A parameter verification method for distribution transformers based on machine vision

Abstract

The present invention relates to the field of power management technology, and more specifically, to a method for parameter calibration of distribution transformers based on machine vision. The method comprises: obtaining an appearance image of a distribution transformer, first classifying the image by a K-means clustering algorithm, and then extracting image features to obtain feature information; based on a grey correlation model, performing grey correlation analysis on the feature information extracted above to obtain theoretical parameters of the distribution transformer and a parameter calibration result of the distribution transformer. Compared with the direct feature extraction of image information in the prior art, the design of the present invention can be more efficient and representative; it can effectively improve the accuracy of parameter calibration of distribution transformers, and provide the possibility for machine calibration of distribution transformer parameters; using the distribution transformer parameters obtained by calibration, it can effectively solve the problems in the prior art, provide a reliable means for distribution transformer management and control, maintain the safe operation of the power system, and reduce economic losses.

Description

Technical Field

[0001] The present invention relates to the technical field of power management, and in particular to a method for detecting parameters of a distribution transformer based on machine vision. Background Art

[0002] The rated capacity of a transformer is one of the main parameters that characterizes the transformer's ability to transmit electrical energy. Its value is a conventional quantity that reflects the transformer's ability to transmit power. The power supply department selects the transformer capacity based on comprehensive factors such as the transformer's installation location, local power load, total load power, demand coefficient, normal overload capacity, accident overload tolerance, and the requirements for the transformer's economic operation.

[0003] Large industrial users in my country, defined as those with transformers over 315 kVA, are subject to a two-part electricity price system. Some electricity users reduce their basic electricity bills by changing the nameplate capacity of their distribution transformers, replacing large-capacity transformers with smaller ones. Uninformed by this, the power supply department still determines the power supply load based on the transformer's nameplate capacity. This increases the transformer's metering ratio and the protection settings. Long-term overloaded operation of the transformer, coupled with inoperative protection, can damage the distribution transformer and, in severe cases, even burn out. This not only impacts the safe operation of the power system but also causes economic losses and even threatens the lives of residents in the affected areas. With the advancement of power grid digitalization, the power system is moving towards the goal of being "comprehensive, objective, precisely measurable, and highly controllable." To address this, the design and application of an efficient and accurate distribution transformer parameter verification method is urgently needed. To address this, we propose a machine vision-based distribution transformer parameter verification method. Summary of the Invention

[0004] The purpose of the present invention is to provide a distribution transformer parameter verification method based on machine vision to solve the problems raised in the above background technology.

[0005] To solve the above technical problems, one of the objectives of the present invention is to provide a method for parameter verification of a distribution transformer based on machine vision, comprising the following steps:

[0006] S1. Obtain the appearance image of the distribution transformer, classify the image using the K-means clustering algorithm, and then extract image features to obtain feature information;

[0007] S2. Based on the grey correlation model, the feature information obtained by first classifying and then extracting image features using the K-means clustering algorithm is subjected to grey correlation analysis to obtain the theoretical parameters of the distribution transformer and the distribution transformer parameter verification results;

[0008] Among them: feature information is the feature information of the image, including the color feature, geometric feature, and texture feature of the image;

[0009] The grey relational model is trained based on a large number of historical images of distribution transformers whose actual parameters match the nameplate parameters and historical images of distribution transformers whose actual parameters do not match the nameplate parameters;

[0010] The distribution transformer parameter verification results include the distribution transformer rated capacity verification results, the distribution transformer rated voltage verification results and the distribution transformer impedance voltage verification results.

[0011] As a further improvement of the present technical solution, in S1, the appearance image of the distribution transformer is obtained, the image is first classified by the K-means clustering algorithm, and then the image features are extracted. The specific method flow for obtaining feature information includes the following steps:

[0012] S1.1. Preprocess the image to be tested and set the initial cluster center;

[0013] S1.2. Perform K-means clustering on the images to be tested based on the set cluster centers to obtain the clustering results of the images to be tested and the representative images of each cluster;

[0014] S1.3. Extract image features from the representative image of each cluster to obtain image feature information.

[0015] As a further improvement of the present technical solution, in S1.1, the specific method flow for preprocessing the image to be inspected includes the following steps:

[0016] S1.1.1. Convert the image to a grayscale image. Subtract the grayscale value of the representative image from the grayscale value of the reference image and multiply the difference by a gain factor to obtain a clear grayscale-corrected subtracted image.

[0017] S1.1.2. Design a program to remove jitter based on the reference image to reduce interference points in the test image removal result, and define the image absolute error to evaluate the degree of overlap between the test image and the reference image;

[0018] S1.1.3. Binarize the representative image, further reduce noise on the representative image to reduce interference points in the image, and then render to obtain a pre-processed image.

[0019] As a further improvement of the present technical solution, in S1.2, the specific method flow for performing K-means clustering on the image to be inspected includes the following steps:

[0020] S1.2.1. Perform pixel-space classification on the preprocessed image. Randomly select a pixel in each pixel cluster as the center point, calculate the distance between the pixels and the center point, and cluster the pixels according to the minimum distance principle to obtain a subset of pixels.

[0021] S1.2.2. Calculate the grayscale mean of the pixel subsets, and use the calculated grayscale value as the label of each pixel subset. Then use the local clustering classification method to obtain the final clustering result and the representative image of each cluster.

[0022] As a further improvement of the present technical solution, in S1.3, the specific method flow for extracting image features from the representative image of each cluster includes the following steps:

[0023] S1.3.1. Select the HSV color space as the analysis model, determine the quantization range of hue and saturation, obtain the quantized color histogram, and thus extract the image color features;

[0024] S1.3.2. Use a Gaussian filter to smooth the image, calculate the gradient magnitude and direction through finite differences of first-order partial derivatives, suppress the gradient magnitude using non-maximum methods, and use a double-threshold algorithm for edge detection to extract image geometric features.

[0025] S1.3.3. Calculate the gradient size and direction of the representative image, divide the image into cell units, construct a gradient direction histogram, and thus extract the texture features of the image.

[0026] As a further improvement of the present technical solution, in S2, based on the grey correlation model, the feature information obtained by first classifying and then extracting image features using the K-means clustering algorithm is subjected to grey correlation analysis to obtain the theoretical parameters of the distribution transformer and the parameter verification results of the distribution transformer. The grey correlation model based on this step needs to be pre-constructed and trained, wherein the specific method flow for constructing the grey correlation model includes the following steps:

[0027] S2.a.1. Determine a reference sequence that reflects the characteristics of the distribution transformer based on the distribution transformer's rated capacity, rated voltage, and impedance voltage verification results;

[0028] S2.a.2. Determine a comparison sequence of factors affecting distribution transformer parameters based on the color, geometric, and texture features of the image to be tested;

[0029] S2.a.3. Calculate the initial image of each sequence (perform dimensionless processing);

[0030] S2.a.4. Determine the absolute value sequence of the differences between the initial image components of the reference sequence and the comparison sequence, and calculate the minimum and maximum values ​​of the absolute value sequence.

[0031] S2.a.5. Calculate the correlation coefficient and correlation degree and establish a grey correlation model.

[0032] As a further improvement of the present technical solution, in said S2, in the process of pre-constructing and training the grey relational model, the specific method flow of training the grey relational model includes the following steps:

[0033] S2.b.1. Construct positive samples based on historical images of distribution transformers whose actual parameters match the nameplate parameters, and construct negative samples based on historical images of distribution transformers whose actual parameters do not match the nameplate parameters.

[0034] S2.b.2. Classify and extract features from the positive and negative images to obtain feature information of the positive and negative samples. Use part of the feature information of the positive and negative samples as a training set and the other part as a test set.

[0035] S2.b.3. Use the training set to train the established grey relational model to obtain multiple groups of grey relational models.

[0036] As a further improvement of the present technical solution, in S2.b.3, after the established grey relational model is trained using the training set and multiple groups of grey relational models are obtained, the optimal grey relational model needs to be selected through training. The specific method flow includes:

[0037] The test set is used to test each group of grey relational models, the accuracy of each group of grey relational models is calculated, and the model with an accuracy higher than the set threshold is selected as the optimal grey relational model.

[0038] As a further improvement of the present technical solution, in S2, based on the grey correlation model, grey correlation analysis is performed on the feature information obtained by first classifying and then extracting image features using the K-means clustering algorithm to obtain the theoretical parameters of the distribution transformer. After obtaining the distribution transformer parameter verification results, it is necessary to correct and improve the grey correlation model after verifying and accumulating the distribution transformer parameters. The specific method flow includes:

[0039] S2.c.1. Determine whether the verification result of the grey relational model is correct;

[0040] S2.c.2. Correct the incorrectly detected feature information and mark it as the correct sample category label. Use it as a training sample to continue training the grey relational model and update the grey relational model.

[0041] The second object of the present invention is to provide an operating platform device for a calibration method, comprising a processor, a memory, and a computer program stored in the memory and running on the processor, wherein the processor is configured to implement the steps of the above-mentioned machine vision-based distribution transformer parameter calibration method when executing the computer program.

[0042] A third object of the present invention is to provide a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned distribution transformer parameter verification method based on machine vision are implemented.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] 1. This machine vision-based distribution transformer parameter verification method uses a K-means clustering algorithm to first classify images and then extract representative image features. Compared with previous technologies that directly extract image information, this method is more efficient and representative.

[0045] 2. In this machine vision-based distribution transformer parameter verification method, the gray correlation model is used to verify the distribution transformer parameters, which can effectively improve the accuracy of distribution transformer parameter verification and provide the possibility for machine verification of distribution transformer parameters;

[0046] 3. In this machine vision-based distribution transformer parameter verification method, the distribution transformer parameters obtained by verification can effectively solve the problem of some power users changing the nameplate capacity of distribution transformers from large-capacity transformers to small-capacity transformers, providing a reliable means for distribution transformer management and control, maintaining the safe operation of the power system, and reducing economic losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 Schematic diagram of the overall verification method of the present invention;

[0048] Figure 2 This is an exemplary image classification and feature information extraction flow chart of the local detection method in the present invention;

[0049] Figure 3 Schematic diagram of the grey model training process of the local verification method of the present invention;

[0050] Figure 4 Schematic diagram of the flow of distribution transformer parameter verification and grey relational model retraining according to the local verification method of the present invention;

[0051] Figure 5 This is a flow chart of an exemplary embodiment of a method for parameter verification of a distribution transformer based on machine vision in the present invention;

[0052] Figure 6 It is a schematic diagram of the structure of an exemplary electronic computer platform device in the present invention. DETAILED DESCRIPTION

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments 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 making creative efforts are within the scope of protection of the present invention.

[0054] like Figures 1-6 As shown, this embodiment provides the following solutions:

[0055] Example 1

[0056] like Figure 1 As shown, this embodiment provides a method for parameter verification of a distribution transformer based on machine vision, comprising the following steps:

[0057] S1. Obtain the appearance image of the distribution transformer, classify the image using the K-means clustering algorithm, and then extract image features to obtain feature information;

[0058] S2. Based on the grey correlation model, the feature information obtained by first classifying and then extracting image features using the K-means clustering algorithm is subjected to grey correlation analysis to obtain the theoretical parameters of the distribution transformer and the distribution transformer parameter verification results;

[0059] Among them: feature information is the feature information of the image, including the color feature, geometric feature, and texture feature of the image;

[0060] The grey relational model is trained based on a large number of historical images of distribution transformers whose actual parameters match the nameplate parameters and historical images of distribution transformers whose actual parameters do not match the nameplate parameters;

[0061] The distribution transformer parameter verification results include the distribution transformer rated capacity verification results, the distribution transformer rated voltage verification results and the distribution transformer impedance voltage verification results.

[0062] In this embodiment, Figure 2 As shown, in S1, the appearance image of the distribution transformer is obtained, the image is first classified using the K-means clustering algorithm, and then image features are extracted. The specific method flow for obtaining feature information includes the following steps:

[0063] S1.1. Preprocess the image to be tested and set the initial cluster center;

[0064] S1.2. Perform K-means clustering on the images to be tested based on the set cluster centers to obtain the clustering results of the images to be tested and the representative images of each cluster;

[0065] S1.3. Extract image features from the representative image of each cluster to obtain image feature information.

[0066] Furthermore, in S1.1, the specific method flow for preprocessing the image to be inspected includes the following steps:

[0067] S1.1.1. Convert the image to a grayscale image. Subtract the grayscale value of the representative image from the grayscale value of the reference image and multiply the difference by a gain factor to obtain a clear grayscale-corrected subtracted image.

[0068] S1.1.2. Design a program to remove jitter based on the reference image to reduce interference points in the test image removal result, and define the image absolute error to evaluate the degree of overlap between the test image and the reference image;

[0069] S1.1.3. Binarize the representative image, further reduce noise on the representative image to reduce interference points in the image, and then render to obtain a pre-processed image.

[0070] Furthermore, in S1.2, the specific method flow for performing K-means clustering on the image to be inspected includes the following steps:

[0071] S1.2.1. Perform pixel-space classification on the preprocessed image. Randomly select a pixel in each pixel cluster as the center point, calculate the distance between the pixels and the center point, and cluster the pixels according to the minimum distance principle to obtain a subset of pixels.

[0072] S1.2.2. Calculate the grayscale mean of the pixel subsets, and use the calculated grayscale value as the label of each pixel subset. Then use the local clustering classification method to obtain the final clustering result and the representative image of each cluster.

[0073] Furthermore, in S1.3, the specific method flow for extracting image features from the representative image of each cluster includes the following steps:

[0074] S1.3.1. Select the HSV color space as the analysis model, determine the quantization range of hue and saturation, obtain the quantized color histogram, and thus extract the image color features;

[0075] S1.3.2. Use a Gaussian filter to smooth the image, calculate the gradient magnitude and direction through finite differences of first-order partial derivatives, suppress the gradient magnitude using non-maximum methods, and use a double-threshold algorithm for edge detection to extract image geometric features.

[0076] S1.3.3. Calculate the gradient size and direction of the representative image, divide the image into cell units, construct a gradient direction histogram, and thus extract the texture features of the image.

[0077] In this embodiment, in S2, based on the grey correlation model, grey correlation analysis is performed on the feature information obtained by first classifying and then extracting image features using the K-means clustering algorithm to obtain the theoretical parameters of the distribution transformer and the distribution transformer parameter verification results. The grey correlation model based on this step needs to be pre-constructed and trained. The specific method flow for constructing the grey correlation model includes the following steps:

[0078] S2.a.1. Determine a reference sequence that reflects the characteristics of the distribution transformer based on the distribution transformer's rated capacity, rated voltage, and impedance voltage verification results;

[0079] S2.a.2. Determine a comparison sequence of factors affecting distribution transformer parameters based on the color, geometric, and texture features of the image to be tested;

[0080] S2.a.3. Calculate the initial image of each sequence (perform dimensionless processing);

[0081] S2.a.4. Determine the absolute value sequence of the differences between the initial image components of the reference sequence and the comparison sequence, and calculate the minimum and maximum values ​​of the absolute value sequence.

[0082] S2.a.5. Calculate the correlation coefficient and correlation degree and establish a grey correlation model.

[0083] Furthermore, if Figure 3 As shown, in S2, in the process of pre-constructing and training the grey relational model, the specific method flow of training the grey relational model includes the following steps:

[0084] S2.b.1. Construct positive samples based on historical images of distribution transformers whose actual parameters match the nameplate parameters, and construct negative samples based on historical images of distribution transformers whose actual parameters do not match the nameplate parameters.

[0085] S2.b.2. Classify and extract features from the positive and negative images to obtain feature information of the positive and negative samples. Use part of the feature information of the positive and negative samples as a training set and the other part as a test set.

[0086] S2.b.3. Use the training set to train the established grey relational model to obtain multiple groups of grey relational models.

[0087] Furthermore, in S2.b.3, the established grey relational model is trained using the training set. After obtaining multiple groups of grey relational models, the optimal grey relational model needs to be selected through training. The specific method flow includes:

[0088] The test set is used to test each group of grey relational models, the accuracy of each group of grey relational models is calculated, and the model with an accuracy higher than the set threshold is selected as the optimal grey relational model.

[0089] Specifically, the establishment process of the grey relational model can be summarized as follows:

[0090] Based on the distribution transformer rated capacity, rated voltage, and impedance voltage test results, a reference sequence X0 reflecting the characteristics of the distribution transformer is determined. Based on the color features, geometric features, and texture features of the image to be tested, a comparison sequence X consisting of factors affecting the distribution transformer parameters is determined. m ;

[0091] The reference sequence reflecting the characteristics of the distribution transformer is: X0 = {X0(1), X0(2), ..., X0(n)}, and the comparison sequence of the factors affecting the parameters of the distribution transformer is: X1 = {X1(1), X1(2), ..., X1(n)}, X2 = {X2(1), X2(2), ..., X2(n)}, ..., X m ={X m (1),X m (2),...,X m (n)};

[0092] Find the initial image of each sequence (perform dimensionless processing): Let X′=X i / X i (1) = {X′ i (1),X′ i (2),...,X′ i (n)}, where i = 0, 1, 2, ..., m, we get X′0, X′1, ..., X′ m ;

[0093] Find the reference sequence X0 and the comparison sequence X m The absolute value sequence of the difference between the initial value image and the corresponding components, and calculate the minimum and maximum values ​​of the absolute value sequence; record Δ i (k) = |X′0(k)-X′ i (k)|,Δ i =(Δ i (1),Δ i(2),...,Δ i (n)), i=1,2,...,m;k=1,2,...,n;the minimum and maximum values ​​of the absolute value sequence are respectively recorded as: Δ min =min i min k Δ i (k), Δ max =max i max k Δ i (k);

[0094] Find the correlation coefficient and correlation degree: correlation coefficient ξ 0i (k)=(Δ min +pΔ max ) / (Δ i (k)+pΔ max ), where p is the resolution coefficient, usually p = 0.5; correlation

[0095] Furthermore, part of the feature information of the positive samples and negative samples is used as a training set, and the other part is used as a test set;

[0096] Input the training set into the grey relational model for training to obtain multiple groups of grey relational models;

[0097] The accuracy of each group of grey relational models is verified using the test set, the accuracy threshold is set, and the optimal grey management model is selected.

[0098] Among them, it refers to the fact that as the number of distribution transformer installations continues to increase, the number of distribution transformer images collected also continues to increase. Self-learning and improvement are achieved through the verification of the grey correlation model, and the erroneous samples are marked and changed into training samples to continue training and updating the model.

[0099] Furthermore, if Figure 4 As shown in S2, based on the grey correlation model, the feature information obtained by first classifying and then extracting image features using the K-means clustering algorithm is subjected to grey correlation analysis to obtain the theoretical parameters of the distribution transformer. After obtaining the distribution transformer parameter verification results, it is necessary to correct and improve the grey correlation model after verifying and accumulating the distribution transformer parameters. The specific method flow includes:

[0100] S2.c.1. Determine whether the test results of the grey relational model are correct;

[0101] S2.c.2. Correct the incorrectly detected feature information and mark it as the correct sample category label. Use it as a training sample to continue training the grey relational model and update the grey relational model.

[0102] The purpose of this scheme is to first classify the image through the K-means clustering algorithm, and then extract representative image features, which can more efficiently and representatively extract image feature information. The calibration of distribution transformer parameters through the grey correlation model can effectively improve the accuracy of distribution transformer parameter calibration, provide the possibility for machine calibration of distribution transformer parameters, and effectively solve the problem of some power users changing the nameplate capacity of distribution transformers and replacing large-capacity transformers with small-capacity transformers. It provides a reliable means for distribution transformer management and control, maintains the safe operation of the power system, and prevents economic losses.

[0103] Furthermore, the above solution can be simplified as follows:

[0104] 1-1. First, it is necessary to collect image information of the distribution transformer and obtain image information using a camera and a photo taking device;

[0105] 1-2. Perform image preprocessing and format unification. The purpose is: first, to eliminate images with high similarity and reduce the number of image samples; second, to unify the image format conversion;

[0106] 1-3. Preprocess the image to be tested, convert it into a grayscale image, design a program to remove jitter based on the reference image to reduce interference points in the image to be tested, perform binarization on the image, and then render it to obtain the preprocessed image;

[0107] 1-4. Perform K-means clustering on the image to obtain the clustering results of the image to be tested and the representative image of each cluster;

[0108] 1-5. Extract the image features of the distribution transformer to obtain color feature information, geometric feature information, and texture feature information.

[0109] The images in 1-1 to 1-5 above include both images to be tested and sample images; the sample images are used to train and test the grey relational model.

[0110] Among them, unlike the images to be tested, the sample images need to be divided into two categories after 1-2, and then 1-3 are performed. Specifically, construct positive samples, use images whose distribution transformer parameters are consistent with the nameplate parameters as positive samples, and normalize the sample images; construct negative samples, use images whose distribution transformer parameters are consistent with the nameplate parameters as negative samples, and normalize them according to the size of the positive samples.

[0111] Example 2

[0112] like Figure 5As shown, this embodiment also provides a flow chart of an embodiment of a method for parameter verification of a distribution transformer based on machine vision, and the parameter verification of the distribution transformer specifically includes:

[0113] Step 1: Collect distribution transformer images;

[0114] Step 2: Image preprocessing and format unification;

[0115] Step 3: Image K-means clustering;

[0116] Step 4: Extract image feature information;

[0117] Step 5: Establish and train the grey relational model;

[0118] Step 6: Verify the parameters of the distribution transformer;

[0119] Step 7: Improvement of grey relational model.

[0120] Through steps 1 to 7, distribution transformer image information collection, image preprocessing and format unification are realized. The images are first classified through K-means clustering, and then feature information is extracted from representative images to obtain a distribution transformer parameter sequence. The feature information parameter sequence of the sample image is input into the grey relational model. After meeting the accuracy threshold requirement, a trained grey relational model is obtained to verify the parameters of the distribution transformer. At the same time, based on the verification results, the grey relational model is retrained to improve the accuracy and reliability of detection.

[0121] like Figure 6 As shown, this embodiment also provides an operating platform device for the verification method, which includes a processor, a memory, and a computer program stored in the memory and running on the processor.

[0122] The processor includes one or more processing cores, and the processor is connected to the memory through a bus. The memory is used to store program instructions. When the processor executes the program instructions in the memory, the above-mentioned distribution transformer parameter verification method based on machine vision is realized.

[0123] Alternatively, the memory may be implemented by any type of volatile or non-volatile memory device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0124] In addition, the present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the above-mentioned distribution transformer parameter verification method based on machine vision are implemented.

[0125] Optionally, the present invention further provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute the steps of the above-mentioned method for parameter verification of distribution transformers based on machine vision.

[0126] Those skilled in the art will appreciate that the process of implementing all or part of the steps of the above embodiments may be accomplished by hardware, or may be accomplished by instructing the relevant hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk, or an optical disk, etc.

[0127] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for parameter verification of distribution transformers based on machine vision, characterized in that: The steps include: S1. Obtain the appearance image of the distribution transformer, classify the image using the K-means clustering algorithm, and then extract image features to obtain feature information; S2. Based on the grey correlation model, the feature information obtained by first classifying and then extracting image features using the K-means clustering algorithm is subjected to grey correlation analysis to obtain the theoretical parameters of the distribution transformer and the distribution transformer parameter verification results; Among them: feature information is the feature information of the image, including the color feature, geometric feature, and texture feature of the image; The grey relational model is trained based on a large number of historical images of distribution transformers whose actual parameters match the nameplate parameters and historical images of distribution transformers whose actual parameters do not match the nameplate parameters; Distribution transformer parameter verification results, including distribution transformer rated capacity verification results, distribution transformer rated voltage verification results, and distribution transformer impedance voltage verification results; In S1, the appearance image of the distribution transformer is obtained, the image is first classified using the K-means clustering algorithm, and then image features are extracted. The specific method flow for obtaining feature information includes the following steps: S1.

1. Preprocess the image to be tested and set the initial cluster center; S1.

2. Perform K-means clustering on the images to be tested based on the set cluster centers to obtain the clustering results of the images to be tested and the representative images of each cluster; S1.

3. Extract image features from the representative image of each cluster to obtain image feature information; In S1.2, the specific method flow for performing K-means clustering on the image to be tested includes the following steps: S1.2.

1. Perform pixel-space classification on the preprocessed image. Randomly select a pixel in each pixel cluster as the center point, calculate the distance between the pixels and the center point, and cluster the pixels according to the minimum distance principle to obtain a subset of pixels. S1.2.

2. Calculate the grayscale mean of each pixel subset, use the calculated grayscale value as a label for each pixel subset, and perform local clustering classification to obtain the final clustering results and a representative image for each cluster; In S1.3, the specific method flow for extracting image features from the representative image of each cluster includes the following steps: S1.3.

1. Select the HSV color space as the analysis model, determine the quantization range of hue and saturation, obtain the quantized color histogram, and thus extract the image color features; S1.3.

2. Use a Gaussian filter to smooth the image, calculate the gradient magnitude and direction through finite differences of first-order partial derivatives, suppress the gradient magnitude using non-maximum methods, and use a double-threshold algorithm for edge detection to extract image geometric features. S1.3.

3. Calculate the gradient size and direction of the representative image, divide the image into cell units, construct a gradient direction histogram, and thus extract the texture features of the image.

2. The method for parameter verification of distribution transformers based on machine vision according to claim 1, characterized in that: In S1.1, the specific method flow for preprocessing the image to be tested includes the following steps: S1.1.

1. Convert the image to a grayscale image. Subtract the grayscale value of the representative image from the grayscale value of the reference image and multiply the difference by a gain factor to obtain a clear grayscale-corrected subtracted image. S1.1.

2. Design a program to remove jitter based on the reference image to reduce interference points in the test image removal result, and define the image absolute error to evaluate the degree of overlap between the test image and the reference image; S1.1.

3. Binarize the representative image, further reduce noise on the representative image to reduce interference points in the image, and then render to obtain a pre-processed image.

3. The method for parameter verification of distribution transformers based on machine vision according to claim 2, characterized in that: In S2, based on the grey correlation model, grey correlation analysis is performed on the feature information obtained by first classifying and then extracting image features using the K-means clustering algorithm to obtain the theoretical parameters of the distribution transformer and the distribution transformer parameter verification results. The grey correlation model based on this step needs to be pre-constructed and trained. The specific method flow for constructing the grey correlation model includes the following steps: S2.a.

1. Determine a reference sequence that reflects the characteristics of the distribution transformer based on the distribution transformer's rated capacity, rated voltage, and impedance voltage verification results; S2.a.

2. Determine a comparison sequence of factors affecting distribution transformer parameters based on the color, geometric, and texture features of the image to be tested; S2.a.

3. Find the initial image of each sequence; S2.a.

4. Determine the absolute value sequence of the differences between the initial image components of the reference sequence and the comparison sequence, and calculate the minimum and maximum values ​​of the absolute value sequence. S2.a.

5. Calculate the correlation coefficient and correlation degree and establish a grey correlation model.

4. The method for parameter verification of distribution transformers based on machine vision according to claim 3, characterized in that: In the process of pre-constructing and training the grey relational model in S2, the specific method flow of training the grey relational model includes the following steps: S2.b.

1. Construct positive samples based on historical images of distribution transformers whose actual parameters match the nameplate parameters, and construct negative samples based on historical images of distribution transformers whose actual parameters do not match the nameplate parameters. S2.b.

2. Classify and extract features from the positive and negative images to obtain feature information of the positive and negative samples. Use part of the feature information of the positive and negative samples as a training set and the other part as a test set. S2.b.

3. Use the training set to train the established grey relational model to obtain multiple groups of grey relational models.

5. The method for parameter verification of distribution transformers based on machine vision according to claim 4, characterized in that: In S2.b.3, the established grey relational model is trained using the training set to obtain multiple groups of grey relational models. Then, the optimal grey relational model needs to be selected through training. The specific method flow includes: The test set is used to test each group of grey relational models, the accuracy of each group of grey relational models is calculated, and the model with an accuracy higher than the set threshold is selected as the optimal grey relational model.

6. The method for parameter verification of distribution transformers based on machine vision according to claim 5, characterized in that: In S2, based on the grey correlation model, the feature information obtained by first classifying and then extracting image features using the K-means clustering algorithm is subjected to grey correlation analysis to obtain the theoretical parameters of the distribution transformer. After obtaining the distribution transformer parameter verification results, it is necessary to correct and improve the grey correlation model by verifying and accumulating the distribution transformer parameters. The specific method flow includes: S2.c.

1. Determine whether the verification result of the grey relational model is correct; S2.c.

2. Correct the incorrectly detected feature information and mark it as the correct sample category label. Use it as a training sample to continue training the grey relational model and update the grey relational model.

Images