A transformer fault detection method, device, equipment and medium based on KFWAdaBoost

Abstract

The present invention provides a transformer fault detection method, device, equipment and medium based on KFWAdaBoost, which relates to the technical field of transformers. The method generates clustering features using the KPP algorithm by combining feature generation technology and integrated learning strategy, and encodes the distribution pattern of samples by analyzing the relationship between samples and cluster centers. The model generates weighted soft labels for each sample during the classification process, and uses soft labels to express the ambiguity and similarity between categories, effectively improving the information expression ability of the data set. The sample weights are also adjusted through multiple rounds of iterations to strengthen the learning of difficult-to-classify samples, and the final classification results are output in combination with a voting mechanism. In addition, the generated new data set is not limited to the KFWAdaBoost model, but can be widely used in the training of various classifiers, thereby improving the utilization efficiency of data and the accuracy of diagnosis.

Description

Technical Field

[0001] The present invention relates to the technical field of transformers, and in particular to a transformer fault detection method, device, equipment and medium based on KFWAdaBoost. Background Art

[0002] In the power system, the health status of the transformer is crucial to ensure the continuity and safety of power supply. As a key component in the power system, the early diagnosis of internal faults of oil-immersed transformers is crucial to preventing potential large-scale power outages and economic losses. Due to internal faults of the transformer (such as insulation aging and cracking), specific gases (such as hydrogen, methane, ethylene, acetylene, etc.) are generated in the insulation material, which are then dissolved in the transformer oil. Therefore, by analyzing the composition and concentration of these dissolved gases, the health status of the transformer can be evaluated.

[0003] Traditional dissolved gas analysis (DGA) methods, such as the three-ratio method and the Rogers ratio method, are helpful for fault diagnosis to a certain extent, but their limitations gradually become apparent when dealing with complex fault modes and data. With the development of artificial intelligence technology, researchers have begun to explore machine learning-based fault detection methods, such as neural networks and support vector machines, in order to improve the accuracy of diagnosis. However, these methods often find it difficult to effectively capture and utilize the complex relationships between gas features when faced with limited sample numbers and insufficient data dimensions, resulting in limited performance and low accuracy of fault detection.

[0004] In view of this, this application is filed. Summary of the invention

[0005] The present invention provides a transformer fault detection method, device, equipment and medium based on KFWAdaBoost, which can at least partially improve the above-mentioned problems.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] A transformer fault detection method based on KFWAdaBoost, comprising:

[0008] Obtaining original training data, preprocessing the original training data, and generating a model training set;

[0009] Using the KPP algorithm to perform clustering and splicing preprocessing on the model training set and the original feature matrix to generate a training feature matrix;

[0010] Training the FWAdaBoost model according to the training feature matrix to generate training results;

[0011] Based on the training results and the FWAdaBoost model, soft label feature learning is performed to generate soft label features, and the soft label features are fused to obtain a classification feature enhanced data set;

[0012] Performing performance evaluation processing on the classification feature enhanced data set, and ending the soft label feature learning process when the number of base classifiers reaches a preset value;

[0013] The transformer data to be detected is obtained, and the transformer data is input into the trained FWAdaBoost model, and model prediction is performed based on the classification feature enhanced data set to generate a fault detection result.

[0014] The present invention also provides a transformer fault detection device based on KFWAdaBoost, which comprises:

[0015] A preprocessing unit, used to obtain original training data, preprocess the original training data, and generate a model training set;

[0016] A clustering unit, used for performing clustering and splicing preprocessing on the model training set and the original feature matrix using a KPP algorithm to generate a training feature matrix;

[0017] A training unit, used to train the FWAdaBoost model according to the training feature matrix and generate a training result;

[0018] A soft label feature learning unit, used to perform soft label feature learning based on the training result and the FWAdaBoost model, generate soft label features, and perform fusion processing on the soft label features to obtain a classification feature enhanced data set;

[0019] An evaluation unit, used to perform performance evaluation processing on the classification feature enhanced data set, and when the number of base classifiers reaches a preset value, end the soft label feature learning process;

[0020] The detection unit is used to obtain transformer data to be detected, input the transformer data into the trained FWAdaBoost model, perform model prediction based on the classification feature enhanced data set, and generate a fault detection result.

[0021] The present invention also provides a transformer fault detection device based on KFWAdaBoost, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the transformer fault detection method based on KFWAdaBoost as described in any one of the above.

[0022] The present invention also provides a readable storage medium storing a computer program, wherein the computer program can be executed by a processor of a device where the storage medium is located to implement a transformer fault detection method based on KFWAdaBoost as described in any one of the above.

[0023] In summary, the transformer fault detection method based on KFWAdaBoost significantly improves the accuracy and robustness of oil-immersed transformer fault detection through advanced feature generation and ensemble learning strategies. The core method includes clustering the data set using the KPP (K-Means Plus Plus) algorithm to generate cluster features, and analyzing the relationship between samples and cluster centers to encode sample distribution patterns and enhance the expression ability of gas features. In this process, the KFWAdaBoost model generates weighted soft labels for samples, which not only capture the fuzziness and similarity between samples, but also strengthen the learning of difficult-to-classify samples through multiple rounds of iterative adjustment of sample weights, and improve the classification ability of the model for unbalanced data sets. Compared with the traditional AdaBoost algorithm, this method can more effectively tap the data potential and improve the accuracy and stability of fault classification when dealing with limited samples and complex feature data. In addition, the new data set generated by this method is not only suitable for the KFWAdaBoost model, but can also be widely used in the training of various classifiers, thereby improving data utilization efficiency and diagnostic accuracy, and providing strong technical support for the stable operation and fault prevention of power systems. It aims to improve the accuracy and robustness of oil-immersed transformer fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a flowchart of a transformer fault detection method based on KFWAdaBoost provided by the first embodiment of the present invention;

[0025] Figure 2 It is a flowchart of a transformer fault detection method based on KFWAdaBoost provided by the first embodiment of the present invention;

[0026] Figure 3 is a similarity matrix generation flow chart provided by an embodiment of the present invention;

[0027] Figure 4 is a schematic diagram of the KFAdaBoost model structure provided by an embodiment of the present invention;

[0028] Figure 5 It is a module schematic diagram of a transformer fault detection device based on KFWAdaBoost provided by the second embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0030] refer to Figure 1 , Figure 2 As shown, the first embodiment of the present invention discloses a transformer fault detection method based on KFW AdaBoost, which can be executed by a transformer fault detection device based on KFW AdaBoost (hereinafter referred to as a fault detection device), and in particular, executed by one or more processors in the fault detection device to implement the following method:

[0031] S1, obtaining original training data, preprocessing the original training data, and generating a model training set;

[0032] Preferably, the preprocessing includes feature encoding processing, feature scaling processing, label encoding and Z-score normalization processing.

[0033] Specifically, in this embodiment, for the input transformer fault detection data set, preprocessing operations such as feature encoding and feature scaling are performed, including label encoding and Z-score standardization, to obtain a sample set for learning. This step ensures the consistency and comparability of the data, laying the foundation for subsequent model training.

[0034] S2, using the KPP algorithm to perform clustering and splicing preprocessing on the model training set and the original feature matrix to generate a training feature matrix;

[0035] Specifically, step S2 includes: using a KPP algorithm with a Rand coefficient as a clustering value k to perform clustering processing on the model training set to obtain cluster centers;

[0036] Calculate each sample data in the model training set and the cluster center, calculate the corresponding similarity, and obtain a similarity matrix;

[0037] An original feature matrix is ​​obtained, and the original feature matrix is ​​concatenated with the similarity matrix to generate a training feature matrix.

[0038] In this embodiment, the KPP algorithm is used to perform clustering and splicing preprocessing on the model training set and the original feature matrix to generate a training feature matrix. This step is a key link in improving the accuracy of transformer fault detection, because it can reveal the potential structure and pattern in the sample data through cluster analysis, thereby providing richer feature information for subsequent model training.

[0039] First, the Rand coefficient is used to determine the optimal value of the clustering value k. The Rand coefficient is an indicator to measure the clustering effect. It evaluates the effect of different clustering numbers k by comparing the similarity between samples. The k value with the largest Rand coefficient is selected to ensure the rationality and effectiveness of the clustering results. This selection process not only improves the accuracy of clustering, but also enhances the model's ability to recognize data structure. Next, the similarity between each sample data in the model training set and the cluster center obtained by the KPP algorithm is calculated. This calculation process involves measuring the distance between the feature vector of each sample data and the cluster center vector, usually using methods such as Euclidean distance or cosine similarity. By calculating the similarity, a similarity matrix is ​​obtained, which records the similarity between each sample and the center of its cluster. The introduction of the similarity matrix not only increases the data dimension, but also provides the model with additional information about the relationship between samples, which is particularly important for dealing with complex fault detection tasks.

[0040] The original feature matrix is ​​concatenated with the similarity matrix to generate a training feature matrix. This step combines the original fault features with the similarity information between samples to form a more comprehensive feature representation. This feature representation not only includes the individual features of the samples, but also the relationships between samples, which has significant benefits for improving the generalization ability and classification accuracy of the model.

[0041] In short, the transformer fault detection method based on KFWAdaBoost generates features for the original data through the similarity matrix obtained by KPP and performs feature fusion. The number of clusters k is determined by the optimal Rand coefficient, the k value is adaptively selected, and the appropriate cluster center is found to obtain the similarity matrix; thereby enriching the sample data dimension in the transformer fault detection task.

[0042] Specifically, Figure 3 As shown, Figure 3 The process of the KPP method for similarity matrix is ​​shown: First, the training samples are divided into m clusters (a). Then, based on the similarity between the samples and the cluster centers, new features are created and merged with the existing features of the samples (b). Next, this new set of samples is used to train the classifier (c), and the calculated clusters are saved during the training process (d). Through this process, a trained classification model can be obtained (e). When there are new unclassified samples that need to be predicted, I first calculate the similarity between these samples and the center of each cluster and encode these similarities as new features (g). Finally, these new samples are input into the model together with their new features for classification (h), and the model predicts the samples accordingly (i). This process effectively uses the similarity matrix to enhance the feature representation of the model, thereby improving the accuracy of classification.

[0043] See also Figure 4, S3, training the FWAdaBoost model according to the training feature matrix to generate a training result;

[0044] Specifically, step S3 includes: iteratively inputting the training feature matrix into the FWAdaBoost model for training, and determining whether a training termination condition is met, wherein the training termination condition is that the number of iterated base classifiers reaches a set number;

[0045] If yes, generate training results;

[0046] If not, continue to iteratively input the training feature matrix into the FWAdaBoost model for training.

[0047] In this embodiment, the FWAdaBoost (Fuzzy Weight Adaptive Boosting Algorithm) model is an integrated learning method that can improve the accuracy and robustness of fault detection by combining multiple weak classifiers. Specifically, the training feature matrix is ​​first iteratively input into the FWAdaBoost model for training. The training feature matrix is ​​generated by the aforementioned step S2, which not only contains the original feature information, but also integrates the similarity information between samples, thereby providing a more comprehensive sample feature representation for the model.

[0048] During the training process, the training feature matrix is ​​continuously and iteratively input into the model, and the base classifiers are gradually constructed. Each base classifier is optimized and adjusted based on the previous classifier to address the shortcomings of the current model performance. This iterative process continues until the preset number of iterations is reached or the number of base classifiers reaches the set value, which is the training termination condition. The purpose of setting the number of iterations is to control the time and complexity of model training, avoid overfitting, and ensure that the model can achieve optimal performance within a reasonable time. In each iteration, the FWAdaBoost model adjusts the sample weights based on the previous classification results, paying special attention to those samples that are misclassified, that is, difficult to classify samples. In this way, the model can gradually improve its recognition ability for these samples, thereby improving the overall classification accuracy. In addition, the use of fuzzy weights allows the model to take into account the fuzziness and similarity between categories when classifying, which is particularly important when dealing with complex data in transformer fault detection.

[0049] When the training termination condition is met, the training results are generated, that is, an optimized FWAdaBoost model is obtained. This model can effectively handle unbalanced data sets and provide probability information about samples belonging to each category, which is of significant benefit for improving boundary sample identification and classification accuracy in fault detection.

[0050] See also Figure 2, S4, based on the training results and the FWAdaBoost model, performing soft label feature learning, generating soft label features, and fusing the soft label features to obtain a classification feature enhanced data set;

[0051] Specifically, step S4 includes: inputting the training results into the FWAdaBoost model, and the base classifier in the FWAdaBoost model selects representative samples in each category;

[0052] Performing calculations on each sample in the model training set and each representative sample to generate a degree of membership;

[0053] All membership degrees are integrated to generate soft label features, and the soft label features are fused with the original feature matrix to obtain a classification feature enhanced data set.

[0054] In this embodiment, if Figure 1 As shown in the figure, the process starts with the input of the data set. The data set is clustered by KPP and then enters the FWAdaBoost integrated framework for soft label learning. The similar feature matrix is ​​obtained by the KPP method and fused with the data set to increase the amount of information contained in each sample in the data set, so as to achieve the effect of improving the classification performance of the FWAdaBoost model. Repeat the operation twenty times and take the average value to avoid the error caused by chance that the result is not authentic.

[0055] Specifically, considering that data splitting has a great impact on the model's soft label information learning, this method uses a five-fold cross-validation method to obtain the soft label features of each sample and increase the amount of information in the data set. This method uses the training results of the FWAdaBoost model to learn soft label features, which is a key step in improving the accuracy of transformer fault detection. The purpose of soft label feature learning is to generate a soft label for each sample that represents the probability of it belonging to each category, rather than a traditional hard label (i.e., a certain category affiliation). This method can more accurately capture the similarities between samples and the ambiguity between categories, thereby improving the level of detail and accuracy of classification.

[0056] Specifically, the training results are input into the FWAdaBoost model, and each base classifier in the model will select representative samples from the category it is responsible for. These representative samples are the most characteristic and discriminative samples in the category, and they can well represent the characteristics of the category to which they belong. In this way, key information can be extracted from each category, laying the foundation for subsequent soft label generation. Next, each sample in the model training set and each representative sample are calculated and processed to generate membership. Membership refers to the degree to which a sample belongs to a certain category. This is a value between 0 and 1 that can quantify the similarity between the sample and the category. By calculating the membership of each sample to all representative samples, a comprehensive membership set can be obtained, which reflects the distribution of samples in multiple categories.

[0057] Subsequently, all the memberships are integrated to generate soft label features. These soft label features contain the probability information of the samples belonging to each category, providing the model with richer information and helping to improve the accuracy of fault detection. Compared with traditional hard labels, soft labels can better handle the problems of boundary samples and overlapping categories, thereby improving the performance of the classifier. Finally, the generated soft label features are fused with the original feature matrix to obtain a classification feature enhanced dataset; the classification feature enhanced dataset is not limited to a certain type of classifier, but can be widely used in the training of various classifiers. This enhanced dataset not only contains the original fault features, but also adds additional information about the relationship between samples and the probability information of the samples belonging to each category. The application of this feature enhancement technology enables the model to better adapt to unbalanced data sets and improves the recognition ability of minority category samples, which has important practical significance for fault detection in actual power systems.

[0058] In short, the transformer fault detection method based on KFWAdaBoost uses the established model to learn soft label features, and fuses the soft label features with the original feature matrix to form a feature matrix with soft label information, which provides information about the probability that the sample belongs to each category, which helps to process boundary samples and improve the accuracy of classification. This method enhances the model's recognition ability for unbalanced data sets through cluster information and soft label information, improves the accuracy of transformer fault detection, can be widely used in various classifiers, and provides more reliable and accurate fault detection results for the stability and safety of power systems.

[0059] S5, performing performance evaluation processing on the classification feature enhanced data set, and when the number of base classifiers reaches a preset value, ending the soft label feature learning process;

[0060] Preferably, the performance evaluation process includes average accuracy and average ranking, the average accuracy is used to evaluate the ability of the FWAdaBoost model to correctly classify, and the calculation formula of the average accuracy is: , where C is the number of categories, is a sample correctly classified as the positive class, For samples correctly classified as negative classes, is a positive sample that is misclassified as a negative class, is a positive sample correctly classified as a negative class.

[0061] The average ranking is used to evaluate the average performance ranking of different classifiers on multiple data sets. The calculation formula of the average ranking is: , where D is the number of data sets, The ranking is determined based on performance indicators.

[0062] Specifically, in this embodiment, the performance evaluation process of the classification feature enhancement dataset is a crucial step, which can quantify the performance of the FWAdaBoost model and determine whether the soft label feature learning process should be terminated. The performance evaluation process includes calculating two key indicators, the average accuracy and the average ranking, which work together to evaluate the classification ability of the model and its performance compared with other classifiers.

[0063] First, the average accuracy is calculated, which is an important indicator to measure the correct classification ability of the FWAdaBoost model. The accuracy of the model can be obtained based on the average accuracy, that is, the proportion of samples correctly classified by the model to the total number of samples. The higher this proportion, the stronger the classification ability of the model. This calculation formula not only provides an intuitive understanding of the overall performance of the model, but also helps to identify whether the model's performance on positive and negative classes is balanced, which is especially important for transformer fault detection, because accurately identifying fault (positive class) and non-fault (negative class) states is crucial to preventing potential power system failures.

[0064] Secondly, the average rank is calculated, which is the average of the performance rankings of different classifiers on multiple data sets; the ranking is determined based on the performance index, and the lower the ranking value, the better the performance. This metric provides a quantitative measure of the relative performance of the model on different data sets, which helps to understand the generalization ability of the model under diverse data conditions. By comparing the average rankings of different classifiers, the advantages and disadvantages of the FWAdaBoost model compared with other existing technologies can be evaluated. When the results of the performance evaluation process reach the preset value, that is, the average accuracy and average ranking meet the set performance standards, it is considered that the soft label feature learning process has been sufficiently optimized, and the soft label feature learning process is terminated at this time. This decision point is based on the quantitative evaluation of the model performance, which ensures the effectiveness and reliability of the model in practical applications.

[0065] S6, obtaining transformer data to be detected, inputting the transformer data into the trained FWAdaBoost model, performing model prediction based on the classification feature enhanced data set, and generating a fault detection result.

[0066] In this embodiment, the model obtained by training the training set will predict the data set by inputting the test set data into the model verified by step 4, and the model will predict by averaging the experimental results of twenty times. In the transformer fault detection task, the combination of soft label learning and clustering algorithm improves the model's ability to recognize difficult samples and enhances the accuracy of classification. It is not limited to a certain type of classifier, but can be widely used in the training of various classifiers, providing more reliable and accurate fault detection results for the stability and safety of the power system.

[0067] Specifically, in this embodiment, the transformer fault detection method based on KFWAdaBoost is specifically used for transformer fault detection tasks. By combining clustering algorithms with ensemble learning methods, soft label learning is performed on the input data to generate a feature matrix that incorporates soft label information to improve the accuracy of transformer fault detection. Specifically, by combining multiple weak classifiers to improve the accuracy and robustness of diagnosis, the accuracy of classification can be improved, especially when dealing with difficult sample data sets, the sample data dimension in the transformer fault detection task is increased through cluster information and soft label information, and the classification ability of the model for unbalanced data sets is enhanced, which can be widely used in various classifiers.

[0068] In order to further verify the performance of the KFWAdaBoost-based transformer fault detection method, two representative data sets were selected: the IEC TC 10 standard data set and the transformer fault data set of actual engineering. These two data sets have the characteristics of international standardization and actual engineering background, respectively, which provide diversity and comprehensiveness for the verification of the model. The IECTC 10 standard data set contains 117 groups of fault samples. It is a standard data set for transformer fault detection issued by the International Electrotechnical Commission. It has authoritativeness and wide application value and is often used as a benchmark data set for research model comparison analysis. The transformer fault data set contains 419 groups of samples, which are derived from long-term monitoring and recording in the actual operation of the power system. The data is authentic and reliable, covering complex and diverse fault types and working conditions, and can be closer to actual application scenarios. According to the DL / T 722-2000 guide and the IEC 60599-2015 standard, the samples in these two data sets are marked as five typical transformer internal fault types, including medium and low temperature overheating (T1-T2), high temperature overheating (T3), low energy discharge (D1), high energy discharge (D2) and partial discharge (PD). These fault types are identified by analyzing the concentration characteristics of dissolved gases, including hydrogen (H 2 ), methane (CH 4 ), ethane (C 2 H 6 ), ethylene (C 2 H 4 ), acetylene (C 2 H 2 ), carbon monoxide (CO) and carbon dioxide (CO 2 ).

[0069] In order to ensure the quality and consistency of the data, the dataset was strictly preprocessed, including removing duplicate samples, deleting outliers and non-numeric value samples, and normalizing all input variables. These operations not only reduce redundant information, but also balance the impact of different dimensional features on model training. After preprocessing, the IEC TC 10 standard dataset was divided into 87 groups of training samples and 30 groups of test samples, mainly used for preliminary verification of model performance. The transformer fault dataset was used for further model testing to evaluate its generalization ability in actual engineering scenarios. Table 1 shows in detail the statistical distribution of the five fault types in the two datasets. It can be seen that the sample distribution is relatively balanced, which can meet the needs of model training for multi-fault type data. At the same time, the rich data volume also provides sufficient support for the stability verification of the model. The authority of the IEC TC 10 standard dataset enables this method to be fairly compared with other research methods, and the actual engineering background of the dataset effectively demonstrates the application value of the model in a real environment. Table 1 shows the fault type distribution of the IEC TC10 standard dataset and 419 sets of domestic transformer fault datasets.

[0070] Table 1 Fault type distribution

[0071]

[0072] In summary, the transformer fault detection method based on KFWAdaBoost obtains the soft label features of the transformer fault detection dataset samples through the KFWAdaBoost model and fuses them with the original feature set. This process can effectively improve the classification accuracy. In the transformer fault detection task, the application of this method can obtain the soft label features of the samples, which contains the probability information of the samples belonging to each category, provides more abundant information for the model, and helps to improve the accuracy of fault detection. Through the KPP (Kmeans Plus Plus) clustering method, the similarity between samples is calculated, the similarity matrix is ​​constructed, and the obtained soft label features are fused with the original feature set to form a new feature matrix. This step enhances the information content of the dataset and provides more comprehensive input data for the transformer fault detection model. The new dataset containing the similarity matrix encoding is input into the KFWAdaBoost model for learning. By learning these new features, the model can capture the complex patterns and relationships in the transformer fault data. The weighted label information of each label finally learned by the KFWAdaBoost model provides important label information for the classification of the transformer fault dataset. This information reflects the distinction between different fault types and helps to improve the accuracy of fault detection. The new data set with label information is not limited to the KFWAdaBoost model, but can be widely used in the training and prediction of various classifiers, which means that the data set processed by the method of the present invention can be used by different fault detection models, improving the utilization efficiency of the data and the accuracy of diagnosis. In summary, the application of this method in the transformer fault detection task, through the acquisition and fusion of soft label features and the construction of a similarity matrix, not only improves the accuracy of fault detection, but also enhances the model's ability to learn and classify data. The wide applicability of this method provides strong technical support for the stable operation and fault prevention of power systems.

[0073] See also Figure 5 The second embodiment of the present invention provides a transformer fault detection device based on KFWAdaBoost, which includes:

[0074] A preprocessing unit 201 is used to obtain original training data, preprocess the original training data, and generate a model training set;

[0075] A clustering unit 202 is used to perform clustering and splicing preprocessing on the model training set and the original feature matrix using a KPP algorithm to generate a training feature matrix;

[0076] A training unit 203, used to train the FWAdaBoost model according to the training feature matrix and generate a training result;

[0077] A soft label feature learning unit 204 is used to perform soft label feature learning based on the training result and the FWAdaBoost model, generate soft label features, and perform fusion processing on the soft label features to obtain a classification feature enhanced data set;

[0078] An evaluation unit 205 is used to perform performance evaluation processing on the classification feature enhanced data set, and when the number of base classifiers reaches a preset value, the soft label feature learning process is terminated;

[0079] The detection unit 206 is used to obtain transformer data to be detected, input the transformer data into the trained FWAdaBoost model, perform model prediction based on the classification feature enhanced data set, and generate a fault detection result.

[0080] The third embodiment of the present invention provides a transformer fault detection device based on KFW AdaBoost, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the transformer fault detection method based on KFW AdaBoost as described in any one of the above.

[0081] A fourth embodiment of the present invention provides a readable storage medium storing a computer program, wherein the computer program can be executed by a processor of a device where the storage medium is located to implement a transformer fault detection method based on KFWAdaBoost as described in any one of the above.

[0082] Exemplarily, the above-mentioned devices and process steps can be implemented by a computer program. The computer program can be divided into one or more units. The one or more units are stored in the memory and executed by the processor to complete the present invention.

[0083] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0084] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the present invention by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0085] Wherein, if the electronic device or printer integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0086] It should be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art may understand and implement it without paying any creative effort.

[0087] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A transformer fault detection method based on KFWAdaBoost, characterized in that: include: Obtaining original training data, preprocessing the original training data, and generating a model training set; Using the KPP algorithm to perform clustering and splicing preprocessing on the model training set and the original feature matrix to generate a training feature matrix; Training the FWAdaBoost model according to the training feature matrix to generate training results; Based on the training results and the FWAdaBoost model, soft label feature learning is performed to generate soft label features, and the soft label features are fused to obtain a classification feature enhanced data set; Performing performance evaluation processing on the classification feature enhanced data set, and ending the soft label feature learning process when the number of base classifiers reaches a preset value; Obtain transformer data to be detected, input the transformer data into the trained FWAdaBoost model, perform model prediction based on the classification feature enhanced data set, and generate fault detection results; Based on the training results and the FWAdaBoost model, soft label feature learning is performed to generate soft label features, and the soft label features are fused to obtain a classification feature enhanced data set, specifically: Inputting the training results into the FWAdaBoost model, the base classifier in the FWAdaBoost model screens out representative samples in each category; Performing calculations on each sample in the model training set and each representative sample to generate a degree of membership; Integrate all membership degrees to generate soft label features, and fuse the soft label features with the original feature matrix to obtain a classification feature enhanced data set; The performance evaluation process includes average accuracy and average ranking. The average accuracy is used to evaluate the ability of the FWAdaBoost model to correctly classify. The calculation formula of the average accuracy is: , where C is the number of categories, is a sample correctly classified as the positive class, For samples correctly classified as negative classes, is a positive sample that is misclassified as a negative class, is a positive sample correctly classified as a negative class; The average ranking is used to evaluate the average performance ranking of different classifiers on multiple data sets. The calculation formula of the average ranking is: , where D is the number of data sets, The ranking is determined based on performance indicators.

2. The transformer fault detection method based on KFWAdaBoost according to claim 1 is characterized in that: The preprocessing includes feature encoding processing, feature scaling processing, label encoding and Z-score normalization processing.

3. The transformer fault detection method based on KFWAdaBoost according to claim 1 is characterized in that: The KPP algorithm is used to perform clustering and splicing preprocessing on the model training set and the original feature matrix to generate a training feature matrix, specifically: Using the KPP algorithm with the Rand coefficient as the clustering value k, clustering the model training set is performed to obtain the cluster center; Calculate each sample data in the model training set and the cluster center, calculate the corresponding similarity, and obtain a similarity matrix; An original feature matrix is ​​obtained, and the original feature matrix is ​​concatenated with the similarity matrix to generate a training feature matrix.

4. The transformer fault detection method based on KFWAdaBoost according to claim 1 is characterized in that: The FWAdaBoost model is trained according to the training feature matrix to generate training results, specifically: Iteratively input the training feature matrix into the FWAdaBoost model for training, and determine whether a training termination condition is met, wherein the training termination condition is that the number of iterated base classifiers reaches a set number; If yes, generate training results; If not, continue to iteratively input the training feature matrix into the FWAdaBoost model for training.

5. A transformer fault detection device based on KFWAdaBoost, characterized in that: The transformer fault detection method based on KFWAdaBoost as claimed in any one of claims 1 to 4 is applied, comprising: A preprocessing unit, used to obtain original training data, preprocess the original training data, and generate a model training set; A clustering unit, used for performing clustering and splicing preprocessing on the model training set and the original feature matrix using a KPP algorithm to generate a training feature matrix; A training unit, used to train the FWAdaBoost model according to the training feature matrix and generate a training result; A soft label feature learning unit, used to perform soft label feature learning based on the training result and the FWAdaBoost model, generate soft label features, and perform fusion processing on the soft label features to obtain a classification feature enhanced data set; An evaluation unit, used to perform performance evaluation processing on the classification feature enhanced data set, and when the number of base classifiers reaches a preset value, end the soft label feature learning process; The detection unit is used to obtain transformer data to be detected, input the transformer data into the trained FWAdaBoost model, perform model prediction based on the classification feature enhanced data set, and generate a fault detection result.

6. A transformer fault detection device based on KFWAdaBoost, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the transformer fault detection method based on KFWAdaBoost as described in any one of claims 1 to 4 is implemented.

7. A readable storage medium, characterized in that: A computer program is stored, and the computer program can be executed by a processor of the device where the storage medium is located to implement the transformer fault detection method based on KFWAdaBoost as described in any one of claims 1 to 4.

Images