A transformer fault detection method, device, equipment and medium

Abstract

The embodiment of the application discloses a transformer fault detection method, device, equipment and medium. The method comprises the following steps: collecting a set of oil-soluble gas parameters in a transformer; inputting the set of oil-soluble gas parameters into a generative adversarial network to obtain target gas parameter features; inputting the target gas parameter features into a feature decision fusion model to obtain target fusion features; and determining a fault type of the transformer according to the target fusion features. According to the technical scheme, the set of oil-soluble gas parameters is processed by the generative adversarial network model and the feature decision fusion model respectively, and then the fault type of the transformer is determined, so that the accuracy and reliability of the transformer fault detection are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power equipment, and particularly relates to a transformer fault detection method, device, equipment and medium. BACKGROUND

[0002] As one of the core devices of the power system, the operation state of the power transformer directly affects the safety and stability of the entire power grid. The insulation materials inside the transformer may appear aging, overheating and other phenomena during long-term operation, resulting in various faults.

[0003] In order to timely discover and handle these faults, dissolved gas analysis (DGA) has become an effective monitoring means. By analyzing the concentrations of various gases dissolved in transformer oil, it can be determined whether there is an abnormal situation inside the transformer.

[0004] The traditional fault diagnosis method based on DGA mainly relies on rules and standards formulated by artificial experience, such as IEC60599 three-ratio method. However, this method often has difficulty in accurately distinguishing different types of faults, especially in the early stage of the fault, when the gas content changes little, the diagnosis result is easily affected by subjective factors. In addition, with the continuous expansion of the scale of the power system and the continuous development of technology, the traditional fault diagnosis method has been difficult to meet the needs of modern power systems. SUMMARY

[0005] The present application provides a transformer fault detection method, device, equipment and medium, which generates a generative adversarial network model and a feature decision fusion model to process the dissolved gas parameter set in the oil, and then determines the fault type of the transformer, thereby improving the accuracy and reliability of the transformer fault detection.

[0006] According to an aspect of the present application, a transformer fault detection method is provided, comprising:

[0007] Collecting a set of dissolved gas parameters in the oil in the transformer;

[0008] Inputting the set of dissolved gas parameters in the oil into a generative adversarial network to obtain target gas parameter features;

[0009] Inputting the target gas parameter features into a feature decision fusion model to obtain target fusion features;

[0010] Determining the fault type of the transformer according to the target fusion features.

[0011] Optionally, the generative adversarial network includes a plurality of convolution kernels;

[0012] Inputting the set of dissolved gas parameters in the oil into a generative adversarial network to obtain target gas parameter features, comprising:

[0013] convolve the set of oil-dissolved gas parameters by a plurality of convolution kernels to obtain a multi-level feature vector;

[0014] perform pooling processing and normalization processing on the multi-level feature vector to obtain a target gas parameter feature.

[0015] Optionally, the plurality of convolution kernels includes a first convolution kernel, a second convolution kernel, and a third convolution kernel.

[0016] convolve the set of oil-dissolved gas parameters by a plurality of convolution kernels to obtain a multi-level feature vector, including:

[0017] convolve the set of oil-dissolved gas parameters by the first convolution kernel to obtain a first feature vector;

[0018] convolve the first feature vector by the second convolution kernel to obtain a second feature vector;

[0019] convolve the second feature vector by the third convolution kernel to obtain a third feature vector;

[0020] channel-connect the first feature vector, the second feature vector, and the third feature vector to obtain a multi-level feature vector.

[0021] Optionally, the feature decision fusion model includes a feature fusion layer.

[0022] input the target gas parameter feature into a feature decision fusion model to obtain a target fusion feature, including:

[0023] perform feature fusion processing on the target gas parameter feature by the feature fusion layer to obtain initial fusion feature data;

[0024] channel-connect the initial fusion feature data and the target gas parameter feature to obtain a target fusion feature.

[0025] Optionally, performing feature fusion processing on the target gas parameter feature by the feature fusion layer to obtain initial fusion feature data, includes:

[0026] determining a gas change characteristic of the target gas parameter feature;

[0027] determining a first-level gas feature and a second-level gas feature according to the gas change characteristic;

[0028] performing feature fusion on the first-level gas feature and the second-level gas feature to obtain initial fusion feature data.

[0029] Optionally, the first-level gas feature and the second-level gas feature are feature fused to obtain initial fusion feature data, including:

[0030] The first-level gas feature is subjected to a scale reconstruction operation to obtain a reconstructed gas feature.

[0031] The reconstructed gas feature is fused with the second-level gas feature to obtain initial fusion feature data.

[0032] Optionally, the target fusion feature is used to determine the fault type of the transformer, including:

[0033] The target fusion feature is processed by a classification function to obtain the fault type of the transformer.

[0034] According to another aspect of the present application, a transformer fault detection device is provided, including:

[0035] A data acquisition module is configured to acquire a set of oil-soluble gas parameters in a transformer.

[0036] A gas parameter feature acquisition module is configured to input the set of oil-soluble gas parameters into a generative adversarial network to obtain a target gas parameter feature.

[0037] A target fusion feature acquisition module is configured to input the target gas parameter feature into a feature decision fusion model to obtain a target fusion feature.

[0038] A fault type determination module is configured to determine the fault type of the transformer according to the target fusion feature.

[0039] According to another aspect of the present application, an electronic device is provided, including:

[0040] At least one processor; and

[0041] A memory in communication connection with the at least one processor; wherein

[0042] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the transformer fault detection method according to any one of the embodiments of the present application.

[0043] According to another aspect of the present application, a computer readable storage medium is provided, and the computer readable storage medium stores computer instructions for enabling a processor to execute the transformer fault detection method according to any one of the embodiments of the present application.

[0044] The technical scheme of the embodiment of the present application comprises the following steps: collecting a set of oil-dissolved gas parameters in a transformer; inputting the set of oil-dissolved gas parameters into a generative adversarial network to obtain target gas parameter features; inputting the target gas parameter features into a feature decision fusion model to obtain target fusion features; and determining the fault type of the transformer according to the target fusion features. The technical scheme processes the set of oil-dissolved gas parameters through the generative adversarial network model and the feature decision fusion model respectively, and then determines the fault type of the transformer, thereby improving the accuracy and reliability of the transformer fault detection.

[0045] It should be understood that the content described in this part is not intended to identify key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0047] Figure 1 is a flow chart of a transformer fault detection method provided by the first embodiment of the present application;

[0048] Figure 2 is a structural example diagram of a generative adversarial network provided by the first embodiment of the present application;

[0049] Figure 3 is a flow chart of a transformer fault detection method provided by the second embodiment of the present application;

[0050] Figure 4 is a structural schematic diagram of a transformer fault detection device provided by the third embodiment of the present application;

[0051] Figure 5 is a structural schematic diagram of an electronic device provided by the fourth embodiment of the present application. DETAILED DESCRIPTION

[0052] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0053] It should be noted that the terms "first", "second", "initial", and "target" and the like in the description and claims of the application and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a particular order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily limit to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0054] Embodiment one

[0055] Figure 1 It is a flow chart of a transformer fault detection method provided according to the first embodiment of the application. The embodiment can be applicable to the case of detecting the fault of the transformer. The method can be executed by a transformer fault detection device which can be realized in the form of hardware and / or software and can be configured in an electronic device with data processing capability. As shown in the figure, the method comprises: Figure 1

[0056] S110, collecting a set of oil-soluble gas parameters in the transformer.

[0057] The transformer can be a power transformer device in a power system. The set of oil-soluble gas parameters can be understood as a set formed by various gas concentration parameters generated during the operation of the transformer. The set of oil-soluble gas parameters can include gas parameters such as hydrogen (H2), methane (CH4), ethane (C2H6), ethylene (C2H4), and acetylene (C2H2). Specifically, in the embodiment, due to the effects of electric, thermal, mechanical and chemical stresses, the insulation materials and oil inside the device will gradually age and decompose, generating various gas concentrations such as hydrogen (H2), methane (CH4), ethane (C2H6), ethylene (C2H4), acetylene (C2H2), carbon monoxide (CO), and carbon dioxide (CO2).

[0058] In the embodiment, a multi-parameter coupling oil-immersed transformer intelligent sensing device based on a micro-electromechanical system can be used to collect multiple operating parameters of the transformer and form a set of oil-soluble gas parameters. Different types of gases and their concentration changes in the set of oil-soluble gas parameters often correspond to different types of faults inside the transformer device. ​

[0059] S120, input the oil-dissolved gas parameter set into the generative adversarial network to obtain a target gas parameter feature.

[0060] The generative adversarial network can be a trained neural network model. In this embodiment, the generative adversarial network can be used for feature extraction and feature scale processing of the oil-dissolved gas parameter set. Specifically, the generative adversarial network can be constructed based on a generative adversarial network structure that integrates feature extraction and feature merging processing. In this embodiment, the generative adversarial network can be composed of a generator G, a discriminator D, and a reconstruction network R. The target gas parameter feature can be understood as a fault gas parameter feature obtained by the generative adversarial network extracting features from the oil-dissolved gas parameter set that obviously has a fault. It can be understood that in this embodiment, different oil-dissolved gas concentrations have corresponding normal value ranges. If each gas concentration in the oil-dissolved gas parameter set is obviously not in the normal value range, it can be considered as a gas parameter with a fault. The target gas parameter feature can include multiple fault gas parameter features of different characteristics. In this embodiment, the feature extraction layer included in the generative adversarial network can be used to extract features and perform feature scale processing on the oil-dissolved gas parameter set to obtain fault gas parameter features of different characteristics.

[0061] In this embodiment, the generative adversarial network includes multiple convolution kernels. The oil-dissolved gas parameter set is input into the generative adversarial network to obtain the target gas parameter feature, including: performing convolution processing on the oil-dissolved gas parameter set through multiple convolution kernels to obtain a multi-level feature vector; and performing pooling processing and normalization processing on the multi-level feature vector to obtain the target gas parameter feature.

[0062] The multiple convolution kernels can include a set number of convolution kernels, which can be used to perform convolution processing on the input oil-dissolved gas parameter set. In this embodiment, the generative adversarial network can include multiple convolution kernels of different sizes. The multi-level feature vector can be a feature vector of different levels obtained by performing convolution operations on the gas parameter through multiple convolution kernels of different sizes. The pooling processing can be maximum pooling processing on the multi-level feature vector, which is used to extract significant features of the multi-level features. The normalization processing can be scaling the multi-level feature vector according to a certain proportion so that it falls within a specific interval, thereby eliminating the data range difference between the multi-level feature vectors. In this embodiment, the target gas parameter feature can be obtained by performing maximum pooling and data normalization processing on the multi-level feature vector.

[0063] In this embodiment, through such a setting, the convolution kernel in the generative adversarial network can be used to perform convolution processing on the set of oil-soluble gas parameters, and the set of oil-soluble gas parameters can be subjected to pooling processing and other operations, thereby obtaining a plurality of fault gas parameter features of different levels, effectively combining a plurality of gas features of different scales, and enhancing the reliability of the data.

[0064] In this embodiment, the plurality of convolution kernels includes a first convolution kernel, a second convolution kernel, and a third convolution kernel. The set of oil-soluble gas parameters is subjected to convolution processing by the plurality of convolution kernels to obtain a multi-level feature vector, including: the set of oil-soluble gas parameters is subjected to convolution processing by the first convolution kernel to obtain a first feature vector; the first feature vector is subjected to convolution processing by the second convolution kernel to obtain a second feature vector; the second feature vector is subjected to convolution processing by the third convolution kernel to obtain a third feature vector; and the first feature vector, the second feature vector, and the third feature vector are connected in a channel to obtain the multi-level feature vector.

[0065] The first feature vector can be a feature vector obtained by the first convolution kernel performing convolution processing on each gas parameter in the set of oil-soluble gas parameters. The second feature vector can be a feature vector obtained by the second convolution kernel performing convolution processing on the first feature vector. The third feature vector can be a feature vector obtained by the third convolution kernel performing convolution processing on the second feature vector.

[0066] In this embodiment, the first convolution kernel, the second convolution kernel, and the third convolution kernel included in the plurality of convolution kernels in the generative adversarial network can be convolution kernels of different weights. The gas parameters in the set of oil-soluble gas parameters are subjected to convolution processing by the convolution kernels of different weights in sequence to obtain feature vectors of different levels, and then the feature vectors of different levels are connected in a channel axis to obtain a multi-level feature vector.

[0067] For example, the structure of the generative adversarial network in this embodiment is shown in Figure 2 The generative adversarial network includes an input layer, a building block 1, a building block 2, a building block 3, a feature merging layer, and an output layer. The building block 1, the building block 2, and the building block 3 respectively include a first convolution kernel W1, a second convolution kernel W2, and a third convolution kernel W3. Feature 1, feature 2, and feature 3 correspond to a first feature vector F1, a second feature vector F2, and a third feature vector F3, respectively. Specifically, a plurality of convolution kernels W1, W2, and W3 of different sizes are used to perform convolution processing on the set of oil-soluble gas parameters I:

[0068] F1 = W1 * I

[0069] F2 = W2 * F1

[0070] F3 = W3 * F2

[0071] Wherein, F1 can be represented as the first feature vector, F2 can be represented as the second feature vector, and F3 can represent the third feature vector, and W1, W2, and W3 can also represent the weights of the convolution operation.

[0072] Then the first feature vector, the second feature vector and the third feature vector are connected through the channel axis to obtain a multi-level feature vector F:

[0073] F = Concat (F1, F2, F3, axis = Channel);

[0074] Wherein, wherein F and Fi respectively represent the feature vectors in different layers, and Concat(·,·,·,axis = Channel) indicates that the connection is performed on the channel axis of the feature.

[0075] Wherein, through the operations such as maximum pooling and batch normalization, it is specifically:

[0076] O = MP (ReLU (BN (W4*F)));

[0077] Wherein, O is the output of the generative adversarial network, BN represents batch normalization, ReLU represents the activation function, and MP represents maximum pooling.

[0078] In addition, in order to ensure the size consistency of different scale features in the embodiment, the step parameter and the padding parameter can be set for scale adjustment, and specifically the convolution operation with the step parameter of 1 and the padding parameter of 2 can be used to ensure the size consistency of the features.

[0079] Through such a setting in the embodiment, different levels of gas features can be effectively extracted, and different scale gas features can be fused to obtain a multi-level feature vector, thereby enhancing the quality of the generated gas data and ensuring the consistency of the data scale.

[0080] S130, input the target gas parameter feature into the feature decision fusion model to obtain a target fusion feature.

[0081] The feature decision fusion model can be understood as a trained network model, which can be used for feature fusion of the extracted different feature gas parameter features to obtain a target fusion feature. The target fusion feature can be understood as a feature obtained by fusing a plurality of different feature gas parameter features. Specifically, the different features can include higher-level and lower-level gas parameter features. In this embodiment, the target gas parameter feature is input into the feature decision fusion model, the feature decision fusion model is used to process the lower-level feature layer as a weight to adjust the scale of the feature, and then the adjusted feature is connected with the higher-level feature to further complete the feature fusion through convolution operation to obtain the target fusion feature.

[0082] S140, determining the fault type of the transformer according to the target fusion feature.

[0083] The fault type can be understood as a fault state corresponding to the transformer. In this embodiment, the fault type can include a partial discharge (PD) type, a low-energy discharge (D1) type, a high-energy discharge (D2) type, a medium-low temperature overheating (T1) type, and a high temperature overheating (T2) type. In this embodiment, the target fusion feature can be classified by a classification function, and the corresponding fault type can be determined according to the classification result.

[0084] The technical scheme of the embodiment of the present application comprises the following steps: collecting a set of oil-soluble gas parameters in a transformer; inputting the set of oil-soluble gas parameters into a generative adversarial network to obtain a target gas parameter feature; inputting the target gas parameter feature into a feature decision fusion model to obtain a target fusion feature; and determining the fault type of the transformer according to the target fusion feature. The technical scheme processes the set of oil-soluble gas parameters through the generative adversarial network model and the feature decision fusion model respectively, and then determines the fault type of the transformer, thereby improving the accuracy and reliability of the transformer fault detection.

[0085] Embodiment two

[0086] Figure 3 is a flowchart of a transformer fault detection method according to the embodiment two of the present application. The embodiment is optimized based on the above-mentioned embodiment. The specific optimization is that the feature decision fusion model comprises a feature fusion layer; the target gas parameter feature is input into the feature decision fusion model to obtain a target fusion feature, which comprises: performing feature fusion processing on the target gas parameter feature through the feature fusion layer to obtain initial fusion feature data; and connecting the initial fusion feature data with the target gas parameter feature in a channel to obtain the target fusion feature. As shown in Figure 3 The method comprises the following steps:

[0087] S310, collecting a set of oil-soluble gas parameters in a transformer.

[0088] S320, input the oil-dissolved gas parameter set into the generative adversarial network to obtain a target gas parameter feature.

[0089] The generative adversarial network in this embodiment is constructed based on a generative adversarial network structure, and specifically, the generative adversarial network can include a generator G, a discriminator D and a reconstruction network R. In the training process of the generative adversarial network, the generator G and the discriminator D are optimized by minimizing the following loss functions:

[0090]

[0091] wherein, L adv represents an adversarial loss, p z (z) is a distribution of a latent variable z, D is a discriminator, G is a generator, and x is a real sample;

[0092]

[0093] wherein, L rec represents a reconstruction loss, R is a reconstruction network, and ||·||2 represents an L2 norm;

[0094]

[0095] wherein, L msff represents a multi-scale feature fusion loss, N iPj and N iTj represent the number of remaining samples of the i-th class in the j-th experiment generated by the proposed method and the traditional method, respectively, C is the number of classes, and M is the number of experiments;

[0096] L total = L adv + λ rec L rec + λ msff L msff ;

[0097] wherein, L total represents a total loss, λ rec and λ msff are weight coefficients of the reconstruction loss and the multi-scale feature fusion loss, respectively; by optimizing the above loss functions, the generator G is trained to generate high-quality fault samples, which are superior to the samples generated by the traditional GAN model in terms of similarity and diversity.

[0098] Further, a regularization technique is introduced in this embodiment to optimize the generative adversarial network to prevent mode collapse problem. This regularization technique aims to encourage the generator to produce more diverse samples, thus avoiding the problem of only learning a few modes in the dataset. This regularization term is achieved by maximizing the distance between generated samples. This regularization term can be defined as:

[0099]

[0100] where z and z' are samples independently drawn from p z (z) and d(·,·) is a metric function used to calculate the distance between two generated samples G(z) and G(z'). The metric function d(·,·) is chosen as the Euclidean distance metric.

[0101] Therefore, the total loss function of the generator can be represented as:

[0102] L G = L adv + λ rec L rec + λ msff L msff + λ div R diversity ;

[0103] where L adv is the adversarial loss, L rec is the reconstruction loss, L msff is the multi-scale feature matching loss, and R diversity is the regularization term we added. Each loss term has its corresponding weight coefficient λ rec , λ msff , λ div , which determines the degree of influence of different loss terms on the final result.

[0104] In this embodiment, the generative adversarial network is trained by minimizing L G , the generator G is trained to generate high-quality and diverse samples that can better cover all modes in the dataset, thus avoiding the problem of mode collapse.

[0105] In this embodiment, the feature decision fusion model includes a feature fusion layer.

[0106] S330, the target gas parameter features are processed by the feature fusion layer to obtain initial fusion feature data.

[0107] The feature fusion layer can be used for feature fusion operation on the input target gas parameter features. The feature fusion processing can be a processing operation of scale reconstruction and fusion on the target gas parameter features. In this embodiment, the target gas parameter features can include gas parameter features with different gas variation characteristics. The feature fusion layer is used to perform feature fusion processing on the gas parameter features with different gas variation characteristics to obtain initial fusion feature data.

[0108] In this embodiment, the feature fusion layer is used to perform feature fusion processing on the target gas parameter features to obtain initial fusion feature data, including: determining gas variation characteristics of the target gas parameter features; determining first-level gas features and second-level gas features according to the gas variation characteristics; and performing feature fusion on the first-level gas features and the second-level gas features to obtain the initial fusion feature data.

[0109] The gas variation characteristics can include rapid variation characteristics and slow variation characteristics. In this embodiment, the rapid variation characteristics and the slow variation characteristics corresponding to the target gas parameter features are received by different receiving domains to determine different gas variation characteristics. The first-level gas features can be shallow-level features, which are used to capture rapid variation characteristics of gas concentration in the target gas parameter features. The second-level gas features can be deep-level features, which are used to capture additional slow variation characteristics of gas concentration in the target gas parameter features. Specifically, in this embodiment, the shallow-level features correspond to small receiving domains, which are used to extract rapid variation characteristics of gas concentration in the target gas parameter features. The deep-level features have large receiving domains, which are used to capture additional slow variation characteristics of gas concentration in the target gas parameter features to determine the first-level gas features and the second-level gas features. In this embodiment, the specific way of performing feature fusion on the first-level gas features and the second-level gas features can be to perform scale adjustment on the first-level gas features, so that the adjusted first-level gas features and the second-level gas features are connected to perform feature fusion processing to obtain the initial fusion feature data.

[0110] In this embodiment, different levels of gas features can be captured according to the gas variation characteristics, and the initial fusion feature data can be obtained through feature fusion, which further improves the classification ability of the features and enhances the accuracy of the feature data.

[0111] In this embodiment, the first-level gas features and the second-level gas features are fused to obtain the initial fusion feature data, including: performing scale reconstruction operation on the first-level gas features to obtain reconstructed gas features; and fusing the reconstructed gas features with the second-level gas features to obtain the initial fusion feature data.

[0112] The scale reconstruction operation can be understood as an operation of adjusting the scale of the first-level gas feature. In this embodiment, the specific manner of performing the scale reconstruction operation on the first-level gas feature is to perform a convolution operation on the first-level gas feature to reconstruct the feature scale, to obtain the reconstructed gas feature. The specific manner of fusing the reconstructed gas feature with the second-level gas feature in this embodiment is to connect the reconstructed gas feature with the second-level gas feature in the channel, and then perform further convolution to complete the feature fusion, thereby obtaining the initial fusion feature data.

[0113] In this embodiment, the first-level gas feature is processed through a lower-level feature layer, and the second-level gas feature is processed through a higher-level feature layer. Specifically, in this embodiment, the feature scale of the first-level gas feature is reconstructed by performing a convolution operation on the lower-level feature layer, to obtain the reconstructed gas feature, and the reconstructed gas feature is connected with the second-level gas feature processed by the high-level feature layer to obtain the initial fusion feature data.

[0114] In this embodiment, the output of the generative adversarial network is used as the input of the decision layer of the feature decision fusion model to integrate features from different scales without significantly increasing the calculation cost and information loss. In addition, in this embodiment, the feature decision fusion model replaces the convolution layer in the traditional discriminator network to significantly increase the number of feature layers and improve the classification ability of the features; the performance of the deep network is improved by fusing features of different scales.

[0115] S340, performing channel connection on the initial fusion feature data and the target gas parameter feature to obtain a target fusion feature.

[0116] The target fusion feature can be understood as a feature data obtained by connecting and processing the initial fusion feature and the target gas parameter feature along the channel. In this embodiment, the initial fusion feature data and the target gas parameter feature can be connected along the channel axis, and then processed by convolution operation to obtain the target fusion feature.

[0117] For example, in this embodiment, the reconstructed gas feature is connected with the second-level gas feature processed by the high-level feature layer to obtain the initial fusion feature data, which is specifically as follows:

[0118] Di' = ReLU(BN(W*Di));

[0119] Then, the initial fusion feature data and the target gas parameter feature are connected along the channel axis to obtain the target fusion feature:

[0120] Di = Concat(F, D, axis = Channel);

[0121] where Concat denotes concatenation along the channel axis, BN denotes batch normalization, ReLU denotes an activation function, and W denotes weights of a convolution operation.

[0122] S350, determining the fault type of the transformer according to the target fusion feature.

[0123] In this embodiment, optionally, the fault type of the transformer is determined according to the target fusion feature, including: processing the target fusion feature through a classification function to obtain the fault type corresponding to the transformer.

[0124] The classification function processing can be an activation classification function, and can also be other classification functions. In this embodiment, the classification function can be a ReLU activation function.

[0125] In this embodiment, after the target fusion feature is classified and processed through the ReLU activation function, each target fusion feature will obtain 5 neurons, each neuron corresponds to a fault type, and each neuron corresponds to different weights. The fault type represented by the neuron with the largest weight is determined as the fault type corresponding to the transformer.

[0126] Through such a setting, the transformer fault type can be determined according to the target fusion feature, and the accuracy of transformer fault detection is improved.

[0127] Further, the present embodiment can form a whole data processing total model by combining the generative adversarial network and the feature fusion model. In the training process of the total model, a reconstruction network is established by minimizing the similarity between the encoding vector of the generated gas parameter sample and the latent vector to suppress the error of the feature vector reconstructed by the generated gas parameter sample. The encoding vector refers to the representation obtained after the encoding of the generated sample. The encoding process is usually a process of mapping data from a high-dimensional space to a low-dimensional space, and the purpose is to extract the key features of the data. The latent vector is a random noise vector used in the generative model. The reconstruction network receives the generated sample and attempts to recover the initial random noise (i.e., the latent vector) used to generate these samples, thereby reducing the difference between the encoding representation of the generated sample and these latent vectors. This method helps to ensure the diversity of generated samples while ensuring that they are sufficiently similar to actual samples.

[0128] In this embodiment, the reconstruction network includes establishing a reconstruction network R to limit the error of the latent vector z reconstructed by the generated sample, to prevent overfitting of the generated sample and improve its diversity; by minimizing the Euclidean distance between the latent vector and the original latent vector:

[0129]

[0130] In addition, in order to improve the stability and convergence speed of the model, spectral normalization is introduced to impose a relatively mild constraint on the model parameters; and different learning rate update rules are used to improve the convergence performance of the model.

[0131] The technical scheme of the embodiment uses the multi-parameter coupling oil-immersed transformer intelligent sensing device based on a micro-electromechanical system to collect a sequence of oil solubility gas parameters. Then, a generative adversarial network structure integrating feature extraction and feature fusion is constructed to extract features of the sequence of oil solubility gas parameters. In order to further improve the diagnostic accuracy, a multi-scale feature decision fusion feature fusion model is designed, which processes the lower-level feature layers through convolution operation, uses the processed features as weights to adjust the scale of the features, and then connects the higher-level features and completes feature fusion to determine the fault type of the transformer. In addition, a reconstruction network is established by minimizing the similarity between the encoding vector and the hidden vector of the generated gas sample to suppress the error of the feature vector reconstructed from the generated sample. In the model training process in the embodiment, the oil solubility gas data sets of power transformers under different environmental working conditions are selected, which are data under various fault types of transformers, including sample data under fault types such as partial discharge (PD), low-energy discharge (D1), high-energy discharge (D2), medium-low temperature overheating (T1), and high temperature overheating (T2). Each fault mode contains 8192 data points and is organized into 100 samples. Therefore, the embodiment has significant advantages in the similarity, diversity, and effectiveness of the generated samples, and can be effectively applied to the fault diagnosis of transformers, greatly improving the accuracy of diagnosis.

[0132] The technical scheme of the embodiment of the application acquires a set of oil solubility gas parameters in a transformer; inputs the set of oil solubility gas parameters into a generative adversarial network to obtain target gas parameter features; processes the target gas parameter features through a feature fusion layer to obtain initial fusion feature data; connects the initial fusion feature data and the target gas parameter features to obtain target fusion features; and determines the fault type of the transformer according to the target fusion features. The technical scheme processes the set of oil solubility gas parameters through a generative adversarial network model and a feature decision fusion model to determine the fault type of the transformer, thereby improving the accuracy and reliability of transformer fault detection.

[0133] Embodiment Three

[0134] Figure 4 is a structural schematic diagram of a transformer fault detection device provided by Embodiment Three of the application. As shown in Figure 4 , the device comprises:

[0135] The data collection module 410 is configured to collect a set of oil-dissolved gas parameters in the transformer.

[0136] The gas parameter feature acquisition module 420 is configured to input the set of oil-dissolved gas parameters into a generative adversarial network to obtain target gas parameter features.

[0137] The target fusion feature acquisition module 430 is configured to input the target gas parameter features into a feature decision fusion model to obtain target fusion features.

[0138] The fault type determination module 440 is configured to determine a fault type of the transformer according to the target fusion features.

[0139] Optionally, the generative adversarial network includes a plurality of convolution kernels, and the gas parameter feature acquisition module 320 includes:

[0140] The convolution processing unit is configured to perform convolution processing on the set of oil-dissolved gas parameters by using the plurality of convolution kernels to obtain a multi-level feature vector.

[0141] The feature vector processing unit is configured to perform pooling processing and normalization processing on the multi-level feature vector to obtain the target gas parameter features.

[0142] Optionally, the plurality of convolution kernels include a first convolution kernel, a second convolution kernel, and a third convolution kernel.

[0143] The convolution processing unit is specifically configured to perform convolution processing on the set of oil-dissolved gas parameters by using the first convolution kernel to obtain a first feature vector, perform convolution processing on the first feature vector by using the second convolution kernel to obtain a second feature vector, perform convolution processing on the second feature vector by using the third convolution kernel to obtain a third feature vector, and perform channel connection on the first feature vector, the second feature vector, and the third feature vector to obtain the multi-level feature vector.

[0144] Optionally, the feature decision fusion model includes a feature fusion layer.

[0145] The target fusion feature acquisition module 430 includes:

[0146] The fusion processing unit is configured to perform feature fusion processing on the target gas parameter features by using the feature fusion layer to obtain initial fusion feature data.

[0147] The feature connection unit is configured to perform channel connection on the initial fusion feature data and the target gas parameter features to obtain the target fusion features.

[0148] Optionally, the fusion processing unit includes:

[0149] The gas change characteristic determination subunit is configured to determine a gas change characteristic of the target gas parameter features.

[0150] a gas feature determination subunit configured to determine a first-level gas feature and a second-level gas feature according to the gas variation characteristic;

[0151] a gas feature fusion subunit configured to perform feature fusion on the first-level gas feature and the second-level gas feature to obtain initial fusion feature data.

[0152] Optionally, the gas feature fusion subunit is specifically configured to perform a scale reconstruction operation on the first-level gas feature to obtain a reconstructed gas feature, and perform fusion on the reconstructed gas feature and the second-level gas feature to obtain the initial fusion feature data.

[0153] Optionally, the fault type determination module 440 is specifically configured to process the target fusion feature through a classification function to obtain a fault type of the transformer.

[0154] The transformer fault detection device provided in the embodiment of the present application can execute the transformer fault detection method provided in any embodiment of the present application, and has the corresponding function modules and beneficial effects of the execution method.

[0155] Embodiment four

[0156] Figure 5 is a structural schematic diagram of an electronic device provided by the embodiment four of the present application. The electronic device 10 is intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular telephones, smart phones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not intended to limit the implementations of the present application described and / or claimed in this document.

[0157] As shown in Figure 5 The electronic device 10 includes at least one processor 11, and a memory, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., which are in communication connection with the at least one processor 11, wherein the memory stores a computer program which can be executed by the at least one processor. The processor 11 can execute various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0158] A plurality of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunication networks.

[0159] The processor 11 can be various general and / or special purpose processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 performs various methods and processes described above, such as the transformer fault detection method.

[0160] In some embodiments, the transformer fault detection method can be implemented as a computer program tangibly embodied in a computer readable storage medium, such as the storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded onto the RAM 13 and executed by the processor 11, one or more steps of the transformer fault detection method described above can be performed. Alternatively, in other embodiments, the processor 11 can be configured to perform the transformer fault detection method by any other appropriate means, such as by means of firmware.

[0161] The various implementations of the systems and techniques described above can be realized in a digital electronic circuit system, an integrated circuit system, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), an application specific standard product (ASSP), a system on a chip system (SOC), a programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0162] Computer programs for implementing the methods of the present application can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the computer program, when executed, can cause instructions defined in the flow charts and / or block diagrams to be implemented. The computer program can be executed entirely on a machine, partially on a machine, partially on a machine as a stand-alone software package, and partially on a remote machine or entirely on a remote machine or server.

[0163] In the context of the present application, a computer-readable storage medium can be a tangible medium that can contain or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. A computer-readable storage medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium can be a machine-readable signal medium. More specific examples of a machine-readable storage medium will include one or more lines of a program of instructions in a transitory signal, 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 foregoing.

[0164] To provide for interaction with a user, the systems and techniques described here can be implemented on an electronic device having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the electronic device. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0165] The systems and techniques described herein can be implemented in a computing system that includes a back end component, e.g., as a data server, or that includes a middleware component, e.g., an application server, or that includes a front end component, e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described herein, or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication, e.g., a communication network. Examples of communication networks include a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0166] The computing system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. A server can be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system, to solve the defects of large management difficulty and weak business scalability in traditional physical host and VPS service.

[0167] It should be understood that the various forms of flow shown above can be re-ordered, added to, or deleted from without departing from the scope of the present application. For example, the steps recited in the present application can be performed in parallel, in series, or in a different order, without departing from the desired results of the technical solutions of the present application, and this is not limited herein.

[0168] The above detailed description does not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A transformer fault detection method, characterized by, The method comprises the following steps: collecting a set of oil-dissolved gas parameters in a transformer; inputting the set of oil-dissolved gas parameters into a generative adversarial network to obtain target gas parameter features; inputting the target gas parameter features into a feature decision fusion model to obtain target fusion features; determining a fault type of the transformer according to the target fusion features; the feature decision fusion model comprises a feature fusion layer; the target gas parameter features are inputted into the feature decision fusion model to obtain the target fusion features, which comprises the following steps: performing feature fusion processing on the target gas parameter features through the feature fusion layer to obtain initial fusion feature data; connecting the initial fusion feature data and the target gas parameter features in a channel to obtain target fusion features; the feature fusion processing on the target gas parameter features through the feature fusion layer to obtain the initial fusion feature data comprises the following steps: determining gas change characteristics of the target gas parameter features; determining first-level gas features and second-level gas features according to the gas change characteristics; wherein the first-level gas features are used to capture rapid change characteristics of gas concentration in the target gas parameter features; and the second-level gas features are used to capture slow change characteristics of gas concentration in the target gas parameter features; performing feature fusion on the first-level gas features and the second-level gas features to obtain the initial fusion feature data; the feature fusion on the first-level gas features and the second-level gas features to obtain the initial fusion feature data comprises the following steps: performing a scale reconstruction operation on the first-level gas features to obtain reconstructed gas features; fusing the reconstructed gas features and the second-level gas features to obtain the initial fusion feature data.

2. The method of claim 1, wherein, the generative adversarial network comprises a plurality of convolution kernels; the set of oil-dissolved gas parameters is inputted into the generative adversarial network to obtain the target gas parameter features, which comprises the following steps: performing convolution processing on the set of oil-dissolved gas parameters through the plurality of convolution kernels to obtain a plurality of hierarchical feature vectors; performing pooling processing and normalization processing on the plurality of hierarchical feature vectors to obtain the target gas parameter features.

3. The method of claim 2, wherein, the plurality of convolution kernels comprises a first convolution kernel, a second convolution kernel and a third convolution kernel; the convolution processing on the set of oil-dissolved gas parameters through the plurality of convolution kernels to obtain the plurality of hierarchical feature vectors comprises the following steps: performing convolution processing on the set of oil-dissolved gas parameters through the first convolution kernel to obtain a first feature vector; performing convolution processing on the first feature vector through the second convolution kernel to obtain a second feature vector; performing convolution processing on the second feature vector through the third convolution kernel to obtain a third feature vector; connecting the first feature vector, the second feature vector and the third feature vector in a channel to obtain the plurality of hierarchical feature vectors.

4. The method of claim 1, wherein, determining the fault type of the transformer according to the target fusion features comprises the following steps: processing the target fusion features through a classification function to obtain the fault type of the transformer.

5. A transformer fault detection apparatus, characterized by, The method comprises the following steps: a data acquisition module is configured to collect a set of oil-dissolved gas parameters in a transformer; The gas parameter feature acquisition module is configured to input the set of the oil-dissolved gas parameters into a generative adversarial network to obtain target gas parameter features. The target fusion feature acquisition module is configured to input the target gas parameter features into a feature decision fusion model to obtain target fusion features. The fault type determination module is configured to determine a fault type of the transformer according to the target fusion features. The feature decision fusion model includes a feature fusion layer. The target fusion feature acquisition module includes: The fusion processing unit is configured to perform feature fusion processing on the target gas parameter features through the feature fusion layer to obtain initial fusion feature data. The feature connection unit is configured to connect the initial fusion feature data and the target gas parameter features in a channel to obtain target fusion features. The fusion processing unit includes: The gas change characteristic determination subunit is configured to determine a gas change characteristic of the target gas parameter features. The gas feature determination subunit is configured to determine a first-level gas feature and a second-level gas feature according to the gas change characteristic; the first-level gas feature is configured to capture a rapid change characteristic of a gas concentration in the target gas parameter features; and the second-level gas feature is configured to capture a slow change characteristic of the gas concentration in the target gas parameter features. The gas feature fusion subunit is configured to perform feature fusion on the first-level gas feature and the second-level gas feature to obtain initial fusion feature data. The gas feature fusion subunit is specifically configured to perform a scale reconstruction operation on the first-level gas feature to obtain a reconstructed gas feature; and perform fusion on the reconstructed gas feature and the second-level gas feature to obtain initial fusion feature data.

6. An electronic device, comprising: The electronic device includes: at least one processor; and a memory connected to the at least one processor in communication; wherein The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the transformer fault detection method of any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions for enabling the processor to execute the transformer fault detection method of any one of claims 1-4 when executed.

Images