A high voltage bushing condition diagnosis method and system

Abstract

The application discloses a high-voltage bushing state diagnosis method and system, relates to the technical field of high-voltage bushing state diagnosis, and aims to solve the technical problem of low accuracy of the existing diagnosis method in high-voltage bushing state diagnosis.The high-voltage bushing state diagnosis method comprises the following steps: obtaining a visible light image and a thermal image of a high-voltage bushing; performing feature fusion on the visible light image and the thermal image to obtain a fused image; extracting feature data in the fused image by using a scale-invariant feature transform; inputting the feature data into a probability neural network optimized by using a salp swarm algorithm; and outputting a high-voltage bushing state diagnosis result represented by the fused image by the probability neural network.The diagnosis result comprises normal, mechanical failure and electrical failure.The technical scheme of the application is used for providing a high-voltage bushing state diagnosis method and system.

Description

Technical Field

[0001] This invention relates to the field of high-voltage bushing condition diagnosis technology, and more specifically, to a high-voltage bushing condition diagnosis method and system. Background Technology

[0002] High-voltage bushings are an important component of substation systems, and faults in high-voltage bushings are mainly classified into two categories: mechanical faults and electrical faults. In the diagnosis of high-voltage bushings, although traditional mathematical models based on multi-scale transformation, sparse representation, and low-rank representation can extract some complementary features, their reliance on manually designed rules makes them difficult to adapt to the feature expression requirements under complex operating conditions. For example, coupling interference caused by contamination on the bushing surface can easily lead to misjudgments.

[0003] In recent years, deep learning methods, such as Convolutional Neural Networks (CNNs) and Generative Adversarial Networks (GANs), have significantly improved fusion performance in the diagnosis of high-voltage bushings through semantic feature extraction. Specifically, CNNs can identify the correlation between micron-level cracks and local overheating, but they require deep stacking to expand the receptive field for feature extraction, and low-level information in the source image (such as edge and line structures) may be lost during convolution. While GANs preserve the realism of thermal distribution, the instability of network training may lead to low accuracy in fault diagnosis. Furthermore, Transformer technology can achieve global context modeling through self-attention mechanisms, associating cross-regional collaborative changes in temperature fields and texture features (such as causal analysis of the temperature rise of the bushing skirt and corrosion of adjacent ceramic bushings), but its high computational complexity makes it difficult to meet the real-time processing requirements of high-resolution images, and its accuracy in diagnosing high-voltage bushing faults remains low. It can be seen that the above-mentioned shortcomings lead to low accuracy in diagnosing the condition of high-voltage bushings, making it difficult to meet practical diagnostic requirements. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for diagnosing the condition of high-voltage bushings, thereby addressing the technical problem of low accuracy in diagnosing the condition of high-voltage bushings using existing diagnostic methods. In view of this, the invention is achieved through the following solution.

[0005] In a first aspect, the present invention provides a method for diagnosing the condition of a high-voltage bushing, comprising:

[0006] Acquire visible light and thermal images of the high-voltage bushing;

[0007] The visible light image and the thermal image are fused to obtain a fused image;

[0008] Feature data in the fused image is extracted using scale-invariant feature transform.

[0009] The feature data is input into a probabilistic neural network optimized using the SAP bee colony algorithm, and the probabilistic neural network outputs a high-voltage bushing status diagnosis result represented by the fused image; the diagnosis result includes normal, mechanical fault, and electrical fault.

[0010] Compared with existing technologies, the high-voltage bushing condition diagnosis method of the present invention, after acquiring the visible light image and thermal image of the high-voltage bushing, performs feature fusion processing to obtain the fused image. The feature data in the fused image is then extracted using Scale-Invariant Feature Transform (SIFT) and input into a probabilistic neural network optimized by the SAP bee colony algorithm to obtain the diagnostic result of the high-voltage bushing condition. Specifically, the feature fusion process in the above technical solution of the present invention can be implemented using a cross-modal Transformer fusion algorithm. This algorithm can capture global interactions in the spatial and channel domains through spatial transformers and channel transformers. Furthermore, by optimizing the parameters of the probabilistic neural network (PNN) based on the SAP bee colony algorithm, the correlation between fault features and classification results can be effectively strengthened, improving the accuracy of the PNN model for high-voltage bushing fault diagnosis. Simultaneously, this PNN model can not only effectively analyze fault conditions but also accelerate the accurate location of the cause of high-voltage bushing anomalies through its ideal model size. Its rapid response characteristics can meet the needs of intelligent real-time inspection. The above-mentioned technical solution of the present invention solves the technical problem of low accuracy in high-voltage bushing condition diagnosis of existing diagnostic methods, while reducing the cost of massive data processing and the delay in the processing process.

[0011] Furthermore, in the high-voltage bushing condition diagnosis method of the present invention, feature fusion of the visible light image and the thermal image includes:

[0012] Given a pair of visible light images and thermal images, the L-level feature pyramid is extracted by stacking convolutional layers using the cross-modal Transformer fusion algorithm.

[0013] The two sets of feature maps from the Lth level are input into the thinning branch to capture global information from the source image;

[0014] The initial feature map of the current level is concatenated with the enhanced features of the previous level and then input into the residual feature distillation block. An optimized feature map is generated through feature learning.

[0015] The optimized feature maps of each layer are fused using self-fusion convolutional blocks to obtain the fused image.

[0016] Furthermore, in the high-voltage bushing condition diagnosis method of the present invention, after generating the optimized feature map through feature learning, it further includes:

[0017] By mining complementary information across channels to capture the global contextual relationships of the source image, redundant information is removed from the feature maps in both the spatial and channel domains.

[0018] Furthermore, in the high-voltage bushing condition diagnosis method of the present invention, during the feature fusion process of the visible light image and the thermal image, a total loss function is constructed by optimizing data loss, spatial loss, and frequency loss. The expression of the total loss function is:

[0019] ;

[0020] in, Represents the total loss function. This represents the loss function for fused data. Represents the space loss function, Represents the perceptual loss function. Represents the frequency loss function. , , These represent the respective hyperparameters, used to control the weights between the four losses.

[0021] Furthermore, in the high-voltage bushing condition diagnosis method of the present invention, the step of extracting feature data from the fused image using scale-invariant feature transform includes:

[0022] The scale-invariant feature transform is used to perform multi-scale representation of the input fused image, and a Gaussian pyramid is constructed.

[0023] Local extrema are detected in scale space by Gaussian difference, parameters are optimized for the image characteristics of the high-voltage bushing, and extrema are selected as candidate key points by comparing each pixel with 26 points in its adjacent scale and spatial neighborhood.

[0024] Taylor expansion is used to perform sub-pixel level correction on the candidate keypoint positions to eliminate unstable points;

[0025] The high-voltage bushing is accurately located, and the neighborhood of the key point is divided into 4×4 sub-regions. The gradient histogram of each sub-region is calculated in 8 directions to form a 128-dimensional descriptor; feature data extraction is completed.

[0026] Furthermore, in the high-voltage bushing condition diagnosis method of the present invention, the probabilistic neural network optimized by the SAP bee colony algorithm includes:

[0027] The neurons of the probabilistic neural network receive input signals and transmit the input signals to the pattern layer; in the pattern layer, the neurons are grouped by category and used to calculate the matching degree between the category of the corresponding training sample of the neuron in the pattern layer and the input signal.

[0028] Regarding the first k The sequence or feature is related to the first c The neuron output of a category can be obtained using the following formula:

[0029] ;

[0030] in, Indicates the first k The sequence or feature is related to the first c The output value of the neuron of the category, Represents pi (π). Indicates the smoothing constraint parameter, n Indicates the input feature dimension. Represented by natural constant e An exponential function with base 0. X Indicates the input sample. Indicates the first c Category No. k The nucleus center of each neuron, and , express n 3D real space, Indicates input sample X With training samples The Euclidean distance;

[0031] Summing the outputs of the pattern layers for the same category, and using density to calculate the closest class conditional probability distribution;

[0032] The pattern vectors are classified based on the maximum output of the summation unit, and the category with the highest probability is selected as the final classification result according to the Bayesian decision rule.

[0033] With the classification accuracy of the probabilistic neural network as the optimization objective, the parameters of the SAP bee colony algorithm are optimized, and the optimal optimization variables for minimizing the loss function of the bass lead search are updated.

[0034] Furthermore, in the high-pressure bushing condition diagnosis method of the present invention, the optimal optimization variable for updating the bass lead search to minimize the loss function is expressed as:

[0035] ;

[0036] in, Indicates the first leader jWei's new position Indicates the food source in the first place j The position in the dimension, i.e., the reference value. Indicates the first j The upper bound of the dimension, Indicates the first j The lower bound of a dimensional variable. , This represents a random number between 0 and 1, which is used to adjust the position in the update formula. This represents the perturbation intensity coefficient, which adaptively decreases with increasing iteration number. , e Represents the natural constant. t Indicates the number of iterations. Indicates the preset total number of iterations; for the position of the follower Then there is, , Indicates the first j Vidi i The updated position of each follower Indicates the first j The position of the previous follower.

[0037] Furthermore, in the high-pressure bushing condition diagnosis method of the present invention, after updating the optimal optimization variable of the bass lead search to minimize the loss function, it further includes:

[0038] The optimal optimization variables output by the SAP bee colony algorithm are substituted into the model corresponding to the probabilistic neural network to complete the optimization of the probabilistic neural network.

[0039] Furthermore, in the high-pressure bushing condition diagnosis method of the present invention, the mechanical faults include surface contamination, as well as skirt corrosion and skirt damage;

[0040] The electrical faults include dielectric loss, poor contact, oil leakage, and partial discharge.

[0041] Secondly, the present invention provides a high-voltage bushing condition diagnostic system, comprising:

[0042] The image acquisition module is used to acquire visible light and thermal images of the high-voltage bushing;

[0043] The image fusion module is used to perform feature fusion on the visible light image and the thermal image to obtain a fused image;

[0044] The feature data acquisition module is used to extract feature data from the fused image based on scale-invariant feature transformation;

[0045] The status diagnosis module is used to input the feature data into a probabilistic neural network optimized by the SAP bee colony algorithm, and the probabilistic neural network outputs the high-voltage bushing status diagnosis result represented by the fused image; the diagnosis result includes normal, mechanical fault and electrical fault.

[0046] Compared with the prior art, the beneficial effects of the high-voltage bushing condition diagnosis system of the present invention are the same as those of the high-voltage bushing condition diagnosis method described in the above technical solution, and will not be repeated here. Attached Figure Description

[0047] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0048] Figure 1 This is a schematic diagram of the cross-modal Transformer fusion algorithm architecture implemented by capturing global interactions in the spatial and channel domains in this invention;

[0049] Figure 2 This is a schematic diagram of the self-fusion convolutional block architecture in the cross-modal Transformer fusion algorithm of this invention;

[0050] Figure 3 This is a schematic diagram of the architecture of each layer of the probabilistic neural network optimized based on the SAP bee colony algorithm in this invention;

[0051] Figure 4 This is a flowchart illustrating a high-voltage bushing condition diagnosis method according to the present invention. Detailed Implementation

[0052] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0053] It should be noted that when a component is referred to as being "fixed to" or "set on" another component, it can be directly on or indirectly on that other component. When a component is referred to as being "connected to" another component, it can be directly connected to or indirectly connected to that other component.

[0054] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. "Several" means one or more, unless otherwise explicitly specified.

[0055] In recent years, deep learning methods, such as Convolutional Neural Networks (CNNs) and Generative Adversarial Networks (GANs), have significantly improved fusion performance in the diagnosis of high-voltage bushings through semantic feature extraction. Specifically, CNNs can identify the correlation between micron-level cracks and local overheating, but they require deep stacking to expand the receptive field for feature extraction, and low-level information in the source image (such as edge and line structures) may be lost during convolution. While GANs preserve the realism of thermal distribution, the instability of network training may lead to low accuracy in fault diagnosis. Furthermore, Transformer technology can achieve global context modeling through self-attention mechanisms, associating cross-regional collaborative changes in temperature fields and texture features (such as causal analysis of the temperature rise of the bushing skirt and corrosion of adjacent ceramic bushings), but its high computational complexity makes it difficult to meet the real-time processing requirements of high-resolution images, and its accuracy in diagnosing high-voltage bushing faults remains low. It can be seen that the above-mentioned shortcomings lead to low accuracy in diagnosing the condition of high-voltage bushings, making it difficult to meet practical diagnostic requirements.

[0056] To address the above technical problems, this invention provides a method for diagnosing the condition of high-pressure bushings, comprising:

[0057] Acquire visible light and thermal images of the high-voltage bushing;

[0058] The visible light image and the thermal image are fused to obtain a fused image;

[0059] Feature data in the fused image is extracted using scale-invariant feature transform.

[0060] The feature data is input into a probabilistic neural network optimized using the SAP bee colony algorithm, and the probabilistic neural network outputs a high-voltage bushing status diagnosis result represented by the fused image; the diagnosis result includes normal, mechanical fault, and electrical fault.

[0061] In the high-voltage bushing condition diagnosis method of the present invention, after acquiring the visible light image and thermal image of the high-voltage bushing, the fused image is obtained by feature fusion processing. The feature data in the fused image is extracted using Scale Invariant Feature Transform (SIFT), and then input into a probabilistic neural network optimized by the SAP bee colony algorithm to obtain the diagnosis result of the high-voltage bushing condition. Specifically, the feature fusion process in the above technical solution of the present invention can be implemented using a cross-modal Transformer fusion algorithm. This algorithm can capture global interactions in the spatial and channel domains through spatial transformers and channel transformers. Furthermore, by optimizing the parameters of the probabilistic neural network (PNN) based on the SAP bee colony algorithm, the correlation between fault features and classification results can be effectively strengthened, improving the accuracy of the PNN model for high-voltage bushing fault diagnosis. Simultaneously, the PNN model can not only effectively analyze fault conditions but also accelerate the accurate location of the cause of high-voltage bushing anomalies through its ideal model size. Its rapid response characteristics can meet the needs of intelligent real-time inspection. The technical solution of the present invention solves the technical problem of low accuracy in diagnosing the condition of high-voltage bushings using existing diagnostic methods.

[0062] To better understand the present invention, the following specific embodiments further illustrate the content of the present invention, but the content of the present invention is not limited to the following embodiments.

[0063] Example 1

[0064] This embodiment provides a method for diagnosing the condition of high-voltage bushings, including:

[0065] Step 1: Acquire visible light and thermal images of the high-voltage bushing;

[0066] Step 2: Perform feature fusion on the visible light image and the thermal image to obtain a fused image;

[0067] Step 3: Extract feature data from the fused image using Scale Invariant Feature Transform (SIFT);

[0068] Step 4: Using a probabilistic neural network optimized by the SAP bee colony algorithm, the feature data is input into the probabilistic neural network, and the probabilistic neural network outputs the high-voltage bushing status diagnosis result represented by the fused image; the diagnosis result includes three types of status: normal, mechanical fault, and electrical fault.

[0069] Example 2

[0070] This embodiment provides a method for diagnosing the condition of high-voltage bushings, including:

[0071] S100, acquire visible light and thermal images of the high-voltage bushing;

[0072] S200, perform feature fusion on the visible light image and the thermal image to obtain a fused image;

[0073] Furthermore, the feature fusion of the visible light image and the thermal image includes:

[0074] S201, given a pair of visible light images and thermal images, please refer to Figure 1 The cross-modal Transformer fusion algorithm is used to extract an L-level feature pyramid through stacked convolutional layers;

[0075] S202, input the two sets of feature maps of the Lth level into the thinning branch to capture global information of the source image;

[0076] S203: The initial feature map of the current level is concatenated with the enhanced features of the previous level and then input into the residual feature distillation block, and optimized features are generated through feature learning;

[0077] S204, by mining complementary information across channels to capture the global contextual association of the source image, redundant information is removed from the feature maps in the spatial domain and channel domain respectively, further enhancing their complementarity; two visible features and two infrared features are extracted. In this embodiment, the two types of features are merged to convey important information and improve the interactivity within the spatial channel attention so that a single 1×1 convolutional layer can be used to further fuse the convolutional layers.

[0078] S205, please refer to Figure 2 The optimized feature maps of each layer are fused using self-fusion convolutional blocks to obtain the fused image;

[0079] S206, in the process of feature fusion of the visible light image and the thermal image, a total loss function is constructed by optimizing data loss, spatial loss, and frequency loss. The expression of the total loss function is: ;in, Represents the total loss function. This represents the loss function for fused data. Represents the space loss function, Represents the perceptual loss function. Represents the frequency loss function. , , These represent the respective hyperparameters, used to control the weights between the four losses;

[0080] S300, Scale-Invariant Feature Transform (SIFT) is used to extract feature data from the fused image;

[0081] Furthermore, step S300 includes:

[0082] S301, The scale-invariant feature transform (SIFT) is used to perform multi-scale representation on the input fused image to construct a Gaussian pyramid;

[0083] S302, in the scale space, local extreme points are detected by Gaussian difference, the parameters are optimized according to the image characteristics of the high-voltage bushing, and extreme points are selected as candidate key points by comparing each pixel with 26 points in its adjacent scale and spatial neighborhood.

[0084] S303 uses Taylor expansion to perform sub-pixel level correction on the candidate key point positions and eliminates unstable points.

[0085] S304, accurately locates the high-voltage bushing, divides the neighborhood of key points into 4×4 sub-regions, calculates gradient histograms in 8 directions for each sub-region, forming a 128-dimensional descriptor; completes feature data extraction;

[0086] S400 utilizes a probabilistic neural network optimized by the SAP bee colony algorithm. The feature data is input into the probabilistic neural network, and the probabilistic neural network outputs the high-voltage bushing status diagnosis results represented by the fused image. The diagnosis results include normal, mechanical faults, and electrical faults. The mechanical faults include surface contamination, as well as skirt corrosion and skirt damage. The electrical faults include dielectric loss, poor contact, oil leakage, and partial discharge.

[0087] Further, in step S400, the probabilistic neural network optimized by the SAP bee colony algorithm includes:

[0088] S401, the neurons of the probabilistic neural network receive input signals and transmit the input signals to the pattern layer; in the pattern layer, the neurons are grouped by category and used to calculate the matching degree between the category of the corresponding training sample of the neuron in the pattern layer and the input signal;

[0089] S402, for the first k The sequence or feature is related to the first c The neuron output of a category can be obtained using the following formula:

[0090] ;in, Indicates the first k The sequence or feature is related to the first c The output value of the neuron of the category, Represents pi (π). Indicates the smoothing constraint parameter, n Indicates the input feature dimension. Represented by natural constante An exponential function with base 0. X Indicates the input sample. Indicates the first c Category No. k The nucleus center of each neuron, and , express n 3D real space, Indicates input sample X With training samples The Euclidean distance;

[0091] S403 sums the outputs of the pattern layer for the same category and uses density to calculate the closest class conditional probability distribution;

[0092] S404: Based on the maximum output of the summation unit, the pattern vector is classified, and the category with the highest probability is selected as the final classification result according to the Bayesian decision rule.

[0093] S405, with the classification accuracy of the probabilistic neural network as the optimization objective, perform parameter optimization of the SAP bee colony algorithm and update the optimal optimization variable of the bass lead search to minimize the loss function.

[0094] S406, Substitute the optimal optimization variables output by the SAP bee colony algorithm into the model corresponding to the probabilistic neural network to complete the optimization of the probabilistic neural network;

[0095] Furthermore, the optimal optimization variable for updating the bass lead search to minimize the loss function is expressed as:

[0096] ;

[0097] in, Indicates the first leader j Wei's new position Indicates the food source in the first place j The position in the dimension, i.e., the reference value. Indicates the first j The upper bound of the dimension, Indicates the first j The lower bound of a dimensional variable. , This represents a random number between 0 and 1, which is used to adjust the position in the update formula. This represents the perturbation intensity coefficient, which adaptively decreases with increasing iteration number. , e Represents the natural constant. t Indicates the number of iterations. Indicates the preset total number of iterations; for the position of the follower Then there is, , Indicates the first j Vidi i The updated position of each follower Indicates the first j The position of the previous follower.

[0098] Example 3

[0099] Firstly, this embodiment provides a method for diagnosing the condition of a high-voltage bushing, including:

[0100] S100, please refer to Figure 4 To acquire visible light and thermal images of the high-voltage bushing;

[0101] Specifically, this embodiment utilizes an OV5640 camera to capture visible light images for system analysis, and a low-cost infrared sensor, FLIR Lepton 3.5, to capture thermal images for system analysis; the camera provides an automatically calibrated 160×120 pixel output image in a 14-bit data packet acquired by the Raspberry Pi 4 board by establishing an SPI communication protocol between the devices.

[0102] Furthermore, after acquiring the visible light and thermal images of the high-voltage bushing, the images are preprocessed.

[0103] Specifically, images received via the Raspberry Pi 4 board's communication port in 14-bit data packets require a preprocessing stage. This stage remaps the pixel values ​​of the thermal image to 8-bit data packets, that is, remaps the initial range of pixel values ​​from 0-16383 to the target range of 0-255. The image is then subjected to noise reduction, contrast enhancement, and grayscale processing.

[0104] S200, perform feature fusion on the visible light image and the thermal image to obtain a fused image;

[0105] Furthermore, the feature fusion of the visible light image and the thermal image includes:

[0106] S201, given a pair of visible light images and thermal images, please refer to Figure 1 The cross-modal Transformer fusion algorithm is used to extract an L-level feature pyramid through stacked convolutional layers;

[0107] S202, input the two sets of feature maps of the Lth level into the thinning branch to capture global information of the source image;

[0108] S203: The initial feature map of the current level is concatenated with the enhanced features of the previous level and then input into the residual feature distillation block. Optimized feature maps are then generated through feature learning. and ;

[0109] S204 captures the global contextual relationships of the source image by mining complementary information across channels, and modifies the feature maps in both the spatial and channel domains. and Redundant information is removed to further enhance the complementarity between the two; two visible features { , } and two infrared features { , The extracted features are combined in this embodiment to convey important information and improve interactivity within the spatial channel attention, so that a single 1×1 convolutional layer can be used for further fusion; then:

[0110] ;

[0111] ;

[0112] in, The optimized feature map features representing the visible light modes. This represents 1×1 convolutional layer fusion dimensionality reduction. Spatial attention feature map features representing visible light modes. Channel attention feature map features representing visible light modes. This indicates a concatenation operation within the channel dimension. The optimized feature map features representing the infrared modes. Spatial attention feature map features representing infrared modes, Channel attention feature map features representing infrared modes;

[0113] S205, please refer to Figure 2 The optimized feature maps from each layer are fused using self-fusion convolutional blocks to obtain the fused image. ;

[0114] S206, in the process of feature fusion of the visible light image and the thermal image, a total loss function is constructed by optimizing data loss, spatial loss, and frequency loss. The expression of the total loss function is: ;in, Represents the total loss function. This represents the loss function for fused data. Represents the space loss function, Represents the perceptual loss function. Represents the frequency loss function. , , These represent the respective hyperparameters, used to control the weights between the four losses.

[0115] S300, Scale-Invariant Feature Transform (SIFT) is used to extract feature data from the fused image;

[0116] Furthermore, step S300 includes:

[0117] S301, The scale-invariant feature transform (SIFT) is used to perform multi-scale representation on the input fused image to construct a Gaussian pyramid;

[0118] S302, in the scale space, local extreme points are detected by difference of Gaussians (DoG), the parameters are optimized according to the image characteristics of the high-voltage bushing, and extreme points are selected as candidate key points by comparing each pixel with 26 points in its adjacent scale and spatial neighborhood.

[0119] S303 uses Taylor expansion to perform sub-pixel level correction on the candidate key point positions, and removes unstable points with low contrast (e.g., |D(x)| < 0.03, which means that the contrast of the point is too low and the feature point may be caused by noise. D(x) represents the DoG response value of the corrected sub-pixel level precise position) or edge response (judged by the eigenvalue of the Hessian matrix).

[0120] S304, accurately locates the high-voltage bushing, divides the neighborhood of key points into 4×4 sub-regions (16 in total), calculates gradient histograms in 8 directions for each sub-region, forming a 128-dimensional (16×8-dimensional) descriptor; completes feature data extraction.

[0121] S400, please refer to Figure 3 The probabilistic neural network optimized using the SAP bee colony algorithm is used to input the feature data into the probabilistic neural network, and the probabilistic neural network outputs the high-voltage bushing status diagnosis results represented by the fused image; please refer to Table 1, the diagnosis results include normal, mechanical fault and electrical fault; the mechanical fault includes surface contamination, as well as skirt corrosion and skirt damage; the electrical fault includes dielectric loss, poor contact, oil leakage fault and partial discharge;

[0122] Further, in step S400, the probabilistic neural network optimized by the SAP bee colony algorithm includes:

[0123] S401, the neurons of the probabilistic neural network receive input signals and transmit the input signals to the pattern layer; in the pattern layer, the neurons are grouped by category and used to calculate the matching degree between the category of the corresponding training sample of the neuron in the pattern layer and the input signal;

[0124] S402, for the first k The sequence or feature is related to the first cThe neuron output of a category can be obtained using the following formula:

[0125] ;in, Indicates the first k The sequence or feature is related to the first c The output value of the neuron of the category, Represents pi (π). Indicates the smoothing constraint parameter, n Indicates the input feature dimension. Represented by natural constant e An exponential function with base 0. X Indicates the input sample. Indicates the first c Category No. k The nucleus center of each neuron, and , express n 3D real space, Indicates input sample X With training samples The Euclidean distance;

[0126] S403 sums the outputs of the pattern layer for the same category and uses density to calculate the closest class conditional probability distribution;

[0127] S404: Based on the maximum output of the summation unit, the pattern vector is classified, and the category with the highest probability is selected as the final classification result according to the Bayesian decision rule.

[0128] S405, with the classification accuracy of the probabilistic neural network as the optimization objective, perform parameter optimization of the SAP bee colony algorithm and update the optimal optimization variable of the bass lead search to minimize the loss function.

[0129] S406, Substitute the optimal optimization variables output by the SAP bee colony algorithm into the model corresponding to the probabilistic neural network to complete the optimization of the probabilistic neural network;

[0130] Furthermore, the optimal optimization variable for updating the bass lead search to minimize the loss function is expressed as:

[0131] ;

[0132] in, Indicates the first leader j Wei's new position Indicates the food source in the first place j The position in the dimension, i.e., the reference value. Indicates the first j The upper bound of the dimension, Indicates the first jThe lower bound of a dimensional variable. , This represents a random number between 0 and 1, which is used to adjust the position in the update formula. This represents the perturbation intensity coefficient, which adaptively decreases with increasing iteration number. , e Represents the natural constant. t Indicates the number of iterations. Indicates the preset total number of iterations; for the position of the follower Then there is, , Indicates the first j Vidi i The updated position of each follower Indicates the first j The position of the previous follower.

[0133] Table 1 High-voltage bushing condition diagnosis

[0134]

[0135] Secondly, this embodiment provides a high-voltage bushing condition diagnostic system, including:

[0136] The image acquisition module is used to acquire visible light and thermal images of the high-voltage bushing;

[0137] The image fusion module is used to perform feature fusion on the visible light image and the thermal image to obtain a fused image;

[0138] The feature data acquisition module is used to extract feature data from the fused image based on the scale-invariant feature transform (SIFT);

[0139] The status diagnosis module is used to input the feature data into a probabilistic neural network optimized by the SAP bee colony algorithm. The probabilistic neural network outputs the high-voltage bushing status diagnosis result represented by the fused image. The diagnosis result includes three types of status: normal, mechanical fault, and electrical fault.

[0140] Example 4

[0141] This embodiment employs the same high-voltage bushing condition diagnosis method and system as Embodiment 3 above. The purpose of this embodiment is to demonstrate the effectiveness of the technical solution of the present invention through specific numerical examples. Specifically, to test the method proposed in this invention, a self-built dataset is used for training. In this embodiment, image blocks of size 256×256 are randomly cropped from the original image, and the RGB color space of the visible light image is converted to grayscale space. The Adam optimizer is used during training, with parameters set to... =0.9、 =0.999, initial learning rate 10 -4 The training period is 20; hyperparameters , and The weights were fixed at 0.8, 0.02, and 0.05 respectively to balance optimal overall subjective quality and training stability under the constraint of prioritizing spatial information. and All values ​​are set to 0.5 to ensure a balanced contribution of infrared and visible light features during the fusion process; the values ​​of the important weighting parameters mentioned above are shown in Table 2 below.

[0142] Table 2 Important weight parameter values

[0143]

[0144] Furthermore, to verify the effectiveness of the image fusion method mentioned in this invention, it is compared with traditional fusion methods GAN, RFN-Nest, and IVFusion; for ease of comparison, three quality indicators are used to evaluate the image fusion performance: total edge information (... The scores for the three metrics—Multi-scale Structural Similarity (MS-SSIM), Sum of Difference Correlation (SCD)—are calculated between the source and fused images and then averaged. Higher scores indicate better performance. The final test results are shown in Table 3 below, where CMTs represent the cross-modal Transformer fusion algorithm of this invention.

[0145] Table 3 Test Results

[0146]

[0147] As can be seen from Table 3 above, quantitatively comparing the fusion performance on the datasets, for The cross-modal Transformer fusion algorithm (CMTs) mentioned in this invention scores the highest, indicating that this invention effectively preserves complementary information in the source image. In addition, by preserving the structure and fine texture in the source image, it outperforms other algorithms on fidelity-based metrics MS-SSIM and SCD, achieving the best overall performance.

[0148] Furthermore, to verify the effectiveness of the probabilistic neural network (PNN) classification module optimized by the SAP bee colony algorithm mentioned in this invention, it is compared with traditional algorithms SVM and ANN, and the data representation is shown in Table 4 below.

[0149] Table 4 Data Representation

[0150]

[0151] As shown in Table 4, the PNN classifier outperforms both ANN and SVM classifiers in terms of accuracy and response speed. This invention optimizes the PNN parameters based on the SAP bee colony algorithm, effectively strengthening the correlation between fault features and classification results, and improving the accuracy of the PNN model in diagnosing high-voltage bushing faults. Furthermore, the PNN model not only effectively analyzes fault states but also accelerates the precise location of the causes of high-voltage bushing anomalies through its ideal model size. Its rapid response characteristic meets the needs of intelligent real-time inspection. The PNN model in this invention can serve as a highly efficient model for high-voltage bushing image analysis and state classification tasks.

[0152] In the description of the above embodiments, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0153] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for diagnosing the condition of a high-voltage bushing, characterized in that, include: Acquire visible light and thermal images of the high-voltage bushing; The visible light image and the thermal image are fused to obtain a fused image; Feature data in the fused image is extracted using scale-invariant feature transform. A probabilistic neural network optimized using the SAP bee colony algorithm is used to input the feature data. The probabilistic neural network outputs a high-voltage bushing status diagnosis result represented by the fused image. The diagnosis result includes normal operation, mechanical fault, and electrical fault. The feature fusion of the visible light image and the thermal image includes: Given a pair of visible light images and thermal images, the L-level feature pyramid is extracted by stacking convolutional layers using the cross-modal Transformer fusion algorithm. The two sets of feature maps from the Lth level are input into the thinning branch to capture global information from the source image; The initial feature map of the current level is concatenated with the enhanced features of the previous level and then input into the residual feature distillation block. An optimized feature map is generated through feature learning. The optimized feature maps from each layer are fused using self-fusion convolutional blocks to obtain the fused image; The probabilistic neural network optimized using the SAP bee colony algorithm includes: The neurons of the probabilistic neural network receive input signals and transmit the input signals to the pattern layer; in the pattern layer, the neurons are grouped by category and used to calculate the matching degree between the category of the corresponding training sample and the input signal. Summing the outputs of the pattern layers for the same category, and using density to calculate the closest class conditional probability distribution; The pattern vectors are classified based on the maximum output of the summation unit, and the category with the highest probability is selected as the final classification result according to the Bayesian decision rule. With the classification accuracy of the probabilistic neural network as the optimization objective, the parameters of the SAP bee colony algorithm are optimized, and the optimal optimization variables for minimizing the loss function of the bass lead search are updated. Then, the optimal optimization variables output by the SAP bee colony algorithm are substituted into the model corresponding to the probabilistic neural network to complete the optimization of the probabilistic neural network. The extraction of feature data from the fused image using scale-invariant feature transform includes: The scale-invariant feature transform is used to perform multi-scale representation of the input fused image, and a Gaussian pyramid is constructed. Local extrema are detected in scale space by Gaussian difference, parameters are optimized for the image characteristics of the high-voltage bushing, and extrema are selected as candidate key points by comparing each pixel with 26 points in its adjacent scale and spatial neighborhood. Taylor expansion is used to perform sub-pixel level correction on the candidate keypoint positions to eliminate unstable points; Accurately locate the high-voltage bushing, divide the neighborhood of the key point into 4×4 sub-regions, calculate the gradient histogram in 8 directions for each sub-region to form a 128-dimensional descriptor; complete feature data extraction. The optimal optimization variable for updating the bass ray search to minimize the loss function is expressed as: ； in, Indicates the first leader j Wei's new position Indicates the food source in the first place j The position in the dimension, i.e., the reference value. Indicates the first j The upper bound of the dimension, Indicates the first j The lower bound of a dimensional variable. , This represents a random number between 0 and 1, which is used to adjust the position in the update formula. This represents the perturbation intensity coefficient, which adaptively decreases with increasing iteration number. , e Represents the natural constant. t Indicates the number of iterations. Indicates the preset total number of iterations; for the position of the follower Then there is, , Indicates the first j Vidi i The updated position of each follower Indicates the first j The position of the previous follower.

2. The high-voltage bushing condition diagnosis method according to claim 1, characterized in that, After generating the optimized feature map through feature learning, the following is also included: By mining complementary information across channels to capture the global contextual relationships of the source image, redundant information is removed from the feature maps in both the spatial and channel domains.

3. The high-voltage bushing condition diagnosis method according to claim 2, characterized in that, In the process of feature fusion of the visible light image and the thermal image, a total loss function is constructed by optimizing data loss, spatial loss, and frequency loss. The expression of the total loss function is as follows: ； in, Represents the total loss function. This represents the loss function for fused data. Represents the space loss function, Represents the perceptual loss function. Represents the frequency loss function. , , These represent the respective hyperparameters, used to control the weights between the four losses.

4. The high-voltage bushing condition diagnosis method according to claim 1, characterized in that, In the process of optimizing the probabilistic neural network using the SAP bee colony algorithm: Regarding the first k The sequence or feature is related to the first c The neuron output of a category can be obtained using the following formula: ； in, Indicates the first k The sequence or feature is related to the first c The output value of the neuron of the category, Represents pi (π). Indicates the smoothing constraint parameter, n Indicates the input feature dimension. Represented by natural constant e An exponential function with base 0. X Indicates the input sample. Indicates the first c Category No. k The nucleus center of each neuron, and , express n 3D real space, As training samples, Indicates input sample X With training samples The Euclidean distance.

5. The high-voltage bushing condition diagnosis method according to claim 1, characterized in that, The mechanical failures include surface contamination, as well as skirt corrosion and skirt damage; The electrical faults include dielectric loss, poor contact, oil leakage, and partial discharge.

6. A high-voltage bushing condition diagnostic system, characterized in that, The high-voltage bushing condition diagnosis method according to any one of claims 1 to 5 includes: The image acquisition module is used to acquire visible light and thermal images of the high-voltage bushing; The image fusion module is used to perform feature fusion on the visible light image and the thermal image to obtain a fused image; The feature data acquisition module is used to extract feature data from the fused image based on scale-invariant feature transformation; The status diagnosis module is used to input the feature data into a probabilistic neural network optimized by the SAP bee colony algorithm, and the probabilistic neural network outputs the high-voltage bushing status diagnosis result represented by the fused image; the diagnosis result includes normal, mechanical fault and electrical fault.

Images