Method, device and equipment for identifying running state of high-voltage circuit breaker and storage medium

Abstract

The invention relates to the technical field of electric power system monitoring and computer vision, and provides a high-voltage circuit breaker operation state identification method, device and equipment and a storage medium, and the method comprises the steps: collecting the image data of a high-voltage circuit breaker comprising an indication sign, and carrying out generative adversarial network-based style migration on the collected image data, making a target detection data set, training a target detection model constructed based on a structured state space sequence, and identifying the running state of the high-voltage circuit breaker by using the trained target detection model. The method provided by the invention is suitable for the task of automatically identifying the switching position information of the ultra-high-voltage circuit breaker, realizes the discrimination of the switching indication mark and the analysis of the operation state of the circuit breaker, and can improve the accuracy of the visual judgment of the operation state of the ultra-high-voltage circuit breaker.

Description

Technical Field

[0001] The present application relates to the field of power system monitoring and computer vision technology, and specifically to a method, device, equipment and storage medium for identifying the operating status of a high-voltage circuit breaker. Background Art

[0002] In the power system, the main function of the ultra-high voltage circuit breaker is to quickly cut off the fault circuit when a fault occurs in the power grid to prevent the fault from expanding, thereby protecting the safe operation of the power grid and related equipment. As a vital protection device in the power system, the ultra-high voltage circuit breaker's operating status indicator (such as working status indicator light, mechanical position indicator, etc.) is of great significance to ensure the safe and stable operation of the power grid. These indicators convey the working status of the ultra-high voltage circuit breaker, such as whether it is closed, etc., through changes in color or mechanical position.

[0003] Ultra-high voltage circuit breakers are mostly installed in outdoor environments. The surface of the glass shell used for the indicator signs will inevitably become rough under the continuous attack of wind and sand and the baptism of rain. In addition, due to the alternation of temperature differences between day and night and the temperature changes of the four seasons, the glass shell frequently experiences physical stresses of thermal expansion and contraction, which may induce the formation of fine cracks or scratches. In addition, long-term ultraviolet radiation will gradually cause the material of the glass shell to age, causing yellowing or darkening, thereby reducing the clarity of light transmission. Therefore, for ultra-high voltage circuit breakers exposed to natural conditions for a long time, the aging of the glass shell will increase the difficulty of identifying the internal indicator signs.

[0004] In the prior art, there is a lack of methods that can effectively identify the operating status of ultra-high voltage circuit breakers. Summary of the invention

[0005] To solve the above problems, the embodiments of the present application provide a method, device, equipment and storage medium for identifying the operating status of a high-voltage circuit breaker, collect image data of ultra-ultra-high voltage circuit breakers with different degrees of aging under various environmental conditions, and use style transfer technology to effectively transfer the aging style to normal images, thereby establishing a target detection data set, and using the target detection data set to train artificial intelligence and machine learning models to accurately identify and classify the operating status of ultra-ultra-high voltage circuit breakers. Even when the indicator signs become unclear due to aging, accurate judgments can be made, thereby improving the accuracy of identifying the aging status of ultra-ultra-high voltage circuit breakers, thereby helping to quickly locate problems, reduce misjudgments and missed judgments, and improve operation and maintenance efficiency.

[0006] The present application embodiment adopts the following technical solutions:

[0007] In a first aspect, an embodiment of the present application provides a method for identifying an operating state of a high-voltage circuit breaker, the method comprising:

[0008] Acquire image data of a high-voltage circuit breaker including an indicator mark, and pre-process the image data;

[0009] dividing the preprocessed image data into a clear image data set and an aged image data set according to whether the indicator mark is aged, and classifying the aged image data set according to the aged style;

[0010] According to the clear image dataset and the aged image dataset, an image generation model constructed based on a style transfer generative adversarial network model is trained, and the clear image dataset is input into the trained image generation model for style transfer to generate a virtual image dataset;

[0011] Performing box selection and annotation on the aging image dataset and the virtual image dataset to construct an object detection dataset corresponding to the aging style;

[0012] Based on the target detection data set, a target detection model constructed based on a structured state space sequence model is trained, and the trained target detection model is used to identify the operating status of a high-voltage circuit breaker.

[0013] In a second aspect, an embodiment of the present application provides a device for identifying an operating state of a high-voltage circuit breaker, the device comprising:

[0014] An image data acquisition unit, used to acquire image data of the high-voltage circuit breaker including the indicator mark, and pre-process the image data;

[0015] a real image data set establishing unit, configured to divide the preprocessed image data into a clear image data set and an aged image data set according to whether the indicator mark is aged, and classify the aged image data set according to the aged style;

[0016] A virtual image data set establishing unit, configured to train an image generation model constructed based on a style transfer generative adversarial network model according to the aged image data set, and to input the clear image data set into the trained image generation model for style transfer to generate a virtual image data set;

[0017] An object detection data set establishment unit, configured to perform box selection and annotation on the aging image data set and the virtual image data set to establish an object detection data set corresponding to an aging style;

[0018] An operating status recognition unit is used to train a target detection model constructed based on a structured state space sequence model based on the target detection data set, and use the trained target detection model to identify the operating status of the high-voltage circuit breaker.

[0019] In a third aspect, an embodiment of the present application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned method for identifying the operating status of a high-voltage circuit breaker when executing the computer program.

[0020] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program, and when the computer program is instructed by a processor, it implements the steps of the above-mentioned method for identifying the operating status of a high-voltage circuit breaker.

[0021] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects:

[0022] The present application provides a method for identifying the operating status of a high-voltage circuit breaker. First, image data of a high-voltage circuit breaker including an indicator mark is acquired, and the image data is preprocessed. The preprocessed image data is divided into a clear image data set and an aged image data set according to whether the indicator mark is aged, and the aged image data set is classified according to the aging style; then, an image generation model constructed based on a style transfer generative adversarial network model is trained according to the clear image data set and the aged image data set, and the clear image data set is input into the trained image generation model for style transfer to generate a virtual image data set; then, the aged image data set and the virtual image data set are frame-selected and annotated to construct a target detection data set corresponding to the aging style; finally, based on the target detection data set, a target detection model constructed based on a structured state space sequence model is trained, and the trained target detection model is used to identify the operating status of the high-voltage circuit breaker. The method provided in the present application collects image data of ultra-ultra-high voltage circuit breakers with different degrees of aging under various environmental conditions, and uses style transfer technology to effectively transfer the aging style to normal images, thereby establishing a target detection data set, and uses the target detection data set to train artificial intelligence and machine learning models to accurately identify and classify the operating status of ultra-ultra-high voltage circuit breakers. Even when the indicator signs become unclear due to aging, accurate judgments can be made, thereby improving the accuracy of identifying the aging status of ultra-ultra-high voltage circuit breakers, which helps to quickly locate problems, reduce misjudgments and missed judgments, and improve operation and maintenance efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0024] Figure 1 A schematic flow chart showing a method for identifying an operating state of a high-voltage circuit breaker according to an embodiment of the present application;

[0025] Figure 2A schematic diagram showing a framework of a structured state space sequence model according to an embodiment of the present application is shown;

[0026] Figure 3 A schematic structural diagram showing a device for identifying the operating status of a high-voltage circuit breaker according to an embodiment of the present application;

[0027] Figure 4 A schematic structural diagram of a computer device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0029] The present application is described in detail below through specific embodiments.

[0030] Figure 1 A schematic diagram showing a flow chart of a method for identifying the operating status of a high-voltage circuit breaker proposed in an embodiment of the present application is shown. Figure 1 As shown, this embodiment includes steps S110 to S150:

[0031] Step S110, acquiring image data of the high-voltage circuit breaker including the indicator mark, and preprocessing the image data.

[0032] The image data of the ultra-high voltage circuit breaker is collected by using the substation monitoring equipment or the image acquisition equipment. The image acquisition equipment may include a drone equipped with a camera, a jib camera, etc. The image data at least includes the indication mark of the ultra-high voltage circuit breaker, and the indication mark can at least be used to indicate the operating status of the ultra-high voltage circuit breaker, and the indication mark may include a working status indicator light, a mechanical position indicator, etc.

[0033] The collected image data is preprocessed to filter out abnormal image data such as incomplete indication signs in the image, overexposure or underexposure of the image resulting in unrecognizable content, etc.

[0034] Step S120 , the pre-processed image data are divided into a clear image data set and an aged image data set according to whether the indicator mark is aged, and the aged image data set is classified according to the aged style.

[0035] The preprocessed image data is divided into two data sets according to whether it is aged or not. For the new UHV circuit breaker, its image data is clear, so this part of the image data can be divided into a clear image data set. For the other part of the image data, it can be divided into an aged image data set. The images in the aged image data set are classified according to the aging style. The aging style can include crack style, stain style, discoloration style, spot style, etc. according to the different aging conditions of the UHV circuit breaker.

[0036] Step S130, training an image generation model based on a style transfer generative adversarial network model according to the clear image dataset and the aged image dataset, inputting the clear image dataset into the trained image generation model for style transfer to generate a virtual image dataset.

[0037] Style transfer technology is used to simulate the aging of the UHV circuit breaker indicator signs, and the aging style is effectively transferred to the images in the clear image dataset, thereby generating a virtual image dataset to expand the data volume and improve the recognition effect. The image generation model can include multiple sub-models, each of which is constructed by training a style transfer generative adversarial network model using an image of an aging style in the aging image dataset. By establishing an image generation model based on a style transfer generative adversarial network, the adaptive enhancement function can be achieved by introducing an adaptive instance normalization mechanism, so that the generator network can dynamically adjust the content and style of the generated image according to the characteristics of the aging image dataset while performing style transfer, thereby making the generated image more realistic and diverse.

[0038] Step S140 , performing box selection and annotation on the aging image dataset and the virtual image dataset to construct an object detection dataset corresponding to the aging style.

[0039] The aging image dataset and the virtual image dataset are marked by box selection to produce the target detection dataset. The indicator signs in the aging image dataset and the virtual image dataset are marked by box selection to mark the category of the operating state indicated by the indicator sign. Since the images in the aging image dataset and the virtual image dataset correspond to the corresponding aging style, the images in the target detection dataset have the corresponding features of "image-aging style-operating state category".

[0040] Step S150: Based on the target detection data set, a target detection model constructed based on the structured state space sequence model is trained, and the trained target detection model is used to identify the operating state of the high-voltage circuit breaker.

[0041] The structured state space sequence model is trained using the target detection dataset to obtain the target detection model. The trained target detection model can recognize the indicator images of ultra-high voltage circuit breakers with different aging styles and determine the operating status of the ultra-high voltage circuit breakers. The trained target detection model can be embedded in the circuit breaker monitoring system to monitor the operating status of the ultra-high voltage circuit breaker in real time.

[0042] from Figure 1 It can be seen from the method shown that the method for identifying the operating status of a high-voltage circuit breaker provided in the present application first obtains image data of a high-voltage circuit breaker including an indicator mark, and preprocesses the image data, and divides the preprocessed image data into a clear image data set and an aging image data set according to whether the indicator mark is aging, and classifies the aging image data set according to the aging style; then, an image generation model constructed based on a style transfer generative adversarial network model is trained according to the clear image data set and the aging image data set, and the clear image data set is input into the trained image generation model for style transfer to generate a virtual image data set; then, the aging image data set and the virtual image data set are box-selected and annotated to construct a target detection data set corresponding to the aging style; finally, based on the target detection data set, a target detection model constructed based on a structured state space sequence model is trained, and the trained target detection model is used to identify the operating status of the high-voltage circuit breaker. The method provided in the present application collects images of ultra-ultra-high voltage circuit breakers with different degrees of aging under various environmental conditions, and uses style transfer technology to effectively transfer the aging style to normal images, thereby establishing an aging image dataset, and using the aging image dataset to train artificial intelligence and machine learning models to accurately identify and classify the operating status of ultra-ultra-high voltage circuit breakers. Even when the indicator signs become unclear due to aging, accurate judgments can be made, thereby improving the accuracy of identifying the aging status of ultra-ultra-high voltage circuit breakers, which helps to quickly locate problems, reduce misjudgments and missed judgments, and improve operation and maintenance efficiency.

[0043] In some embodiments of the present application, in the above method, step S110, obtaining image data of a high-voltage circuit breaker including an indicator mark, and preprocessing the image data, includes: using a substation monitoring device or an image acquisition device to regularly obtain image data of a high-voltage circuit breaker including an indicator mark; wherein the indicator mark includes an opening and closing indicator mark of the high-voltage circuit breaker; performing data verification and quality inspection on the image data, and removing abnormal image data that does not meet quality rules; extracting a region of interest in the retained image data; wherein the region of interest includes the indicator mark.

[0044] Use substation monitoring equipment or other image acquisition equipment to regularly acquire image data of the ultra-ultra-high voltage circuit breaker indicator. The image data may include the opening and closing indicator of the ultra-ultra-high voltage circuit breaker body. In the acquired image data, the operating status of the ultra-ultra-high voltage circuit breaker is distributed as evenly as possible, that is, the number of image data of the closing and opening states, and the number of image data of different types of ultra-ultra-high voltage circuit breakers are roughly equal.

[0045] The collected image data is verified and quality checked to ensure that the collected image data is complete and of good quality. For the retained image data, the region of interest is extracted, that is, the region irrelevant to the opening and closing indication marks of the ultra-high voltage circuit breaker is removed to obtain the pre-processed image data.

[0046] In some embodiments of the present application, in the above method, step S120, dividing the preprocessed image data into a clear image data set and an aged image data set according to whether the indicator mark is aging, and classifying the aged image data set according to the aging style, includes: dividing the image data into a clear image data set and an aged image data set according to whether the indicator mark in the preprocessed image data is aging; classifying the aged image data set into an aging style according to the aging type; wherein the aging type includes cracks, stains, discoloration, and spots.

[0047] The preprocessed image data is divided into two data sets according to whether there is aging phenomenon. The clear image data set is the image data without aging phenomenon; the aged image data set is the image data with aging phenomenon. According to the aging type, the aged image data set is classified into crack style images, stain style images, discoloration style images, and spot style images.

[0048] In some embodiments of the present application, in the above method, step S130, trains an image generation model constructed based on a style transfer generative adversarial network model according to a clear image dataset and an aging image dataset, inputs the clear image dataset into the trained image generation model for style transfer, and then generates a virtual image dataset, including: forming a content picture based on extracting content features of each image in the clear image dataset, and forming a style picture based on extracting style features of each image in the aging image dataset; inputting the content picture into the generation network to obtain a generated picture; inputting the content picture, the style picture, and the generated picture into the discriminant network, and using an adversarial loss function to measure the difference in content and style; alternately updating the generation network and the discriminant network according to the adversarial loss function until the difference in content and style is minimized, and obtaining a trained image generation model; after inputting the clear image dataset into the trained image generation model for style transfer, a virtual image dataset with an aging style but unchanged original content is generated.

[0049] Image features are extracted based on the clear image dataset. The image features are content features used to represent the content of the image. Image features are extracted based on the aged image dataset. The image features are style features used to represent the style of the image. Content features based on the clear image dataset form a content picture. Style features based on the aged image dataset form a style picture.

[0050] Since the aging image dataset includes images of multiple aging styles, a style transfer generative adversarial network model can be trained based on each aging style image to form a sub-model, and multiple sub-models corresponding to different aging styles constitute a trained image generation model.

[0051] When a clear image dataset is input into a trained image generation model, style transfer can be performed through multiple sub-models to generate a virtual image dataset with an aging style but unchanged original content. In the generated virtual image dataset, the number of images with different aging styles can be roughly equal.

[0052] The style transfer generative adversarial network model (GAN) includes a generator network and a discriminator network. Both the generator network and the discriminator network are deep convolutional networks. In the process of training the image generation model, the generator network and the discriminator network are continuously optimized alternately. Among them, the discriminator network measures the content loss and style loss according to the adversarial loss function, and conducts adversarial training with the generator network until the loss is minimized, and finally obtains the optimal image generation model. The image generation model can make the images in the generated virtual image dataset closer to the content of the images in the input clear picture dataset, and closer to the aging style of the images in the aging image dataset.

[0053] In some cases, in the above method, the content image, the style image and the generated image are input into the discriminator network, and the adversarial loss function is used to measure the differences in content and style, including: measuring the content difference between the content image and the generated image through the content loss function; measuring the style difference between the style image and the generated image through the style loss function; measuring the discriminative difference between the generated image and the content image and the style image through the contrastive learning loss function; determining the adversarial loss function based on the content loss function, the style loss function and the contrastive learning loss function.

[0054] The discriminator network measures the content difference between two images (i.e., the content image and the generated image) through the content loss function. It is defined as follows:

[0055]

[0056] Among them, X and Y represent the feature maps extracted from the two images, L represents the subscript of the convolutional layer, and I and J represent the subscripts of the rows and columns in the matrix, respectively.

[0057] The discriminator network measures the style difference between two images (i.e., the style image and the generated image) through the style loss function.

[0058] The style loss function uses the Gram matrix of the feature map to represent the style features of the image. The l-th layer feature map of the style image is X l , and the corresponding Gram matrix G l The definition is as follows:

[0059]

[0060] in, and Represent the feature map X l The eigenvalues ​​of the i-th channel and the j-th channel at position (h, w) in the image. Represents the feature map X l The correlation between different channels in is used to characterize the style information of the style image at the lth feature layer.

[0061] The lth layer feature map of the generated image is Y l , and the corresponding Gram matrix H l The definition is as follows:

[0062]

[0063] in, and Represents the feature map Y l The eigenvalues ​​of the i-th channel and the j-th channel at position (h, w) in the image. Represents the feature map Y l The correlation between different channels in is used to characterize the style information of the generated image at the lth feature layer.

[0064] By calculating the Gram matrix of the style image and the generated image at multiple levels and using the style loss function to measure the difference between them, image style transfer and generation are achieved. The style loss function is defined as follows:

[0065]

[0066]

[0067] Among them, N l and M l Respectively represent the number of channels and side length of the feature map, ω l Represents the weight coefficient.

[0068] By comparing the discriminability of the image features learned by the contrastive learning model, the features of the generated image are compared with the features of the content image and with the features of the style image, thereby defining the contrastive learning loss function. The contrastive learning loss function is defined as follows:

[0069]

[0070] Among them, z i Represents the characteristics of the content image, z j Represents the features of the generated image, z k Represents the characteristics of the style image, sin() represents the similarity between the features, τ represents the temperature coefficient, and K represents the number of negative samples.

[0071] The adversarial loss function is obtained based on the content loss function, style loss function and contrastive learning loss function. The adversarial loss function is defined as follows:

[0072] L total =αL c +βL s +γL contrastive , (Formula 7);

[0073] Among them, α, β, and γ represent weights respectively.

[0074] The parameters of the generator network are fixed, and only the parameters of the discriminator network are adjusted. When a generated image is input, the discriminator network gives a penalty to determine whether the input image is a virtual image. Then, the parameters of the discriminator network are fixed, and only the parameters of the generator network are adjusted, and the generated image generated by the generator network is input into the discriminator network. By continuously optimizing the parameters of the generator network, the discriminator network gives high scores to the generated images, thereby generating highly realistic generated images.

[0075] In some embodiments of the present application, in the above method, step S140, box-selecting and annotating the aging image dataset and the virtual image dataset to construct a target detection dataset corresponding to the aging style, includes: constructing a target image dataset corresponding to the aging style based on the aging image dataset and the virtual image dataset; box-selecting and annotating the operating status of the indicator sign in the target image dataset to obtain the target detection dataset.

[0076] Based on the aging image dataset and the generated virtual image dataset, a target image dataset corresponding to multiple aging styles is constructed. The operating states of the indication signs in the target image dataset can be distributed as evenly as possible, that is, the number of images of the indication signs with the operating states of closed and opened positions is as even as possible.

[0077] In the target image dataset, the operating status of the indicator sign is marked as the target category. The target category includes closing and opening. The images after the marking constitute the target detection dataset. That is, the images in the target detection dataset have the corresponding features of "image-aging style-operating status category".

[0078] In some embodiments of the present application, in the above method, step S150, based on the target detection data set, trains the target detection model constructed based on the structured state space sequence model, and uses the trained target detection model to identify the operating status of the high-voltage circuit breaker, including: dividing the target detection data set into a training set, a test set and a verification set; using deformable large core attention to improve the structured state space sequence model; using the training set to train the structured state space sequence model, using the verification set to verify the trained structured state space sequence model, and using the test set to test the trained structured state space sequence model to obtain the trained target detection model; embedding the trained target detection model into the high-voltage circuit breaker monitoring system, and establishing a data transmission channel for the current sensor to identify the operating status of the high-voltage circuit breaker.

[0079] The target detection dataset is divided into training set, test set and validation set, and a target detection model based on structured state space sequence model is constructed. The structured state space sequence model is used to extract the features of the target detection dataset and construct the target detection model. The structured state space sequence model is trained using the training set; the performance indicators of the structured state space sequence model are evaluated based on the validation set to understand the performance of the model on the validation set, and the model parameters are adjusted or the model is fine-tuned according to the performance of the model; the performance and generalization ability of the structured state space sequence model are tested based on the test set. When the model performance meets the requirements, the trained target detection model is used to identify the operating status of the ultra-high voltage circuit breaker. Through the current pulse sequence data, the trained target detection model can effectively identify the corresponding pulse features under different operating conditions.

[0080] The structured state space sequence model can be improved by using deformable large kernel attention. The structured state space sequence model has stronger long-sequence long-distance dependency capabilities and faster training speed. Compared with traditional sequence models, the structured state space sequence model has linear complexity when processing long sequence data, which can significantly reduce training time and resource consumption while maintaining model performance. By constructing a structured state space, the target detection model can not only capture local features in sequence data, but also effectively track and model long-distance dependencies in time series. Traditional sequence models such as recurrent neural networks (RNN) or long short-term memory networks (LSTM) are prone to gradient vanishing or exploding problems when facing extremely long sequences.

[0081] The performance evaluation indicators of the trained structured state space sequence model mainly include accuracy, precision, recall and F1 score. Their expressions are:

[0082]

[0083] Among them, TP means that the true category of the sample is a positive example and the result of model recognition is also a positive example, TN means that the true category of the sample is a negative example and the result of model recognition is also a negative example, FP means that the true category of the sample is a negative example but the result of model recognition is a positive example, and FN means that the true category of the sample is a positive example but the result of model recognition is a negative example.

[0084] Figure 2 A schematic diagram of a structured state space sequence model proposed in one embodiment of the present application is shown, referring to Figure 2 As shown in Figure 1, in order to more effectively capture the global context features of images in the target detection dataset, a structured state space sequence model is used. The state space of the structured state space sequence model is represented as:

[0085]

[0086] y t =Ch t , (Formula 13);

[0087] Among them, h represents the hidden state, A, B, C represent parameter matrices, x t Indicates the input parameter, t t Represents an output parameter.

[0088] Among them, the parameter matrix A captures information about the previous state to construct a new state, and the HiPPO matrix helps the parameter matrix A to handle long sequence remote dependencies. The HiPPO matrix is ​​expressed as:

[0089]

[0090] Where n is the index of the state and k is the discretized time step.

[0091] A structured state-space sequence model is employed to label image sequences using position embeddings, which are subsequently fed into a Vim encoder that processes the labeled sequence in both forward and backward directions, yielding a richer understanding of the image context.

[0092] The overall loss of the structured state-space sequence model includes the classification loss function, the localization loss function, and the object boundary localization loss function.

[0093] The binary cross entropy loss function is used as the classification loss function for structured state space sequence model training. The classification loss function is defined as follows:

[0094]

[0095]

[0096] Among them, N represents the total number of categories, x i Represents the current category prediction value, y i Represents the probability of the current category after the activation function Sigmoid(), Indicates the true value of the current category.

[0097] The positioning loss function is defined as follows:

[0098] L local =(x0-x * ) 2 +(y0-y * ) 2 +(w0-w * ) 2 +(h0-h * ) 2 , (Formula 17);

[0099] Among them, x0 and y0 represent the coordinates of the upper left corner of the prediction box, and w0 and h0 represent the width and height of the prediction box respectively. * ,y * Respectively represent the coordinates of the upper left corner of the real box, w * 、h * Represent the width and height of the real box respectively.

[0100] The target boundary localization loss functions include IoU loss, GIoU loss, DIoU loss and CIoU loss.

[0101] The IoU loss function is defined as follows:

[0102] L IoU =1-IoU(B1,B2), (Formula 18);

[0103] in, B1 and B2 represent the predicted box area and the real box area respectively.

[0104] The GIoU loss function is defined as follows:

[0105] L GIoU =1-GIoU(B1,B2), (Formula 19);

[0106] in, C represents the smallest circumscribed rectangular area that can contain both B1 and B2.

[0107] The DIoU loss function is defined as follows:

[0108] L DIoU =1-DIoU(B1,B2), (Formula 20);

[0109] in, b and b gt They represent the center points of the predicted box and the real box respectively, ρ represents the Euclidean distance between the two center points, and c represents the diagonal length of the smallest circumscribed rectangular area that can contain both B1 and B2.

[0110] The CIoU loss function is defined as follows:

[0111] L CIoU =1-CIoU(B1,B2), (Formula 21);

[0112] in, α represents the weight function, and v is used to measure the consistency of aspect ratio.

[0113] Export the trained target detection model in ONNX format and embed it into the high-voltage circuit breaker monitoring system. In the high-voltage circuit breaker monitoring system, integrate appropriate edge computing devices or microprocessors, load the trained target detection model into the monitoring system, and establish a logical process for data acquisition, model calling, and result output. Call the model inference interface to analyze the newly acquired images and output the corresponding ultra-high voltage circuit breaker operating status information. Perform regular maintenance and updates on the monitoring system to ensure the stability and reliability of the system.

[0114] Figure 3 A schematic diagram showing the structure of a device for identifying the operating status of a high-voltage circuit breaker according to an embodiment of the present application is shown. Figure 3 As shown, the device 300 includes:

[0115] An image data acquisition unit 310 is used to acquire image data of the high-voltage circuit breaker including the indicator mark, and pre-process the image data;

[0116] A real image data set establishing unit 320 is used to divide the pre-processed image data into a clear image data set and an aged image data set according to whether the indicator mark is aged, and to classify the aged image data set according to the aged style;

[0117] A virtual image data set establishing unit 330 is used to train an image generation model based on a style transfer generative adversarial network model according to a clear image data set and an aged image data set, and to input the clear image data set into the trained image generation model for style transfer to generate a virtual image data set;

[0118] The target detection data set establishment unit 340 is used to select and annotate the aging image data set and the virtual image data set to establish a target detection data set corresponding to the aging style;

[0119] The operating state recognition unit 350 is used to train a target detection model constructed based on a structured state space sequence model based on a target detection data set, and use the trained target detection model to recognize the operating state of the high-voltage circuit breaker.

[0120] In some embodiments of the present application, in the above-mentioned device 300, the image data acquisition unit 310 is specifically used to: use a substation monitoring device or an image acquisition device to regularly acquire image data of a high-voltage circuit breaker including an indicator mark; wherein the indicator mark includes an opening and closing indicator mark of the high-voltage circuit breaker; perform data verification and quality inspection on the image data, and remove abnormal image data that does not meet quality rules; extract an area of ​​interest in the retained image data; wherein the area of ​​interest includes the indicator mark.

[0121] In some embodiments of the present application, in the above-mentioned device 300, the real image data set establishing unit 320 is specifically used to: divide the image data into a clear image data set and an aged image data set according to whether the indicator signs in the pre-processed image data are aged; classify the aged image data set into aging styles according to the aging type; wherein the aging type includes cracks, stains, discoloration, and spots.

[0122] In some embodiments of the present application, in the above-mentioned device 300, the virtual image dataset establishment unit 330 is specifically used to: form a content picture based on extracting content features of each image in the clear image dataset, and form a style picture based on extracting style features of each image in the aged image dataset; input the content picture into the generation network to obtain a generated picture; input the content picture, the style picture and the generated picture into the discriminant network, and use an adversarial loss function to measure the difference in content and style; alternately update the generation network and the discriminant network according to the adversarial loss function until the difference in content and style is minimized, and a trained image generation model is obtained; after inputting the clear image dataset into the trained image generation model for style transfer, a virtual image dataset with an aged style but the original content remains unchanged is generated.

[0123] In some embodiments of the present application, in the above-mentioned device 300, the virtual image dataset establishing unit 330 is also specifically used to: measure the content difference between the content picture and the generated picture through the content loss function; measure the style difference between the style picture and the generated picture through the style loss function; measure the discriminative difference between the generated picture and the content picture and the style picture through the contrastive learning loss function; determine the adversarial loss function based on the content loss function, the style loss function and the contrastive learning loss function.

[0124] In some embodiments of the present application, in the above-mentioned device 300, the target detection dataset establishment unit 340 is specifically used to: construct a target image dataset corresponding to the aging style based on the aging image dataset and the virtual image dataset; and select and mark the operating status of the indicator sign in the target image dataset to obtain the target detection dataset.

[0125] In some embodiments of the present application, in the above-mentioned device 300, the operating status identification unit 350 is specifically used to: divide the target detection data set into a training set, a test set and a verification set; use deformable large core attention to improve the structured state space sequence model; use the training set to train the structured state space sequence model, use the verification set to verify the trained structured state space sequence model, and use the test set to test the trained structured state space sequence model to obtain a trained target detection model; embed the trained target detection model into the high-voltage circuit breaker monitoring system, and establish a data transmission channel for the current sensor to identify the operating status of the high-voltage circuit breaker.

[0126] It should be noted that the above-mentioned device 300 for identifying the operating status of a high-voltage circuit breaker can implement the above-mentioned method for identifying the operating status of a high-voltage circuit breaker one by one, which will not be described in detail.

[0127] Figure 4 A schematic diagram showing the structure of a computer device according to an embodiment of the present application is shown. Figure 4 As shown, the internal structure of the computer device may include a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client through a network connection. When the computer program is executed by the processor, the functions or steps of the method for identifying the operating state of a high-voltage circuit breaker are realized.

[0128] In one embodiment, the computer device provided by the present application includes a memory and a processor, the memory stores a database and a computer program that can be run on the processor, and the processor implements the following steps when executing the computer program:

[0129] Acquire image data of the high-voltage circuit breaker including the indicator mark, and pre-process the image data;

[0130] The preprocessed image data are divided into a clear image data set and an aged image data set according to whether the indicator mark is aged, and the aged image data set is classified according to the aging style;

[0131] According to the clear image dataset and the aged image dataset, an image generation model based on the style transfer generative adversarial network model is trained, and the clear image dataset is input into the trained image generation model for style transfer to generate a virtual image dataset;

[0132] Perform box selection and annotation on the aging image dataset and the virtual image dataset to construct an object detection dataset corresponding to the aging style;

[0133] Based on the target detection dataset, a target detection model built based on the structured state space sequence model is trained, and the trained target detection model is used to identify the operating status of high-voltage circuit breakers.

[0134] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0135] Acquire image data of the high-voltage circuit breaker including the indicator mark, and pre-process the image data;

[0136] The preprocessed image data are divided into a clear image data set and an aged image data set according to whether the indicator mark is aged, and the aged image data set is classified according to the aging style;

[0137] According to the clear image dataset and the aged image dataset, an image generation model based on the style transfer generative adversarial network model is trained, and the clear image dataset is input into the trained image generation model for style transfer to generate a virtual image dataset;

[0138] Perform box selection and annotation on the aging image dataset and the virtual image dataset to construct an object detection dataset corresponding to the aging style;

[0139] Based on the target detection dataset, a target detection model built based on the structured state space sequence model is trained, and the trained target detection model is used to identify the operating status of high-voltage circuit breakers.

[0140] It should be noted that the above functions or steps that can be implemented by the computer device or computer-readable storage medium can refer to the relevant description in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0141] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0142] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0143] The embodiments described above 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 may be replaced by equivalents. Such 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, and should all be included in the protection scope of the present invention.

Claims

1. A method for identifying the operating status of a high-voltage circuit breaker, characterized in that: The method comprises: Acquire image data of a high-voltage circuit breaker including an indicator mark, and pre-process the image data; dividing the preprocessed image data into a clear image data set and an aged image data set according to whether the indicator mark is aged, and classifying the aged image data set according to the aged style; According to the clear image dataset and the aged image dataset, an image generation model constructed based on a style transfer generative adversarial network model is trained, and the clear image dataset is input into the trained image generation model for style transfer to generate a virtual image dataset; Performing box selection and annotation on the aging image dataset and the virtual image dataset to construct an object detection dataset corresponding to the aging style; Based on the target detection data set, a target detection model constructed based on a structured state space sequence model is trained, and the trained target detection model is used to identify the operating status of a high-voltage circuit breaker.

2. The method for identifying the operating status of a high-voltage circuit breaker according to claim 1, characterized in that: The step of acquiring image data of the high-voltage circuit breaker including the indicator mark and preprocessing the image data comprises: The image data of the high-voltage circuit breaker including the indicator mark is periodically acquired by using a substation monitoring device or an image acquisition device; wherein the indicator mark includes an opening and closing indicator mark of the high-voltage circuit breaker; Performing data verification and quality inspection on the image data to remove abnormal image data that does not meet quality rules; Extracting a region of interest from the retained image data; wherein the region of interest includes the indicator mark.

3. The method for identifying the operating status of a high-voltage circuit breaker according to claim 1, characterized in that: The pre-processed image data is divided into a clear image data set and an aged image data set according to whether the indicator mark is aged, and the aged image data set is classified according to the aged style, including: According to whether the indication mark in the preprocessed image data is aged, dividing the image data into the clear image data set and the aged image data set; The aging image dataset is classified into aging styles according to aging types; wherein the aging types include cracks, stains, discoloration, and spots.

4. The method for identifying the operating status of a high-voltage circuit breaker according to claim 1, characterized in that: The step of training an image generation model constructed based on a style transfer generative adversarial network model according to the clear image dataset and the aged image dataset, and inputting the clear image dataset into the trained image generation model for style transfer to generate a virtual image dataset, comprises: Forming a content picture based on extracting content features of each image in the clear image dataset, and forming a style picture based on extracting style features of each image in the aged image dataset; Inputting the content image into a generation network to obtain a generated image; Inputting the content image, the style image, and the generated image into a discriminant network, and using an adversarial loss function to measure the differences in content and style; Alternately updating the generation network and the discrimination network according to the adversarial loss function until the difference between the content and the style is minimized, thereby obtaining the trained image generation model; After the clear image data set is input into the trained image generation model for style transfer, a virtual image data set with the aging style but the original content unchanged is generated.

5. The method for identifying the operating status of a high-voltage circuit breaker according to claim 4, characterized in that: The step of inputting the content image, the style image, and the generated image into a discriminant network and using an adversarial loss function to measure the differences in content and style includes: Measuring the content difference between the content image and the generated image by a content loss function; Measuring the style difference between the style image and the generated image by a style loss function; Measuring the discriminative difference between the generated image and the content image and the style image by comparing the learning loss function; The adversarial loss function is determined according to the content loss function, the style loss function and the contrastive learning loss function.

6. The method for identifying the operating status of a high-voltage circuit breaker according to claim 1, characterized in that: The step of performing box selection and labeling on the aging image dataset and the virtual image dataset to construct an object detection dataset corresponding to the aging style includes: Based on the aging image dataset and the virtual image dataset, constructing a target image dataset corresponding to the aging style; The running status of the indicator sign in the target image data set is marked by box selection to obtain the target detection data set.

7. The method for identifying the operating status of a high-voltage circuit breaker according to claim 1, characterized in that: The method of training a target detection model constructed based on a structured state space sequence model based on the target detection data set and identifying the operating state of a high-voltage circuit breaker using the trained target detection model comprises: Dividing the target detection data set into a training set, a test set and a validation set; Adopting deformable large kernel attention to improve the structured state-space sequence model; Using the training set to train the structured state space sequence model, using the verification set to verify the trained structured state space sequence model, and using the test set to test the trained structured state space sequence model to obtain the trained target detection model; The trained target detection model is embedded into the high-voltage circuit breaker monitoring system, and a data transmission channel is established for the current sensor to identify the operating status of the high-voltage circuit breaker.

8. A device for identifying the operating status of a high-voltage circuit breaker, characterized in that: The device comprises: An image data acquisition unit, used to acquire image data of the high-voltage circuit breaker including the indicator mark, and pre-process the image data; a real image data set establishing unit, configured to divide the preprocessed image data into a clear image data set and an aged image data set according to whether the indicator mark is aged, and classify the aged image data set according to the aged style; A virtual image data set establishing unit, configured to train an image generation model constructed based on a style transfer generative adversarial network model according to the aged image data set, and to input the clear image data set into the trained image generation model for style transfer to generate a virtual image data set; An object detection data set establishment unit, configured to perform box selection and annotation on the aging image data set and the virtual image data set to establish an object detection data set corresponding to the aging style; An operating status recognition unit is used to train a target detection model constructed based on a structured state space sequence model based on the target detection data set, and use the trained target detection model to identify the operating status of the high-voltage circuit breaker.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method for identifying the operating status of a high-voltage circuit breaker according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: The computer program, when instructed by a processor, implements the steps of the method for identifying the operating status of a high-voltage circuit breaker as claimed in any one of claims 1 to 7.

Images