GIS switchgear operation state evaluation method and system based on image analysis

Abstract

The application relates to the technical field of power equipment monitoring, and discloses a GIS switch equipment operation state evaluation method and system based on image analysis, which adopts a high-definition camera, an infrared thermal imager and an ultraviolet imager to collect initial images of GIS switch equipment; an adaptive median filtering algorithm is used to carry out denoising processing on the collected initial images, a histogram equalization and CLAHE combined mode is used to carry out enhancement processing on the denoised images, and a pretreated image is obtained; the pretreated image is input into a feature extraction model, and the extracted feature vectors are spliced into a multi-dimensional feature matrix, and the weight of the features is dynamically adjusted through a particle swarm optimization algorithm; the multi-dimensional feature matrix is input into an LSTM neural network, the equipment state grade is divided according to the equipment health index, and the operation state evaluation result of the GIS switch equipment is output to carry out graded early warning; and the application effectively improves the recognition accuracy of the operation state of the GIS switch equipment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power equipment monitoring, in particular to a GIS switch equipment operation state evaluation method and system based on image analysis. BACKGROUND

[0002] GIS switch equipment is widely used in modern power systems due to its small footprint, high reliability, and convenient maintenance; however, GIS switch equipment may be affected by various factors during long-term operation, such as internal component aging, poor contact, and insulation performance degradation, which can cause abnormal operation of the equipment and even cause faults, affecting the safe and stable operation of the power system; therefore, timely and accurate evaluation of the operation state of GIS switch equipment is of great significance.

[0003] Currently, the evaluation methods for the operation state of GIS switch equipment mainly include electrical detection, acoustic detection, and chemical detection; electrical detection such as partial discharge detection can effectively detect internal insulation defects of the equipment, but requires high detection equipment and complex detection process; acoustic detection analyzes sound signals during equipment operation to determine the equipment state, but is easily disturbed by environmental noise; chemical detection requires collecting internal gases for analysis, which is time-consuming; in addition, existing evaluation methods mostly rely on a single detection method, which is difficult to comprehensively and accurately reflect the actual operation state of GIS switch equipment. SUMMARY

[0004] The present application is designed to solve the above problems, and provides a GIS switch equipment operation state evaluation method and system based on image analysis.

[0005] The present application provides a GIS switch equipment operation state evaluation method based on image analysis, which comprises the following steps:

[0006] High-definition cameras, infrared thermographs, and ultraviolet imagers are used to collect initial images of GIS switch equipment;

[0007] An adaptive median filter algorithm is used to denoise the collected initial images, and a histogram equalization combined with CLAHE is used to enhance the denoised images to obtain preprocessed images;

[0008] The preprocessed images are input into a feature extraction model, and the extracted feature vectors are spliced into a multi-dimensional feature matrix, and the particle swarm optimization algorithm is used to dynamically adjust the weights of the features, wherein the feature extraction model includes YOLOv8 network, U-Net network, and FasterR-CNN network;

[0009] The multi-dimensional feature matrix is input into an LSTM neural network to obtain a device health index, and a device state level is divided according to the device health index, and an operation state evaluation result of the GIS switch device is output.

[0010] A graded early warning is performed according to the operation state evaluation result of the GIS switch device.

[0011] Optionally, in the first implementation manner of the first aspect of the present application, the adaptive median filtering algorithm is used to perform denoising processing on the collected initial image, and a histogram equalization combined with CLAHE is used to perform enhancement processing on the denoised image to obtain a preprocessed image, including:

[0012] An initial size of a filter window is set for each pixel point in the initial image, a median value of pixels in the window is calculated, and the gray value of the center pixel of the window is compared with the median value and the maximum and minimum values of the gray values in the window;

[0013] If the center pixel is noise, the median value is replaced, and if it is not noise, it is judged whether the window reaches the maximum size, and if not, the window is enlarged and repeated calculation is performed, and if yes, the original pixel value is retained;

[0014] The gray histogram of the denoised image is expanded to the entire gray range, and then the CLAHE algorithm is used to divide the image into multiple sub-blocks, and the histogram equalization is performed on each sub-block to obtain the preprocessed image.

[0015] Optionally, in the second implementation manner of the first aspect of the present application, the preprocessed image is input into a feature extraction model, and the extracted feature vectors are spliced into a multi-dimensional feature matrix, and the weights of the features are dynamically adjusted through a particle swarm optimization algorithm, including:

[0016] The preprocessed image is input into a feature extraction model, the YOLOv8 network extracts an apparent defect feature vector, the U-Net network extracts a temperature feature vector, and the FasterR-CNN network extracts a discharge feature vector;

[0017] The feature vectors output by the three networks are spliced into a multi-dimensional feature matrix, the positions and speeds of the particle swarms are initialized, and each particle represents a group of feature weights;

[0018] The fitness values of each particle are calculated, and the global optimal position and the individual optimal position of the particle are updated according to the contribution of the features to the GIS switch device state evaluation;

[0019] The position and speed of the particle are adjusted according to the particle swarm iteration until a preset iteration number is reached, the optimized feature weights are obtained, and the dynamic weighting of each feature in the multi-dimensional feature matrix is completed.

[0020] Optionally, in a third implementation form of the first aspect of the present application, the YOLOv8 network extracts an apparent defect feature vector, and the extraction comprises:

[0021] The preprocessed image is input into the backbone network of the YOLOv8 architecture, and an initial convolution operation is performed on the image to adjust the channel number of the image and compress the spatial size of the image;

[0022] The backbone network gradually extracts image features through a plurality of CSP modules, each CSP module divides the input features into two branches, one branch undergoes a plurality of convolution operations, and the other branch is directly connected, the features of the two branches are fused, and image features of different scales are extracted;

[0023] The neck network fuses the image features of different scales output by the backbone network through upsampling and downsampling operations to obtain a fused multi-scale feature map;

[0024] The fused feature map is input into the head network, the features are processed through convolution operations, the targets in the image are predicted based on an anchor box mechanism, the position and category information of the apparent defect are determined, and an apparent defect feature vector is obtained.

[0025] Optionally, in a fourth implementation form of the first aspect of the present application, the U-Net network extracts a temperature feature vector, and the extraction comprises:

[0026] The preprocessed image is input into the encoder part of the U-Net network, the encoder is composed of a plurality of convolution blocks and pooling layers, each convolution block extracts image features through convolution operations, and the pooling layer down-samples the feature map to gradually extract deep features of the image;

[0027] The deep features enter the decoder part, the decoder enlarges the spatial size of the deep features through upsampling operations to match the size of the feature map of the corresponding layer of the encoder, and the upsampling features are spliced and fused with the corresponding layer feature map saved by the encoder;

[0028] After a plurality of upsampling and feature fusion operations, the decoder outputs a feature map with the same size as the input image, extracts features including local hot spot temperature and temperature difference gradient, and converts the features into a temperature feature vector.

[0029] Optionally, in a fifth implementation form of the first aspect of the present application, the FasterR-CNN network extracts a discharge feature vector, and the extraction comprises:

[0030] The preprocessed image is input into the feature extraction network of the FasterR-CNN network, and the image is processed through convolution operations and pooling operations to extract a global feature map of the image;

[0031] The global feature map is input into a region proposal network, the region proposal network generates a plurality of candidate anchor boxes on the feature map through sliding windows, and then classifies and performs boundary box regression on the anchor boxes to screen out candidate regions that may contain targets;

[0032] A region of interest pooling operation is performed on the screened candidate regions to convert candidate region feature maps of different sizes into fixed-size feature vectors;

[0033] The fixed-size feature vectors are input into a classifier and a regressor to obtain accurate target regions, count the number of discharge pulses and the energy distribution, and convert them into a discharge feature vector.

[0034] Optionally, in a sixth implementation manner of the first aspect of the present application, the multi-dimensional feature matrix is input into the LSTM neural network to obtain a device health index, and a device state level is divided according to the device health index, and an operation state evaluation result of the GIS switch device is output, including:

[0035] The multi-dimensional feature matrix optimized by the particle swarm is adjusted in format, and the adjusted multi-dimensional feature matrix is input into an input layer of the LSTM neural network in time sequence order, and enters a memory unit of a hidden layer through an input gate, a forgetting gate and an output gate;

[0036] The hidden state is converted into a device health index through a full connection operation, a plurality of health index thresholds are preset, each health index threshold corresponds to a device state level, the calculated device health index is compared with the health index threshold, and the corresponding device state level is divided;

[0037] According to the divided device state level, a GIS switch device operation state evaluation result containing the device state level, the device health index and the feature information is generated.

[0038] The second aspect of the present application provides a GIS switch device operation state evaluation system based on image analysis, which comprises:

[0039] The acquisition module is configured to acquire initial images of the GIS switch device by using a high-definition camera, an infrared thermal imager and an ultraviolet imager respectively.

[0040] The preprocessing module is configured to perform denoising processing on the acquired initial images by using an adaptive median filtering algorithm, and perform enhancement processing on the denoised images by using a combination of histogram equalization and CLAHE to obtain preprocessed images.

[0041] The feature extraction module is configured to input the preprocessed image into a feature extraction model, splice the extracted feature vectors into a multi-dimensional feature matrix, and dynamically adjust the weight of the features by using a particle swarm optimization algorithm, wherein the feature extraction model comprises a YOLOv8 architecture, a U-Net network and a FasterR-CNN network;

[0042] The output module is configured to input the multi-dimensional feature matrix into an LSTM neural network, obtain a device health index, divide a device state level according to the device health index, and output an operation state evaluation result of the GIS switch device.

[0043] The early warning module is configured to perform hierarchical early warning according to the operation state evaluation result of the GIS switch device.

[0044] The third aspect of the present application provides a GIS switch device operation state evaluation device based on image analysis, which comprises a memory and at least one processor, and the memory stores instructions; the at least one processor invokes the instructions in the memory to enable the GIS switch device operation state evaluation device based on image analysis to perform each step of the GIS switch device operation state evaluation method based on image analysis as described above.

[0045] The fourth aspect of the present application provides a computer readable storage medium, which stores instructions, and the instructions are executed by a processor to realize each step of the GIS switch device operation state evaluation method based on image analysis as described above.

[0046] In the technical solution provided by the present application, a high-definition camera, an infrared thermal imager and an ultraviolet imager are respectively used to collect initial images of the GIS switch device; an adaptive median filtering algorithm is used to perform denoising processing on the collected initial images, a histogram equalization combined with CLAHE is used to perform enhancement processing on the denoised images, and preprocessed images are obtained; the preprocessed images are input into a feature extraction model, and the extracted feature vectors are spliced into a multi-dimensional feature matrix, and the weight of the features is dynamically adjusted by using a particle swarm optimization algorithm; the multi-dimensional feature matrix is input into an LSTM neural network to obtain a device health index, and a device state level is divided according to the device health index, and an operation state evaluation result of the GIS switch device is output; hierarchical early warning is performed according to the operation state evaluation result of the GIS switch device; the present application can obtain operation state information of the GIS switch device from multiple angles, and compared with a single detection means, the evaluation result is more comprehensive and accurate, which can effectively improve the recognition accuracy of the operation state of the GIS switch device, reduce the probability of misjudgment and omission, timely discover the change of the operation state of the device, and quickly provide evaluation results and early warning information, which is helpful for power operation and maintenance personnel to take measures in time, and ensures the safe and stable operation of the power system. Attached Figure Description

[0047] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0048] Figure 1 A flowchart of the image analysis-based GIS switchgear operation status assessment method provided in this embodiment of the invention;

[0049] Figure 2 A schematic diagram of the structure of the GIS switchgear operation status assessment system based on image analysis provided in an embodiment of the present invention;

[0050] Figure 3 A schematic diagram of the structure of the GIS switchgear operation status assessment device based on image analysis provided in an embodiment of the present invention. Detailed Implementation

[0051] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0052] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 A flowchart of the image analysis-based GIS switchgear operation status assessment method provided in this embodiment of the invention is shown. The method specifically includes the following steps:

[0053] Step 101: Acquire initial images of the GIS switchgear using a high-definition camera, an infrared thermal imager, and an ultraviolet imager, respectively;

[0054] In this embodiment, in the monitoring field of GIS switch equipment, three kinds of acquisition devices, high-definition camera, infrared thermal imager and ultraviolet imager, are reasonably arranged. The high-definition camera is installed at a position capable of clearly capturing the overall appearance and key component details of the equipment, and visible light initial images containing surface cracks, deformation, rust and other apparent features of the equipment are obtained by continuous shooting. The infrared thermal imager is aimed at the hot areas of the equipment, such as joints and insulating parts, and real-time acquisition of infrared thermal initial images reflecting the temperature distribution of the surface of the equipment can capture local hot spots caused by poor contact and insulation aging. The ultraviolet imager focuses on the parts of the equipment where discharge may occur, and uses its high sensitivity to ultraviolet light to acquire ultraviolet initial images containing phenomena such as corona discharge and arc discharge. The three devices work together to obtain initial image data of GIS switch equipment from different dimensions.

[0055] In step 102, the adaptive median filtering algorithm is used to denoise the collected initial images, and the histogram equalization combined with CLAHE is used to enhance the denoised images, to obtain the preprocessed images.

[0056] In this embodiment, for each pixel point in the initial image, a fixed initial size filter window is first set, which is centered on the current pixel point and covers a certain number of adjacent pixels around it. Then all the pixels in the window are traversed, their gray values are extracted and sorted, and the gray value at the middle position is found as the median value of the pixels in the window. At the same time, the maximum and minimum values of the gray values in the window are recorded, and the gray value of the center pixel is compared with the median value, the maximum value and the minimum value one by one.

[0057] According to the comparison result, it is judged whether the center pixel is noise: if the gray value of the center pixel is less than the minimum value of the gray values in the window or greater than the maximum value, it is determined that the pixel is noise, and the gray value of the center pixel is directly replaced by the median value of the window calculated before. If the gray value of the center pixel is between the minimum value and the maximum value, it is further checked whether the size of the current filter window reaches the preset maximum size. If the maximum size has not been reached, the window size is enlarged according to the set increment, the median value, the maximum value and the minimum value in the new window are recalculated, and the comparison process is repeated. If the maximum size has been reached, it is considered that the pixel is not noise, and the original gray value is kept unchanged.

[0058] The gray scale histogram of the denoised image is expanded, the pixel values originally concentrated in a certain gray scale interval are uniformly distributed in the entire gray scale range of 0-255 by adjusting the gray scale distribution of the image, so as to improve the overall contrast of the image; then the CLAHE algorithm is adopted, the image is divided into a plurality of non-overlapping sub-blocks according to a preset size, and then the histogram equalization processing is performed on each sub-block individually, so that the local contrast of the sub-block is enhanced, and the over-enhancement in the sub-block is avoided by limiting the contrast gain, and finally the preprocessed image with clear details and appropriate contrast is obtained.

[0059] In step 103, the preprocessed image is input into a feature extraction model, and the extracted feature vectors are spliced into a multi-dimensional feature matrix, and the weights of the features are dynamically adjusted by a particle swarm optimization algorithm, wherein the feature extraction model includes a YOLOv8 network, a U-Net network and a FasterR-CNN network.

[0060] In this embodiment, the preprocessed image is input into the feature extraction model, the YOLOv8 network extracts the apparent defect feature vector, the U-Net network extracts the temperature feature vector, and the FasterR-CNN network extracts the discharge feature vector; the apparent defect feature vector output by the YOLOv8 architecture, the temperature-related feature vector output by the U-Net network and the discharge-related feature vector output by the FasterR-CNN network are spliced in order of feature dimension to form a multi-dimensional feature matrix containing multiple types of feature information; at the same time, the particle swarm in the particle swarm optimization algorithm is initialized, and each particle is randomly assigned an initial position and an initial speed, wherein the position of each particle corresponds to a set of feature weight parameters, the number of weight parameters is consistent with the feature dimension of the multi-dimensional feature matrix, so as to establish the correspondence between the feature weight and the particle position.

[0061] The fitness value of each particle is calculated according to a preset fitness function, the function is based on the influence degree of the feature on the evaluation result of the GIS switch device state, the higher the contribution of the feature to the evaluation result, the greater the fitness value of the corresponding particle; after the calculation is completed, the current fitness value of each particle is compared with its historical optimal fitness value, if the current value is better, the individual optimal position of the particle is updated; at the same time, the current fitness value of all particles is compared with the global optimal fitness value, if there is a better value, the global optimal position of the entire particle swarm is updated, so as to realize the dynamic adjustment of the optimization direction of the particle.

[0062] According to the iteration rule of the particle swarm algorithm, the new speed and the new position of each particle in the next iteration are calculated by combining the current position, speed, individual optimal position and global optimal position of the particle, so that the particle gradually approaches the global optimal position; after each iteration, the fitness value of the particle is recalculated and the optimal position is updated, and the process is repeated until the preset number of iterations is reached; at this time, the feature weight corresponding to the global optimal position is the optimized result, and the weights are respectively weighted with the corresponding features in the multi-dimensional feature matrix to complete the dynamic weighting processing of the feature matrix.

[0063] In this embodiment, the preprocessed image is input into the backbone network of the YOLOv8 architecture, and initial convolution operation is performed on the image to adjust the channel number of the image and compress the spatial size of the image; the backbone network gradually extracts image features through multiple CSP modules, each CSP module divides the input features into two branches, one branch undergoes multiple convolution operations, and the other branch is directly connected, the features of the two branches are fused, and image features of different scales are extracted; the neck network fuses the image features of different scales output by the backbone network through upsampling and downsampling operations to obtain a fused multi-scale feature map; the fused feature map is input into the head network, the features are processed through convolution operation, the positions and category information of the apparent defects are determined based on the anchor box mechanism, and the apparent defect feature vector is obtained.

[0064] In this embodiment, the preprocessed image is input into the encoder part of the U-Net network, and the encoder is composed of multiple convolution blocks and pooling layers. Each convolution block extracts image features through convolution operation, and the pooling layer down-samples the feature map to gradually extract deep features of the image; the deep features enter the decoder part, the decoder enlarges the spatial size of the deep features through upsampling operation to match the size of the feature map of the corresponding layer of the encoder, and the upsampling features are spliced and fused with the corresponding layer feature map saved by the encoder; after multiple upsampling and feature fusion operations, the decoder outputs a feature map with the same size as the input image, extracts features including local hotspot temperature and temperature difference gradient, and converts them into a temperature feature vector.

[0065] In this embodiment, the preprocessed image is input into the feature extraction network of the FasterR-CNN network, and the image is processed through convolution operation and pooling operation to extract the global feature map of the image; the global feature map is input into the region proposal network, the region proposal network generates multiple candidate anchor boxes on the feature map through sliding window, and then classifies and regresses the anchor boxes to screen out candidate regions that may contain targets; the region of interest pooling operation is used on the screened candidate regions to convert candidate region feature maps of different sizes into fixed-size feature vectors; the fixed-size feature vectors are input into the classifier and regressor to obtain accurate target regions, count the number of discharge pulses and energy distribution, and convert them into discharge feature vectors.

[0066] Step 104, input the multi-dimensional feature matrix into the LSTM neural network to obtain the device health index, divide the device state level according to the device health index, and output the operation state evaluation result of the GIS switch device;

[0067] In this embodiment, the multi-dimensional feature matrix optimized by the particle swarm is adjusted in format, and the adjusted multi-dimensional feature matrix is input into the input layer of the LSTM neural network in time sequence order, and enters the memory unit of the hidden layer through the input gate, the forgetting gate and the output gate; the hidden state is converted into the device health index through full connection operation, a plurality of health index threshold values are preset, each health index threshold value corresponds to a device state level, the calculated device health index is compared with the health index threshold value to divide the corresponding device state level; according to the divided device state level, the GIS switch device operation state evaluation result containing the device state level, the device health index and the feature information is generated.

[0068] Step 105, hierarchical early warning is performed according to the operation state evaluation result of the GIS switch device.

[0069] In this embodiment, after receiving the operation state evaluation result of the GIS switch device, the early warning module will start the corresponding early warning mechanism according to the preset grading standard: when the evaluation result is that the device is in a healthy state, the system does not issue any early warning signal, and only records the current state information; if the evaluation result shows that the device has slight abnormalities and may have potential failure risks, the system immediately triggers a blue early warning, pops up a prompt window through the monitoring platform, and sends a reminder message to the mobile terminal of the operation and maintenance personnel, prompting regular inspection; when the evaluation result shows that the device is in a moderate abnormal state and the failure risk is higher, the system automatically switches to a yellow early warning, in addition to the above prompts, it will also start the real-time monitoring function of the key parameters of the device, and update the data every hour; if the evaluation result shows that the device has serious abnormalities and may fail at any time, the system will trigger the highest level of red early warning, at this time the monitoring platform will issue an audible and visual alarm, and send an emergency notification to the operation and maintenance personnel and the dispatch center, clearly marking the abnormal position and possible failure type, so as to quickly organize repair and minimize the impact of failure.

[0070] Referring to Figure 2 The GIS switch device operation state evaluation system based on image analysis provided by the embodiment of the application has the advantages that the system can realize real-time monitoring of the operation state of the GIS switch device, and can provide early warning and timely maintenance of the device, thereby improving the safety and reliability of the device.

[0071] The acquisition module is configured to acquire initial images of the GIS switch device by using a high-definition camera, an infrared thermal imager and an ultraviolet imager respectively.

[0072] The preprocessing module is configured to perform denoising processing on the acquired initial images by using an adaptive median filtering algorithm, and perform enhancement processing on the denoised images by using a combination of histogram equalization and CLAHE, to obtain preprocessed images.

[0073] The feature extraction module is configured to input the preprocessed images into a feature extraction model, splice the extracted feature vectors into a multi-dimensional feature matrix, and dynamically adjust the weights of the features by using a particle swarm optimization algorithm, wherein the feature extraction model includes a YOLOv8 architecture, a U-Net network and a FasterR-CNN network.

[0074] The output module is configured to input the multi-dimensional feature matrix into an LSTM neural network to obtain a device health index, divide a device state level according to the device health index, and output an operation state evaluation result of the GIS switch device.

[0075] The early warning module is configured to perform graded early warning according to the operation state evaluation result of the GIS switch device.

[0076] Figure 3is a structural schematic view of a GIS switch device operation state evaluation device based on image analysis provided by an embodiment of the present application. The GIS switch device operation state evaluation device based on image analysis 300 can have great differences due to different configurations or performances, and can include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, one or more storage media 330 (for example, one or more mass storage devices) storing application programs 333 or data 332. The memory 320 and the storage media 330 can be temporary storage or persistent storage. The programs stored in the storage media 330 can include one or more modules (not shown in the figure), and each module can include a series of instruction operations in the GIS switch device operation state evaluation device based on image analysis 300. Further, the processor 310 can be configured to communicate with the storage media 330 and execute a series of instruction operations in the storage media 330 on the GIS switch device operation state evaluation device based on image analysis 300 to implement the method provided by the above embodiment.

[0077] The GIS switch device operation state evaluation device based on image analysis 300 can further include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, and the like. Those skilled in the art can understand that the GIS switch device operation state evaluation device based on image analysis 300 can include more or fewer components than those shown, or some components can be combined, or different components can be arranged. Figure 3 The GIS switch device operation state evaluation device based on image analysis shown in the structure does not constitute a limitation on the computer device provided by the present application, and can include more or fewer components than those shown, or some components can be combined, or different components can be arranged.

[0078] The present application also provides a computer readable storage medium, which can be a non-volatile computer readable storage medium or a volatile computer readable storage medium. The computer readable storage medium stores instructions, and when the instructions are run on a computer, the computer executes the steps of the GIS switch device operation state evaluation method based on image analysis provided by the above embodiments.

[0079] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described device or apparatus, unit can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0080] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or say the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0081] The basic principles, main features and advantages of the present application are shown and described above. Those skilled in the art should understand that the present application is not limited by the above embodiments, and the above embodiments and descriptions in the specification are only preferred examples of the present application and are not intended to limit the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. GIS switchgear operating condition assessment method based on image analysis, characterized in that, The method comprises the following steps: respectively using a high-definition camera, an infrared thermal imager and an ultraviolet imager to collect initial images of GIS switch equipment; using an adaptive median filtering algorithm to perform denoising processing on the collected initial images, using a combination of histogram equalization and CLAHE to perform enhancement processing on the denoised images, and obtaining preprocessed images; inputting the preprocessed images into a feature extraction model, splicing the extracted feature vectors into a multi-dimensional feature matrix, and dynamically adjusting the weights of the features through a particle swarm optimization algorithm, wherein the feature extraction model comprises a YOLOv8 network, a U-Net network and a FasterR-CNN network; inputting the multi-dimensional feature matrix into an LSTM neural network to obtain a device health index, dividing the device state level according to the device health index, and outputting the running state evaluation result of the GIS switch equipment; carrying out hierarchical early warning according to the running state evaluation result of the GIS switch equipment; the step of inputting the preprocessed images into the feature extraction model, splicing the extracted feature vectors into the multi-dimensional feature matrix, and dynamically adjusting the weights of the features through the particle swarm optimization algorithm comprises: inputting the preprocessed images into the feature extraction model, extracting an apparent defect feature vector by the YOLOv8 network, extracting a temperature feature vector by the U-Net network, and extracting a discharge feature vector by the FasterR-CNN network; splicing the feature vectors output by the three networks into the multi-dimensional feature matrix, initializing the position and speed of the particle swarm, and each particle representing a group of feature weights; calculating the fitness value of each particle, updating the global optimal position and individual optimal position of the particle according to the contribution of the features to the GIS switch equipment state evaluation; adjusting the position and speed of the particle according to the particle swarm iteration until a preset iteration number is reached, obtaining the optimized feature weights, and completing the dynamic weighting of each feature in the multi-dimensional feature matrix; the step of extracting the apparent defect feature vector by the YOLOv8 network comprises: inputting the preprocessed images into the backbone network of the YOLOv8 architecture, performing initial convolution operation on the images, adjusting the channel number of the images, and compressing the spatial size of the images at the same time; the backbone network extracts image features step by step through multiple CSP modules, each CSP module divides the input features into two branches, one branch undergoes multiple convolution operations, and the other branch is directly connected, the features of the two branches are fused, and image features of different scales are extracted; the neck network fuses the image features of different scales output by the backbone network through upsampling and downsampling operations, and obtains a fused multi-scale feature map; the fused feature map is input into the head network, the features are processed through convolution operation, the position and category information of the apparent defect are determined based on the anchor box mechanism, and the apparent defect feature vector is obtained; the step of extracting the temperature feature vector by the U-Net network comprises: The preprocessed image is input into the encoder part of the U-Net network, the encoder is composed of multiple convolution blocks and pooling layers, each convolution block extracts image features through convolution operation, and the pooling layer down-samples the feature map to gradually extract deep features of the image; The deep features enter the decoder part, the decoder enlarges the spatial size of the deep features through up-sampling operation to match the feature map size of the corresponding layer of the encoder, and the up-sampled features are spliced and fused with the corresponding layer feature map saved by the encoder; After multiple up-sampling and feature fusion operations, the decoder outputs a feature map with the same size as the input image, extracts features including local hot spot temperature and temperature difference gradient, and converts them into a temperature feature vector; the FasterR-CNN network extracts a discharge feature vector, including: The preprocessed image is input into the feature extraction network of the FasterR-CNN network, and the image is processed through convolution operation and pooling operation to extract a global feature map of the image; The global feature map is input into the region proposal network, which generates multiple anchor boxes on the feature map through sliding window, and then classifies and regresses the anchor boxes to filter out candidate regions that may contain targets; The region of interest pooling operation is performed on the filtered candidate regions to convert candidate region feature maps of different sizes into fixed-size feature vectors; The fixed-size feature vectors are input into the classifier and regressor to obtain accurate target regions, count the number of discharge pulses and energy distribution, and convert them into a discharge feature vector.

2. The image analysis based GIS switchgear operating condition assessment method as claimed in claim 1, wherein, The adaptive median filtering algorithm is used to denoise the collected initial image, and the histogram equalization combined with CLAHE is used to enhance the denoised image to obtain the preprocessed image, including: An initial size filter window is set for each pixel point in the initial image, the median value of the pixels in the window is calculated, and the gray value of the center pixel in the window is compared with the median value and the maximum and minimum values of the gray scale in the window; If the center pixel is noise, replace it with the median value, if it is not noise, determine whether the window has reached the maximum size, if not, increase the window size and repeat the calculation, if yes, keep the original pixel value; The gray scale histogram of the denoised image is expanded to the entire gray scale range, and then the CLAHE algorithm is used to divide the image into multiple sub-blocks, and histogram equalization is performed on each sub-block to obtain the preprocessed image.

3. The image analysis based GIS switchgear operating condition assessment method as claimed in claim 1, wherein, The multi-dimensional feature matrix is input into the LSTM neural network to obtain the device health index, and the device state level is divided according to the device health index to output the running state evaluation result of the GIS switch device, including: The multi-dimensional feature matrix optimized by the particle swarm is adjusted in format, and the adjusted multi-dimensional feature matrix is input into the input layer of the LSTM neural network in time sequence order, and enters the memory unit of the hidden layer through the input gate, the forgetting gate and the output gate; The hidden state is converted into a device health index through full connection operation, multiple health index thresholds are preset, each health index threshold corresponds to a device state level, the calculated device health index is compared with the health index threshold to divide the corresponding device state level; According to the divided equipment state level, a GIS switch equipment operation state evaluation result containing the equipment state level, the equipment health index and the characteristic information is generated.

4. A GIS switchgear operating condition assessment system based on image analysis, characterized in that, The system comprises: The acquisition module is configured to acquire initial images of the GIS switch equipment by using a high-definition camera, an infrared thermal imager and an ultraviolet imager respectively. The preprocessing module is configured to perform denoising processing on the acquired initial images by using an adaptive median filtering algorithm, and perform enhancement processing on the denoised images by using a combination of histogram equalization and CLAHE, to obtain preprocessed images. The feature extraction module is used for inputting the preprocessed image into a feature extraction model and splicing the extracted feature vectors into a multi-dimensional feature matrix, and dynamically adjusting the weight of the features through a particle swarm optimization algorithm, wherein the feature extraction model comprises a YOLOv8 architecture, a U-Net network and a FasterR-CNN network; the preprocessed image is input into the feature extraction model, the YOLOv8 network extracts an apparent defect feature vector, the U-Net network extracts a temperature feature vector, and the FasterR-CNN network extracts a discharge feature vector; the feature vectors output by the three networks are spliced into a multi-dimensional feature matrix, the position and speed of the particle swarm are initialized, and each particle represents a group of feature weights; the fitness value of each particle is calculated, the global optimal position and individual optimal position of the particle are updated according to the contribution of the features to the state evaluation of the GIS switch device; the position and speed of the particle are adjusted according to the iteration of the particle swarm until a preset iteration number is reached, the optimized feature weight is obtained, and the dynamic weighting of each feature in the multi-dimensional feature matrix is completed; the preprocessed image is input into the backbone network of the YOLOv8 architecture, the initial convolution operation is performed on the image first, the channel number of the image is adjusted, and the spatial size of the image is compressed at the same time; the backbone network extracts image features step by step through a plurality of CSP modules, each CSP module divides the input features into two branches, one branch undergoes multiple convolution operations, and the other branch is directly connected, the features of the two branches are fused, and image features of different scales are extracted; the neck network fuses the image features of different scales output by the backbone network through upsampling and downsampling operations to obtain a fused multi-scale feature map; the fused feature map is input into the head network, the features are processed through convolution operation, the position and category information of the apparent defect are determined based on the anchor box mechanism, and the apparent defect feature vector is obtained; the preprocessed image is input into the encoder part of the U-Net network, the encoder comprises a plurality of convolution blocks and pooling layers, each convolution block extracts image features through convolution operation, and the pooling layer down-samples the feature map to gradually extract deep features of the image; the deep features enter the decoder part, the decoder enlarges the spatial size of the deep features through upsampling operation to match the feature map size of the corresponding layer of the encoder, and the up-sampled features are spliced and fused with the corresponding layer feature map saved by the encoder; after a plurality of upsampling and feature fusion operations, the decoder outputs a feature map with the same size as the input image, extracts features including local hot spot temperature and temperature difference gradient, and converts them into a temperature feature vector; the preprocessed image is input into the feature extraction network of the FasterR-CNN network, the image is processed through convolution operation and pooling operation, and the global feature map of the image is extracted; the global feature map is input into the region proposal network, the region proposal network generates a plurality of candidate anchor boxes on the feature map through a sliding window, classifies and regresses the anchor boxes, and screens out candidate regions that may contain targets;The candidate region screened out is subjected to a region of interest pooling operation, and candidate region feature maps of different sizes are converted into fixed-size feature vectors; the fixed-size feature vectors are input into a classifier and a regressor to obtain an accurate target region, count the number of discharge pulses and the energy distribution, and convert into a discharge feature vector; The output module is configured to input the multi-dimensional feature matrix into an LSTM neural network to obtain the equipment health index, divide the equipment state level according to the equipment health index, and output the operation state evaluation result of the GIS switch equipment. The early warning module is configured to perform hierarchical early warning according to the operation state evaluation result of the GIS switch equipment.

5. GIS switchgear operating state evaluation device based on image analysis, characterized in that The image analysis-based GIS switch equipment operation state evaluation device comprises a memory and at least one processor, and the memory stores instructions; the at least one processor invokes the instructions in the memory, so that the image analysis-based GIS switch equipment operation state evaluation device performs each step of the image analysis-based GIS switch equipment operation state evaluation method according to any one of claims 1-3.

6. A computer-readable storage medium having stored thereon instructions, the computer-readable storage medium comprising: The instructions are executed by the processor to implement each step of the image analysis-based GIS switch equipment operation state evaluation method according to any one of claims 1-3.

Images