A Transformer Fault Diagnosis Method, System, Device and Storage Medium Based on Attention Variational Autoencoder

Abstract

The present invention discloses a transformer fault diagnosis method, system, device and storage medium based on attention variational autoencoder. The method includes: obtaining historical concentration data of each characteristic gas of the transformer, historical oil temperature data and historical load data of the transformer; performing normalization processing on the historical data; constructing a self-attention variational autoencoder model; extracting fault features to obtain a fault feature training set, and constructing a probabilistic neural network fault classification model; performing normalization processing on the real-time monitored data, and using the processed real-time data as the input data of the self-attention variational autoencoder model to dynamically predict the dissolved gas concentration, obtaining the prediction deviation of the dissolved gas concentration. If the prediction deviation is greater than a preset threshold, it is used as a fault feature; using the obtained fault features, performing fault diagnosis through the probabilistic neural network fault classification model, and outputting the fault type. The present invention can improve the accuracy of dissolved gas prediction and the interpretability of the model, and accurately diagnose faults.

Description

Technical Field

[0001] The present invention relates to the field of power automation technology, and particularly to a transformer fault diagnosis method, system, device and storage medium based on attention variational autoencoder. Background Art

[0002] The best method for early diagnosis of transformer faults is mainly the dissolved gas analysis in oil. The precursors of transformer equipment faults can be captured by the concentration of dissolved characteristic gases in transformer oil. In practice, the dissolved gas analysis method in transformer oil is commonly used to monitor and judge the operation status of the transformer during oil sample collection, and to detect hidden defects or faults of the transformer in time. The principle is that during the long-term operation of an oil-filled power transformer, it will age and deteriorate under the action of electricity or heat, and a small amount of gas (generally called characteristic gas) will be dissolved in the transformer oil. Generally speaking, the amount of dissolved gas is a gradually stable accumulation process, but when a fault or abnormality occurs, the gas content value will change greatly.

[0003] Considering comprehensively the gas production principle of transformer characteristic gases, as well as the influence of transformer oil temperature and load on characteristic gases, using big data analysis algorithms, explore the mutual correlation relationship between dissolved characteristic gases in transformer oil and the correlation relationship between characteristic gases and transformer equipment load and top oil temperature, excavate the factors with greater correlation with the target gas to be predicted as input variables, accurately predict the future change of oil-dissolved gas concentration, effectively identify the latent faults of transformer equipment, and realize the advanced control of the equipment.

[0004] To realize the prediction of the concentration of dissolved gases in transformer oil, some data-driven prediction methods have emerged. These methods use historical data to establish a model to predict future data. However, there are the following problems: (1) The internal correlation between input variables is not considered, and variable parameters are directly stacked as model inputs, resulting in input redundancy and affecting the generalization ability of the model; (2) The correlation model between input variables and external influencing factors is not established, resulting in poor interpretability of the model; (3) The prediction target only considers the output reconstruction accuracy and ignores the data distribution of low-dimensional latent variable features.

[0005] As the operation time of the transformer increases, the data of characteristic gases also shows explosive growth. How to mine valuable information from a large amount of monitoring data has become an urgent problem to be solved. Traditional manual experience is difficult to quickly and accurately identify the fault types. Accurate fault diagnosis is of great significance for quickly solving fault problems and ensuring the safe and stable operation of the transformer. Summary of the Invention

[0006] Objective of the Invention: The objective of the present invention is to provide a transformer fault diagnosis method, system, device and storage medium based on attention variational autoencoder, which can improve the accuracy of dissolved gas prediction and the interpretability of the model, and accurately diagnose faults.

[0007] Technical Solution: A transformer fault diagnosis method based on attention variational autoencoder of the present invention includes:

[0008] Select characteristic gases from the dissolved gases in transformer oil, obtain the historical concentration data of each characteristic gas of the transformer, obtain the historical oil temperature data and historical load data of the transformer;

[0009] Perform normalization processing on the historical concentration data, historical oil temperature data and historical load data;

[0010] Construct an attention variational autoencoder model and use the normalized historical data to train the model. Obtain the correlation between the dissolved gas concentrations of the transformer through the attention variational autoencoder model, and extract its low-dimensional latent variable features; the loss term of the attention variational autoencoder model includes the reconstruction loss of the transformer dissolved gas concentration, the data distribution consistency loss and the attention prediction loss. This model makes the low-dimensional latent variable features of the model related to external influencing factors by introducing an attention mechanism;

[0011] Extract fault features from the fault database to obtain a fault feature training set, and construct an association network between the fault features and the fault types through a probabilistic neural network to obtain a probabilistic neural network fault classification model;

[0012] Perform normalization processing on the real-time monitored transformer load, oil temperature data and dissolved gas concentration. Use the normalized real-time data as the input data of the attention variational autoencoder model to dynamically predict the dissolved gas concentration, and obtain the prediction deviation of the dissolved gas concentration. If the prediction deviation is greater than the preset threshold, use it as a fault feature;

[0013] Use the fault features output by the attention variational autoencoder model to perform fault diagnosis through the probabilistic neural network fault classification model and output the fault type.

[0014] Further, select H2, CH4, C2H6, C2H4, C2H2, C0, C02 from the dissolved gases in transformer oil as characteristic gases, extract the historical concentration data of each characteristic gas of the transformer in a fixed period before the prediction date from the power OMS system, and extract the historical oil temperature and historical load data of the transformer equipment in a fixed period before the prediction date from the power OMS system.

[0015] Further, the normalization processing includes:

[0016] The centralization and variance normalization methods of data are adopted, and the specific process is as follows:

[0017]

[0018] Among them, x o , respectively represent the original value and the normalized value of the i-th input parameter; m(x) is the mean of the parameter, and s(x) is the standard deviation of the parameter.

[0019] Furthermore, the construction process of the attention variational autoencoder model is as follows:

[0020] Construct a variational autoencoder network. Among them, the encoder adopts a multi-layer perceptron structure, and the input is a variety of dissolved gas concentration samples of the transformer x = [H2, CH4, C2H6, C2H4, C2H2, CO, CO2] T , and after multi-layer non-linear mapping, the output features follow a normal distribution N(μ, σ), and the output z O_mean and z o_var correspond to the mean and covariance of this normal distribution respectively;

[0021] Use the reparameterization trick to sample N(μ, σ), so as to obtain the low-dimensional representation z o in the latent space, and its expression is:

[0022] z o = μ + σ·ε

[0023] In the formula: ε is a random variable obeying the standard normal distribution, that is, ε ∼ N(0, I);

[0024] The decoder structure is similar to that of the encoder. The low-dimensional representation z of the input encoder is reconstructed into the original transformer gas concentration through multi-layer non-linear mapping; the encoder and decoder are jointly optimized through the loss function. For the input x = [H2, CH4, C2H6, C2H4, C2H2, CO, CO2] T , the loss function of the variational autoencoder network

[0025]

[0026] can be described as: is the reconstruction loss term of the variational autoencoder network; is the Kullback-Leibler divergence loss term between the approximate probability distribution q φ (z|x) and the prior probability distribution p(z).

[0027] Furthermore, the input of the attention variational autoencoder model is the low-dimensional latent variable z y extracted by the encoder, zy From z y_mean and z y_var obtained through reparameterization, the model output is external influencing factors, and the external influencing factors include transformer load and oil temperature.

[0028] Furthermore, the loss terms of the attention variational autoencoder model include the reconstruction loss of transformer dissolved gas concentration, the data distribution consistency loss, and the attention prediction loss. Among them, the construction process of the loss terms is as follows:

[0029] The probability distributions of the low-dimensional latent variables z0 and z y follow independent multivariate Gaussian distributions. Therefore, the probability expression of the decoder is as follows:

[0030]

[0031] In the formula: De1 is the decoder, and De2 is the decoder based on the attention mechanism; θ and are the network parameters of De1 and De2 respectively;

[0032] Correspondingly, the probability expression of the encoder is as follows:

[0033] z ~ En(x) = q φ (z o , z y |z) = q φ (z o |x)q φ (z y |x)

[0034] In the formula: q φ (z y |x) is the approximate posterior probability distribution; En is the encoder; φ is the network parameter of En

[0035] Therefore, the total loss function of the model is as follows:

[0036]

[0037] In the formula: the first term is the reconstruction loss of transformer dissolved gas concentration, which is usually calculated using the mean square error between the predicted concentration and the actual concentration; the second term is the data distribution consistency loss, indicating the difference measure between the latent distribution q φ (z o |x) output by the encoder and the prior distribution p(z o ); the third term is also the data distribution consistency loss, indicating the difference measure between the latent distribution q φ (z y |x) output by the encoder and the prior distribution p(z yFor the difference metric between them, the Kullback-Leibler divergence is used as the measurement index; the fourth item is the attention prediction loss, which is usually calculated using the mean square error between the predicted transformer load and oil temperature and the actual transformer load and oil temperature; β1 and β2 are hyperparameters for weighing different loss terms.

[0038] Further, the expression of the prediction deviation is as follows:

[0039]

[0040] In the formula: x and are the actual value and the predicted value respectively;

[0041] Based on the same inventive concept, a transformer fault diagnosis system based on attention variational autoencoder of the present invention includes:

[0042] A data acquisition module, configured to select characteristic gases from the dissolved gases of transformer oil, obtain the historical concentration data of each characteristic gas of the transformer, obtain the historical oil temperature data and historical load data of the transformer;

[0043] A data preprocessing module, configured to perform normalization processing on the historical concentration data, historical oil temperature data and historical load data;

[0044] An attention variational autoencoder model construction module, configured to construct an attention variational autoencoder model and train the model using the normalized historical data, obtain the correlation between the dissolved gas concentrations of the transformer through the attention variational autoencoder model, and extract its low-dimensional latent variable features; the loss terms of the attention variational autoencoder model include the reconstruction loss of the transformer dissolved gas concentration, the data distribution consistency loss and the attention prediction loss, and the model makes the low-dimensional latent variable features of the model have a correlation with external influencing factors by introducing an attention mechanism;

[0045] A probability neural network fault classification model construction module, configured to extract fault features from a fault database to obtain a fault feature training set, and construct an association network between the fault features and the fault types through a probability neural network to obtain a probability neural network fault classification model;

[0046] A fault feature extraction module, configured to perform normalization processing on the real-time monitored transformer load, oil temperature data and dissolved gas concentration, use the normalized real-time data as the input data of the attention variational autoencoder model, dynamically predict the dissolved gas concentration, obtain the prediction deviation of the dissolved gas concentration, and if the prediction deviation is greater than a preset threshold, use it as a fault feature and output it;

[0047] A fault diagnosis module, which is used to utilize the fault features output by the attention variational autoencoder model and perform fault diagnosis through a probabilistic neural network fault classification model to output the fault type.

[0048] Based on the same inventive concept, a transformer fault diagnosis device based on attention variational autoencoder of the present invention includes a processor and a memory. Computer instructions are stored in the memory, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the electronic device implements the steps of the above-mentioned transformer fault diagnosis method based on attention variational autoencoder.

[0049] Based on the same inventive concept, a computer-readable storage medium of the present invention stores a computer program thereon. When the program is executed by a processor, the steps of the above-mentioned transformer fault diagnosis method based on attention variational autoencoder are implemented.

[0050] Advantageous effects: Compared with the prior art, the remarkable technical effects of the present invention are as follows:

[0051] The variational autoencoder is used to automatically learn the correlation between different oil-gas concentrations, extract low-dimensional representations, and eliminate the redundancy of input variables.

[0052] The attention mechanism is introduced to establish an interpretable association model between the input and external influencing factors (transformer load and oil temperature), strengthen the model's ability to capture key features in the input data, and improve the model's representation ability and prediction accuracy.

[0053] The loss term of the attention variational autoencoder model comprehensively combines different losses to improve the model's generalization ability and prediction interpretability, accurately predict the change trend of the dissolved gas concentration, and provide highly reliable fault features for transformer fault diagnosis.

[0054] Integrating the two functions of prediction and diagnosis into one, a complete transformer fault diagnosis system is constructed, which can realize the intelligence from data acquisition to fault diagnosis. Description of the Drawings

[0055] Figure 1 is a schematic flow chart of a transformer fault diagnosis method based on attention variational autoencoder disclosed in an embodiment of the present invention;

[0056] Figure 2 is a schematic structural diagram of an attention variational autoencoder model disclosed in an embodiment of the present invention;

[0057] Figure 3 is a schematic structural diagram of a transformer fault diagnosis system based on attention variational autoencoder disclosed in an embodiment of the present invention;

[0058] Figure 4It is a schematic structural diagram of a transformer fault diagnosis device based on attention variational autoencoder disclosed in an embodiment of the present invention. Detailed implementation manners

[0059] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. Those skilled in the art will understand that the objectives and advantages that can be achieved by the present invention are not limited to the specific beneficial effects described above, and the above and other objectives that the present invention can achieve will be more clearly understood from the following detailed description.

[0060] Those of ordinary skill in the art should understand that the various exemplary components, systems, and methods described in combination with the embodiments disclosed in the present invention can be implemented in hardware, software, or a combination of both. Specifically, whether to execute in hardware or software depends on the specific application and design and tree conditions of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0061] When the present invention mentions "embodiment", it means that the specific features, structures, or characteristics described in combination with the embodiment can be included in at least one embodiment of the present invention. The appearance of this phrase at various positions in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0062] Embodiment 1

[0063] Please refer to Figure 1 , Figure 1 It is a schematic flowchart of a transformer fault diagnosis method based on attention variational autoencoder disclosed in an embodiment of the present invention. Among them, Figure 1 The described transformer fault diagnosis method based on attention variational autoencoder is applied in a power system, such as for transformer fault diagnosis, etc., which is not limited in the embodiments of the present invention. As Figure 1 shown, the transformer fault diagnosis method based on attention variational autoencoder may include the following operations:

[0064] S1. Select characteristic gases from the dissolved gases in the transformer oil, obtain the historical concentration data of each characteristic gas of the transformer, obtain the historical oil temperature data and historical load data of the transformer.

[0065] In this embodiment, H2 (hydrogen), CH4 (methane), C2H6 (ethane), C2H4 (ethylene), C2H2 (acetylene), CO (carbon monoxide), and CO2 (carbon dioxide) are selected as characteristic gases from the dissolved gases in transformer oil. Historical concentration data of each characteristic gas of the transformer for the 30 days before the prediction date are extracted from the power OMS system (grid dispatching operation management system), and historical oil temperature and historical load data of the transformer equipment for the 30 days before the prediction date are extracted from the power OMS system.

[0066] S2. Normalize the historical concentration data, historical oil temperature data, and historical load data.

[0067] The collected historical data needs to be preprocessed by data normalization to eliminate the influence of the measurement dimensions of different parameters on the clustering results. Z-score standardization is adopted, that is, the data centering and variance normalization method. The specific process is as follows:

[0068]

[0069]

[0070] Among them, x i , respectively represent the original value and the normalized value of the i-th input parameter; m(x) is the mean of the parameter, and s(x) is the standard deviation of the parameter.

[0071] S3. Construct an attention variational autoencoder model and train the model using the normalized historical data. Obtain the correlation between the dissolved gas concentrations of the transformer through the attention variational autoencoder model, and extract its low-dimensional latent variable features; the loss term of the attention variational autoencoder model includes the reconstruction loss of the transformer dissolved gas concentration, the data distribution consistency loss, and the attention prediction loss. This model makes the low-dimensional latent variable features of the model related to external influencing factors (transformer load and oil temperature) by introducing an attention mechanism.

[0072] In this embodiment, the construction process of the attention variational autoencoder model is as follows:

[0073] Construct a variational autoencoder network. Among them, the encoder adopts a multi-layer perceptron structure, and the input is a sample of various dissolved gas concentrations of the transformer x = [H2, CH4, C2H6, C2H4, C2H2, CO, CO2] T , after multiple non-linear mappings, the output features follow a normal distribution N(μ, σ), and the output z O_mean and z o_var correspond to the mean and covariance of this normal distribution respectively;

[0074] Use the reparameterization technique to sample N(μ,σ) to obtain a low-dimensional representation z of the latent space o , whose expression is:

[0075] z o =μ+σ·ε

[0076] Where: ε is a random variable that obeys the standard normal distribution, that is, ε~N(0,I);

[0077] The decoder structure is similar to the encoder. The low-dimensional representation z of the input encoder is used to reconstruct the original transformer gas concentration through multi-layer nonlinear mapping. The encoder and decoder are jointly optimized through the loss function. For the input x = [H2, CH4, C2H6, C2H4, C2H2, CO, CO2] T , the loss function of the variational autoencoder network It can be described as:

[0078]

[0079] Where: is the reconstruction loss term of the variational autoencoder network; is the approximate probability distribution q φ The Kullback-Leibler divergence loss term between (z|x) and the prior probability distribution p(z).

[0080] In this embodiment, the attention variational autoencoder model input is the low-dimensional latent variable z extracted by the encoder y , z y By z y_mean and z y_var It is obtained by re-parameterization; through the decoding network based on the attention mechanism, the model output is external influencing factors such as transformer load and oil temperature.

[0081] In this embodiment, the loss term of the attention variational autoencoder model includes the reconstruction loss of the transformer dissolved gas concentration, the data distribution consistency loss and the attention prediction loss, wherein the construction process of the loss term is as follows:

[0082] The network structure of the attention variational autoencoder model is as follows Figure 2 As shown. Low-dimensional latent variable z o and z y The probability distribution of follows a mutually independent multivariate Gaussian distribution. Therefore, the probability expression of the decoder is as follows:

[0083]

[0084] Where: De1 is the decoder, De2 is the decoder based on the attention mechanism; θ and These are the network parameters of De1 and De2 respectively;

[0085] Accordingly, the probability expression of the encoder is as follows:

[0086] z ∼ En(x) = q φ (z o , z y |x) = q φ (z o |x)q φ (z y |x)

[0087] Where: q φ (z y |x) is the approximate posterior probability distribution; En is the encoder; φ is the network parameter of En

[0088] Therefore, the total loss function of the model is as follows:

[0089]

[0090] Where: the first term is the reconstruction loss of the transformer dissolved gas concentration, usually calculated using the mean square error between the predicted concentration and the actual concentration; the second term is the data distribution consistency loss, indicating the difference measure between the latent distribution q φ (z o |x) output by the encoder and the prior distribution p(z o ); the third term is also the data distribution consistency loss, indicating the difference measure between the latent distribution q φ (z y |x) output by the encoder and the prior distribution p(z y ), and the Kullback-Leibler divergence is used as the measurement index for both; the fourth term is the attention prediction loss, usually calculated using the mean square error between the predicted transformer load and oil temperature and the actual transformer load and oil temperature; β1 and β2 are hyperparameters for weighing different loss terms.

[0091] S4. Extract fault features from the fault database to obtain a fault feature training set, and construct an association network between the fault features and the fault types through a probabilistic neural network to obtain a probabilistic neural network fault classification model.

[0092] S5. Normalize the real-time monitored transformer load, oil temperature data, and dissolved gas concentration, use the normalized real-time data as the input data of the attention variational autoencoder model, dynamically predict the dissolved gas concentration, obtain the prediction deviation of the dissolved gas concentration, and if the prediction deviation is greater than the preset threshold, use it as a fault feature and output it.

[0093] Among them, the prediction deviation of the dissolved gas concentration is the deviation between the predicted value and the actual value, and the expression of the prediction deviation is as follows:

[0094]

[0095] In the formula: x and are the actual value and the predicted value respectively.

[0096] S6. Use the fault features output by the attention variational autoencoder model for fault diagnosis through the probabilistic neural network fault classification model, and output the fault type.

[0097] Among them, the fault categories include: high-temperature overheating, medium-temperature overheating, low-temperature overheating, partial discharge, low-energy discharge, and high-energy discharge.

[0098] The method provided by the present invention can effectively improve the interpretability and generalization ability of the prediction model, accurately predict the change trend of the oil-soluble gas concentration, provide highly credible fault features based on the prediction deviation, and realize the accurate diagnosis of transformer faults, providing a reliable basis for the health management and maintenance decision-making of transformers.

[0099] Embodiment 2

[0100] Please refer to Figure 3 , Figure 3 which is a schematic structural diagram of a transformer fault diagnosis system based on attention variational autoencoder disclosed in an embodiment of the present invention. This system can realize transformer fault diagnosis, specifically including:

[0101] A data acquisition module, which is used to select characteristic gases from the dissolved gases of transformer oil, obtain the historical concentration data of each characteristic gas of the transformer, obtain the historical oil temperature data and historical load data of the transformer;

[0102] A data preprocessing module, which is used to perform normalization processing on the historical concentration data, historical oil temperature data and historical load data;

[0103] An attention variational autoencoder model construction module, which is used to construct an attention variational autoencoder model and train the model using the normalized historical data, obtain the correlation between the dissolved gas concentrations of the transformer through the attention variational autoencoder model, and extract its low-dimensional latent variable features; the loss term of the attention variational autoencoder model includes the reconstruction loss of the transformer dissolved gas concentration, the data distribution consistency loss and the attention prediction loss. This model makes the low-dimensional latent variable features of the model have a correlation with external influencing factors by introducing an attention mechanism;

[0104] The probability neural network fault classification model construction module is used to extract fault features from the fault database to obtain a fault feature training set, and construct an association network between the fault features and the fault types through the probability neural network to obtain a probability neural network fault classification model;

[0105] The fault feature extraction module is used to normalize the real-time monitored transformer load, oil temperature data, and dissolved gas concentration, use the normalized real-time data as the input data of the attention variational autoencoder model, dynamically predict the dissolved gas concentration, obtain the prediction deviation of the dissolved gas concentration, and if the prediction deviation is greater than the preset threshold, use it as a fault feature and output it;

[0106] The fault diagnosis module is used to use the fault features output by the attention variational autoencoder model to perform fault diagnosis through the probability neural network fault classification model and output the fault type.

[0107] In an optional embodiment, the transformer fault diagnosis method based on the attention variational autoencoder includes: a) obtaining the historical concentration data of each characteristic gas, the historical oil temperature data, and the historical load data of the transformer; b) normalizing the historical concentration data, historical oil temperature data, and historical load data; c) constructing an attention variational autoencoder model; d) extracting fault features from the fault database to obtain a fault feature training set, and constructing an association network between the fault features and the fault types through the probability neural network to obtain a probability neural network fault classification model; e) normalizing the real-time monitored transformer load, oil temperature data, and dissolved gas concentration, using the normalized real-time data as the input data of the attention variational autoencoder model, dynamically predicting the dissolved gas concentration, obtaining the prediction deviation of the dissolved gas concentration, and if the prediction deviation is greater than the preset threshold, using it as a fault feature; f) using the obtained fault features to perform fault diagnosis through the probability neural network fault classification model and output the fault type.

[0108] Embodiment 3

[0109] Please refer to Figure 4 , Figure 4 which is a schematic structural diagram of a transformer fault diagnosis device disclosed in an embodiment of the present invention. Among them, Figure 4 the described device can be applied to the power system, such as for transformer fault diagnosis, etc., and the embodiments of the present invention do not make limitations.

[0110] Such as Figure 4As shown, the device may include a processor and a memory. Computer instructions are stored in the memory, and the processor is configured to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the electronic device implements the steps of the method described in the above embodiments and can achieve the same technical effects as the above method.

[0111] The memory may include a computer system readable medium in the form of volatile memory, such as random access memory (RAM) and / or cache memory. The device may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the memory may be used to read and write a non-removable, non-volatile magnetic medium (commonly referred to as a "hard disk drive"). A program / utility having a set (at least one) of program modules may be stored, for example, in the memory. Such program modules include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Implementations of a network environment may be included in each or some combination of these examples. The program modules generally execute the functions and / or methods in the embodiments described in the present invention.

[0112] The processor executes various functional applications and data processing by running the programs stored in the memory, such as implementing the method provided in Embodiment 1 of the present invention.

[0113] Embodiment 4

[0114] Embodiment 4 of the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the steps of the method described in the above embodiments and can achieve the same technical effects as the above method.

[0115] The computer storage medium of the embodiments of the present invention may employ any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (a non-exhaustive list) of the computer-readable storage medium include: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device.

[0116] A computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal may take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. The computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device.

[0117] The program code contained on a computer-readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wire, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0118] The computer program code for performing the operations of the present invention may be written in one or more programming languages or combinations thereof. The programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, executed as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., by using an Internet service provider to connect through the Internet).

[0119] Of course, a storage medium containing computer-executable instructions provided by an embodiment of the present invention, the computer-executable instructions are not limited to the above method operations, and may also execute related operations in the methods provided by any embodiment of the present invention.

[0120] The specific embodiments described above further elaborate on the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A transformer fault diagnosis method based on attention variational autoencoder, characterized in that: include: Select characteristic gases from the dissolved gases in transformer oil, obtain historical concentration data of each characteristic gas of the transformer, and obtain historical oil temperature data and historical load data of the transformer; Normalize historical concentration data, historical oil temperature data and historical load data; An attention variational autoencoder model is constructed and trained using normalized historical data. The correlation between the concentrations of dissolved gases in transformers is obtained through the attention variational autoencoder model, and its low-dimensional latent variable characteristics are extracted. The loss terms of the attention variational autoencoder model include the reconstruction loss of the concentration of dissolved gases in transformers, the data distribution consistency loss, and the attention prediction loss. The model introduces an attention mechanism to make the low-dimensional latent variable characteristics of the model correlated with external influencing factors. The loss term of the attention variational autoencoder model includes the reconstruction loss of the transformer dissolved gas concentration, the data distribution consistency loss and the attention prediction loss, wherein the construction process of the loss term is as follows: The low-dimensional latent variable is z o and z y It consists of two parts: o and z y The probability distribution of follows a mutually independent multivariate Gaussian distribution. Therefore, the probability expression of the decoder is as follows: Where: De1 is the decoder, De2 is the decoder based on the attention mechanism; θ and These are the network parameters of De1 and De2 respectively; Accordingly, the probability expression of the encoder is as follows: z~En(x)=q φ (z o ,z y |x)=q φ (z o |x)q φ (z y |x) Where: q φ (z y |x) is The approximate posterior probability distribution of ; En is the encoder; φ is the network parameter of En; Therefore, the overall loss function of the model is as follows: Where: the first term is the reconstruction loss of the transformer dissolved gas concentration, which is usually calculated using the mean square error between the predicted concentration and the actual concentration; the second term is the data distribution consistency loss, which represents the potential distribution q of the encoder output φ (z o |x) and the prior distribution p(z o ) between them; the third term is also the data distribution consistency loss, which represents the potential distribution q output by the encoder φ (z y |x) and the prior distribution p(z y ) are measured by Kullback-Leibler divergence; the fourth term is the attention prediction loss, which is usually calculated by the mean square error between the predicted transformer load and oil temperature and the actual transformer load and oil temperature; β1 and β2 are hyperparameters for weighing different loss terms; Extract fault features from the fault database to obtain a fault feature training set, build an association network between fault features and fault types through a probabilistic neural network, and obtain a probabilistic neural network fault classification model; The real-time monitored transformer load, oil temperature data and dissolved gas concentration are normalized, and the normalized real-time data is used as the input data of the attention variational autoencoder model to dynamically predict the dissolved gas concentration and obtain the prediction deviation of the dissolved gas concentration. If the prediction deviation is greater than the preset threshold, it is used as a fault feature and output; Using the fault features output by the attention variational autoencoder model, fault diagnosis is performed through the probabilistic neural network fault classification model to output the fault type.

2. The transformer fault diagnosis method based on attention variational autoencoder according to claim 1 is characterized in that: H2, CH4, C2H6, C2H4, C2H2, C0, and C02 are selected as characteristic gases from the dissolved gases in the transformer oil, and the historical concentration data of each characteristic gas of the transformer in a fixed period before the forecast date is extracted from the power OMS system. The historical oil temperature and historical load data of the transformer equipment in a fixed period before the forecast date are extracted from the power OMS system.

3. The transformer fault diagnosis method based on attention variational autoencoder according to claim 1 is characterized in that: The normalization process includes: The z-score standardization is adopted, and the specific process is as follows: Among them, x i , They represent the original value and normalized value of the i-th input parameter respectively; m(x) is the mean of the parameter, and s(x) is the standard deviation of the parameter.

4. The transformer fault diagnosis method based on attention variational autoencoder according to claim 1 is characterized in that: The construction process of the attention variational autoencoder model is as follows: Construct a variational autoencoder network, in which the encoder adopts a multi-layer perceptron structure and inputs a transformer with multiple dissolved gas concentration samples x = [H2, CH4, C2H6, C2H4, C2H2, CO, CO2] T , after multiple layers of nonlinear mapping, the output features obey the normal distribution N(μ,σ), and the output z O_mean and z o_var They correspond to the mean and covariance of the normal distribution respectively; Use the reparameterization technique to sample N(μ,σ) to obtain a low-dimensional representation z of the latent space o , whose expression is: z o =μ+σ·e Where: ε is a random variable that obeys the standard normal distribution, that is, ε~N(0,I); The decoder structure is similar to the encoder. The low-dimensional representation z of the input encoder is used to reconstruct the original transformer gas concentration through multi-layer nonlinear mapping. The encoder and decoder are jointly optimized through the loss function. For the input x = [H2, CH4, C2H6, C2H4, C2H2, CO, CO2] T , the loss function of the variational autoencoder network It can be described as: Where: is the reconstruction loss term of the variational autoencoder network; is the approximate probability distribution q φ The Kullback-Leibler divergence loss term between (z|x) and the prior probability distribution p(z).

5. The transformer fault diagnosis method based on attention variational autoencoder according to claim 1 is characterized in that: The attention variational autoencoder model input is the low-dimensional latent variable z extracted by the encoder y , z y By z y_mean and z y_var The model output obtained after re-parameterization is the external influencing factors, which include transformer load and oil temperature.

6. The transformer fault diagnosis method based on attention variational autoencoder according to claim 1 is characterized in that: The expression of the prediction deviation is as follows: Where: x′ and are the actual value and the predicted value respectively.

7. A transformer fault diagnosis system based on attention variational autoencoder, characterized in that: include: A data acquisition module is used to select characteristic gases from the dissolved gases in the transformer oil, obtain historical concentration data of each characteristic gas of the transformer, and obtain historical oil temperature data and historical load data of the transformer; Data preprocessing module, used to normalize historical concentration data, historical oil temperature data and historical load data; An attention variational autoencoder model construction module is used to construct an attention variational autoencoder model and train the model using normalized historical data. The attention variational autoencoder model is used to obtain the correlation between the concentrations of dissolved gases in transformers and extract their low-dimensional latent variable features. The loss terms of the attention variational autoencoder model include the reconstruction loss of the concentration of dissolved gases in transformers, the data distribution consistency loss, and the attention prediction loss. The model introduces an attention mechanism to make the low-dimensional latent variable features of the model correlated with external influencing factors. The loss term of the attention variational autoencoder model includes the reconstruction loss of the transformer dissolved gas concentration, the data distribution consistency loss and the attention prediction loss, wherein the construction process of the loss term is as follows: The low-dimensional latent variable is z o and z y It consists of two parts: o and z y The probability distribution of follows a mutually independent multivariate Gaussian distribution. Therefore, the probability expression of the decoder is as follows: Where: De1 is the decoder, De2 is the decoder based on the attention mechanism; θ and These are the network parameters of De1 and De2 respectively; Accordingly, the probability expression of the encoder is as follows: z~En(x)=q φ (z o ,z y |x)=q φ (z o |x)q φ (z y |x) Where: q φ (z y |x) is The approximate posterior probability distribution of ; En is the encoder; φ is the network parameter of En; Therefore, the overall loss function of the model is as follows: Where: the first term is the reconstruction loss of the transformer dissolved gas concentration, which is usually calculated using the mean square error between the predicted concentration and the actual concentration; the second term is the data distribution consistency loss, which represents the potential distribution q of the encoder output φ (z o |x) and the prior distribution p(z o ) between them; the third term is also the data distribution consistency loss, which represents the potential distribution q output by the encoder φ (z y |x) and the prior distribution p(z y ) are measured by Kullback-Leibler divergence; the fourth term is the attention prediction loss, which is usually calculated by the mean square error between the predicted transformer load and oil temperature and the actual transformer load and oil temperature; β1 and β2 are hyperparameters for weighing different loss terms; A probabilistic neural network fault classification model building module is used to extract fault features from the fault database to obtain a fault feature training set, and to build an association network between fault features and fault types through a probabilistic neural network to obtain a probabilistic neural network fault classification model; The fault feature extraction module is used to normalize the real-time monitored transformer load, oil temperature data and dissolved gas concentration, and use the normalized real-time data as the input data of the attention variational autoencoder model to dynamically predict the dissolved gas concentration and obtain the prediction deviation of the dissolved gas concentration. If the prediction deviation is greater than a preset threshold, it is used as a fault feature and output; The fault diagnosis module is used to utilize the fault features output by the attention variational autoencoder model to perform fault diagnosis through the probabilistic neural network fault classification model and output the fault type.

8. A transformer fault diagnosis device based on attention variational autoencoder, characterized in that: The device comprises a processor and a memory, wherein the memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the device implements the steps of the transformer fault diagnosis method based on attention variational autoencoding as claimed in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the transformer fault diagnosis method based on attention variational autoencoding as claimed in any one of claims 1 to 6.

Images