Transformer fault diagnosis method and system based on IGWO-GBDT

Abstract

The invention relates to the technical field of transformer fault diagnosis, in particular to an IGWO-GBDT-based transformer fault diagnosis method and system, and the method comprises the steps: constructing an improved grey wolf optimization algorithm, and optimizing a grey wolf population position updating strategy through a nonlinear convergence factor, a dynamic weight distribution mechanism and a Levy flight disturbance mechanism; performing global optimization on the hyper-parameter combination of the gradient boosting tree by using an improved grey wolf optimization algorithm to determine an optimal hyper-parameter combination; and training the optimized gradient boosting tree by adopting a dissolved gas analysis data set, and performing transformer fault type diagnosis by using the trained gradient boosting tree. The improved grey wolf optimization algorithm (IGWO) is adopted to optimize GBDT hyper-parameters, and the diagnosis precision and efficiency are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power equipment fault diagnosis, and specifically relates to a transformer fault diagnosis method and system based on IGWO-GBDT. Background Technique

[0002] As a core device in the power system, the operating state of a transformer is directly related to the safety and stability of the power grid. During the operation of the transformer, due to the long-term action of multiple stresses such as electricity, heat, and mechanics, various faults may occur inside it, such as partial discharge, overheating fault, and insulation aging. If these faults are not diagnosed and processed in time, it may lead to equipment damage and even serious power accidents. Therefore, accurate and efficient transformer fault diagnosis technology is an important means to ensure the reliable operation of the power system. Dissolved gas analysis (DGA), as a technology widely used in transformer fault diagnosis, can effectively reflect the operating state and potential fault types of the transformer by monitoring the content change of dissolved gases in the oil. However, in practical applications, DGA data usually has problems such as sample imbalance, complex features, and diverse fault categories, which pose severe challenges to traditional diagnosis methods.

[0003] Existing transformer fault diagnosis methods mainly include rule-based methods, statistical analysis methods, and machine learning methods. Rule-based methods rely on criteria formulated based on expert experience, such as the IEC three-ratio method and the Duval triangle method. However, these methods have poor adaptability to complex faults and are prone to misjudgment or missed judgment. Statistical analysis methods analyze data through mathematical modeling, but their assumptions about data distribution may not match the actual situation, resulting in limited diagnostic accuracy. In recent years, machine learning methods have received extensive attention in transformer fault diagnosis due to their powerful non-linear modeling ability and adaptability to complex data. However, traditional machine learning models, such as support vector machines (SVM) and random forests (RF), have limited performance in dealing with high-dimensional data and sample imbalance problems, and their hyperparameter optimization process usually relies on manual experience and is difficult to achieve global optimality. In addition, the diagnostic performance of existing methods often drops significantly when facing insufficient diversity of fault samples or high-uncertainty samples.

[0004] To address the above problems, there is an urgent need for a transformer fault diagnosis method that can effectively solve sample imbalance, improve the generalization ability of the model, and achieve automatic hyperparameter optimization. Specifically, a method that can generate high-quality synthetic samples in the data preprocessing stage to balance the dataset is required, and at the same time, an advanced optimization algorithm is combined to globally optimize the hyperparameters of the machine learning model, thereby improving the diagnostic accuracy and reliability of the model. In addition, a more comprehensive evaluation mechanism needs to be designed to comprehensively consider the diagnostic difficulty of different fault categories and the performance of the model on high-uncertainty samples to further improve the credibility and practicality of the diagnostic results. Summary of the Invention

[0005] Aiming at the problems of data imbalance, low efficiency of model hyperparameter optimization, and insufficient diagnostic accuracy in the existing transformer fault diagnosis methods, the present invention proposes a transformer fault diagnosis method and system based on IGWO-GBDT. The method performs data balancing processing by introducing an Auxiliary Classifier Generative Adversarial Network (ACGAN), optimizes the hyperparameters of the Gradient Boosting Decision Tree (GBDT) model by combining an Improved Grey Wolf Optimization algorithm (IGWO), and improves the fitness function by using the weighted F1 score and uncertainty evaluation, thereby overcoming the deficiencies of the existing technologies and realizing efficient and accurate diagnosis of transformer faults.

[0006] The present invention is realized through the following technical solutions. A transformer fault diagnosis method based on IGWO-GBDT comprises the following steps: S1: Construct a dissolved gas analysis data set; S2: Construct an Improved Grey Wolf Optimization algorithm (IGWO), and optimize the grey wolf population position update strategy through a non-linear convergence factor, a dynamic weight allocation mechanism, and a Lévy flight perturbation mechanism; S3: Use the Improved Grey Wolf Optimization algorithm to globally optimize the hyperparameter combination of the Gradient Boosting Decision Tree (GBDT), and determine the optimal hyperparameter combination; S4: Use the dissolved gas analysis data set to train the optimized Gradient Boosting Decision Tree, and use the trained Gradient Boosting Decision Tree to diagnose the transformer fault type.

[0007] Further preferably, step S1 includes: preprocessing the original transformer dissolved gas analysis data to obtain the preprocessed dissolved gas analysis data, performing sample balancing processing on the preprocessed dissolved gas analysis data through an Auxiliary Classifier Generative Adversarial Network (ACGAN) to generate synthetic samples with the same distribution as the real fault data, forming a dissolved gas analysis data set with the synthetic samples and the dissolved gas analysis data, and dividing the dissolved gas analysis data set into a training set and a test set.

[0008] Further preferably, the process of performing sample balancing processing on the preprocessed dissolved gas analysis data through an Auxiliary Classifier Generative Adversarial Network (ACGAN) is as follows: Concatenate the random noise vector z and the fault category label c as the input of the generator to guide the generator to generate synthetic samples, and the synthetic samples are represented as , where G represents the generator; The discriminator not only includes a binary classifier for judging the authenticity of samples, but also adds an auxiliary classifier for predicting the fault category of samples; The optimization objective of the Auxiliary Classifier Generative Adversarial Network (ACGAN) consists of two parts. One is to measure the ability of the discriminator to distinguish real samples from generated samples , and the other is to ensure the consistency of the discriminator and the generator in recognizing the categories of samples ; meanwhile, during the optimization process, a feature matching term is introduced , and the expression is: ; where is the feature matching term, represents the feature extraction function of the th layer in the discriminator, n is the number of layers of the discriminator, and respectively represent the rd real sample and synthetic sample, represents the number of samples; the total loss function is: ; where is the total loss of the Auxiliary Classifier Generative Adversarial Network, and are the weight coefficients of and respectively.

[0009] Furthermore, preferably, the calculation formula of the non-linear convergence factor is: ; where is the non-linear convergence factor, is the initial convergence factor, and its value range is [0, 2]; is the current iteration number; is the maximum iteration number; and are two non-linear adjustment parameters, and , , and e is the natural constant.

[0010] Furthermore, preferably, a dynamic weight allocation mechanism is introduced to dynamically adjust the weights of α wolf, β wolf, and δ wolf according to the current iteration number. The weight calculation formulas are as follows: ; ; ; where , , respectively represent the weights of α wolf, β wolf, and δ wolf.

[0011] The process of optimizing using the Lévy flight perturbation mechanism is as follows: During the process of updating the position of the wolves, Lévy flight perturbation is introduced, and its step size follows the following distribution: ; Among them, represents the step size that follows the Lévy distribution, is the scale parameter, is the position parameter, is a random variable, represents the characteristic exponent of the Lévy distribution, represents a normal distribution with a mean of 0 and a variance of ; represents the standard normal distribution, and the standard deviation is calculated by the gamma function; After each position update, a Lévy flight perturbation is applied to the position of the wolves, and the perturbation intensity coefficient is set to to control the jump amplitude; the position update formula is: ; ; Among them, represents the Lévy flight perturbation applied to the position of the wolves, is the updated population position, is the population position before update.

[0012] Further preferably, the hyperparameters described in step S3 include the number of trees, learning rate, maximum tree depth, subsample ratio, minimum sample split, and minimum sample leaf node.

[0013] Further preferably, the gradient boosting decision tree (GBDT) introduces L1 regularization parameter and L2 regularization parameter during the training process, and adopts an adaptive regularization parameter adjustment mechanism to dynamically adjust the regularization parameter according to the performance of the gradient boosting decision tree during the training process.

[0014] Further preferably, when globally optimizing the hyperparameter combination of the gradient boosting decision tree, the weighted F1 score is used as the fitness function, and a class weight adjustment mechanism and gradient boosting decision tree uncertainty evaluation are introduced, and its calculation formula is as follows: ; Among them, C represents the number of fault categories, represents the number of samples of the c-th type of fault, represents the F1 score of the c-th type of fault, represents the class weight of the c-th type of fault; Meanwhile, the fitness function combines the uncertainty assessment predicted by the gradient boosting tree and gives an additional reward term R to the performance of the gradient boosting tree on high-uncertainty samples. The calculation formula is as follows: ; Wherein, represents the prediction uncertainty of the gradient boosting tree for the c-th type of fault sample, represents the corresponding weight coefficient; the final fitness function is: ; Wherein, represents the reward term weight.

[0015] The present invention also provides a transformer fault diagnosis system based on IGWO-GBDT, including a data acquisition and processing module, a model construction module, and a fault diagnosis module; The data acquisition and processing module acquires the original dissolved gas analysis data and processes it to construct a dissolved gas analysis data set; The model construction module optimizes the position update strategy of the gray wolf population through a non-linear convergence factor, a dynamic weight allocation mechanism, and a Levy flight perturbation mechanism, thereby constructing an improved gray wolf optimization algorithm (IGWO); uses the improved gray wolf optimization algorithm to globally optimize the hyperparameter combination of the gradient boosting tree (GBDT) to determine the optimal hyperparameter combination; and uses the dissolved gas analysis data set to train the optimized gradient boosting tree; The fault diagnosis module deploys the trained gradient boosting tree for transformer fault type diagnosis.

[0016] By adopting the technical solution of combining the improved gray wolf optimization algorithm (IGWO) with the gradient boosting tree (GBDT), the present invention solves the data imbalance problem in transformer fault diagnosis and improves the generalization ability and diagnosis accuracy of the model. The method enhances the efficiency and accuracy of hyperparameter optimization through a non-linear convergence factor, a dynamic weight allocation mechanism, and a Levy flight perturbation mechanism. At the same time, high-quality synthetic samples are generated through ACGAN to improve the class balance of the data set, providing reliable support for the state assessment and maintenance of transformers. Brief Description of the Drawings

[0017] Figure 1 is the overall flow schematic diagram of the transformer fault diagnosis method based on IGWO-GBDT of the present invention. Detailed Embodiments

[0018] The present invention will be further described in detail below with reference to the drawings.

[0019] Referring toFigure 1 , a transformer fault diagnosis method based on IGWO-GBDT, the steps are as follows: S1: Construct a dissolved gas analysis data set; preprocess the original dissolved gas analysis data of the transformer (including outlier processing and feature standardization) to obtain the preprocessed dissolved gas analysis data, and perform sample balancing on the preprocessed dissolved gas analysis data through an auxiliary classifier generative adversarial network (ACGAN) to generate synthetic samples with the same distribution as the real fault data. The dissolved gas analysis data set composed of the synthetic samples and the dissolved gas analysis data is divided into a training set and a test set; S2: Construct an improved grey wolf optimization algorithm (IGWO), and optimize the grey wolf population position update strategy through a non-linear convergence factor, a dynamic weight allocation mechanism, and a Lévy flight perturbation mechanism; S3: Use the improved grey wolf optimization algorithm to globally optimize the hyperparameter combination of the gradient boosting decision tree (GBDT) to determine the optimal hyperparameter combination; S4: Use the dissolved gas analysis data set to train the optimized gradient boosting decision tree, and use the trained gradient boosting decision tree to diagnose the transformer fault type.

[0020] In step S1, first in the data collection and preprocessing stage, the original dissolved gas analysis data generated during the operation of the oil-immersed transformer is processed to generate the original dissolved gas analysis data set, and preprocessing (including outlier processing and feature standardization) is performed to obtain the preprocessed dissolved gas analysis data set. Subsequently, an auxiliary classifier generative adversarial network (ACGAN model) is used to perform sample balancing on the preprocessed dissolved gas analysis data set to generate a balanced dissolved gas analysis data set. Next, the balanced dissolved gas analysis data set is divided into a training set, a validation set, and a test set to prepare for subsequent model training and parameter optimization.

[0021] In practical applications, the original dissolved gas analysis data (DGA) usually contains hydrogen ( ), methane ( ), ethane ( ), ethylene ( ), acetylene ( )and other gas content information. After the original dissolved gas analysis data is collected by sensors, there may be outliers, such as data deviations caused by equipment aging or external interference. To eliminate outliers, statistical-based anomaly detection methods, such as the box plot method or the Z-score method, are used to identify and remove data points outside the normal range. After the outlier processing is completed, the data is subjected to feature standardization processing to make each feature value have the same dimension and distribution range. This process uses the min-max normalization or Z-score normalization method to ensure good numerical stability of the data in the subsequent modeling process.

[0022] Since the number of samples of different fault categories in the original dissolved gas analysis data may vary significantly, direct use will result in insufficient learning of the model for samples of minority categories, thus affecting the diagnostic accuracy. Therefore, the sample balancing process of the preprocessed dissolved gas analysis data is carried out through an Auxiliary Classifier Generative Adversarial Network (ACGAN), and the specific process is as follows: Generator input: The random noise vector z is concatenated with the fault category label c as the input of the generator to guide the generator to generate synthetic samples of specific categories. The synthetic samples are represented as , where G represents the generator. At the same time, a feature fusion network is used to perform deep fusion processing on the random noise vector z and the fault category label c, extract the key features in the random noise vector z and the fault category label c, and perform feature cross-interaction to form a more representative feature vector, which is then input into the main network of the generator to generate higher-quality synthetic samples.

[0023] Discriminator: The discriminator not only includes a binary classifier for judging the authenticity of samples, but also adds an auxiliary classifier for predicting the fault category of samples. In the discriminator, in addition to retaining the original tasks of sample authenticity judgment and category label prediction, a new fault degree prediction branch is added to predict the severity of the fault corresponding to the input sample. The loss term related to fault degree prediction is added to the optimization objective accordingly, so that the discriminator learns the authenticity, category and fault degree characteristics of samples simultaneously during training, and improves the comprehensive discrimination performance for fault samples.

[0024] Optimization objective: The optimization objective of the Auxiliary Classifier Generative Adversarial Network (ACGAN) consists of two parts. One is to measure the ability of the discriminator to distinguish real samples from generated samples , and the other is to ensure the consistency of the discriminator and the generator in identifying the sample categories , and its expression is as follows: ; ; Among them, represents the expectation of the real data distribution of, Denotes the joint expectation of the noise distribution and the class distribution . Denotes sampling real samples from the real dataset and their corresponding fault class labels c, and calculating the expected value for the paired data ( ). Denotes real samples Denotes synthetic samples generated by the generator according to the random noise z and the fault class label c. The goal of model training is to maximize for the generator G and maximize for the discriminator D. At the same time, during the optimization process, a feature matching term is introduced to make the features of the samples generated by the generator closer to those of the real samples in statistical distribution, thereby improving the quality and diversity of the generated samples. Its expression is: ; where is the feature matching term, denotes the feature extraction function of the th layer in the discriminator, n is the number of layers of the discriminator, and respectively denote the th real sample and synthetic sample, denotes the number of samples. The total loss function is: ; where is the total loss of the auxiliary classifier generative adversarial network, and are the weight coefficients of and respectively.

[0025] Through the above improvements, the optimized auxiliary classifier generative adversarial network can generate higher-quality synthetic samples, while improving the comprehensive discrimination performance of the discriminator for fault samples, providing a more balanced and accurate data basis for the subsequent transformer fault diagnosis based on IGWO-GBDT, and enhancing the performance and reliability of the entire fault diagnosis method.

[0026] After the dataset is partitioned, it enters the model parameter optimization and verification stage. The core of this stage is to globally optimize the hyperparameter combination of the GBDT model using the improved grey wolf optimization algorithm (IGWO). The improved grey wolf optimization algorithm (IGWO) has been improved in three aspects based on the standard grey wolf optimization algorithm: non-linear convergence factor, dynamic weight allocation mechanism, and Lévy flight perturbation mechanism.

[0027] Replace the linear convergence factor in the standard Grey Wolf Optimization (GWO) algorithm with a non - linear convergence factor. The calculation formula of the non - linear convergence factor is as follows: ; Where, is the non - linear convergence factor, is the initial convergence factor, and its value range is [0, 2]; is the current iteration number; is the maximum iteration number; and are two non - linear adjustment parameters, and , , and e is the natural constant.

[0028] This non - linear convergence factor can provide a larger exploration space at the initial stage of algorithm iteration, enhancing the global search ability; as the number of iterations increases, the convergence factor gradually decreases, enabling the algorithm to focus on local search in the later stage and improving the convergence accuracy.

[0029] Replace the fixed weight allocation in the standard Grey Wolf Optimization (GWO) algorithm with a dynamic weight allocation mechanism, as follows: Introduce a dynamic weight allocation mechanism to dynamically adjust the weights of α - wolf, β - wolf, and δ - wolf according to the current iteration number to balance the global exploration and local exploitation capabilities. The weight calculation formula is as follows: ; ; ; Where, , , represent the weights of α - wolf, β - wolf, and δ - wolf respectively.

[0030] At the initial stage of iteration, assign a larger weight to α - wolf to highlight its global guiding role; as the number of iterations increases, gradually reduce the weight of α - wolf, and at the same time increase the weights of β - wolf and δ - wolf to enhance the local search ability, accelerate the convergence speed, and improve the optimization accuracy. Through the above improvements, the dynamic weight allocation mechanism can better adapt to the different stage requirements in the optimization process, balance the global exploration and local exploitation capabilities, and enhance the optimization performance of the IGWO algorithm, making it more suitable for the optimization problem of GBDT hyperparameters in transformer fault diagnosis.

[0031] Adopt the Levy flight perturbation mechanism to enhance population diversity and help the algorithm jump out of the local optimal solution, as follows: During the process of updating the position of wolves, introduce the Levy flight perturbation, and its step size follows the following distribution: ; wherein, represents the step size subject to the Lévy distribution, is the scale parameter, is the location parameter, is a random variable, represents the characteristic exponent of the Lévy distribution, = 1.5, represents a normal distribution with a mean of 0 and a variance of ; represents the standard normal distribution, and the standard deviation is calculated by the gamma function, specifically: ; wherein, is the gamma function.

[0032] After each position update, a Lévy flight perturbation is applied to the position of the wolf, and the perturbation intensity coefficient is set to to control the jump amplitude. The specific position update formula is: ; ; wherein, represents the Lévy flight perturbation applied to the position of the wolf, is the updated population position, is the population position before update.

[0033] According to the diversity index of the current population (such as the average distance between population individuals, the standard deviation of fitness values, etc.), the perturbation intensity coefficient is dynamically adjusted. When the population diversity is low, is increased to enhance the global search ability; when the population diversity is high, is decreased to refine the local search, so as to more effectively jump out of the local optimal solution and enhance the global optimization ability and convergence speed of the algorithm.

[0034] Through the above improvement, the Lévy flight perturbation mechanism can more flexibly adapt to different stages in the optimization process, improve population diversity, enhance the ability of the algorithm to jump out of the local optimal solution, and improve the efficiency and effect of the improved grey wolf optimization algorithm (IGWO) in optimizing the hyperparameter combination of the gradient boosting decision tree (GBDT) in transformer fault diagnosis.

[0035] In step S3, the improved grey wolf optimization algorithm (IGWO) is used to globally optimize the hyperparameter combination of the gradient boosting decision tree (GBDT). The hyperparameters include the number of trees, learning rate, maximum tree depth, subsample ratio, minimum samples for split, and minimum samples for leaf node. The search ranges of each hyperparameter are as follows: Number of trees ; Learning rate ; Maximum depth ; Subsample ratio ; Minimum samples for split ; Minimum samples for leaf node ; L1 regularization parameter ; L2 regularization parameter ; The termination condition for the hyperparameter optimization is that the performance of the gradient boosting decision tree does not improve significantly for 10 consecutive iterations, and in the hyperparameter optimization process, the hyperparameters that have a greater impact on the performance of the gradient boosting decision tree are adjusted first.

[0036] In the training process of the gradient boosting decision tree (GBDT), L1 regularization parameter and L2 regularization parameter are introduced, and an adaptive regularization parameter adjustment mechanism is adopted to dynamically adjust the regularization parameter according to the performance of the gradient boosting decision tree in the training process. The regularization term The calculation formula is: ; where M is the number of leaf nodes of the decision tree, is the predicted value of the leaf node. The adjustment formula of the regularization parameter is: ; where, is the L1 regularization parameter of the t-th iteration, is the L1 regularization parameter of the (t + 1)-th iteration, is the L2 regularization parameter of the t-th iteration, is the L2 regularization parameter of the (t + 1)-th iteration, is the adjustment rate, is the performance change rate of the gradient boosting decision tree in consecutive iterations.

[0037] In step S3, the weighted F1 score of the validation set is used as the fitness function, and a class weight adjustment mechanism and gradient boosting decision tree uncertainty evaluation are introduced. The calculation formula is as follows: ; Where C represents the number of fault categories, represents the number of samples of the c-th type of fault, represents the F1 score of the c-th type of fault, represents the class weight of the c-th type of fault, which is used to reflect the importance of this class in fault diagnosis. The weight coefficient can be manually set or automatically adjusted according to the risk level or diagnostic difficulty of the fault type.

[0038] Meanwhile, the fitness function combines the uncertainty assessment of the gradient boosting tree prediction and gives an additional reward term R to the performance of the gradient boosting tree on high-uncertainty samples. Its calculation formula is: ; Where represents the prediction uncertainty of the gradient boosting tree for the c-th type of fault sample, represents the corresponding weight coefficient. The final fitness function is: ; Where represents the reward term weight. By combining the class weight adjustment mechanism and the gradient boosting tree prediction uncertainty assessment, the improved fitness function can more comprehensively reflect the comprehensive performance of the gradient boosting tree on different fault categories, and at the same time encourage the gradient boosting tree to improve the diagnostic accuracy on high-uncertainty samples, providing more reliable results for transformer fault diagnosis.

[0039] After completing the hyperparameter optimization, it enters the model testing and diagnostic result output stage. In this stage, the test set is used to test the gradient boosting tree with optimized parameters to evaluate the performance of the gradient boosting tree in actual fault diagnosis tasks. The test metrics include key performance metrics such as accuracy, recall, and F1 score. Accuracy reflects the correct classification ratio of the gradient boosting tree for all samples, recall measures the recognition ability of the gradient boosting tree for a certain type of fault sample, and the F1 score comprehensively considers the performance of accuracy and recall. During the test, by comparing the prediction results of the gradient boosting tree with the true labels, the values of each metric are calculated, and the performance of the gradient boosting tree on different fault categories is analyzed. Finally, according to the test results, the transformer fault diagnosis results are output to provide decision support for the condition assessment and maintenance of the transformer.

[0040] Another embodiment of the present invention provides a transformer fault diagnosis system based on IGWO-GBDT, including a data acquisition and processing module, a model construction module, and a fault diagnosis module; The data acquisition and processing module acquires the original dissolved gas analysis data and processes it to construct a dissolved gas analysis data set; The model construction module optimizes the position update strategy of the grey wolf population through a non-linear convergence factor, a dynamic weight allocation mechanism, and a Lévy flight perturbation mechanism, thereby constructing an improved grey wolf optimization algorithm (IGWO); uses the improved grey wolf optimization algorithm to globally optimize the hyperparameter combination of the gradient boosting decision tree (GBDT), and determines the optimal hyperparameter combination; and trains the optimized gradient boosting decision tree using the dissolved gas analysis dataset. The fault diagnosis module deploys the trained gradient boosting decision tree for transformer fault type diagnosis.

[0041] During the whole implementation process, the technical solution of the present invention solves the key problems in the prior art through multiple innovative points. First, the ACGAN model is used to perform sample equalization processing on the dataset to generate high-quality synthetic samples, solving the problem of class imbalance in the dataset. Second, the hyperparameter combination of the gradient boosting decision tree is globally optimized through the IGWO algorithm, significantly improving the generalization ability and diagnostic accuracy of the gradient boosting decision tree. Finally, the L1 regularization parameter and the L2 regularization parameter are introduced during the training process of the gradient boosting decision tree, and an adaptive regularization parameter adjustment mechanism is adopted to dynamically adjust the regularization parameter according to the performance of the gradient boosting decision tree during the training process, further enhancing the stability and robustness of the gradient boosting decision tree. The combination of the above technical solutions enables the present invention to effectively meet the complex and changeable transformer fault diagnosis requirements in practical application scenarios, providing reliable technical support for the operation and maintenance of power equipment.

[0042] The above only expresses the preferred embodiments of the present invention and does not limit the present invention in other forms. Any person skilled in the art may use the disclosed content to make changes or modifications into equivalent embodiments with equivalent changes. However, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A transformer fault diagnosis method based on IGWO-GBDT, characterized in that the steps are as follows As follows: S1: Construct a dissolved gas analysis data set; S2: Construct an improved grey wolf optimization algorithm, and optimize the position update strategy of the grey wolf population by means of a non-linear convergence factor, a dynamic weight allocation mechanism and a Lévy flight perturbation mechanism; S3: Use the improved grey wolf optimization algorithm to globally optimize the hyperparameter combination of the gradient boosting tree and determine the optimal hyperparameter combination; S4: Use the dissolved gas analysis data set to train the optimized gradient boosting tree, and use the trained gradient boosting tree for transformer fault type diagnosis.

2. The transformer fault diagnosis method according to claim 1, characterized in that the steps S1 includes: preprocessing the original dissolved gas analysis data of the transformer to obtain the preprocessed dissolved gas analysis data, performing sample balancing processing on the preprocessed dissolved gas analysis data through an auxiliary classifier generative adversarial network, generating synthetic samples consistent with the distribution of real fault data, the dissolved gas analysis data set composed of the synthetic samples and the dissolved gas analysis data, and dividing the dissolved gas analysis data set into a training set and a test set.

3. The transformer fault diagnosis method according to claim 2, characterized in that Performing sample balancing processing on the preprocessed dissolved gas analysis data through an auxiliary classifier generative adversarial network, the specific process is as follows: Concatenate the random noise vector z with the fault class label c as the input to the generator to guide the generator to generate synthetic samples, and the synthetic samples are denoted as , where G represents the generator; The discriminator not only includes a binary classifier for judging the authenticity of samples, but also adds an auxiliary classifier for predicting the fault category of samples; The optimization objective of the auxiliary classifier generative adversarial network consists of two parts. One is to measure the ability of the discriminator to distinguish real samples from generated samples , and the other is to ensure the consistency of the discriminator and the generator in recognizing the sample categories ; meanwhile, during the optimization process, a feature matching term is introduced , and the expression is as follows: ； Among them, is a feature matching item, represents the feature extraction function of the th layer in the discriminator, n is the number of discriminator layers, and respectively represent the th real sample and synthetic sample, represents the number of samples; the total loss function is: ； Among them, is the total loss of the auxiliary classifier generative adversarial network, and are respectively and weight coefficients of 4. The transformer fault diagnosis method according to claim 1, characterized in that The calculation formula of the non-linear convergence factor is: ； Among them, is a non-linear convergence factor, is the initial convergence factor, and its value range is [0, 2]; is the current iteration number; is the maximum iteration number; and are two non-linear adjustment parameters, and , , where e is the natural constant.

5. The transformer fault diagnosis method according to claim 4, characterized in that Dynamic weight allocation mechanism, dynamically adjusting the weights of α wolf, β wolf, and δ wolf according to the current iteration number, and the weight calculation formula is as follows: ； ； ； Among them, , , represent the weights of alpha wolf, beta wolf, and delta wolf respectively.

6. The transformer fault diagnosis method according to claim 5, characterized in that, The process of optimizing by using the Lévy flight perturbation mechanism is: Introduce Lévy flight perturbation in the position update process of ω wolf, and its step size follows the following distribution: ； Among them, represents the step size following the Lévy distribution, is the scale parameter, is the location parameter, is the random variable, represents the characteristic exponent of the Lévy distribution, represents a normal distribution with a mean of 0 and a variance of , represents the standard normal distribution, and the standard deviation is calculated by the gamma function; After each position update, Levy flight perturbation is applied to the position of ω wolves, and the perturbation intensity coefficient is set to , to control the jump amplitude; the position update formula is: ； ； Among them, represents the Lévy flight perturbation applied to the position of ω wolves, is the updated population position, is the population position before update.

7. The transformer fault diagnosis method according to claim 1, characterized in that, The hyperparameters described in step S3 include the number of trees, learning rate, maximum tree depth, subsample ratio, minimum sample split, and minimum sample leaf node.

8. The transformer fault diagnosis method according to claim 1, characterized in that, L1 regularization parameter and L2 regularization parameter are introduced in the training process of the gradient boosting tree, and an adaptive regularization parameter adjustment mechanism is adopted to dynamically adjust the regularization parameter according to the performance of the gradient boosting tree in the training process.

9. The transformer fault diagnosis method according to claim 1, characterized in that When globally optimizing the hyperparameter combination of the gradient boosting tree, a weighted F1 score is used as the fitness function, and a class weight adjustment mechanism and a gradient boosting tree uncertainty evaluation are introduced, and the calculation formula is as follows: ； Among them, C represents the number of fault categories, represents the number of fault samples of the c-th category, represents the F1 score of the c-th category of faults, represents the category weight of the c-th category of faults; At the same time, the fitness function combines the uncertainty evaluation of the gradient boosting tree prediction, and gives an additional reward term R to the performance of the gradient boosting tree on high-uncertainty samples, and the calculation formula is: ； Among them, represents the prediction uncertainty of the gradient boosting tree for the fault samples of the c-th class, represents the corresponding weight coefficient; the final fitness function is: ； Among them, represents the weight of the reward item.

10. A transformer fault diagnosis system based on IGWO-GBDT, including a data acquisition and processing module, a model construction module and a fault diagnosis module, characterized in that The data acquisition and processing module acquires the original dissolved gas analysis data and processes it to construct a dissolved gas analysis data set; The model construction module optimizes the position update strategy of the grey wolf population through a non-linear convergence factor, a dynamic weight allocation mechanism and a Lévy flight perturbation mechanism, so as to construct an improved grey wolf optimization algorithm; use the improved grey wolf optimization algorithm to globally optimize the hyperparameter combination of the gradient boosting tree and determine the optimal hyperparameter combination; And the optimized gradient boosting tree is trained using the dissolved gas analysis dataset; The trained gradient boosting tree is deployed in the fault diagnosis module for diagnosing the transformer fault types.

Images