Multi-spectral cooperative sensing and adaptive diagnosis method and device for extra-high voltage valve hall

Abstract

This invention relates to a multispectral collaborative sensing and adaptive diagnostic method for ultra-high voltage valve halls, comprising: acquiring high-resolution three-spectrum images captured at preset sampling points and forming a dataset; constructing a multispectral collaborative sensing model and processing the images output from the three branches; inputting the preprocessed training set into the multispectral collaborative sensing model for training; inputting the parameters of the sampling points of the valve tower under test and the high-resolution three-spectrum images under test into the trained multispectral collaborative sensing model to calculate the fault status of the valve tower. This invention overcomes the bottleneck of monitoring single physical quantities through collaborative sensing of visible light, infrared, and ultraviolet bands, effectively identifying complex defects such as valve tower discharge and overheating of dissimilar copper-aluminum metal contact; constructing a physical-driven real-time diagnostic system, coupled with physics knowledge, and through multi-parameter correlation analysis of mechanical deformation, temperature change, and discharge intensity, it can effectively distinguish defects from interference signals, eliminate spatial monitoring blind spots, and enhance anti-interference capabilities.

Description

Technical Field

[0001] This invention relates to the field of monitoring technology for high voltage direct current transmission equipment, and in particular to a multispectral collaborative sensing and adaptive diagnostic method and equipment for ultra-high voltage valve halls. Background Technology

[0002] As a core component of DC transmission systems, ultra-high voltage converter valves endure high voltage, high current, and complex electromagnetic environments for extended periods. The operational status of the equipment within the valve chamber directly impacts system reliability. Statistical data shows that typical faults are numerous, including trigger plate burnout, loose capacitor wiring, and discharge of equalizing electrodes. These faults also include overheating caused by poor contact and insulation degradation caused by partial discharge.

[0003] Traditional detection methods have the following significant limitations:

[0004] First, single-spectrum detection is difficult to fully characterize the equipment status. For example, infrared thermometry cannot capture ultraviolet discharge characteristics, and visible light inspection is difficult to detect early insulation degradation.

[0005] Second, fixed monitoring has blind spots, and overheating of the connectors often occurs in hidden locations such as the junction of the busbar and the shield, making it difficult to detect.

[0006] Third, manual inspections suffer from poor timeliness and insufficient quantification; despite daily temperature checks, timely warnings are still not issued. Summary of the Invention

[0007] To address the problems of difficulty in comprehensively characterizing equipment status, the existence of visual blind spots, and poor timeliness, the primary objective of this invention is to provide a multispectral collaborative sensing and adaptive diagnostic method for ultra-high voltage valve halls that can effectively identify complex defects such as valve tower discharge and copper-aluminum dissimilar metal contact overheating, eliminate spatial monitoring blind spots, and enhance anti-interference capabilities.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: a multispectral collaborative sensing and adaptive diagnosis method for ultra-high voltage valve halls, the method comprising the following sequential steps:

[0009] (1) A gimbal camera equipped with a high-pixel, zoom three-light camera is used. For the range of valve hall and valve tower, the gimbal camera is planned to set up the sampling points and paths. Then the gimbal camera collects images according to the preset sampling points and paths.

[0010] (2) Acquire high-resolution three-light images taken at preset sampling points and form a dataset. Divide the dataset into training set, validation set and test set and perform preprocessing.

[0011] (3) Construct a multispectral collaborative sensing model, process the images output from the three branches, and obtain weighted fusion features. ;

[0012] (4) Input the preprocessed training set into the multispectral collaborative sensing model for training to obtain the trained multispectral collaborative sensing model;

[0013] (5) Input the parameters of the sampling points of the valve tower to be tested and the high-resolution three-light image to be tested into the trained multispectral collaborative sensing model, and perform weighted fusion based on the features. The failure status of the valve tower, i.e., the failure probability matrix, is calculated. .

[0014] In step (1), the gimbal camera acquires visible light, infrared, and ultraviolet image information; the sampling points are set using a global sampling method, with 300 sampling points set according to a preset position; the parameters of the sampling points include valve tower number, valve tower layer number, camera position, height, orientation angle, and zoom magnification.

[0015] Step (2) specifically includes the following steps:

[0016] (2a) Collect images at each preset sampling point;

[0017] (2b) Label the effective regions in the acquired high-resolution tri-light images, and the labeled images form a dataset;

[0018] (2c) Divide the dataset into training set, validation set and test set in units of sampling points; perform data augmentation operations such as rotation, flipping, scaling and mosaic on the training set; and perform normalization on the validation set and test set.

[0019] Step (3) specifically includes the following steps:

[0020] (3a) Improve the YOLOv5 model by replacing the backbone network of the YOLOv5 model with a parallel ResNet-50 network, ConvLSTM network and FFT-CNN fusion network, and replacing the feature extraction network of the YOLOv5 model with a physical law feature extraction module; the ResNet-50 network processes the visible light image to form the visible light branch; the ConvLSTM network processes the infrared image to form the infrared branch; and the FFT-CNN fusion network processes the ultraviolet image to form the ultraviolet branch.

[0021] (3b) The physical law feature extraction module processes the images output from the three branches, specifically as follows:

[0022] The visible light branch outputs a visible light image F1(B, 3, H, W), where B is the batch size, 3 is the number of channels, H is the image height, and W is the image width. After normalization and standardization, F1 is converted into a grayscale image F2(B, 1, H, W). Then, through a convolutional layer and progressive downsampling pooling, a grayscale feature map F3(B, 1, H / 8, W / 8) is obtained. Laplacian kernel is used for Laplacian calculation to obtain a transformed feature map F4(B, 64, H / 8, W / 8). Subsequently, elastic module constraints are applied, and a convolution operation, W*F4, is added to extract features. The elastic module constraints, W*F4, are then added to the transformed feature map F4, and the ReLU activation function is used to output the first feature map Fv.

[0023] The infrared branch outputs a single-channel temporal infrared image sequence F5(B, T, 1, H, W), where T is time. This sequence is then normalized, and spatiotemporal features are extracted using three layers of 3D convolution and 3D max pooling, outputting spatiotemporal features F6(B, 128, T, H / 8, W / 8). A ConvLSTM network is then used for temporal processing, outputting temporal features F7(B, 128, T, H / 8, W / 8). The temporal derivative is then calculated, outputting the temporal derivative F8(B, 128, T-1, H / 8, W / 8). Spatial computation is performed using a Laplacian kernel, outputting the spatial Laplacian result F9(B, 128, T, H / 8, W / 8). The first T-1 frames are truncated from the spatial Laplacian result F9 to obtain the aligned spatial temperature distribution features F. 10 (B, 128, T-1, H / 8, W / 8);

[0024] After aligning the time dimension according to the following formula, physical constraints are calculated, and the temporal and spatial features of the ConvLSTM network are added to output an intermediate feature map F. 11 (B, 128, T-1, H / 8, W / 8):

[0025] ;

[0026] In the formula, α is the thermal diffusivity coefficient and β is the constraint strength coefficient;

[0027] For intermediate feature map F 11 Average pooling is performed along the time dimension to output the second feature map Fir(B, 128, H / 8, W / 8);

[0028] The ultraviolet branch outputs a single-channel ultraviolet image sequence F. 12 The image (B, T, 1, H, W) is first normalized, and then converted to the frequency domain using a Fast Fourier Transform to obtain the first frequency domain feature F. 13(B, T, 2, H, W), frequency domain features are extracted through a two-dimensional convolutional layer to obtain the downsampled frequency domain features F. 14 (B, 128, T, H / 8, W / 8); Calculate the potential gradient. By applying the constraints of Maxwell's equations, the second frequency domain feature F is obtained. 15 (B, 128, T, H / 8, W / 8):

[0029] ;

[0030] In the formula, W grad Learnable convolutional kernels are used to approximate the frequency domain gradient.

[0031] Similarly, based on the downsampling frequency domain characteristic F 14 The electromagnetic confinement characteristic F is obtained. 16 (B, 128, T, H / 8, W / 8):

[0032] ;

[0033] In the formula, ϵ is the spatial distribution diagram of dielectric constant, and γ is the learnable constraint strength coefficient; The kernel is a Laplace convolution.

[0034] Fusion of second frequency domain features F 15 Electromagnetic confinement characteristics F 16 The third feature map, Fuv(B, 128, H / 8, W / 8), is then output.

[0035] The first feature map Fv, the second feature map Fir, and the third feature map Fuv are concatenated to obtain the weighted fused feature. .

[0036] Step (5) specifically includes the following steps:

[0037] (5a) Input weighted fusion features The design environment is adaptively compensated to obtain the filtered enhanced features. :

[0038] ;

[0039]

[0040] In the formula, This is a physical constraint term used to calculate the multiphysics coupling residual under the current environment, and to sense and quantify environmental disturbances; This represents the displacement field of visible light; The temperature change field of the infrared branch; The coefficient of thermal expansion; The vibration discharge coupling coefficient; The discharge density distribution; The electric field intensity distribution in three-dimensional space; The distribution of magnetic induction intensity; The compensation strength coefficient; The learnable environment filtering convolution kernel; ReLU is the activation function;

[0041] (5b) Based on the enhanced features after filtering Calculate the output failure probability matrix :

[0042] ;

[0043] In the formula, It is a 1x1 convolutional kernel used to... Mapped to the fault category space; Let be an arbitrarily small positive number.

[0044] Another object of the present invention is to provide an electronic device comprising:

[0045] Processor; and

[0046] A memory storing computer program instructions, which, when executed by the processor, cause the processor to perform the multispectral collaborative sensing and adaptive diagnostic method for ultra-high voltage valve halls as described above.

[0047] The present invention also provides a computer-readable storage medium having stored thereon computer program instructions, which, when executed by a processor, cause the processor to perform the multispectral collaborative sensing and adaptive diagnostic method for ultra-high pressure valve halls as described above.

[0048] As can be seen from the above technical solution, the beneficial effects of the present invention are as follows: First, the present invention breaks through the bottleneck of monitoring a single physical quantity by using the coordinated sensing of visible light, infrared and ultraviolet three bands, and can effectively identify composite defects such as valve tower discharge and overheating of copper-aluminum dissimilar metal contact; Second, it constructs a physical-driven real-time diagnostic system, which overcomes the pain points of poor timeliness and insufficient quantification of manual inspection. At the same time, based on neural networks and coupled with physical knowledge, through multi-parameter correlation analysis of mechanical deformation, temperature change and discharge intensity, it can effectively distinguish defects from interference signals, eliminate spatial monitoring blind spots and enhance anti-interference capabilities. Attached Figure Description

[0049] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0050] like Figure 1 As shown, a multispectral collaborative sensing and adaptive diagnostic method for ultra-high voltage valve halls includes the following sequential steps:

[0051] (1) A gimbal camera equipped with a high-pixel, zoom three-light camera is used. For the range of valve hall and valve tower, the gimbal camera is planned to set up the sampling points and paths. Then the gimbal camera collects images according to the preset sampling points and paths.

[0052] (2) Acquire high-resolution three-light images taken at preset sampling points and form a dataset. Divide the dataset into training set, validation set and test set and perform preprocessing.

[0053] (3) Construct a multispectral collaborative sensing model, process the images output from the three branches, and obtain weighted fusion features. ;

[0054] (4) Input the preprocessed training set into the multispectral collaborative sensing model for training to obtain the trained multispectral collaborative sensing model;

[0055] (5) Input the parameters of the sampling points of the valve tower to be tested and the high-resolution three-light image to be tested into the trained multispectral collaborative sensing model, and perform weighted fusion based on the features. The failure status of the valve tower, i.e., the failure probability matrix, is calculated. .

[0056] In step (1), the gimbal camera acquires visible light, infrared, and ultraviolet image information; the sampling points are set using a global sampling method, with 300 sampling points set according to a preset position; the parameters of the sampling points include valve tower number, valve tower layer number, camera position, height, orientation angle, and zoom magnification.

[0057] Step (2) specifically includes the following steps:

[0058] (2a) Collect images at each preset sampling point;

[0059] (2b) Label the effective regions in the acquired high-resolution tri-light images, and the labeled images form a dataset;

[0060] (2c) Divide the dataset into training set, validation set and test set in units of sampling points; perform data augmentation operations such as rotation, flipping, scaling and mosaic on the training set; and perform normalization on the validation set and test set.

[0061] Step (3) specifically includes the following steps:

[0062] (3a) Improve the YOLOv5 model by replacing the backbone network of the YOLOv5 model with a parallel ResNet-50 network, ConvLSTM network and FFT-CNN fusion network, and replacing the feature extraction network of the YOLOv5 model with a physical law feature extraction module; the ResNet-50 network processes the visible light image to form the visible light branch; the ConvLSTM network processes the infrared image to form the infrared branch; and the FFT-CNN fusion network processes the ultraviolet image to form the ultraviolet branch.

[0063] (3b) The physical law feature extraction module processes the images output from the three branches, specifically as follows:

[0064] The visible light branch outputs a visible light image F1(B, 3, H, W), where B is the batch size, 3 is the number of channels, H is the image height, and W is the image width. After normalization and standardization, F1 is converted into a grayscale image F2(B, 1, H, W). Then, through a convolutional layer and progressive downsampling pooling, a grayscale feature map F3(B, 1, H / 8, W / 8) is obtained. Laplacian kernel is used for Laplacian calculation to obtain a transformed feature map F4(B, 64, H / 8, W / 8). Subsequently, elastic module constraints are applied, and a convolution operation, W*F4, is added to extract features. The elastic module constraints, W*F4, are then added to the transformed feature map F4, and the ReLU activation function is used to output the first feature map Fv.

[0065] The infrared branch outputs a single-channel temporal infrared image sequence F5(B, T, 1, H, W), where T is time. This sequence is then normalized, and spatiotemporal features are extracted using three layers of 3D convolution and 3D max pooling, outputting spatiotemporal features F6(B, 128, T, H / 8, W / 8). A ConvLSTM network is then used for temporal processing, outputting temporal features F7(B, 128, T, H / 8, W / 8). The temporal derivative is then calculated, outputting the temporal derivative F8(B, 128, T-1, H / 8, W / 8). Spatial computation is performed using a Laplacian kernel, outputting the spatial Laplacian result F9(B, 128, T, H / 8, W / 8). The first T-1 frames are truncated from the spatial Laplacian result F9 to obtain the aligned spatial temperature distribution features F. 10 (B, 128, T-1, H / 8, W / 8);

[0066] After aligning the time dimension according to the following formula, physical constraints are calculated, and the temporal and spatial features of the ConvLSTM network are added to output an intermediate feature map F. 11 (B, 128, T-1, H / 8, W / 8):

[0067] ;

[0068] In the formula, α is the thermal diffusivity coefficient and β is the constraint strength coefficient;

[0069] For intermediate feature map F 11 Average pooling is performed along the time dimension to output the second feature map Fir(B, 128, H / 8, W / 8);

[0070] The ultraviolet branch outputs a single-channel ultraviolet image sequence F. 12 The image (B, T, 1, H, W) is first normalized, and then converted to the frequency domain using a Fast Fourier Transform to obtain the first frequency domain feature F. 13 (B, T, 2, H, W), frequency domain features are extracted through a two-dimensional convolutional layer to obtain the downsampled frequency domain features F. 14 (B, 128, T, H / 8, W / 8); Calculate the potential gradient. By applying the constraints of Maxwell's equations, the second frequency domain feature F is obtained. 15 (B, 128, T, H / 8, W / 8):

[0071] ;

[0072] In the formula, W grad Learnable convolutional kernels are used to approximate the frequency domain gradient.

[0073] Similarly, based on the downsampling frequency domain characteristic F 14 The electromagnetic confinement characteristic F is obtained. 16 (B, 128, T, H / 8, W / 8):

[0074] ;

[0075] In the formula, ϵ is the spatial distribution diagram of dielectric constant, and γ is the learnable constraint strength coefficient; The kernel is a Laplace convolution.

[0076] Fusion of second frequency domain features F 15 Electromagnetic confinement characteristics F 16 The third feature map, Fuv(B, 128, H / 8, W / 8), is then output.

[0077] The first feature map Fv, the second feature map Fir, and the third feature map Fuv are concatenated to obtain the weighted fused feature. .

[0078] Step (5) specifically includes the following steps:

[0079] (5a) Input weighted fusion features The design environment is adaptively compensated to obtain the filtered enhanced features. :

[0080] ;

[0081]

[0082] In the formula, This is a physical constraint term used to calculate the multiphysics coupling residual under the current environment, and to sense and quantify environmental disturbances; This represents the displacement field of visible light; The temperature change field of the infrared branch; The coefficient of thermal expansion; The vibration discharge coupling coefficient; The discharge density distribution; The electric field intensity distribution in three-dimensional space; The distribution of magnetic induction intensity; The compensation strength coefficient; The learnable environment filtering convolution kernel; ReLU is the activation function;

[0083] (5b) Based on the enhanced features after filtering Calculate the output failure probability matrix :

[0084] ;

[0085] In the formula, It is a 1x1 convolutional kernel used to... Mapped to the fault category space; Let be an arbitrarily small positive number.

[0086] In summary, this invention overcomes the bottleneck of monitoring single physical quantities by using the coordinated sensing of visible light, infrared, and ultraviolet bands. It can effectively identify complex defects such as valve tower discharge and overheating of dissimilar copper-aluminum metal contact. It constructs a physical-driven real-time diagnostic system, overcoming the pain points of poor timeliness and insufficient quantification of manual inspection. At the same time, based on neural networks and coupled with physical knowledge, it can effectively distinguish defects from interference signals through multi-parameter correlation analysis of mechanical deformation, temperature change, and discharge intensity, eliminating spatial monitoring blind spots and enhancing anti-interference capabilities.

[0087] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A multispectral collaborative sensing and adaptive diagnostic method for ultra-high voltage valve halls, characterized in that: The method includes the following steps in sequence: (1) A gimbal camera equipped with a high-pixel, zoom three-light camera is used. For the range of valve hall and valve tower, the gimbal camera is planned to set up the sampling points and paths. Then the gimbal camera collects images according to the preset sampling points and paths. (2) Acquire high-resolution three-light images taken at preset sampling points and form a dataset. Divide the dataset into training set, validation set and test set and perform preprocessing. (3) Construct a multispectral collaborative sensing model, process the images output from the three branches, and obtain weighted fusion features. ; (4) Input the preprocessed training set into the multispectral collaborative sensing model for training to obtain the trained multispectral collaborative sensing model; (5) Input the parameters of the sampling points of the valve tower to be tested and the high-resolution three-light image to be tested into the trained multispectral collaborative sensing model, and perform weighted fusion based on the features. The failure status of the valve tower, i.e., the failure probability matrix, is calculated. ; Step (3) specifically includes the following steps: (3a) Improve the YOLOv5 model by replacing the backbone network of the YOLOv5 model with a parallel ResNet-50 network, ConvLSTM network and FFT-CNN fusion network, and replacing the feature extraction network of the YOLOv5 model with a physical law feature extraction module; the ResNet-50 network processes the visible light image to form the visible light branch; the ConvLSTM network processes the infrared image to form the infrared branch; and the FFT-CNN fusion network processes the ultraviolet image to form the ultraviolet branch. (3b) The physical law feature extraction module processes the images output from the three branches, specifically as follows: The visible light branch outputs a visible light image F1(B, 3, H, W), where B is the batch size, 3 is the number of channels, H is the image height, and W is the image width. After normalization and standardization, F1 is converted into a grayscale image F2(B, 1, H, W). Then, through a convolutional layer and progressive downsampling pooling, a grayscale feature map F3(B, 1, H / 8, W / 8) is obtained. Laplacian kernel is used for Laplacian calculation to obtain a transformed feature map F4(B, 64, H / 8, W / 8). Subsequently, elastic module constraints are applied, and a convolution operation, W*F4, is added to extract features. The elastic module constraints, W*F4, are then added to the transformed feature map F4, and the ReLU activation function is used to output the first feature map Fv. The infrared branch outputs a single-channel temporal infrared image sequence F5(B, T, 1, H, W), where T is time. This sequence is then normalized, and spatiotemporal features are extracted using three layers of 3D convolution and 3D max pooling, outputting spatiotemporal features F6(B, 128, T, H / 8, W / 8). A ConvLSTM network is then used for temporal processing, outputting temporal features F7(B, 128, T, H / 8, W / 8). The temporal derivative is then calculated, outputting temporal derivative F8(B, 128, T-1, H / 8, W / 8). Spatial computation is performed using a Laplacian kernel, outputting the spatial Laplacian result F9(B, 128, T, H / 8, W / 8). The first T-1 frames are truncated from the spatial Laplacian result F9 to obtain the aligned spatial temperature distribution features F. 10 (B, 128, T-1, H / 8, W / 8); After aligning the time dimension according to the following formula, physical constraints are calculated, and the temporal and spatial features of the ConvLSTM network are added to output an intermediate feature map F. 11 (B, 128, T-1, H / 8, W / 8): ； In the formula, α is the thermal diffusivity coefficient and β is the constraint strength coefficient; For intermediate feature map F 11 Average pooling is performed along the time dimension to output the second feature map Fir(B, 128, H / 8, W / 8); The ultraviolet branch outputs a single-channel ultraviolet image sequence F. 12 The image (B, T, 1, H, W) is first normalized, and then converted to the frequency domain using a Fast Fourier Transform to obtain the first frequency domain feature F. 13 (B, T, 2, H, W), frequency domain features are extracted through a two-dimensional convolutional layer to obtain the downsampled frequency domain features F. 14 (B, 128, T, H / 8, W / 8); Calculate the potential gradient. By applying the constraints of Maxwell's equations, the second frequency domain feature F is obtained. 15 (B, 128, T, H / 8, W / 8): ； In the formula, W grad Learnable convolutional kernels are used to approximate the frequency domain gradient. Similarly, based on the downsampling frequency domain characteristic F 14 The electromagnetic confinement characteristic F is obtained. 16 (B, 128, T, H / 8, W / 8): ； In the formula, ϵ is the spatial distribution diagram of dielectric constant, and γ is the learnable constraint strength coefficient; The kernel is a Laplace convolution. Fusion of second frequency domain features F 15 Electromagnetic confinement characteristics F 16 The third feature map, Fuv(B, 128, H / 8, W / 8), is then output. The first feature map Fv, the second feature map Fir, and the third feature map Fuv are concatenated to obtain the weighted fused feature. .

2. The multispectral collaborative sensing and adaptive diagnosis method for ultra-high voltage valve halls according to claim 1, characterized in that: In step (1), the gimbal camera acquires visible light, infrared, and ultraviolet image information; the sampling points are set using a global sampling method, with 300 sampling points set according to a preset position; the parameters of the sampling points include valve tower number, valve tower layer number, camera position, height, orientation angle, and zoom magnification.

3. The multispectral collaborative sensing and adaptive diagnosis method for ultra-high voltage valve halls according to claim 1, characterized in that: Step (2) specifically includes the following steps: (2a) Collect images at each preset sampling point; (2b) Label the effective regions in the acquired high-resolution tri-light images, and the labeled images form a dataset; (2c) Divide the dataset into training set, validation set and test set in units of sampling points; perform data augmentation operations such as rotation, flipping, scaling and mosaic on the training set; and perform normalization on the validation set and test set.

4. The multispectral collaborative sensing and adaptive diagnosis method for ultra-high voltage valve halls according to claim 1, characterized in that: Step (5) specifically includes the following steps: (5a) Input weighted fusion features The design environment is adaptively compensated to obtain the filtered enhanced features. : ； ； In the formula, This is a physical constraint term used to calculate the multiphysics coupling residual under the current environment, and to sense and quantify environmental disturbances; This represents the displacement field of visible light; The temperature change field of the infrared branch; The coefficient of thermal expansion; The vibration discharge coupling coefficient; This represents the discharge density distribution. The electric field intensity distribution in three-dimensional space; The distribution of magnetic induction intensity; The compensation strength coefficient; The learnable environment filtering convolution kernel; ReLU is the activation function; (5b) Based on the enhanced features after filtering Calculate the output failure probability matrix : ； In the formula, It is a 1x1 convolutional kernel used to... Mapped to the fault category space; Let be an arbitrarily small positive number.

5. An electronic device, comprising: processor; as well as A memory storing computer program instructions that, when executed by the processor, cause the processor to perform the multispectral collaborative sensing and adaptive diagnostic method for ultra-high voltage valve halls as described in any one of claims 1-4.

6. A computer-readable storage medium having stored thereon computer program instructions, which, when executed by a processor, cause the processor to perform the multispectral collaborative sensing and adaptive diagnostic method for ultra-high voltage valve halls as described in any one of claims 1-4.

Images