Transformer oil aging detection method, device, equipment and medium

Abstract

The invention discloses an aging detection method, device and equipment for transformer oil and a medium, and the method comprises the steps: obtaining offline detection data and oil visual data corresponding to a plurality of aging stages of the transformer oil by responding to a modeling request; carrying out pixel classification and labeling on the oil visual data to obtain a classification labeling image; training a preset feature segmentation model by using the offline detection data and the associated feature pair of the classification annotation image to obtain a target feature segmentation model; when the sample data and the sample image of the to-be-detected transformer oil are received, the target feature segmentation model is called to be combined with the quantitative weight, and the aging grade of the to-be-detected transformer oil is determined. Therefore, the machine learning method is applied to realize the identification of the aging characteristics of the transformer oil and the judgment of the aging grade, and the aging characteristics can be identified and judged more comprehensively and more accurately in a mode of combining the aging characteristic detection indexes and the visual images.

Description

Technical Field

[0001] The invention relates to the technical field of transformer oil aging detection, and in particular to a transformer oil aging detection method, device, equipment and medium. Background Art

[0002] Transformer oil is an important guarantee for the safe operation of transformers. It mainly has the functions of cooling, insulation, protection, etc., and plays a very important role in substation equipment. However, during the operation of transformer oil, as the operation time increases, the transformer oil will gradually age, and the performance of the aged transformer oil will change, resulting in a decrease in the insulation performance of the substation equipment.

[0003] To this end, the current method of studying transformer oil aging is usually to use an aging box to heat the transformer oil for accelerated aging, take out the transformer oil at a specified time for offline testing, and analyze the transformer oil's breakdown voltage, volume resistivity, acid value and other indicators to preliminarily determine the degree of transformer oil aging.

[0004] However, in the above method, the number of offline detection cycles is large and the aging time is long. Each time the aging degree of the transformer oil is determined, it is necessary to re-detect it offline, and the detection efficiency is low. Summary of the invention

[0005] The present invention provides a transformer oil aging detection method, device, equipment and medium, which solves the technical problems that the existing transformer oil aging detection method has many offline detection cycles and long aging time, and each time the aging degree of the transformer oil is determined, it needs to be re-offline detected, resulting in low detection efficiency.

[0006] A first aspect of the present invention provides a method for detecting aging of transformer oil, comprising:

[0007] Respond to modeling requests and obtain offline detection data and oil visual data corresponding to multiple aging stages of transformer oil;

[0008] Classifying and labeling the oil visual data by pixels to obtain a classified and labeled image;

[0009] Using the associated features of the offline detection data and the classified and annotated image to train a preset feature segmentation model to obtain a target feature segmentation model;

[0010] When the sample data and sample image of the transformer oil to be detected are received, the target feature segmentation model is called in combination with the quantization weight to determine the aging grade of the transformer oil to be detected.

[0011] Optionally, before executing the response modeling request and acquiring multiple sets of transformer oil offline detection data and oil visual data, the method further includes:

[0012] Performing an aging test on the transformer oil of the same transformer according to multiple groups of different time and temperature conditions through an aging box, and detecting the offline detection data corresponding to each group of the transformer oil respectively;

[0013] Using at least one image forming device to detect the oil visual data corresponding to each group of transformer oil;

[0014] The offline detection data include chromaticity, acid value, breakdown voltage, volume resistivity and dielectric constant, and the oil visual data include transformer oil color, metal particle information, non-metal particle information, bubble information and water content information.

[0015] Optionally, the pixel classification and labeling of the oil visual data to obtain a classified and labeled image includes:

[0016] If the oil visual data does not meet the labeling condition, labeling the oil visual data in response to external input to obtain a classified labeling image;

[0017] If the oil visual data meets the labeling condition, selecting at least one image segmentation method to segment the oil visual data according to the color, brightness or texture in the oil visual data to obtain a plurality of region segmentation images;

[0018] Each of the region segmentation images is labeled respectively to obtain a classified labeled image.

[0019] Optionally, the adopting of the associated features of the offline detection data and the classified annotated image to train a preset feature segmentation model to obtain a target feature segmentation model includes:

[0020] Performing network segmentation according to the association relationship between the offline detection data and the classified and labeled image to obtain associated feature pairs;

[0021] Inputting the associated feature pairs into a preset feature segmentation model and outputting training feature categories;

[0022] Calculating the loss function value between the training feature category and the actual feature category;

[0023] If the loss function value exceeds the preset loss threshold, an optimizer is used to perform parameter gradient adjustment on the preset aging detection model until the loss function value is less than the preset loss threshold, thereby obtaining a target feature segmentation model.

[0024] Optionally, performing network segmentation according to the association relationship between the offline detection data and the classified annotated image to obtain associated feature pairs includes:

[0025] Creating feature nodes respectively according to the offline detection data and the classified and labeled image;

[0026] Calculating the correlation between the feature nodes;

[0027] Connect each of the feature nodes according to the correlation degree to generate a feature network;

[0028] A preset feature association model is used to perform network segmentation on the feature network to generate associated feature pairs.

[0029] Optionally, when receiving the sample data and sample image of the transformer oil to be detected, calling the target feature segmentation model in combination with the quantization weight to determine the aging level of the transformer oil to be detected includes:

[0030] When receiving sample data and sample images of transformer oil to be tested, performing network segmentation on the sample data and the sample image to obtain sample association features;

[0031] Classifying according to the sample-related features through the target aging detection model to determine the first aging feature category to which the sample data belongs and the second aging feature category to which the sample image belongs;

[0032] Determine a first quantization level according to the first aging feature category and a preset determination criterion, and determine a second quantization level according to the second aging feature category and a preset determination criterion;

[0033] The first quantization level and the second quantization level are weighted and superimposed to determine an aging level.

[0034] Optionally, the weighted superposition of the first quantization level and the second quantization level to determine the aging level includes:

[0035] Calculate the first discrimination accuracy corresponding to the first aging feature category based on the confusion matrix of KNN supervised learning;

[0036] Calculating a second discrimination accuracy corresponding to the second aging feature category according to a plurality of image indicators;

[0037] Determining a quantization weight according to the first discrimination accuracy and the second discrimination accuracy;

[0038] The first quantization level and the second quantization level are weightedly superimposed according to the quantization weight to determine an aging level.

[0039] A second aspect of the present invention provides a transformer oil aging detection device, comprising:

[0040] A data acquisition module, used to respond to modeling requests and obtain offline detection data and oil visual data corresponding to transformer oil at multiple aging stages;

[0041] An image classification and labeling module is used to classify and label the oil visual data pixels to obtain a classified and labeled image;

[0042] A model training module, used to train a preset feature segmentation model using the associated features of the offline detection data and the classified and annotated image to obtain a target feature segmentation model;

[0043] The aging grade determination module is used to call the target feature segmentation model in combination with the quantization weight to determine the aging grade of the transformer oil to be detected when receiving the sample data and sample image of the transformer oil to be detected.

[0044] A third aspect of the present invention provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the transformer oil aging detection method as described in any one of the first aspect of the present invention.

[0045] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed, implements the transformer oil aging detection method as described in any one of the first aspects of the present invention.

[0046] It can be seen from the above technical solutions that the present invention has the following advantages:

[0047] The present invention obtains offline detection data and oil visual data corresponding to multiple aging stages of transformer oil by responding to modeling requests; classifies and labels the oil visual data pixels to obtain classified and labeled images; uses the associated features of the offline detection data and the classified and labeled images to train a preset feature segmentation model to obtain a target feature segmentation model; when receiving sample data and sample images of the transformer oil to be detected, calls the target feature segmentation model in combination with the quantization weights to determine the aging level of the transformer oil to be detected. Thus, the machine learning method is applied to realize the recognition of transformer oil aging features and the determination of aging levels, and the recognition and judgment of aging features can be realized more comprehensively and accurately by combining aging feature detection indicators with visual images. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0049] Figure 1A flowchart of the steps of a transformer oil aging detection method provided by an embodiment of the present invention;

[0050] Figure 2 A flowchart of processing offline detection data provided by an embodiment of the present invention;

[0051] Figure 3 A flow chart of processing oil visual data provided by an embodiment of the present invention;

[0052] Figure 4 This is a structural block diagram of a transformer oil aging detection device in an embodiment of the present application. DETAILED DESCRIPTION

[0053] The embodiments of the present invention provide a transformer oil aging detection method, device, equipment and medium, which are used to solve the technical problems of many offline detection cycles and long aging time, and the need for re-offline detection each time the aging degree of the transformer oil is determined, resulting in low detection efficiency.

[0054] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] See also Figure 1 , Figure 1 A flowchart of the steps of a transformer oil aging detection method provided by an embodiment of the present invention.

[0056] The present invention provides a transformer oil aging detection method, comprising:

[0057] Step 101, in response to a modeling request, obtaining offline detection data and oil visual data corresponding to transformer oil at multiple aging stages;

[0058] Modeling request refers to the request information that triggers or starts model training, so as to train the feature segmentation model using offline detection data of transformer oil and oil vision data to obtain the target feature segmentation model.

[0059] Offline detection data refers to the information obtained by performing offline detection on the transformer oil of the same transformer after multiple groups of aging tests at different times and temperatures are carried out using an aging box to characterize the aging information of the aged transformer oil at different stages, which may include but is not limited to chromaticity, acid value, breakdown voltage, volume resistivity and dielectric constant.

[0060] Oil visual data refers to the surface image or video data of transformer oil when the transformer oil is tested offline, which is used to characterize transformer oil of different aging degrees, and may include but is not limited to transformer oil color, metal particle information, non-metal particle information, bubble information and water content information.

[0061] In an embodiment of the present application, the method can be implemented by a terminal device with data processing capabilities such as a central processing unit, a cloud or a server, which receives a modeling request input by an external terminal or a user, and obtains offline detection data and oil visual data corresponding to the transformer oil of the same transformer at multiple aging stages from a pre-built database.

[0062] In an example of the present application, before executing step 101, in order to generate offline detection data and oil visual data as a data basis for subsequent model training, the method may further include the following steps S11-S12:

[0063] S11, performing an aging test on the transformer oil of the same transformer according to multiple groups of different time and temperature conditions through an aging box, and detecting the offline detection data corresponding to each group of transformer oil respectively;

[0064] S12, using at least one imaging device to detect the oil visual data corresponding to each group of transformer oil;

[0065] Among them, the offline detection data includes chromaticity, acid value, breakdown voltage, volume resistivity and dielectric constant, and the oil visual data includes transformer oil color, metal particle information, non-metallic particle information, bubble information and water content information.

[0066] An aging chamber is a device used to simulate various environmental conditions to accelerate the aging process of a product. Since transformers have a variety of different use environments and different model structures, in order to ensure the accuracy of information, the transformer oil used in the aging test is the transformer oil of the same transformer. The aging chamber controls parameters such as temperature, humidity, time and ventilation in real time to quickly simulate the aging process of transformer oil during use.

[0067] In this embodiment, in order to facilitate subsequent data acquisition for model training, offline detection data and oil visual data can be stored by building a database. Specifically, with time and temperature as variables, an aging box is used to perform an aging test on the transformer oil of the same transformer according to different temperature gradients and different time gradients. After each aging test, the transformer oil is taken out for offline testing to detect the offline detection data of each group of transformer oil. For the offline detection data, the aging test process can be referred to Figure 2As shown in the figure, after the aging test, an offline data set of aging characteristics is obtained, and the data is preprocessed and divided into a training set and a test set. The training set performs cross-validation training on the convolutional neural network, and the test set performs optimal parameter screening by loss function comparison and verification. The target model is obtained and then quantitatively analyzed to determine the aging feature recognition and aging level.

[0068] When detecting offline detection data, at least one imaging device can be used to shoot the transformer oil to obtain visual data of the oil under corresponding time and temperature conditions. The imaging device can include a lensless microscope, strobe lighting, CMOS imager, etc. The obtained visual image of the oil can be processed as follows: Figure 3 The oil visual image is used as the data for image preprocessing such as denoising, contrast adjustment and normalization. The feature data obtained by image segmentation can be subjected to morphological and feature processing, and further classified to obtain the judgment result.

[0069] In the specific implementation, offline detection data is used as a kind of aging feature. The aging feature of the transformer oil is obtained by accumulating data from multiple sets of transformer oil aging tests, and a database system of typical aging features is obtained through a large amount of training with the help of neural networks and machine learning. At the same time, the transformer oil with different aging characteristics is collected by combining continuous image video, and the image database is formed by marking and annotating the images with aging characteristics and training them.

[0070] Step 102, classifying and labeling the oil visual data pixels to obtain a classified and labeled image;

[0071] In this embodiment, the image segmentation technology selects one or more combinations of threshold segmentation, edge detection segmentation, cluster segmentation, graph segmentation, morphological feature segmentation, etc. according to the color, brightness or texture of the transformer oil. In the process of pixel classification of aging images, when the number of features is small, the features are obvious, the shapes are regular, and the sizes are large, the manual labeling method is selected; when the number of features is large, the features are not obvious, the shapes are irregular, and the sizes are small, the segmentation technology combined with the manual labeling method is selected.

[0072] In an example of the present application, step 102 may include the following sub-steps:

[0073] If the oil visual data does not meet the labeling conditions, the oil visual data is labeled in response to the external input to obtain a classified labeled image;

[0074] If the oil visual data meets the labeling conditions, at least one image segmentation method is selected to segment the oil visual data according to the color, brightness or texture in the oil visual data to obtain multiple region segmentation images;

[0075] Each region segmentation image is labeled separately to obtain a classified labeled image.

[0076] In this embodiment, the labeling condition is judged for each oil visual data. If the oil visual data does not meet the labeling condition, it indicates that the oil visual data has few features, obvious features, regular shapes and large sizes, and can only be labeled through external input. Then, the input information of the external terminal or peripheral is received, and each feature area in the oil visual data is labeled to obtain a classified labeled image. If the oil visual data meets the labeling condition, it indicates that the content in the oil visual data is relatively rich, and labeling in response to external input may be inefficient. For example, if there are many features, unclear features, irregular shapes, and small sizes, segmentation technology combined with manual marking can be used for labeling.

[0077] Specifically, for oil visual data that meets the labeling conditions, one or more combinations of image annotation tools such as threshold segmentation, edge detection segmentation, clustering segmentation, graph segmentation, morphological feature segmentation, etc. can be selected according to the color, brightness or texture of the transformer oil to perform image segmentation on the oil visual data to obtain multiple region segmentation images, and then each region segmentation image is labeled in response to external input or through an image annotation model to obtain a classified labeled image.

[0078] Among them, the labeling conditions can be set according to the number of features, feature brightness, feature shape and feature size in the oil visual data, and specifically, corresponding thresholds can be set respectively for representation.

[0079] Step 103, using the offline detection data and the associated features of the classified and annotated image to train a preset feature segmentation model to obtain a target feature segmentation model;

[0080] After the generation of the classified and annotated images is completed, the offline detection data and the classified and annotated images can be combined for network segmentation to determine the association relationship and association information between the two and generate associated feature pairs. The associated feature pairs are used to train the preset feature segmentation model to obtain the target feature segmentation model.

[0081] It should be noted that network segmentation can be implemented using YOLOv8. The feature segmentation model can use the convolutional neural network (CNN) model architecture to extract high-level features from the input data. The convolution blocks are connected by Max Pooling to achieve the purpose of screening the main features and reducing the secondary features. The optimizer uses gradient descent or stochastic gradient descent (SGD), and the loss function uses mean square error (MSE) and cross entropy (Cross-Entropy). At the same time, a part of the data is used as a validation set to evaluate the model performance during the training process and verify the reliability of the model. By adjusting hyperparameters (such as learning rate, batch size, etc.), the model performance is continuously optimized.

[0082] In an example of the present application, step 103 may include the following sub-steps S21-S24:

[0083] S21, performing network segmentation according to the correlation relationship between the offline detection data and the classified annotated image to obtain correlation feature pairs;

[0084] Furthermore, S21 may include the following sub-steps:

[0085] Create feature nodes according to offline detection data and classified annotated images;

[0086] Calculate the correlation between feature nodes;

[0087] Connect each feature node according to the degree of association to generate a feature network;

[0088] The preset feature association model is used to segment the feature network and generate associated feature pairs.

[0089] In this embodiment, feature nodes can be created separately from offline detection data and classified annotated images, and the correlation between feature nodes is calculated using correlation coefficients or other correlation calculation methods, so that the feature nodes are connected according to the correlation to generate a feature network.

[0090] Further, feature association models such as network segmentation tools or YOLOv8 models are used to perform network segmentation on the feature network and generate associated feature pairs.

[0091] It should be noted that if the traffic of a characteristic node in a certain area exceeds the abnormal threshold, it indicates that the area where the characteristic node is located can be divided out and processed separately.

[0092] S22, inputting the associated feature pairs into a preset feature segmentation model, and outputting training feature categories;

[0093] S23, calculating the loss function value between the training feature category and the actual feature category;

[0094] S24. If the loss function value exceeds a preset loss threshold, an optimizer is used to perform parameter gradient adjustment on the preset aging detection model until the loss function value is less than the preset loss threshold, thereby obtaining a target feature segmentation model.

[0095] In this embodiment, the feature segmentation model can be constructed by a convolutional neural network architecture, in which multiple convolution layers are connected by a maximum pooling layer, high-dimensional features are extracted from associated feature pairs and reduced in dimension through the maximum pooling layer, so as to achieve the purpose of screening the main features and reducing the secondary features. Finally, the fully connected layer is used to classify the extracted features to obtain the training feature categories corresponding to the associated feature pairs.

[0096] All training feature categories are used to calculate the loss function values ​​between them and the actual feature categories, such as mean square error (MSE), cross-entropy, etc. If the loss function value exceeds the preset loss threshold, the optimizer is used to adjust the parameter gradient of the preset aging detection model, such as gradient descent or stochastic gradient descent (SGD), until the loss function value is less than the preset loss threshold, and the target feature segmentation model is obtained.

[0097] Among them, multiple feature categories in the fully connected layer can be set according to the quantitative judgment level of single transformer oil aging characteristics. The specific judgment criteria include one or more combinations of threshold method, edge method, clustering method, and morphology. The judgment factors of the level judgment criteria include chromaticity, acid value, breakdown voltage, volume resistivity, dielectric constant, etc. in offline detection data, and transformer oil color, metal particles, non-metallic particles, bubbles, water content, etc. in the image visual data.

[0098] It should be noted that the accuracy of the training feature categories can also be determined. For the offline detection data in the associated feature pairs, the offline detection feature information such as the chromaticity, acid value, breakdown voltage, volume resistivity, and dielectric constant of each matrix can be extracted through digital processing as the input data of the KNN nearest neighbor algorithm estimation model. The input training set is:

[0099]

[0100] in, represents the feature vector of the training set data, represents the label of the training set data, , X is the feature space, which is an n-dimensional real vector space, is the labeled category in the training set, x is the offline detection data, and is also a real feature quantity with n-dimensional data. According to the given distance metric, find the nearest points to x in the training set T, and the neighborhood of x covering k points is recorded as ,exist The category y of x is determined according to the classification decision rule:

[0101]

[0102] Among them, I is the indicator function, that is, when When I is 1, otherwise it is 0. Thus, the recognition and estimation of offline detection characteristic information such as chromaticity, acid value, breakdown voltage, volume resistivity, and dielectric constant can be realized.

[0103] For the classified annotated images in the associated feature pairs, the pixel accuracy expression is:

[0104]

[0105] Among them, k is the number of template categories to be segmented, including the background P, there are a total of k+1 categories, Indicates the number of pixels that identify the i-th feature as the j-th feature. Represents the result of network segmentation, the number of pixels that identify the i-th feature as the i-th feature.

[0106] The intersection-and-union ratio expression is:

[0107]

[0108] The IOU represents the overlap between the predicted result and the true value. Indicates the number of pixels where the prediction results P and T intersect. It represents the total number of pixels occupied by P and T in the image, and T represents the true value; the DICE expression is:

[0109]

[0110] Among them, P and T represent the prediction result and GT respectively. Reflects the overlap between two samples. When the two samples completely overlap, Dice is equal to 1. When the two samples have no overlap at all, is 0.

[0111] Step 104, when the sample data and sample image of the transformer oil to be tested are received, the target feature segmentation model is called in combination with the quantization weight to determine the aging level of the transformer oil to be tested.

[0112] In an example of the present application, step 104 may include the following sub-steps S31-S34:

[0113] S31, when receiving sample data and sample images of transformer oil to be tested, performing network segmentation on the sample data and sample images to obtain sample association features;

[0114] S32, classifying according to sample-related features through the target aging detection model to determine the first aging feature category to which the sample data belongs and the second aging feature category to which the sample image belongs;

[0115] In this embodiment, the specific process of network segmentation and feature classification is similar to the above S21-S22 and will not be repeated here.

[0116] S33, determining a first quantization level according to the first aging feature category and a preset determination criterion, and determining a second quantization level according to the second aging feature category and a preset determination criterion;

[0117] In this embodiment, the preset determination criteria are one or a combination of threshold method, edge method, clustering method, and morphology to cope with different data types.

[0118] Among them, the threshold method can set a specific threshold range for each parameter in the offline detection data (such as chromaticity, acid value, breakdown voltage, volume resistivity, dielectric constant, etc.) and the features in the image video (such as transformer oil color, etc.). For example, for the acid value, an upper limit value can be determined. When the acid value exceeds this value, it indicates that the transformer oil may have aging or pollution problems and needs further treatment or replacement. In image and video analysis, the edge method can be used to detect the boundaries between metal particles, non-metallic particles, bubbles, etc. in transformer oil and the surrounding environment. Through the edge detection algorithm, the contours of these particles or bubbles can be determined, and then their quantity, size and distribution can be analyzed. For offline detection data, the edge method can be used to detect the trend of parameter values. For example, when the breakdown voltage changes sharply within a certain range, it can be considered that there is a potential risk of failure. The clustering method can group offline detection data or image video features with similar characteristics. For example, transformer oil samples can be divided into different categories according to the different colors of transformer oil, the size of metal particles and other characteristics. In image and video processing, morphological methods can be used to remove noise, fill holes, extract objects of specific shapes, etc. For example, morphological operations can be used to remove noise from transformer oil images to more accurately detect particles and bubbles in the oil. For offline detection data, morphological methods can be used to analyze the morphological characteristics of parameter values, such as peaks, valleys, trends, etc.

[0119] S34 , weightedly superimpose the first quantization level and the second quantization level to determine an aging level.

[0120] Further, S34 may include the following sub-steps:

[0121] Calculate the first discrimination accuracy corresponding to the first aging feature category based on the confusion matrix of KNN supervised learning;

[0122] Calculating a second discrimination accuracy corresponding to a second aging feature category according to a plurality of image indicators;

[0123] Determining a quantization weight according to the first discrimination accuracy and the second discrimination accuracy;

[0124] The first quantization level and the second quantization level are weightedly superimposed according to the quantization weight to determine the aging level.

[0125] In this embodiment, after the first aging feature category is classified and predicted by the KNN algorithm, a confusion matrix can be constructed. The first discrimination accuracy can be measured by calculating indicators such as accuracy, precision, and recall. Image indicators may include color features (such as mean, standard deviation, color histogram, etc.), texture features (such as grayscale co-occurrence matrix, local binary pattern, etc.), shape features (such as edge contour, area, etc.), etc. According to the selected image indicators, the second aging feature category is feature extracted and analyzed. Machine learning algorithms (such as support vector machines, random forests, etc.) or deep learning models (such as convolutional neural networks, etc.) can be used to classify and predict the extracted image features. Similar to the calculation of the first discrimination accuracy, the second discrimination accuracy can be calculated by constructing a confusion matrix or other evaluation indicators.

[0126] By adjusting the quantization weights, the discrimination results of the two categories can be reasonably integrated according to different application scenarios and data characteristics. Among them, the first discrimination accuracy and the second discrimination accuracy can be directly used as quantization weights, that is, the weight is proportional to the accuracy. More complex methods can also be used, such as multi-criteria decision-making methods such as entropy weight method and hierarchical analysis method to determine the quantization weights, comprehensively considering the influence of multiple factors.

[0127] Assuming that the first quantization level is L1, the second quantization level is L2, and the quantization weights are w1 and w2 respectively, the aging level L = w1 * L1 + w2 * L2.

[0128] The aging level obtained by weighted superposition can be a continuous value, or it can be discretized according to certain rules to obtain different aging level intervals. For example, the aging level can be divided into several level intervals such as mild aging, moderate aging, and severe aging, and the aging level interval to which the sample belongs can be determined based on the result of weighted superposition.

[0129] In the embodiment of the present application, by responding to the modeling request, the offline detection data and oil visual data corresponding to the transformer oil in multiple aging stages are obtained; the oil visual data is pixel-classified and labeled to obtain the classified and labeled image; the associated features of the offline detection data and the classified and labeled image are used to train the preset feature segmentation model to obtain the target feature segmentation model; when the sample data and sample image of the transformer oil to be detected are received, the target feature segmentation model is called in combination with the quantization weight to determine the aging level of the transformer oil to be detected. Thus, the machine learning method is applied to realize the recognition of the transformer oil aging characteristics and the determination of the aging level, and the aging characteristics can be more comprehensively and accurately recognized and judged by combining the aging feature detection index with the visual image.

[0130] See also Figure 4 , Figure 4 A structural block diagram of a transformer oil aging detection device in an embodiment of the present application is shown.

[0131] The embodiment of the present invention provides a transformer oil aging detection device, comprising:

[0132] The data acquisition module 401 is used to respond to the modeling request and acquire the offline detection data and oil visual data corresponding to the transformer oil at multiple aging stages;

[0133] An image classification and labeling module 402 is used to classify and label the oil visual data pixels to obtain a classified and labeled image;

[0134] A model training module 403 is used to train a preset feature segmentation model using the associated features of the offline detection data and the classified and annotated images to obtain a target feature segmentation model;

[0135] The aging level determination module 404 is used to determine the aging level of the transformer oil to be detected by calling the target feature segmentation model in combination with the quantization weight when receiving the sample data and sample image of the transformer oil to be detected.

[0136] Optionally, before executing the response modeling request and obtaining multiple sets of offline detection data and oil visual data of transformer oil, the device further includes:

[0137] The test simulation module is used to perform an aging test on the transformer oil of the same transformer according to multiple groups of different time and temperature conditions through an aging box, and respectively detect the offline test data corresponding to each group of transformer oil;

[0138] A visual data acquisition module, used to detect the oil visual data corresponding to each group of transformer oil by using at least one image imaging device;

[0139] Among them, the offline detection data includes chromaticity, acid value, breakdown voltage, volume resistivity and dielectric constant, and the oil visual data includes transformer oil color, metal particle information, non-metallic particle information, bubble information and water content information.

[0140] Optionally, the image classification and labeling module 402 is specifically used for:

[0141] If the oil visual data does not meet the labeling conditions, the oil visual data is labeled in response to the external input to obtain a classified labeled image;

[0142] If the oil visual data meets the labeling conditions, at least one image segmentation method is selected to segment the oil visual data according to the color, brightness or texture in the oil visual data to obtain multiple region segmentation images;

[0143] Each region segmentation image is labeled separately to obtain a classified labeled image.

[0144] Optionally, the model training module 403 includes:

[0145] The training association submodule is used to perform network segmentation according to the association relationship between offline detection data and classified annotated images to obtain associated feature pairs;

[0146] A training submodule is used to input the associated feature pairs into a preset feature segmentation model and output training feature categories;

[0147] The loss function calculation submodule is used to calculate the loss function value between the training feature category and the actual feature category;

[0148] The parameter adjustment submodule is used to use the optimizer to adjust the parameter gradient of the preset aging detection model if the loss function value exceeds the preset loss threshold, until the loss function value is less than the preset loss threshold, thereby obtaining the target feature segmentation model.

[0149] Optionally, the training association submodule is specifically used for:

[0150] Create feature nodes according to offline detection data and classified annotated images;

[0151] Calculate the correlation between feature nodes;

[0152] Connect each feature node according to the degree of association to generate a feature network;

[0153] The preset feature association model is used to segment the feature network and generate associated feature pairs.

[0154] Optionally, the aging level determination module 404 includes:

[0155] The sample association submodule is used to perform network segmentation on the sample data and the sample image when receiving the sample data and the sample image of the transformer oil to be tested, so as to obtain the sample association features;

[0156] A feature category determination submodule, used to classify the sample data according to the sample-related features through the target aging detection model, and determine the first aging feature category to which the sample data belongs and the second aging feature category to which the sample image belongs;

[0157] A level determination submodule, used to determine a first quantization level according to a first aging feature category and a preset determination criterion, and to determine a second quantization level according to a second aging feature category and a preset determination criterion;

[0158] The aging level determination submodule is used to perform weighted superposition of the first quantization level and the second quantization level to determine the aging level.

[0159] Optionally, the aging level determination submodule is specifically used for:

[0160] Calculate the first discrimination accuracy corresponding to the first aging feature category based on the confusion matrix of KNN supervised learning;

[0161] Calculating a second discrimination accuracy corresponding to a second aging feature category according to a plurality of image indicators;

[0162] Determining a quantization weight according to the first discrimination accuracy and the second discrimination accuracy;

[0163] The first quantization level and the second quantization level are weightedly superimposed according to the quantization weight to determine the aging level.

[0164] An embodiment of the present invention provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the transformer oil aging detection method as described in any embodiment of the present invention.

[0165] An embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed, the transformer oil aging detection method according to any embodiment of the present invention is implemented.

[0166] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices, modules and sub-modules can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0167] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0168] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0169] In addition, each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of software functional modules.

[0170] If the integrated module is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for an electronic device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0171] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting aging of transformer oil, characterized in that: include: Respond to modeling requests and obtain offline detection data and oil visual data corresponding to multiple aging stages of transformer oil; Classifying and labeling the oil visual data by pixels to obtain a classified and labeled image; Using the associated features of the offline detection data and the classified and annotated image to train a preset feature segmentation model to obtain a target feature segmentation model; When the sample data and sample image of the transformer oil to be detected are received, the target feature segmentation model is called in combination with the quantization weight to determine the aging grade of the transformer oil to be detected.

2. The method according to claim 1, characterized in that Before executing the response modeling request and obtaining multiple sets of transformer oil offline detection data and oil visual data, the method further includes: Performing an aging test on the transformer oil of the same transformer according to multiple groups of different time and temperature conditions through an aging box, and detecting the offline detection data corresponding to each group of the transformer oil respectively; Using at least one image forming device to detect the oil visual data corresponding to each group of transformer oil; The offline detection data include chromaticity, acid value, breakdown voltage, volume resistivity and dielectric constant, and the oil visual data include transformer oil color, metal particle information, non-metal particle information, bubble information and water content information.

3. The method according to claim 1, characterized in that The pixel classification and labeling of the oil visual data to obtain a classified and labeled image includes: If the oil visual data does not meet the labeling condition, labeling the oil visual data in response to external input to obtain a classified labeling image; If the oil visual data meets the labeling condition, at least one image segmentation method is selected to perform image segmentation on the oil visual data according to the color, brightness or texture in the oil visual data to obtain a plurality of region segmentation images; Each of the region segmentation images is labeled respectively to obtain a classified labeled image.

4. The method according to claim 1, characterized in that: The method of using the associated features of the offline detection data and the classified and annotated image to train a preset feature segmentation model to obtain a target feature segmentation model includes: Performing network segmentation according to the association relationship between the offline detection data and the classified and labeled image to obtain associated feature pairs; Inputting the associated feature pairs into a preset feature segmentation model and outputting training feature categories; Calculating the loss function value between the training feature category and the actual feature category; If the loss function value exceeds the preset loss threshold, an optimizer is used to perform parameter gradient adjustment on the preset aging detection model until the loss function value is less than the preset loss threshold, thereby obtaining a target feature segmentation model.

5. The method according to claim 4, characterized in that The network segmentation is performed according to the association relationship between the offline detection data and the classified annotated image to obtain the associated feature pair, including: Creating feature nodes respectively according to the offline detection data and the classified and labeled image; Calculating the correlation between the feature nodes; Connect each of the feature nodes according to the degree of association to generate a feature network; A preset feature association model is used to perform network segmentation on the feature network to generate associated feature pairs.

6. The method according to claim 1, characterized in that When the sample data and sample image of the transformer oil to be detected are received, the target feature segmentation model is called in combination with the quantization weight to determine the aging level of the transformer oil to be detected, including: When receiving sample data and sample images of transformer oil to be tested, performing network segmentation on the sample data and the sample image to obtain sample association features; Classifying according to the sample-related features through the target aging detection model to determine the first aging feature category to which the sample data belongs and the second aging feature category to which the sample image belongs; Determine a first quantization level according to the first aging feature category and a preset determination criterion, and determine a second quantization level according to the second aging feature category and a preset determination criterion; The first quantization level and the second quantization level are weighted and superimposed to determine an aging level.

7. The method according to claim 6, characterized in that The weighted superposition of the first quantization level and the second quantization level to determine the aging level includes: Calculate the first discrimination accuracy corresponding to the first aging feature category based on the confusion matrix of KNN supervised learning; Calculating a second discrimination accuracy corresponding to the second aging feature category according to a plurality of image indicators; Determining a quantization weight according to the first discrimination accuracy and the second discrimination accuracy; The first quantization level and the second quantization level are weightedly superimposed according to the quantization weight to determine an aging level.

8. A transformer oil aging detection device, characterized in that: include: A data acquisition module, used to respond to modeling requests and obtain offline detection data and oil visual data corresponding to transformer oil at multiple aging stages; An image classification and labeling module is used to classify and label the oil visual data pixels to obtain a classified and labeled image; A model training module, used to train a preset feature segmentation model using the associated features of the offline detection data and the classified and annotated image to obtain a target feature segmentation model; The aging grade determination module is used to call the target feature segmentation model in combination with the quantization weight to determine the aging grade of the transformer oil to be detected when receiving the sample data and sample image of the transformer oil to be detected.

9. An electronic device, characterized in that: The method comprises a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the transformer oil aging detection method according to any one of claims 1 to 7.

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

Images