the Leaky ReLU, which is usually a small positive number. It can be considered to be a learnable parameter or a function based on μ and x to adjust adaptively.
[0077] On the basis of the above embodiments, in order to make the network learn nonlinear features more flexibly, an adaptive mechanism can be introduced to adjust the negative slope parameter of Leaky ReLU. During the training process of the network, the negative slope is not fixed, but dynamically adjusted according to the characteristics of the training data and the training process. By adjusting the negative slope parameter of Leaky ReLU through the adaptive mechanism, the network can dynamically adjust the shape of the activation function according to the characteristics of the input data during the training process, thereby improving the training efficiency, stability and model performance.
[0078] In order to better adapt to the needs of different training stages, the transformation and optimization of the temporal convolutional network also includes: adding the HER regularization method to the temporal convolutional network and introducing a dynamic adjustment mechanism for the regularization strength.
[0079] Specifically, the HER regularization function can more flexibly adapt to the needs of different training stages. It can not only encourage model learning in the early stage, but also prevent overfitting in the later stage. In addition, HER can adjust the regularization strategy according to the actual performance of the model, which helps to improve the generalization ability of the model.
[0080] Furthermore, the introduction of a dynamic adjustment mechanism for regularization strength includes: adding a dynamic adjustment factor γ on the basis of Elastic Net to dynamically adjust the strength of regularization. The adjustment formula is as follows:
[0081]
[0082] Among them, L0 is the original loss function, λ is the regularization coefficient, and w i is the model weight, γ1 and γ2 are dynamic adjustment factors.
[0083] In this embodiment, the HER regularization function adds a dynamic adjustment factor based on Elastic Net, which dynamically adjusts the strength of L1 and L2 regularization according to certain indicators in the training process, such as the loss or accuracy on the validation set.
[0084] On the basis of the above embodiment, building a power transformer fault diagnosis model includes:
[0085] S41: Build the power transformer fault diagnosis model architecture, design the number of layers and parameters of each layer of the power transformer fault diagnosis model;
[0086] In this step, the appropriate number of enhanced temporal convolutional network layers and parameters of each layer can be designed according to the characteristics of power transformer data and the requirements of fault identification. These layers will contain customized ALSU activation functions, which are embedded in the neurons of each layer to improve the learning ability of nonlinear features, thereby improving the training efficiency and stability of the network.
[0087] S42: Design the output layer of the power transformer fault diagnosis model and add a fully connected layer before the output layer;
[0088] Specifically, according to the specific tasks of fault diagnosis and prediction, a suitable output layer can be designed. For fault diagnosis tasks, the output layer may be a multi-classification layer corresponding to different fault types; for fault prediction tasks, the output layer may be a regression layer to predict a key indicator of the transformer state in the future, such as remaining life, fault probability, etc. To further improve the generalization ability of the model, a fully connected layer can be added before the output layer for feature fusion and dimensionality reduction.
[0089] S43: Integrating HER regularization method in power transformer fault diagnosis model;
[0090] During the model training process, the HER regularization method can be integrated. After being integrated into the model training process, the HER regularization method is used to dynamically adjust the strength of L1 and L2 regularization, thereby optimizing model parameters and avoiding overfitting. In the power transformer fault diagnosis model, the HER regularization method is integrated through the loss function, combining the advantages of L1 and L2 regularization, and through the mechanism of dynamically adjusting the regularization strength, the model weight distribution is optimized during the training process, thereby improving the sparsity and generalization ability of the model.
[0091] S44: dividing the transformer operation status feature data set into a training set, a validation set and a test set, and using the training set and the validation set to train a power transformer fault diagnosis model;
[0092] Specifically, the processed data set is divided into a training set, a validation set, and a test set. The training set is used for model training, the validation set is used to adjust model parameters and prevent overfitting, and the test set is used to evaluate the final performance of the model. During the training process, appropriate optimization algorithms, such as Adam and RMSprop, can be used to minimize loss functions, such as cross entropy loss and mean square error, to adjust model parameters; in addition, the performance on the validation set can be used to adjust the HER regularization parameters to find the optimal regularization strength combination.
[0093] S45: Use the test set to evaluate the trained power transformer fault diagnosis model, and optimize the power transformer fault diagnosis model according to the evaluation results.
[0094] Specifically, after the model training is completed, the trained power transformer fault diagnosis model can be evaluated using the test set to verify its generalization ability on unknown data. Based on the evaluation results, including accuracy, recall, F1 score, etc., the model is optimized, including adjusting hyperparameters such as learning rate, regularization strength, or modifying the model structure such as the number of layers and convolution kernel size. The optimized model will further improve the fault diagnosis performance. Finally, the final model is deployed to the actual power system to achieve real-time diagnosis and prediction of power transformer faults. At the same time, a feedback mechanism can be established to collect actual operation data for continuous optimization and updating of the model.
[0095] Embodiment 2:
[0096] like Figure 3 As shown, a power transformer fault diagnosis system based on enhanced temporal convolutional network includes:
[0097] A data collection module collects data related to the operating status of power transformers in the power system and obtains a transformer operating status data set;
[0098] A feature screening module performs feature screening on a transformer operation status data set to obtain a transformer operation status feature data set;
[0099] Network design module, designs the temporal convolutional network and transforms and optimizes the temporal convolutional network;
[0100] Model building module, based on the modified temporal convolutional network, builds a power transformer fault diagnosis model, and trains and optimizes the power transformer fault diagnosis model through the transformer operation status feature data set;
[0101] The model deployment module deploys the optimized power transformer fault diagnosis model into the power system to perform fault diagnosis on the power transformer.
[0102] The above-mentioned adjustment system in the present invention can effectively implement the power transformer fault diagnosis method based on the enhanced temporal convolutional network, and the technical effects that can be achieved are as described in the above-mentioned embodiments and will not be repeated here.
[0103] Furthermore, the model building module includes:
[0104] Model architecture design unit: build the power transformer fault diagnosis model architecture, design the number of layers and parameters of each layer of the power transformer fault diagnosis model;
[0105] Output layer design unit, designs the output layer of the power transformer fault diagnosis model, and adds a fully connected layer before the output layer;
[0106] Regularization integration unit, integrating HER regularization method in power transformer fault diagnosis model;
[0107] A diagnosis model training unit divides the transformer operation status feature data set into a training set, a validation set, and a test set, and uses the training set and the validation set to train a power transformer fault diagnosis model;
[0108] The model evaluation and optimization unit uses the test set to evaluate the trained power transformer fault diagnosis model and optimizes the power transformer fault diagnosis model according to the evaluation results.
[0109] 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.
[0110] 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 power transformer fault diagnosis method based on enhanced temporal convolutional network, characterized in that: The method comprises: Collect data related to the operating status of power transformers in the power system to obtain a transformer operating status data set; Performing feature screening on the transformer operation status data set to obtain a transformer operation status feature data set; Designing a temporal convolutional network, and transforming and optimizing the temporal convolutional network; Based on the modified temporal convolutional network, a power transformer fault diagnosis model is built, and the power transformer fault diagnosis model is trained and optimized through the transformer operation status feature data set; The optimized power transformer fault diagnosis model is deployed in the power system to perform fault diagnosis on the power transformer.
2. The power transformer fault diagnosis method based on enhanced temporal convolutional network according to claim 1 is characterized in that: The transformer operation status data set is subjected to feature screening to obtain a transformer operation status feature data set including: Preprocessing the data in the transformer operation status data set; The preprocessed data is screened by feature selection algorithm to obtain key features; The key features are transformed and encoded to obtain a transformer operation status feature data set.
3. The power transformer fault diagnosis method based on enhanced temporal convolutional network according to claim 1 is characterized in that: The temporal convolutional network adopts a one-dimensional dilated convolutional network and a residual connection.
4. The power transformer fault diagnosis method based on enhanced temporal convolutional network according to claim 1 is characterized in that: The transformation and optimization of the temporal convolutional network includes: introducing a new activation function ALSU into the temporal convolutional network, combining the immortality of Leaky ReLU and the smoothness of Sigmoid function.
5. The power transformer fault diagnosis method based on enhanced temporal convolutional network according to claim 4 is characterized in that: The transformation and optimization of the temporal convolutional network also includes: introducing an adaptive mechanism in the temporal convolutional network to adjust the negative slope parameter of the Leaky ReLU, and the form of the adaptive mechanism is as follows: Among them, μ is a learnable threshold parameter; α is the scaling factor of the Sigmoid function; β is the negative slope of the Leaky ReLU.
6. The power transformer fault diagnosis method based on enhanced temporal convolutional network according to claim 1 is characterized in that: The transformation and optimization of the temporal convolutional network also includes: adding a HER regularization method to the temporal convolutional network and introducing a regularization strength dynamic adjustment mechanism.
7. The power transformer fault diagnosis method based on enhanced temporal convolutional network according to claim 6 is characterized in that: The introduction of the regularization strength dynamic adjustment mechanism includes: adding a dynamic adjustment factor γ on the basis of Elastic Net to dynamically adjust the regularization strength. The adjustment formula is as follows: Among them, L0 is the original loss function, λ is the regularization coefficient, and w i is the model weight, γ1 and γ2 are dynamic adjustment factors.
8. The power transformer fault diagnosis method based on enhanced temporal convolutional network according to claim 1 is characterized in that: The construction of the power transformer fault diagnosis model includes: Build a power transformer fault diagnosis model architecture, and design the number of layers and parameters of each layer of the power transformer fault diagnosis model; Designing an output layer of the power transformer fault diagnosis model, and adding a fully connected layer before the output layer; Integrating the HER regularization method into the power transformer fault diagnosis model; Dividing the transformer operating state feature data set into a training set, a validation set and a test set, and using the training set and the validation set to train the power transformer fault diagnosis model; The trained power transformer fault diagnosis model is evaluated using the test set, and the power transformer fault diagnosis model is optimized according to the evaluation result.
9. A power transformer fault diagnosis system based on enhanced temporal convolutional network, characterized in that: The system comprises: A data collection module collects data related to the operating status of power transformers in the power system and obtains a transformer operating status data set; A feature screening module performs feature screening on the transformer operation status data set to obtain a transformer operation status feature data set; A network design module designs a temporal convolutional network and transforms and optimizes the temporal convolutional network; A model building module, based on the modified temporal convolutional network, builds a power transformer fault diagnosis model, and trains and optimizes the power transformer fault diagnosis model through the transformer operation status feature data set; The model deployment module deploys the optimized power transformer fault diagnosis model into the power system to perform fault diagnosis on the power transformer.
10. The power transformer fault diagnosis system based on enhanced temporal convolutional network according to claim 9 is characterized in that: The model building module includes: A model architecture design unit, which builds a power transformer fault diagnosis model architecture and designs the number of layers and parameters of each layer of the power transformer fault diagnosis model; An output layer design unit designs the output layer of the power transformer fault diagnosis model and adds a fully connected layer before the output layer; A regularization integration unit, integrating the HER regularization method in the power transformer fault diagnosis model; A diagnosis model training unit, which divides the transformer operation status feature data set into a training set, a verification set and a test set, and uses the training set and the verification set to train the power transformer fault diagnosis model; A model evaluation and optimization unit is used to evaluate the trained power transformer fault diagnosis model using the test set, and optimize the power transformer fault diagnosis model according to the evaluation result.