A transformer fault diagnosis method and device based on a semi-supervised self-training model

Abstract

The application relates to a transformer fault diagnosis method and device based on a semi-supervised self-training model, content data of characteristic gases in various operating states of a transformer are acquired as sample data to generate an initial fault diagnosis sample set, and the sample data is linearly normalized; feature extraction is performed on the normalized sample data, and the sample data is divided into a labeled data set, an unlabeled data set, a verification set and a test set; artificial synthetic points are combined in an improved semi-supervised self-training SMOTE model; the labels of the labeled data points are assigned to the artificial synthetic points and the nearest unlabeled data points to generate a first labeled data set; the first labeled data set is input into a classifier model for training, and the performance of the classifier model is verified by using the verification set to obtain an optimal classifier model; and the test set is input into the optimal classifier model to obtain a classification result. The application embodiment solves the problem of few labeled samples in transformer fault diagnosis and improves the classification efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of transformer fault diagnosis, and in particular to a transformer fault diagnosis method and device based on a semi-supervised self-training model. BACKGROUND

[0002] Power transformers are one of the most important devices in power systems, and their operation reliability directly affects the safety and stability of power systems. The state detection and fault diagnosis of transformers have become the main direction of current research. Among numerous detection methods, the dissolved gas analysis (DGA) technology can effectively diagnose the latent faults of oil-immersed power transformers, which diagnoses the operation state of the transformer by analyzing the composition and content of the dissolved gas in the transformer oil. Since DGA is an online monitoring method, it has become the mainstream fault diagnosis method for oil-filled power equipment.

[0003] At present, most of the diagnosis methods using DGA data for analysis are supervised methods. The information used is limited to only a small amount of labeled data. In actual life, it is not only difficult to obtain labeled data, but also time-consuming and laborious to obtain a small amount of labeled data. In this case, the effect of the classifier trained is often not ideal. Therefore, the traditional supervised algorithm cannot effectively diagnose the faults of transformers. SUMMARY

[0004] The present application provides a transformer fault diagnosis method and device based on a semi-supervised self-training model. The technical solution of the present application is as follows:

[0005] According to a first aspect of an embodiment of the present application, a transformer fault diagnosis method based on a semi-supervised self-training model is provided, comprising:

[0006] Obtaining the content data of characteristic gases in multiple operating states of a transformer as sample data to generate an initial fault diagnosis sample set, and performing linear normalization on the sample data so that the sample data is mapped to [0, 1] to form normalized sample data;

[0007] Performing feature extraction on the normalized sample data to generate a first fault diagnosis sample set;

[0008] Dividing the normalized sample data in the first fault diagnosis sample set into a labeled data set, an unlabeled data set, a validation set, and a test set;

[0009] Inputting the normalized sample data in the labeled data set and the unlabeled data set into a semi-supervised self-training SMOTE model, combining artificial synthetic points based on a label data point and the nearest same-class label data point and the nearest unlabeled data point of the label data point.

[0010] assigning a label of the label data point to the synthetic point and a nearest unlabeled data point of the label data point, and generating a first label data set;

[0011] inputting the first label data set into a classifier model for training, and verifying performance of the classifier model by using the verification set to obtain an optimal classifier model;

[0012] inputting the test set into the optimal classifier model to obtain a classification result.

[0013] Optionally, the content data of the characteristic gas in the transformer under multiple operating states is acquired as sample data to generate an initial fault diagnosis sample set, and specifically includes:

[0014] The content of the characteristic gas in the transformer oil under various operating states is collected in real time to form a fault diagnosis sample set;

[0015] A rule of the fault characteristic gas is constructed by referring to a relevant ratio method, and a relevant ratio of the characteristic gas and an overall proportion between the gases are selected as a candidate feature set, wherein the total hydrocarbon content is represented by a unit CH.

[0016] Optionally, the characteristic gas includes hydrogen H2, methane CH4, ethane C2H6, ethylene C2H4, and acetylene C2H2, and components of the fault characteristic gas include H2, CH4, C2H6, C2H4, C2H2, CH, C2H2 / H2, C2H2 / C2H4, C2H2 / C2H6, C2H2 / CH4, C2H4 / C2H6, C2H4 / CH4, C2H4 / H2, C2H6 / CH4, C2H6 / H2, CH4 / H2, C2H2 / CH, H2 / CH, C2H4 / CH, CH4 / CH, C2H6 / CH, and CH4+C2H4 / CH.

[0017] Optionally, the synthetic point is composed of the label data point, a nearest same-label data point of the label data point, and a nearest unlabeled data point, and specifically includes:

[0018] The selected label data point is filtered by the classifier;

[0019] The label data point, the nearest same-label data point of the label data point, and the nearest unlabeled data point are input into a synthetic point synthesis algorithm:

[0020] X syn = x + 0.5 × (rand·(x nearest -x) + rand·(x nearest-un -x)·θ) (1)

[0021] wherein X syn characterizes an artificial point, x characterizes a selected labeled sample point, x nearest is the nearest neighbor homogeneous labeled sample point of x, x nearest-un is the nearest neighbor unlabeled point of x, and θ is a decay factor, rand characterizes a random factor, and r characterizes a ratio of the number of the labeled data set to the number of the unlabeled data set.

[0022] Optionally, the feature extraction on the normalized sample data to generate a first fault diagnosis sample set specifically comprises:

[0023] screening an optimal feature dimension using a random forest model;

[0024] selecting an optimal feature group combination under the optimal feature dimension;

[0025] saving the normalized sample data set of the optimal feature group combination as the first fault diagnosis sample set.

[0026] Optionally, the screening of the optimal feature dimension using the random forest model specifically comprises:

[0027] dividing the initial fault diagnosis sample set into a second training set and a second test set;

[0028] obtaining importance data of each input feature by using a random forest algorithm on the second training set, and sorting the importance data from large to small;

[0029] assuming that an optimal input feature dimension set is an empty set;

[0030] according to the feature importance sorting, using a sequential forward selection method, and selecting the first feature in the sorting without replacement to add to the optimal input feature dimension set;

[0031] training the random forest model on the second test set to obtain a classification accuracy rate of the optimal feature dimension set;

[0032] after traversing all the features, the optimal input feature dimension set is the feature dimension corresponding to the highest classification accuracy rate.

[0033] Optionally, the selecting of the optimal feature group combination under the optimal feature dimension specifically comprises:

[0034] selecting different feature combinations under the feature dimension to form different second training subsets;

[0035] training the random forest model using the different second training subsets to obtain different models;

[0036] Test different models by using the second test set to obtain a model with the highest classification accuracy;

[0037] Save a feature group combination of the second training subset of the model with the highest classification accuracy.

[0038] Optionally, the screening of the selected label data points by the classifier further comprises:

[0039] Select each label sample point in turn, train the classifier model by using the label data set, and determine the label of the nearest unlabeled data point of the selected label sample point.

[0040] If the label of the nearest unlabeled data point to the label data point is consistent with the label of the selected label data point, the selected label sample point is input into an artificial synthesis algorithm, and the label sample point is saved.

[0041] According to a second aspect of the embodiment of the present application, a transformer fault diagnosis device based on a semi-supervised self-training model is provided, which comprises:

[0042] An initial fault diagnosis sample set generation module is configured to obtain content data of characteristic gases in multiple operating states of a transformer as sample data to generate an initial fault diagnosis sample set, and perform linear normalization on the sample data so that the sample data is mapped to [0, 1].

[0043] A feature extraction module is configured to perform feature extraction on the normalized fault diagnosis sample data to generate a first fault diagnosis sample set.

[0044] A data set classification module is configured to divide the normalized sample data in the first fault diagnosis sample set into a label data set, an unlabeled data set, a validation set, and a test set.

[0045] An artificial synthesis point synthesis module is configured to input the normalized sample data in the label data set and the unlabeled data set into a semi-supervised self-training SMOTE algorithm, synthesize artificial synthesis points based on label data points, the nearest same-class label data points of the label data points, and the nearest unlabeled data points.

[0046] A first label data set generation module is configured to assign labels of the label data points to the artificial synthesis points and the nearest unlabeled data points of the label data points, and generate a first label data set.

[0047] A classifier model training module is configured to input the first label data set into a classifier model for training, and use the validation set to verify the performance of the classifier model to obtain an optimal classifier model.

[0048] The classification result testing module is configured to input the test set into the optimal classifier model to obtain a classification result.

[0049] According to a third aspect of the embodiments of the present application, a non-volatile storage device is provided, including a processor and a memory connected to the processor in communication;

[0050] The memory stores computer-executable instructions.

[0051] The processor executes the computer-executable instructions stored in the memory to implement the method provided in the first aspect.

[0052] According to a fourth aspect of the embodiments of the present application, a computer-readable storage medium is provided, which stores computer-executable instructions, and the computer-executable instructions are executed by a processor to implement the method provided in the first aspect.

[0053] According to a fifth aspect of the embodiments of the present application, a computer program product is provided, which includes a computer program, and the computer program is executed by a processor to implement the method provided in the first aspect.

[0054] Advantages:

[0055] The transformer fault diagnosis method and device based on a semi-supervised self-training model provided in the present application are related to a transformer fault diagnosis method based on a semi-supervised self-training model. The content data of characteristic gases in various operating states of a transformer are acquired as sample data to generate an initial fault diagnosis sample set, and the sample data is linearly normalized. The normalized sample data is subjected to feature extraction and is divided into a labeled data set, an unlabeled data set, a verification set, and a test set. Artificial synthetic points are synthesized in an improved semi-supervised self-training SMOTE model. The labels of the labeled data points are assigned to the artificial synthetic points and the nearest unlabeled data points to generate a first labeled data set. The first labeled data set is input into a classifier model for training, and the performance of the classifier model is verified by using the verification set to obtain an optimal classifier model. The test set is input into the optimal classifier model to obtain a classification result. The present application solves the problem of a small number of labeled samples in transformer fault diagnosis and improves the classification efficiency.

[0056] It should be understood that the foregoing general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0057] The accompanying drawings incorporated in and forming a part of the specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the application, but are not to be construed as setting forth any limitation of the application.

[0058] Figure 1 is a flowchart of a transformer fault diagnosis method based on a semi-supervised self-training model according to an example embodiment of the present application;

[0059] Figure 2 is a structural diagram of a transformer fault diagnosis device based on a semi-supervised self-training model according to an example embodiment of the present application. DETAILED DESCRIPTION

[0060] In order to make the ordinary people in the art better understand the technical solutions of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings.

[0061] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are only examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0062] The number of electronic devices and servers in the embodiments of the present application is not limited, and the server can provide services for multiple electronic devices at the same time. In the embodiments of the present application, the electronic device can be a personal computer (PC), a notebook computer or a smart phone, and can also be an all-in-one machine, a palm computer, a pad, a smart television playing terminal, a vehicle-mounted terminal or a portable device, etc. The operating system of the electronic device of the PC end, such as an all-in-one machine, can include but is not limited to Linux system, Unix system, Windows series system (such as Windows xp, Windows 7, etc.), Mac OS X system (operating system of Apple computer) and other operating systems. The operating system of the electronic device of the mobile end, such as a smart phone, can include but is not limited to Android system, IOS (operating system of Apple phone), Window system and other operating systems.

[0063] The server can be a stand-alone server, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery network (CDN) and basic cloud computing services such as big data and artificial intelligence platforms. The server can also be realized by a server cluster composed of multiple servers.

[0064] In daily transformer use, the probability of failure exists in a small amount, which belongs to a small sample classification problem. In actual life, not only is it difficult to obtain labeled data, but also it takes time and effort to obtain a small amount of labeled data. The effect of the classifier trained under such circumstances is often unsatisfactory. Therefore, the traditional supervised algorithm cannot effectively perform transformer fault diagnosis. On the contrary, a large amount of unlabeled data exists and is easy to obtain, almost without any cost, and the unlabeled data contains a large amount of available information.

[0065] Semi-supervised technology is an algorithm that uses limited labeled data and unlabeled data. It is particularly efficient and superior when the labeled data is limited.

[0066] Figure 1 According to an example embodiment, the present application provides a flowchart of a transformer fault diagnosis method based on a semi-supervised self-training model. As shown in Figure 1 The present application provides a transformer fault diagnosis method based on a semi-supervised self-training model, which includes the following steps:

[0067] S1, obtaining content data of characteristic gases under multiple operating states of the transformer as sample data to generate an initial fault diagnosis sample set, and performing linear normalization on the sample data so that the sample data is mapped to [0, 1] to form normalized sample data.

[0068] In some embodiments, the characteristic gases include hydrogen H2, methane CH4, ethane C2H6, ethylene C2H4, and acetylene C2H2.

[0069] In some embodiments, in step S1, the content data of the characteristic gases under multiple operating states of the transformer is obtained as sample data to generate an initial fault diagnosis sample set, which specifically includes:

[0070] Real-time collection of the content of the characteristic gases in the transformer oil under various operating states constitutes a fault diagnosis sample set; a rule of the fault characteristic gases is constructed by referring to a relevant ratio method, and a relevant ratio of the characteristic gases and an overall proportion between the gases are selected as a candidate feature set, wherein the total hydrocarbon content is represented by a unit CH.

[0071] In some embodiments, the fault characteristic gas components include H2, CH4, C2H6, C2H4, C2H2, CH, C2H2 / H2, C2H2 / C2H4, C2H2 / C2H6, C2H2 / CH4, C2H4 / C2H6, C2H4 / CH4, C2H4 / H2, C2H6 / CH4, C2H6 / H2, CH4 / H2, C2H2 / CH, H2 / CH, C2H4 / CH, CH4 / CH, C2H6 / CH, CH4+C2H4 / CH.

[0072] S2, performing feature extraction on the normalized fault diagnosis sample data to generate a first fault diagnosis sample set.

[0073] Specifically, the process of generating the first fault diagnosis sample set includes:

[0074] S21, screening an optimal feature dimension using a random forest model;

[0075] S22, selecting an optimal feature group combination under the optimal feature dimension;

[0076] S23, selecting the normalized sample data set of the saved optimal feature group combination as the first fault diagnosis sample set.

[0077] In step S21, the optimal feature dimension is screened using a random forest model, specifically including:

[0078] S211, dividing the initial fault diagnosis sample set into a second training set and a second test set;

[0079] S212, obtaining importance data of each input feature by using a random forest algorithm on the second training set, and sorting the importance from large to small;

[0080] S213, assuming that the optimal input feature dimension set is an empty set;

[0081] S214, using a sequential forward selection method to add the first sorted feature to the optimal input feature dimension set without replacement according to the feature importance sorting;

[0082] S215, training the random forest model on the second test set to obtain the classification accuracy of the optimal feature dimension set;

[0083] S216, after traversing all features, the optimal input feature dimension set is the feature dimension corresponding to the highest classification accuracy.

[0084] In step S22, the optimal feature group combination under the optimal feature dimension is selected, specifically including:

[0085] S221, selecting different feature combinations under the feature dimension to form different second training subsets;

[0086] S222, training the random forest model using different second training subsets to obtain different models;

[0087] S223, testing the different models using the second test set to obtain the model with the highest classification accuracy;

[0088] S224, save the feature combination of the second training subset of the model with the highest classification accuracy.

[0089] S3, divide the normalized sample data in the first fault diagnosis sample set into a labeled data set, an unlabeled data set, a validation set, and a test set.

[0090] S4, input the normalized sample data in the labeled data set and the unlabeled data set into a semi-supervised self-training SMOTE algorithm, and combine artificial synthetic points based on labeled data points, nearest same-labeled data points of the labeled data points, and nearest unlabeled data points.

[0091] In some embodiments, after dividing the normalized sample data in the first fault diagnosis sample set into a labeled data set, an unlabeled data set, a validation set, and a test set, selected labeled data points are filtered by the classifier;

[0092] The labeled data points, nearest same-labeled data points of the labeled data points, and unlabeled data points are input into an artificial synthetic point synthesis algorithm:

[0093] X syn = x + 0.5 * (rand * (x nearest -x) + rand * (x nearest-un -x) * theta) (1)

[0094] wherein X syn represents an artificial synthetic point, x represents a selected labeled sample point, x nearest is a nearest neighbor same-labeled sample point of x, x nearest-un is a nearest neighbor unlabeled point of x, theta is a decay factor, rand represents a random factor, and r represents the ratio of the number of the labeled data set to the number of the unlabeled data set.

[0095] S5, assign the label of the labeled data point to the artificial synthetic point and the nearest unlabeled data point of the labeled data point, and generate a first labeled data set.

[0096] S6, input the first labeled data set into a classifier model for training, and verify the performance of the classifier model using the validation set to obtain an optimal classifier model.

[0097] In some embodiments, the classifier model is trained using a new label data set and validated using a validation set to verify its classification performance; the classifier model is trained using the last label data set and validated using a validation set to verify its classification performance; the classifier model performance is compared, the label data set and the unlabeled data set with better performance are saved, if the classifier model trained by the label data set this time has better performance, the iteration is repeated, otherwise, the iteration is terminated until the best classifier is determined.

[0098] S7, inputting the test set into the best classifier model to obtain a classification result.

[0099] Further, the selected label data points are screened by the classifier, further comprising:

[0100] Each label sample point is selected in turn, the classifier model is trained by the label data set, and the label of the nearest unlabeled data point of the selected label sample point is determined.

[0101] If the label of the nearest unlabeled data point to the label data point is consistent with the label of the selected label data point, the selected label sample point is input into an artificial synthesis algorithm, and the label sample point is saved.

[0102] In addition, the content data of the characteristic gas in multiple operating states of the transformer is obtained as sample data to generate an initial fault diagnosis sample set, and the sample data is linearly normalized to map the sample data to [0, 1] to form normalized sample data, and the normalization algorithm is:

[0103]

[0104] Wherein, x represents sample data, x min The minimum value of the sample data in the table, x max The maximum value of the sample data in the table, x new The normalized sample data, wherein x new ∈[0,1].

[0105] The application relates to a transformer fault diagnosis method based on a semi-supervised self-training model, content data of characteristic gases in various operating states of a transformer are acquired as sample data to generate an initial fault diagnosis sample set, and the sample data is linearly normalized; feature extraction is performed on the normalized sample data, and the sample data is divided into a labeled data set, an unlabeled data set, a verification set and a test set; artificial synthetic points are synthesized in an improved semi-supervised self-training SMOTE model; the labels of the labeled data points are assigned to the artificial synthetic points and the nearest unlabeled data points to generate a first labeled data set; the first labeled data set is input into a classifier model for training, and the performance of the classifier model is verified by using the verification set to obtain an optimal classifier model; and the test set is input into the optimal classifier model to obtain a classification result. The application embodiment solves the problem of few labeled samples in transformer fault diagnosis and improves the classification efficiency.

[0106] Figure 2 A structural schematic diagram of a transformer fault diagnosis device based on a semi-supervised self-training model is provided for an exemplary embodiment of the application. The transformer fault diagnosis device based on the semi-supervised self-training model provided by the application embodiment can execute the processing flow provided by the transformer fault diagnosis method based on the semi-supervised self-training model. As shown in the figure, the transformer fault diagnosis device based on the semi-supervised self-training model provided by the application 20 comprises: Figure 2

[0107] An initial fault diagnosis sample set generation module 201 is configured to acquire content data of characteristic gases in various operating states of a transformer as sample data to generate an initial fault diagnosis sample set, and perform linear normalization on the sample data so that the sample data is mapped to between 0 and 1.

[0108] A feature extraction module 202 is configured to perform feature extraction on the fault diagnosis sample data after normalization to generate a first fault diagnosis sample set.

[0109] A data set classification module 203 is configured to divide the normalized sample data in the first fault diagnosis sample set into a labeled data set, an unlabeled data set, a verification set and a test set.

[0110] An artificial synthetic point synthesis module 204 is configured to input the normalized sample data in the labeled data set and the unlabeled data set into a semi-supervised self-training SMOTE algorithm, and synthesize artificial synthetic points based on labeled data points, the nearest same-labeled data points of the labeled data points and the nearest unlabeled data points.

[0111] A first labeled data set generation module 205 is configured to assign the labels of the labeled data points to the artificial synthetic points and the nearest unlabeled data points of the labeled data points, and generate a first labeled data set.​

[0112] The classifier model training module 206 is configured to input the first label data set into a classifier model for training, and verify the performance of the classifier model by using the verification set to obtain an optimal classifier model.

[0113] The classification result testing module 207 is configured to input the test set into the optimal classifier model to obtain a classification result.

[0114] The apparatus provided by the embodiments of the present application can be specifically used for executing the above-mentioned Figure 1 The schemes provided by the method embodiments correspond to the above-mentioned schemes, and the specific functions and the technical effects that can be achieved will not be repeated here.

[0115] The embodiments of the present application further provide a non-volatile storage device, comprising a processor and a memory connected with the processor in communication;

[0116] The memory stores computer execution instructions;

[0117] The processor executes the computer execution instructions stored in the memory to implement the schemes provided by any of the above-mentioned method embodiments, and the specific functions and the technical effects that can be achieved will not be repeated here. The electronic device can be the server mentioned above.

[0118] The embodiments of the present application further provide a computer readable storage medium, which stores computer execution instructions, and the computer execution instructions are executed by a processor to implement the schemes provided by any of the above-mentioned method embodiments, and the specific functions and the technical effects that can be achieved will not be repeated here.

[0119] The embodiments of the present application further provide a computer program product, which comprises a computer program stored in a readable storage medium, and at least one processor of an electronic device can read the computer program from the readable storage medium, and the at least one processor executes the computer program to make the electronic device execute the schemes provided by any of the above-mentioned method embodiments, and the specific functions and the technical effects that can be achieved will not be repeated here.

[0120] The application scenarios described in the embodiments of the present application are used to more clearly illustrate the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided by the embodiments of the present application. It can be known by those skilled in the art that, with the appearance of new application scenarios, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems.

[0121] Those skilled in the art can understand that each aspect of the present application can be implemented as a system, a method or a program product. Therefore, each aspect of the present application can be embodied in a form of an entirely hardware embodiment, an entirely software embodiment (including firmware, microcode, etc.), or an embodiment combining software and hardware aspects, which can be generally referred to as a "circuit", a "module" or a "system".

[0122] In some possible implementation, the electronic device according to the present application can include at least one processor, and at least one memory. Wherein, the memory stores program codes, when the program codes are executed by the processor, the processor executes the operation data management method according to various exemplary embodiments of the present application described above in the specification. For example, the processor can execute the steps as in the operation data management method.

[0123] It should be noted that although several units or sub-units of the apparatus are mentioned in the above detailed description, such division is merely exemplary and not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided into units for embodiment.

[0124] In addition, although the operations of the method according to the present application are described in a particular order in the accompanying drawings, this does not require or imply that the operations must be performed in that particular order, or that all of the illustrated operations must be performed to achieve desirable results. Additionally or alternatively, certain steps can be omitted, combined, performed simultaneously, or performed separately from other steps.

[0125] Those skilled in the art will appreciate that embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk memory, CD-ROM, optical memory, etc.) containing computer-usable program codes.

[0126] The computer program instructions can also be loaded onto a computer or other programmable image processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0127] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable image processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0128] The computer program instructions can also be loaded onto a computer or other programmable image processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0129] While the preferred embodiments of the application have been described, additional variations and modifications can be employed by those skilled in the art. Therefore, the appended claims intend to cover all such modifications and variations as fall within the true spirit and scope of the application.

[0130] Obviously, numerous modifications and variations of the present application are possible in light of the above teachings. It is therefore to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. A transformer fault diagnosis method based on a semi-supervised self-training model, characterized in that, The method comprises: obtaining content data of characteristic gases in multiple operating states of a transformer as sample data to generate an initial fault diagnosis sample set, and linearly normalizing the sample data so that the sample data is mapped to between 0 and 1 to form normalized sample data; extracting features from the normalized sample data to generate a first fault diagnosis sample set; dividing the normalized sample data in the first fault diagnosis sample set into a labeled data set, an unlabeled data set, a validation set, and a test set; inputting the normalized sample data in the labeled data set and the unlabeled data set into a semi-supervised self-training SMOTE model, combining artificial synthetic points based on a labeled data point and a same-class labeled data point nearest to the labeled data point and a nearest unlabeled data point; assigning a label of the labeled data point to the artificial synthetic points and the nearest unlabeled data point to the labeled data point, and generating a first labeled data set; inputting the first labeled data set into a classifier model for training, and verifying performance of the classifier model using the validation set to obtain an optimal classifier model; inputting the test set into the optimal classifier model to obtain a classification result; the combining of the artificial synthetic points based on the labeled data point and the same-class labeled data point nearest to the labeled data point and the nearest unlabeled data point comprises: selecting a labeled data point through the classifier; inputting the labeled data point and the same-class labeled data point nearest to the labeled data point and the nearest unlabeled data point into an artificial synthetic point synthesis algorithm: ， （1） where X syn characterizes the synthetic points, x characterizes the selected labeled sample points, x nearest is the nearest neighbor homogeneous labeled sample point of x nearest-un is the nearest neighbor unlabeled point of x, and θ is a decay factor, , rand characterizes a random factor, and r characterizes the ratio of the number of labeled data set to the number of unlabeled data set.

2. The method of claim 1, wherein, the obtaining of the content data of the characteristic gases in the multiple operating states of the transformer as the sample data to generate the initial fault diagnosis sample set comprises: real-time collection of content of the characteristic gases in transformer oil in various operating states to form a fault diagnosis sample set; reference to a relevant ratio method to construct rules of fault characteristic gases, selection of relevant ratios of the characteristic gases and overall proportions between the gases as a candidate feature set, wherein a total hydrocarbon content is represented by a unit CH.

3. The method of claim 2, wherein, wherein the characteristic gases include hydrogen H2, methane CH4, ethane C2H6, ethylene C2H4, and acetylene C2H2, and components of the fault characteristic gases include H2, CH4, C2H6, C2H4, C2H2, CH, C2H2 / H2, C2H2 / C2H4, C2H2 / C2H6, C2H2 / CH4, C2H4 / C2H6, C2H4 / CH4, C2H4 / H2, C2H6 / CH4, C2H6 / H2, CH4 / H2, C2H2 / CH, H2 / CH, C2H4 / CH, CH4 / CH, C2H6 / CH, and CH4+C2H4 / CH.

4. The method of claim 1, wherein, the extracting of features from the normalized sample data to generate the first fault diagnosis sample set comprises: selecting an optimal feature dimension using a random forest model; selecting an optimal feature component combination under the optimal feature dimension; selecting and saving a normalized sample data set of the optimal feature component combination as the first fault diagnosis sample set.

5. The method of claim 4, wherein, The optimal feature dimension is screened by using a random forest model, and specifically includes the following steps: The initial fault diagnosis sample set is divided into a second training set and a second test set; The importance data of each input feature is obtained by using a random forest algorithm on the second training set, and the importance is sorted from large to small; It is assumed that the optimal input feature dimension set is an empty set; According to the feature importance ranking, the sequence forward selection method is used to select the top-ranked features each time without replacement to add to the optimal input feature dimension set; The random forest model is trained on the second test set to obtain the classification accuracy of the optimal feature dimension set; After traversing all features, the optimal input feature dimension set is the feature dimension corresponding to the highest classification accuracy.

6. The method of claim 5, wherein, The optimal feature group combination under the optimal feature dimension is selected, and specifically includes the following steps: Different feature combinations are selected under the feature dimension to form different second training subsets; Different models are obtained by training the random forest model using different second training subsets; The highest classification accuracy model is obtained by testing different models using the second test set; The feature group combination of the second training subset of the highest classification accuracy model is saved.

7. The method of claim 1, wherein, The selected label data points are screened by the classifier, and further include the following steps: Each label sample point is selected in turn, the classifier model is trained by the label data set, and the label of the nearest unlabeled data point of the selected label sample point is determined; If the label of the nearest unlabeled data point to the label data point is consistent with the label of the selected label data point, the selected label sample point is input into the artificial synthesis algorithm, and the label sample point is saved.

8. A transformer fault diagnosis device based on a semi-supervised self-training model, characterized in that, The device includes: An initial fault diagnosis sample set generation module is configured to obtain content data of feature gases in multiple operating states of a transformer as sample data to generate an initial fault diagnosis sample set, and perform linear normalization on the sample data to map the sample data to [0, 1]; A feature extraction module is configured to perform feature extraction on the normalized sample data to generate a first fault diagnosis sample set; A data set classification module is configured to divide the normalized sample data in the first fault diagnosis sample set into a label data set, an unlabeled data set, a validation set, and a test set; An artificial synthesis point synthesis module is configured to input the normalized sample data in the label data set and the unlabeled data set into a semi-supervised self-training SMOTE model, and synthesize artificial synthesis points based on label data points, nearest same-class label data points, and nearest unlabeled data points; A first label data set generation module is configured to assign labels of the label data points to the artificial synthesis points and the nearest unlabeled data points of the label data points, and generate a first label data set; A classifier model training module is configured to input the first label data set into a classifier model for training, and verify the performance of the classifier model using the validation set to obtain an optimal classifier model; A classification result testing module is configured to input the test set into the optimal classifier model to obtain a classification result. The artificial synthetic point is composed of the label data point and the nearest same kind label data point and the nearest unlabel data point of the label data point, and specifically comprises: filtering the selected label data point through the classifier; inputting the label data point and the nearest same kind label data point and the nearest unlabel data point of the label data point into an artificial synthetic point synthesis algorithm: ， （1） where X syn characterizes the synthetic points, x characterizes the selected labeled sample points, x nearest is the nearest neighbor homogeneous labeled sample point of x, x nearest-un is the nearest neighbor unlabeled point of x, and θ is a decay factor, , rand characterizes a random factor, and r characterizes the ratio of the number of labeled data set to the number of unlabeled data set.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed by the processor to implement the method in any one of claims 1-7.

Images