Transformer fault diagnosis method, system, equipment and medium

Abstract

The invention discloses a transformer fault diagnosis method, system and equipment and a medium, and the method comprises the steps: through the linkage design of an oil pumping sampling module and a transparent sampling channel, combining high-speed camera shooting collection and synchronous light source illumination, and obtaining the image of particles in a transformer oil sample in real time; secondary pollution and personal errors possibly caused by traditional off-line sampling are avoided, and the detection efficiency and the objectivity of data are improved; the ResNet neural network is introduced to carry out feature extraction on the oil sample image, the particle number and particle size distribution information are automatically identified, the complex process of manually designing features in a traditional image processing method is avoided, and the accuracy and robustness of diagnosis are greatly improved; by setting judgment threshold values for indexes such as the particle number, the average particle size and the large particle size proportion, logic judgment is conducted in combination with the actual operation state, potential abrasion, pollution and other abnormalities can be recognized in the early stage, the capacity of quick response and advanced intervention is achieved, and the operation safety and the intelligent operation and maintenance level of the transformer are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer operating status monitoring, and in particular to a transformer fault diagnosis method, system, equipment and medium. Background Art

[0002] Transformers, as a core component of power systems, are widely used in various fields, including power transmission and distribution, industrial manufacturing, and renewable energy generation. Their operating status directly impacts the safety and stability of the entire power system. Over long-term operation, transformers experience varying degrees of aging and degradation due to the combined effects of electric fields, thermal fields, mechanical vibration, and environmental factors. This can lead to a series of potential hazards, including reduced insulation performance, increased temperature rise, and partial discharge. Transformer oil, a crucial component of the insulation system, not only provides excellent insulation and heat dissipation but also serves as a crucial indicator of the transformer's operating status. During transformer operation, various types of impurity particles, such as metal particles, carbides, fibers, and water condensation, are often introduced into the oil due to factors such as metal wear, insulation material aging, overheating, or electrical breakdown. These particles not only affect the oil's breakdown voltage and insulation performance but can also migrate under the influence of electric fields and accumulate in areas of high field strength, inducing partial discharge and, in severe cases, even leading to insulation breakdown and equipment failure. Therefore, timely and accurate detection and diagnosis of the types and distribution characteristics of impurity particles in transformer oil are crucial for early warning of equipment failures and extending operational life.

[0003] Currently, the main methods for detecting impurity particles in transformer oil, both domestically and internationally, include laser particle size analysis, light obscuration analysis, centrifugal sedimentation, and imaging methods. Image recognition methods, among others, are gaining increasing attention in research and application due to their intuitiveness, visualization, and ease of further classification and analysis. Traditional image-based methods typically rely on manual or semi-automatic image processing techniques to extract morphological features such as particle area, roundness, and edge roughness, and then use classifiers such as support vector machines and random forests to identify particles. However, these methods suffer from poor recognition performance in cases of overlapping particles, blurred edges, or complex shapes. Furthermore, the feature extraction process relies on manual experience, resulting in limited versatility and robustness. In recent years, with the development of deep learning technology, convolutional neural networks have demonstrated outstanding performance in image recognition tasks, possessing excellent feature extraction capabilities for complex object recognition. However, in the field of transformer oil impurity particle identification, a comprehensive solution is still lacking that tightly couples high-speed image acquisition systems with deep neural network diagnostic models to achieve real-time data processing and integrated fault diagnosis. Therefore, it is urgent to propose a transformer fault diagnosis system and method based on deep learning to realize the automatic extraction of image features and accurate identification of fault types, and provide more reliable technical support for transformer status assessment and fault warning. Summary of the Invention

[0004] Based on the problems raised by the above background technology, the purpose of the present invention is to provide a transformer fault diagnosis method, system, equipment and medium, which solves the problem in the research field of impurity particle identification in transformer oil, which is the lack of a high-speed image acquisition system and a deep neural network diagnostic model to realize the linkage between real-time data processing and fault diagnosis.

[0005] The present invention is achieved through the following technical solutions: A first aspect of the present invention provides a transformer fault diagnosis method, comprising the following steps: Perform image acquisition on the extracted oil sample to obtain particle image data; Preprocessing the particle image data to obtain a preprocessed particle image; Performing feature extraction on the preprocessed particle image based on a ResNet network model to obtain particle features; A multi-level threshold is constructed, the particle characteristics are judged with the multi-level threshold, and fault information is generated according to the judgment result.

[0006] In the above technical solution, after the image of the extracted oil sample is captured, the particle image data is preprocessed to remove noise in the image and enhance the feature contrast of the particles, thereby making the subsequent feature extraction more accurate.

[0007] In view of the fact that the powerful feature extraction capabilities of deep neural networks are not fully utilized in existing diagnosis, this method extracts features from preprocessed particle images based on the ResNet network model, which can deeply explore the complex characteristics of particles. The ResNet network, through its unique residual structure, can effectively solve the problem of gradient disappearance during deep network training, so that a deeper network can be constructed to extract subtle features of particles; through a large amount of training data, the model can learn the complex nonlinear relationship between particle characteristics and faults, thereby improving the accuracy of fault diagnosis.

[0008] In existing diagnosis, there is only a single judgment standard or a simple threshold judgment, which does not take into account the complex association between particles of different sizes, shapes and other characteristics and different fault types. This method proposes a method to construct a multi-level threshold. Through the multi-level threshold, the fault type and severity can be judged according to the different characteristic combinations of particles, realizing the linkage between real-time data processing and fault diagnosis.

[0009] In an optional embodiment, preprocessing the particle image data includes the following steps: grayscale processing is performed on the particle image data to obtain an oil sample grayscale image; Smoothing the oil sample grayscale image using a two-dimensional Gaussian kernel to obtain a smoothed oil sample grayscale image; Redistributing the image grayscale values ​​in the smoothed oil-like grayscale image using a histogram equalization method to obtain a uniform oil-like grayscale image; The uniform oil sample grayscale image is size-normalized to obtain a normalized oil sample grayscale image.

[0010] In an optional embodiment, smoothing the oil sample grayscale image using a two-dimensional Gaussian kernel includes: Obtaining the relative position and standard deviation of the filtered center pixel in the grayscale image of the oil sample, wherein the standard deviation is used to determine the degree of image blur; The standard deviation and the relative position are subjected to Gaussian weighted convolution using a two-dimensional Gaussian kernel to obtain a smoothed oil sample grayscale image; wherein the Gaussian weighted convolution process is as follows: ; In the above formula, is the standard deviation, ( x , y ) is the relative position of the pixel in the oil sample grayscale image to the filter center pixel.

[0011] In an optional embodiment, a histogram equalization method is used to redistribute the image grayscale values ​​in the smoothed oil-like grayscale image, including: The grayscale mapping formula is used to calculate the grayscale of the smoothed oil sample grayscale image to obtain a new grayscale; wherein the grayscale mapping formula is expressed as follows: ; In the above formula, s k is the new grayscale, L is the maximum gray level, M 、 N is the image size of the smoothed oil-like grayscale image, n j The gray level is j The number of pixels; Grayscale values ​​of the smoothed oil-like grayscale image are redistributed based on the new grayscale level.

[0012] In an optional embodiment, the uniform oil sample grayscale image is size normalized, including: linearly normalizing the pixel values ​​in the uniform oil sample grayscale image, and mapping the linearly normalized pixel values ​​to the [0,1] interval to obtain a normalized oil sample grayscale image.

[0013] In an optional embodiment, feature extraction is performed on the pre-processed particle image based on a ResNet network model, comprising the following steps: Inputting the preprocessed particle image into the input layer of the ResNet network model; Performing multi-layer convolution feature extraction on the pre-processed particle image through the residual module of the ResNet network model; wherein the residual module includes a plurality of residual blocks, each residual block includes a convolution layer, a batch normalization layer and a ReLU activation function, and cross-layer connection is achieved through equal mapping; using global average pooling instead of a fully connected layer in the middle layer of the ResNet network model to output the global average of each channel; A particle feature vector is generated through the output layer of the ResNet network model. The particle feature vector includes a statistical value of the number of impurity particles in the oil and particle size distribution information, which is used to characterize the state of the oil sample.

[0014] In an optional embodiment, constructing a multi-level threshold value and judging the particle characteristics against the multi-level threshold value includes: Extracting the total number of particles, average particle size, and proportion of large-size particles from the particle characteristics; Establishing a particle number threshold, if the total particle number is greater than the particle number threshold, an abnormality occurs; Establishing a particle size threshold, if the average particle size is greater than the particle size threshold, an abnormality exists; Establishing a ratio threshold, if the proportion of the large-size particles exceeds the ratio threshold, an abnormality occurs; If any of the above anomalies exists, an abnormal state judgment is triggered and fault information is generated.

[0015] A second aspect of the present invention provides a transformer fault diagnosis system, comprising: An image acquisition module is used to acquire images of the extracted oil samples to obtain particle image data; An image preprocessing module, configured to preprocess the particle image data to obtain a preprocessed particle image; A feature extraction module is used to extract features from the pre-processed particle image based on a ResNet network model to obtain particle features; The fault diagnosis and early warning module is used to construct a multi-level threshold, judge the particle characteristics with the multi-level threshold, and generate fault information according to the judgment result.

[0016] A third aspect of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements a transformer fault diagnosis method when executing the computer program.

[0017] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements a transformer fault diagnosis method when executed by a processor.

[0018] Compared with the prior art, the present invention has the following advantages and beneficial effects: 1. Through the linkage design of the oil sampling module and the transparent sampling channel, combined with high-speed camera acquisition and synchronous light source illumination, the particle image in the transformer oil sample can be obtained in real time, avoiding the secondary contamination and human error that may be caused by traditional offline sampling, and improving the detection efficiency and data objectivity; 2. The ResNet neural network is introduced to extract features from oil sample images, automatically identifying the number of particles and particle size distribution information. This avoids the complex process of manually designing features in traditional image processing methods, significantly improving the accuracy and robustness of diagnosis; 3. By setting judgment thresholds for indicators such as particle number, average particle size, and proportion of large particles, and performing logical judgment based on the actual operating status, potential abnormalities such as wear and contamination can be identified at an early stage, with the ability to respond quickly and intervene in advance, significantly enhancing the operating safety and intelligent operation and maintenance level of the transformer. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the examples. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be considered as limiting the scope. A person of ordinary skill in the art can also derive other relevant drawings based on these drawings without inventive effort. In the drawings: Figure 1 A schematic diagram of the structure of the ResNet network model provided in Example 1 of the present invention; Figure 2 A schematic structural diagram of a transformer fault diagnosis system provided in Example 2 of the present invention; Figure 3 This is a structural diagram of an electronic device provided in Example 3 of the present invention. DETAILED DESCRIPTION

[0020] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with examples and drawings. The exemplary embodiments of the present invention and their descriptions are only used to explain the present invention and are not intended to limit the present invention.

[0021] The present invention relates to the field of transformer operating status monitoring technology, which includes video engine technology, deep neural network technology, image processor technology, graphics processor technology and visual detection technology.

[0022] Example 1 Embodiment 1 of the present invention provides a transformer fault diagnosis method, which includes the following steps: Perform image acquisition on the extracted oil sample to obtain particle image data; Preprocessing the particle image data to obtain a preprocessed particle image; Performing feature extraction on the preprocessed particle image based on a ResNet network model to obtain particle features; A multi-level threshold is constructed, the particle characteristics are judged with the multi-level threshold, and fault information is generated according to the judgment result.

[0023] It should be noted that in the field of impurity particle identification in transformer oil, traditional data acquisition methods may not be efficient enough and the data quality is uneven. Therefore, after image acquisition of the extracted oil sample, the particle image data is preprocessed to remove noise in the image and enhance the feature contrast of the particles, so that subsequent feature extraction is more accurate.

[0024] In view of the fact that the powerful feature extraction capabilities of deep neural networks are not fully utilized in existing diagnosis, this method extracts features from preprocessed particle images based on the ResNet network model, which can deeply explore the complex characteristics of particles. The ResNet network, through its unique residual structure, can effectively solve the problem of gradient disappearance during deep network training, so that a deeper network can be constructed to extract subtle features of particles; through a large amount of training data, the model can learn the complex nonlinear relationship between particle characteristics and faults, thereby improving the accuracy of fault diagnosis.

[0025] In existing diagnosis, there is only a single judgment standard or a simple threshold judgment, which does not take into account the complex association between particles of different sizes, shapes and other characteristics and different fault types. This method proposes a method to construct a multi-level threshold. Through the multi-level threshold, the fault type and severity can be judged according to the different characteristic combinations of particles, realizing the linkage between real-time data processing and fault diagnosis.

[0026] In an optional embodiment, preprocessing the particle image data includes the following steps: grayscale processing is performed on the particle image data to obtain an oil sample grayscale image; Smoothing the oil sample grayscale image using a two-dimensional Gaussian kernel to obtain a smoothed oil sample grayscale image; Redistributing the image grayscale values ​​in the smoothed oil-like grayscale image using a histogram equalization method to obtain a uniform oil-like grayscale image; The uniform oil sample grayscale image is size-normalized to obtain a normalized oil sample grayscale image.

[0027] It should be noted that the captured oil sample image is grayscaled to obtain an oil sample grayscale image containing impurity particles; the oil sample grayscale image is smoothed using a two-dimensional Gaussian kernel, and each pixel in the image is denoised by convolving with its neighboring pixels according to Gaussian weights.

[0028] The histogram equalization method is used to redistribute the grayscale values ​​of the image so that the grayscale values ​​are distributed as evenly as possible in the entire dynamic range.

[0029] Linearly normalize the pixel values ​​and map them to the [0,1] interval to satisfy , where I ( x , y ) is the original pixel value, I min 、 I max are the maximum and minimum values ​​in the image, I norm ( x , y ) is the normalized pixel value.

[0030] In an optional embodiment, smoothing the oil sample grayscale image using a two-dimensional Gaussian kernel includes: Obtaining the relative position and standard deviation of the filtered center pixel in the grayscale image of the oil sample, wherein the standard deviation is used to determine the degree of image blur; The standard deviation and the relative position are subjected to Gaussian weighted convolution using a two-dimensional Gaussian kernel to obtain a smoothed oil sample grayscale image; wherein the Gaussian weighted convolution process is as follows: ; In the above formula, is the standard deviation, ( x , y ) is the relative position of the pixel in the oil sample grayscale image to the filter center pixel.

[0031] In an optional embodiment, a histogram equalization method is used to redistribute the image grayscale values ​​in the smoothed oil-like grayscale image, including: The grayscale mapping formula is used to calculate the grayscale of the smoothed oil sample grayscale image to obtain a new grayscale; wherein the grayscale mapping formula is expressed as follows: ; In the above formula, s k is the new grayscale, L is the maximum gray level, M 、 Nis the image size of the smoothed oil-like grayscale image, n j The gray level is j The number of pixels; Grayscale values ​​of the smoothed oil-like grayscale image are redistributed based on the new grayscale level.

[0032] The maximum grayscale value is usually 256.

[0033] In an optional embodiment, the uniform oil sample grayscale image is size normalized, including: linearly normalizing the pixel values ​​in the uniform oil sample grayscale image, and mapping the linearly normalized pixel values ​​to the [0,1] interval to obtain a normalized oil sample grayscale image.

[0034] In an optional embodiment, feature extraction is performed on the pre-processed particle image based on a ResNet network model, comprising the following steps: Inputting the preprocessed particle image into the input layer of the ResNet network model; Performing multi-layer convolution feature extraction on the pre-processed particle image through the residual module of the ResNet network model; wherein the residual module includes a plurality of residual blocks, each residual block includes a convolution layer, a batch normalization layer and a ReLU activation function, and cross-layer connection is achieved through equal mapping; using global average pooling instead of a fully connected layer in the middle layer of the ResNet network model to output the global average of each channel; A particle feature vector is generated through the output layer of the ResNet network model. The particle feature vector includes a statistical value of the number of impurity particles in the oil and particle size distribution information, which is used to characterize the state of the oil sample.

[0035] The structural diagram of the ResNet network model is as follows Figure 2 As shown in the figure, the ResNet neural network includes multiple convolutional layers, residual modules and global average pooling layers, which extract and quantify the number, particle size and distribution characteristics of impurity particles in the image, and upload the extracted feature information to the host computer for further display and processing.

[0036] Specifically, the normalized particle image output by the image preprocessing module is input to the input layer of the ResNet network model. Multi-layer convolutional feature extraction is performed on the image using the ResNet residual module. The residual module contains several residual blocks, each consisting of a convolutional layer, a batch normalization layer, and a ReLU activation function. Cross-layer connections are achieved through identity mapping, enhancing the network's deep feature extraction capabilities and preventing gradient vanishing. Global Average Pooling is used in the middle layer of the network instead of the fully connected layer to retain important features and reduce the number of parameters. Its output is the global average of each channel. , where f k ( i , j ) is the k Channels at position ( i , j ) at the eigenvalue, H 、 W are the height and width of the feature map, z k For the k The global average value of the channels is obtained by the output layer; a particle feature vector is generated through the output layer, and the feature vector includes the number statistics of impurity particles in the oil and the particle size distribution information, which is used to characterize the state of the oil sample.

[0037] In an optional embodiment, constructing a multi-level threshold value and judging the particle characteristics against the multi-level threshold value includes: Extracting the total number of particles, average particle size, and proportion of large-size particles from the particle characteristics; Establishing a particle number threshold, if the total particle number is greater than the particle number threshold, an abnormality occurs; Establishing a particle size threshold, if the average particle size is greater than the particle size threshold, an abnormality exists; Establishing a ratio threshold, if the proportion of the large-size particles exceeds the ratio threshold, an abnormality occurs; If any of the above anomalies exists, an abnormal state judgment is triggered and fault information is generated.

[0038] It should be noted that if the total number of particles is greater than the particle number threshold, the oil is considered to be seriously abnormal; if the average particle size is greater than the particle size threshold, it indicates that the oil is severely abnormal; if the proportion of large-size particles exceeds the proportion threshold, it is judged that there is a risk of potential failure, indicating that the oil is slightly abnormal; other situations that do not exceed the set threshold are judged to be normal.

[0039] If any feature exceeds the set threshold, the system will trigger an abnormal state judgment. The result output unit outputs the judgment result in the form of a digital signal or a graphical interface, and can also issue an alarm command or upload it to the remote monitoring center to provide early warning of potential transformer failures.

[0040] Example 2 Figure 2 This is a schematic diagram of the structure of a transformer fault diagnosis system provided by Example 2 of the present invention, as shown in FIG. Figure 2 As shown, a transformer fault diagnosis system includes: An image acquisition module is used to acquire images of the extracted oil samples to obtain particle image data; An image preprocessing module, configured to preprocess the particle image data to obtain a preprocessed particle image; A feature extraction module is used to extract features from the pre-processed particle image based on a ResNet network model to obtain particle features; The fault diagnosis and early warning module is used to construct a multi-level threshold, judge the particle characteristics with the multi-level threshold, and generate fault information according to the judgment result.

[0041] It should be noted that this embodiment also includes an oil sampling module, which includes a sampling line, a vacuum oil pump, a solenoid valve, a sampling chamber, and connecting sealing pipes. One end of the sampling line is sealed to the oil outlet of the transformer tank, and the other end is connected to the sampling chamber through the oil pump. The sampling chamber is equipped with a transparent pipe leading to the image acquisition module. Specifically, in this embodiment, the sampling line is preferably a polytetrafluoroethylene (PTFE) or stainless steel oil-resistant hose with an inner diameter of 6-10 mm. One end is sealed to the oil outlet of the transformer tank through a threaded seal or flange interface, and the other end is connected to the inlet of the vacuum oil pump. The vacuum oil pump is preferably a micro-diaphragm pump with a pumping rate of 2-5 L / min and a vacuum degree of 0.05-0.08 MPa. Its outlet is connected to the solenoid valve through a sealing pipe. The solenoid valve is normally closed and is opened and closed at a time by the control system to control the flow of oil into the sampling chamber. The sampling chamber is a transparent, sealed container with a capacity of 30 to 100 mL, made of glass or polycarbonate. It is used for short-term oil storage and stable oil flow. The chamber outlet is connected to the image acquisition module via a transparent tube, forming an oil sampling channel. All interfaces in the module are airtightly connected using O-rings or quick connectors with sealing washers, ensuring leak-free and bubble-free operation, thereby protecting the integrity and cleanliness of the collected oil samples and improving image recognition accuracy.

[0042] The image acquisition module includes a transparent sampling pipe, a high-speed camera, an image acquisition card, and a synchronous light source system. The transparent sampling pipe is connected to the oil extraction sampling module. The camera is positioned outside the pipe, and the synchronous light source is positioned on the opposite side of the pipe or around the pipe's circumference, forming a particle image acquisition area. The image acquisition card is connected to the high-speed camera to transmit the captured particle image data to the image preprocessing module. Specifically, in this embodiment, the transparent sampling pipe preferably utilizes a highly transparent polycarbonate (PC) or quartz glass tube with an inner diameter of 8-10 mm and a length of 50-100 mm. One end of the transparent sampling pipe is connected to the sampling chamber outlet via a sealed threaded joint, and the other end is connected to the input channel of the image preprocessing module, allowing the oil sample to be introduced to form a stable flow area. The high-speed camera preferably utilizes an industrial-grade CMOS imaging module with a resolution of at least 1920×1080 and a frame rate of preferably 500-1000 frames per second. It is mounted on the outside of the transparent sampling pipe using a fixed bracket or linear track, maintaining a stable focusing distance from the sampling area to ensure clear imaging of particle motion trajectories. The synchronous light source system comprises multiple high-brightness LED or laser backlight sources, preferably with a wavelength of 550 to 600 nm and adjustable brightness. These are positioned on opposite sides of a transparent sampling pipe or arranged in a ring around the pipe to create a uniform, high-contrast background illumination area. This enhances the optical contrast of particle edges, suppresses reflection interference, and establishes a high-precision particle image acquisition area. The image acquisition card is a high-speed image acquisition interface module that supports industrial communication protocols such as USB 3.0, CameraLink, or GigE. Its input is connected to the video output of a high-speed camera, and its output communicates with the host computer or processing terminal of the image preprocessing module. It transmits the collected particle image data in real-time as a digital signal to subsequent modules for processing. This structure enables the image acquisition module to achieve high temporal and spatial resolution and low latency image acquisition without degrading the state of the oil sample particles, meeting the data quality input requirements of deep learning models.

[0043] The image preprocessing module includes a grayscale module, a denoising module, an image enhancement module, and an image size normalization module. Each module processes the particle image in turn to improve the particle boundary clarity and image quality.

[0044] The feature extraction module includes a host computer and a deep learning-based neural network. The image preprocessing module is connected to the host computer. The host computer deploys a ResNet network structure to extract image features. The ResNet neural network includes multiple convolutional layers, residual modules and global average pooling layers. It extracts and quantifies the number, particle size and distribution characteristics of impurity particles in the image, and uploads the extracted feature information to the host computer for further display and processing.

[0045] The fault diagnosis and warning module includes a threshold judgment unit and a result output unit. The threshold judgment unit is used to perform multi-level threshold comparison on the number of impurity particles and particle size distribution information output by the feature extraction module, and set multiple fault warning levels, including normal, mild abnormality, moderate abnormality and severe abnormality. The result output unit outputs the judgment result in the form of status indication, alarm signal and numerical data to the host computer for display.

[0046] Example 3 Figure 3 This is a schematic diagram of the structure of an electronic device provided in Example 3 of the present invention, such as Figure 3 As shown, the electronic device includes a processor 21, a memory 22, an input device 23 and an output device 24; the number of processors 21 in the computer device can be one or more. Figure 3 In the figure, a processor 21 is taken as an example; the processor 21, memory 22, input device 23 and output device 24 in the electronic device can be connected by a bus or other means. Figure 3 The bus connection is taken as an example.

[0047] Memory 22, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules. Processor 21 executes the software programs, instructions, and modules stored in memory 22 to perform various functional applications and data processing of the electronic device, thereby implementing the transformer fault diagnosis method of Example 1.

[0048] The memory 22 may primarily include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function; the data storage area may store data created based on the use of the terminal, etc. Furthermore, the memory 22 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state memory device. In some instances, the memory 22 may further include a memory remotely located relative to the processor 21, and these remote memories may be connected to the electronic device via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0049] The input device 23 can be used to receive the ID and password input by the user. The output device 24 is used to output the network configuration page.

[0050] Example 4 Embodiment 4 of the present invention further provides a computer-readable storage medium, wherein the computer-executable instructions, when executed by a computer processor, are used to implement a transformer fault diagnosis method as provided in embodiment 1.

[0051] An embodiment of the present invention provides a storage medium containing computer-executable instructions, and its computer-executable instructions are not limited to the method operations provided in Example 1, but can also execute related operations in a transformer fault diagnosis method provided in any embodiment of the present invention.

[0052] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A transformer fault diagnosis method, characterized in that: The steps include: Perform image acquisition on the extracted oil sample to obtain particle image data; Preprocessing the particle image data to obtain a preprocessed particle image; Performing feature extraction on the preprocessed particle image based on a ResNet network model to obtain particle features; A multi-level threshold is constructed, the particle characteristics are judged with the multi-level threshold, and fault information is generated according to the judgment result.

2. A transformer fault diagnosis method according to claim 1, characterized in that: Preprocessing the particle image data includes the following steps: grayscale processing is performed on the particle image data to obtain an oil sample grayscale image; Smoothing the oil sample grayscale image using a two-dimensional Gaussian kernel to obtain a smoothed oil sample grayscale image; Redistributing the image grayscale values ​​in the smoothed oil-like grayscale image using a histogram equalization method to obtain a uniform oil-like grayscale image; The uniform oil sample grayscale image is subjected to size normalization processing to obtain a normalized oil sample grayscale image.

3. A transformer fault diagnosis method according to claim 2, characterized in that: The oil sample grayscale image is smoothed using a two-dimensional Gaussian kernel, including: Obtaining the relative position and standard deviation of the filtered center pixel in the grayscale image of the oil sample, wherein the standard deviation is used to determine the degree of image blur; The standard deviation and the relative position are subjected to Gaussian weighted convolution using a two-dimensional Gaussian kernel to obtain a smoothed oil sample grayscale image; wherein the Gaussian weighted convolution process is as follows: ； In the above formula, is the standard deviation, ( x , y ) is the relative position of the pixel in the oil sample grayscale image to the filter center pixel.

4. A transformer fault diagnosis method according to claim 2, characterized in that: Redistributing the image grayscale values ​​in the smoothed oil-like grayscale image using a histogram equalization method, including: The grayscale mapping formula is used to calculate the grayscale of the smoothed oil sample grayscale image to obtain a new grayscale; wherein the grayscale mapping formula is expressed as follows: ； In the above formula, s k is the new grayscale, L is the maximum gray level, M 、 N is the image size of the smoothed oil-like grayscale image, n j The gray level is j The number of pixels; Grayscale values ​​of the smoothed oil-like grayscale image are redistributed based on the new grayscale level.

5. A transformer fault diagnosis method according to claim 2, characterized in that: The uniform oil sample grayscale image is size normalized, including: linearly normalizing pixel values ​​in the uniform oil sample grayscale image, and mapping the linearly normalized pixel values ​​to the [0, 1] interval to obtain a normalized oil sample grayscale image.

6. A transformer fault diagnosis method according to claim 1, characterized in that: Feature extraction of the pre-processed particle image is performed based on the ResNet network model, including the following steps: Inputting the preprocessed particle image into the input layer of the ResNet network model; Performing multi-layer convolution feature extraction on the pre-processed particle image through the residual module of the ResNet network model; wherein the residual module includes a plurality of residual blocks, each residual block includes a convolution layer, a batch normalization layer and a ReLU activation function, and cross-layer connection is achieved through equal mapping; using global average pooling instead of a fully connected layer in the middle layer of the ResNet network model to output the global average of each channel; A particle feature vector is generated through the output layer of the ResNet network model. The particle feature vector includes a statistical value of the number of impurity particles in the oil and particle size distribution information, which is used to characterize the state of the oil sample.

7. A transformer fault diagnosis method according to claim 1, characterized in that: Constructing a multi-level threshold and judging the particle characteristics with the multi-level threshold, including: Extracting the total number of particles, average particle size, and proportion of large-size particles from the particle characteristics; Establishing a particle number threshold, if the total particle number is greater than the particle number threshold, an abnormality occurs; Establishing a particle size threshold, if the average particle size is greater than the particle size threshold, an abnormality exists; Establishing a ratio threshold, if the proportion of the large-size particles exceeds the ratio threshold, an abnormality occurs; If any of the above anomalies exists, an abnormal state judgment is triggered and fault information is generated.

8. A transformer fault diagnosis system, the system being used to implement a transformer fault diagnosis method according to any one of claims 1 to 7, characterized in that: include: An image acquisition module is used to acquire images of the extracted oil samples to obtain particle image data; An image preprocessing module, configured to preprocess the particle image data to obtain a preprocessed particle image; A feature extraction module is used to extract features from the pre-processed particle image based on a ResNet network model to obtain particle features; The fault diagnosis and early warning module is used to construct a multi-level threshold, judge the particle characteristics with the multi-level threshold, and generate fault information according to the judgment result.

9. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, a transformer fault diagnosis method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the transformer fault diagnosis method according to any one of claims 1 to 7 is implemented.

Images