A temperature measurement method and system for substation equipment based on visible light guidance

Abstract

The present invention relates to the field of pattern recognition technology, and specifically to a temperature measurement method and system for substation equipment based on visible light guidance. First, visible light images, infrared thermal images and vibration spectrum data are collected synchronously, and a dynamic compensation coefficient matrix is ​​constructed through light intensity analysis, which is applied to the infrared thermal image to achieve light interference compensation. Then, the corrected infrared thermal image and visible light image are input into a pre-trained heterogeneous graph matching model to identify the target equipment category and its three-dimensional coordinates. Based on the equipment category information, the standard temperature field template in the digital twin model library is retrieved, and a heat conduction path model is established in combination with the vibration spectrum data to generate a temperature field prediction reference map for the equipment. Finally, through pixel-by-pixel deviation analysis, the temperature deviation value, image features and vibration data are input into a probabilistic graphical model to output equipment fault information. This method can achieve high-precision temperature measurement and fault diagnosis.

Description

Technical Field

[0001] The present invention relates to the field of pattern recognition, and in particular to a temperature measurement method and system for substation equipment based on visible light guidance. Background Art

[0002] With the development of smart grids and digital substations, real-time monitoring and intelligent diagnosis of equipment operating status have become crucial for ensuring the safe and stable operation of power systems. Over the long term, substation equipment often experiences localized heating due to factors such as aging, poor contact, and overload, which can lead to equipment failures and even accidents. Therefore, accurate and timely temperature monitoring and fault warning of key substation equipment are crucial.

[0003] In existing technologies, infrared thermal imaging technology is widely used for non-contact temperature measurement of substation equipment. However, due to issues such as ambient light interference, insufficient target recognition accuracy, and low infrared image resolution, it often suffers from large temperature measurement errors and high false alarm rates. Especially under complex background conditions, it is difficult to accurately identify equipment types and perform precise temperature measurement relying solely on infrared images. In recent years, multimodal perception methods that integrate visible light and infrared information have gradually gained attention. By using visible light images to assist infrared image correction, positioning, and recognition, the accuracy and robustness of temperature measurement can be improved to a certain extent. However, existing solutions are mostly limited to image-level registration and temperature threshold judgment, and lack in-depth modeling of equipment operating conditions and state prediction mechanisms.

[0004] To this end, a temperature measurement method and system for substation equipment based on visible light guidance are proposed. Summary of the Invention

[0005] This invention provides a visible light-guided temperature measurement method and system for substation equipment. By synchronously collecting visible light images, infrared thermal images, and vibration spectrum data, the visible light images are illuminated and the infrared thermal images are corrected. A heterogeneous pattern matching model is used to identify the device type and its three-dimensional coordinates, and a temperature field prediction benchmark is generated in conjunction with a digital twin model. The system then compares the deviation between real-time and predicted temperatures and, in combination with vibration data, uses a probabilistic graphical model to output device fault information.

[0006] To achieve the above objectives, the present invention provides the following technical solutions.

[0007] A method for measuring temperature of substation equipment based on visible light guidance, comprising:

[0008] Synchronously collect multimodal data of target substation equipment, including visible light images, infrared thermal images, and vibration spectrum data;

[0009] Performing light intensity distribution analysis on the visible light image, constructing a dynamic compensation coefficient matrix based on the analysis results, and applying the compensation coefficient matrix to the infrared thermal image to generate a corrected infrared thermal image;

[0010] Input the corrected infrared thermal image and visible light image into the pre-trained heterogeneous image matching model, and output the category information of the target device and the 3D spatial coordinates of the target device in the image space;

[0011] Based on the target device's category information, the corresponding standard temperature field template is retrieved from the digital twin model library. Combined with the vibration spectrum data, a heat conduction path model is established to generate a temperature field prediction benchmark map for the target device in its current state.

[0012] Perform pixel-by-pixel deviation analysis on the corrected infrared thermal image and the temperature field prediction reference image to calculate the temperature deviation between the real-time temperature field and the predicted temperature field;

[0013] The temperature deviation value, image feature information, and vibration spectrum data are input into the probabilistic graphical model as observation variables, and the fault information of the target equipment is output.

[0014] Furthermore, the step of generating the corrected infrared thermal image includes:

[0015] Performing grayscale transformation and illumination mean filtering on the visible light image to obtain an image illumination intensity distribution map;

[0016] Constructing a dynamic compensation coefficient matrix based on the light intensity distribution map, wherein the compensation coefficient of each pixel is determined according to the deviation between the light brightness at the corresponding position of the pixel and the set reference brightness;

[0017] The dynamic compensation coefficient matrix is ​​applied to the original temperature pixel values ​​of the infrared thermal image, and light interference compensation is performed through pixel-level dot multiplication to obtain the corrected infrared thermal image.

[0018] Furthermore, the heterogeneous graph matching model includes:

[0019] The device recognition unit is used to combine the visible light image and the rectified infrared thermal image to identify the category information of the target device through a convolutional neural classification model;

[0020] Image feature extraction unit, used to extract key points and their corresponding descriptors from visible light images and rectified infrared thermal images through the SuperPoint model;

[0021] Feature matching unit, used to perform feature matching on key point descriptors using Euclidean distance and eliminate false matches through cross-validation;

[0022] The spatial geometry estimation unit is used to output the three-dimensional spatial coordinates of the target device in the image space using a geometric model fitting method based on a random sampling consistency algorithm.

[0023] Furthermore, the step of generating the temperature field prediction reference map includes:

[0024] According to the category information of the target device, a standard temperature field template matching the structure, material and operating parameters of the target device is retrieved from the digital twin model library;

[0025] Extracting the operating state parameters of the target device using the vibration spectrum data, including frequency peak, spectrum width, and spectral line energy distribution, and inputting the operating state parameters into the device heat conduction simulation module to calculate the temperature conduction path of the device in the current operating state using the finite element method;

[0026] The temperature field is reconstructed and mapped based on the temperature conduction path and the standard temperature field template to generate a temperature field prediction reference map that is suitable for the current working state.

[0027] Furthermore, the step of outputting current device fault information includes:

[0028] The temperature deviation value, image feature information and vibration spectrum data are used as observation variables to construct an observation vector and input it into a predefined probabilistic graphical model to perform joint probabilistic inference on the health status of the target device;

[0029] The maximum a posteriori estimation method is used to solve the occurrence probability of various typical failure modes and output the probability distribution vector of the target equipment under various failure types;

[0030] Fault information including a fault risk level and a most likely fault type is generated according to the probability distribution vector.

[0031] The present invention also provides a substation equipment temperature measurement system based on visible light guidance, comprising:

[0032] The data acquisition module is used to synchronously collect multimodal data of the target substation equipment, including visible light images, infrared thermal images, and vibration spectrum data;

[0033] The infrared thermal image correction module is used to analyze the light intensity distribution of the visible light image, construct a dynamic compensation coefficient matrix based on the analysis results, and apply the compensation coefficient matrix to the infrared thermal image to generate a corrected infrared thermal image;

[0034] An information acquisition module is used to input the corrected infrared thermal image and visible light image into a pre-trained heterogeneous image matching model, and output the category information of the target device and the three-dimensional spatial coordinates of the target device in the image space;

[0035] The simulation model building module is used to retrieve the corresponding standard temperature field template from the digital twin model library based on the category information of the target device, and combine it with the vibration spectrum data to establish a heat conduction path model to generate a temperature field prediction benchmark map for the target device in its current state;

[0036] The temperature difference analysis module is used to perform pixel-by-pixel deviation analysis on the corrected infrared thermal image and the temperature field prediction reference image, and calculate the temperature deviation value between the real-time temperature field and the predicted temperature field;

[0037] The faulty equipment acquisition module is used to input the temperature deviation value, image feature information, and vibration spectrum data as observation variables into the probabilistic graphical model and output the fault information of the target equipment.

[0038] Furthermore, the step of generating the corrected infrared thermal image includes:

[0039] Performing grayscale transformation and illumination mean filtering on the visible light image to obtain an image illumination intensity distribution map;

[0040] Constructing a dynamic compensation coefficient matrix based on the light intensity distribution map, wherein the compensation coefficient of each pixel is determined according to the deviation between the light brightness at the corresponding position of the pixel and the set reference brightness;

[0041] The dynamic compensation coefficient matrix is ​​applied to the original temperature pixel values ​​of the infrared thermal image, and light interference compensation is performed through pixel-level dot multiplication to obtain the corrected infrared thermal image.

[0042] Furthermore, the heterogeneous graph matching model includes:

[0043] The device recognition unit is used to combine the visible light image and the rectified infrared thermal image to identify the category information of the target device through a convolutional neural classification model;

[0044] Image feature extraction unit, used to extract key points and their corresponding descriptors from visible light images and rectified infrared thermal images through the SuperPoint model;

[0045] Feature matching unit, used to perform feature matching on key point descriptors using Euclidean distance and eliminate false matches through cross-validation;

[0046] The spatial geometry estimation unit is used to output the three-dimensional spatial coordinates of the target device in the image space using a geometric model fitting method based on a random sampling consistency algorithm.

[0047] Furthermore, the step of generating the temperature field prediction reference map includes:

[0048] According to the category information of the target device, a standard temperature field template matching the structure, material and operating parameters of the target device is retrieved from the digital twin model library;

[0049] Extracting the operating state parameters of the target device using the vibration spectrum data, including frequency peak, spectrum width, and spectral line energy distribution, and inputting the operating state parameters into the device heat conduction simulation module to calculate the temperature conduction path of the device in the current operating state using the finite element method;

[0050] The temperature field is reconstructed and mapped based on the temperature conduction path and the standard temperature field template to generate a temperature field prediction reference map that is suitable for the current working state.

[0051] Furthermore, the step of outputting current device fault information includes:

[0052] The temperature deviation value, image feature information and vibration spectrum data are used as observation variables to construct an observation vector and input it into a predefined probabilistic graphical model to perform joint probabilistic inference on the health status of the target device;

[0053] The maximum a posteriori estimation method is used to solve the occurrence probability of various typical failure modes and output the probability distribution vector of the target equipment under various failure types;

[0054] Fault information including a fault risk level and a most likely fault type is generated according to the probability distribution vector.

[0055] The beneficial effects of the present invention are:

[0056] 1. By analyzing the light intensity of visible light images, constructing a dynamic compensation coefficient matrix and applying it to infrared thermal images, the interference of ambient light changes on infrared imaging results is effectively reduced, the accuracy and consistency of temperature images are improved, and more reliable basic data is provided for subsequent image fusion and equipment status assessment.

[0057] 2. By combining the target equipment's category information with vibration spectrum data, a personalized heat conduction path model is constructed, and a prediction benchmark map is generated based on the standard temperature field template in the digital twin model library. This not only improves the accuracy and pertinence of temperature distribution prediction, but also enhances the system's dynamic perception of the equipment's operating status, providing a scientific basis for fault diagnosis and temperature anomaly detection.

[0058] 3. Inputting temperature deviation values, image feature information, and vibration spectrum data as multi-source observation variables into the probabilistic graphical model can realize joint probabilistic reasoning and analysis of equipment fault status based on the fusion of multimodal perception information, effectively improving the accuracy and robustness of fault identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0060] Figure 1 This is a flow chart of a method for measuring temperature of substation equipment based on visible light guidance provided by the present invention;

[0061] Figure 2 This is a structural diagram of a temperature measurement system for substation equipment based on visible light guidance provided by the present invention. DETAILED DESCRIPTION

[0062] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0063] Example 1

[0064] A temperature measurement method for substation equipment based on visible light guidance, such as Figure 1 As shown, including:

[0065] S100: Synchronously collect multimodal data of target substation equipment, including visible light images, infrared thermal images, and vibration spectrum data;

[0066] Specifically, within the target substation area, high-resolution industrial cameras or smart video cameras are used to obtain equipment appearance images and visible light information, and infrared thermal imagers are used to capture thermal images of the equipment surface temperature distribution in real time. A three-axis MEMS accelerometer is used to collect vibration spectrum data during equipment operation. To achieve time consistency between data sources, a GPS time synchronization module is used to synchronize the clocks of each acquisition module to ensure that the collected visible light images, thermal images, and vibration data match in the time dimension, which is conducive to subsequent fusion processing.

[0067] S200: Analyze the light intensity distribution of the visible light image, construct a dynamic compensation coefficient matrix based on the analysis results, and apply the compensation coefficient matrix to the infrared thermal image to generate a corrected infrared thermal image;

[0068] Furthermore, the step of generating the corrected infrared thermal image includes:

[0069] Performing grayscale transformation and illumination mean filtering on the visible light image to obtain an image illumination intensity distribution map;

[0070] Constructing a dynamic compensation coefficient matrix based on the light intensity distribution map, wherein the compensation coefficient of each pixel is determined according to the deviation between the light brightness at its corresponding position and a set reference brightness;

[0071] The dynamic compensation coefficient matrix is ​​applied to the original temperature pixel values ​​of the infrared thermal image, and light interference compensation is performed through pixel-level dot multiplication to obtain the corrected infrared thermal image.

[0072] Specifically, a visible light image corresponding to the infrared thermal image space is obtained and grayscale processed to remove color information interference. Preferably, a standard grayscale conversion algorithm (such as weighted average method) is used to obtain a grayscale image. , for grayscale images Apply Gaussian blur filter to obtain the image light intensity distribution map , which is used to reflect the brightness of each position in the image; for each pixel in the visible light image, according to its local illumination value With reference brightness The deviation between them is used to calculate the dynamic compensation coefficient matrix , where the reference brightness is the overall mean of the image, and the calculation formula of the dynamic compensation coefficient matrix is:

[0073] ;

[0074] in, Represents the pixel coordinates, Indicates the illumination compensation adjustment factor, which is generally in the range of [0.2, 1.0]. In this embodiment, the value is 0.75; for each temperature pixel value in the infrared thermal image , perform pixel-by-pixel multiplication with the corresponding elements of the compensation coefficient matrix to obtain the corrected temperature value:

[0075] ;

[0076] Among them, each temperature pixel value in the infrared thermal image is corrected, and the generated image is the corrected infrared thermal image, which is used to reflect the actual temperature distribution state after eliminating the influence of visible light interference.

[0077] By analyzing the illumination intensity distribution of visible light images, constructing a dynamic compensation coefficient matrix and applying it to infrared thermal images, this method effectively reduces the impact of ambient lighting changes on infrared temperature measurement accuracy. Compared to traditional static calibration or single filtering methods, this method enables pixel-by-pixel adaptive correction, improving the accuracy and spatial consistency of temperature representation in infrared images, and providing more stable and reliable basic data support for subsequent device identification and temperature field analysis.

[0078] S300: Inputting the corrected infrared thermal image and visible light image into a pre-trained heterogeneous image matching model, outputting category information of the target device and the three-dimensional spatial coordinates of the target device in the image space;

[0079] Furthermore, the heterogeneous graph matching model includes:

[0080] The device recognition unit is used to combine the visible light image and the rectified infrared thermal image to identify the category information of the target device through a convolutional neural classification model;

[0081] Image feature extraction unit, used to extract key points and their corresponding descriptors from visible light images and rectified infrared thermal images through the SuperPoint model;

[0082] Feature matching unit, used to perform feature matching on key point descriptors using Euclidean distance and eliminate false matches through cross-validation;

[0083] The spatial geometry estimation unit is used to output the three-dimensional spatial coordinates of the target device in the image space using a geometric model fitting method based on a random sampling consistency algorithm.

[0084] Specifically, the corrected infrared thermal image and the corresponding visible light image are used as input, and the category information of the target device is identified through a convolutional neural classification model. The convolutional neural classification model is not limited here. It can be a single-channel fusion network that fuses the infrared thermal image and the visible light image, or it can be a dual-channel model that inputs the infrared thermal image and the visible light image into different channels respectively. In this embodiment, a dual-channel Daul-YOLO model is preferably used; the SuperPoint model is used to extract key points and calculate descriptors for the two images respectively. SuperPoint is a lightweight neural network architecture, which includes a shared encoder and two branch decoding heads, which generate key point heat maps and descriptor tensors respectively; based on the Euclidean distance metric, the descriptors extracted by SuperPoint are matched pairwise, and the cross-validation method is used to eliminate mismatched pairs, and only the key point pairs that are the nearest neighbors are retained as matching points to enhance the accuracy of matching; the random sampling consistency algorithm is used to fit the spatial geometric model of the matching point pairs, and the three-dimensional spatial coordinates of the target device in the two-dimensional image space are further calculated according to the image shooting parameters.

[0085] By constructing a heterogeneous image matching model and fusing visible light images with calibrated infrared thermal images, the system not only accurately identifies the device category but also locates the device's three-dimensional coordinates within the image space. The SuperPoint keypoint extraction and matching mechanism improves feature consistency across multimodal images. Combined with the RANSAC algorithm's geometric fitting method, this significantly enhances matching robustness and positioning accuracy. This method effectively improves the automation level of device identification and positioning, as well as the spatial accuracy of temperature measurement tasks, laying a high-quality data foundation for subsequent thermal field analysis and fault diagnosis.

[0086] S400: Based on the category information of the target device, the corresponding standard temperature field template is retrieved from the digital twin model library, and combined with the vibration spectrum data, a heat conduction path model is established to generate a temperature field prediction benchmark map of the target device in its current state;

[0087] Furthermore, the step of generating the temperature field prediction reference map includes:

[0088] According to the category information of the target device, a standard temperature field template that matches the structure, material, and operating parameters of the device is retrieved from the digital twin model library;

[0089] Extracting the operating state parameters of the device using the vibration spectrum data, including frequency peak, spectrum width, and spectral line energy distribution, and inputting the operating state parameters into the device heat conduction simulation module to calculate the temperature conduction path of the device in the current operating state using the finite element method;

[0090] The temperature field is reconstructed and mapped based on the temperature conduction path and the standard temperature field template to generate a temperature field prediction reference map that is suitable for the current working state.

[0091] Specifically, a standard temperature field template that matches the device is retrieved from the pre-built digital twin model library. The template is constructed by the actual device structure, material properties (such as thermal conductivity, specific heat capacity), workload and historical operating condition data; the synchronously collected vibration spectrum data is subjected to frequency domain analysis to extract the key operating status parameters of the current device, including frequency peak, spectrum width and spectral line energy distribution. The above operating status parameters are used as input and passed to the device heat conduction simulation module. The module establishes a heat conduction path model of the device based on the finite element method (FEM), including the thermal coupling relationship of the internal structure, the external heat exchange boundary conditions and the time. The temperature evolution in the inter-dimensional space is analyzed; the heat conduction path map is integrated with the standard temperature field template. In the process of generating the temperature field prediction benchmark map, the temperature field distribution map under the current operating state is first reconstructed through the finite element simulation method based on the heat conduction path model and the standard temperature field template in the digital twin model library to achieve temperature field image reconstruction; then, the spatial projection registration and numerical amplitude matching methods are used to map the reconstructed temperature field map to the infrared image coordinate system, and the temperature amplitude is calibrated and adjusted to achieve spatial and numerical consistency between the prediction map and the infrared image, thereby generating a temperature field prediction benchmark map that matches the actual observation conditions. This prediction benchmark map reflects the theoretical temperature distribution of the equipment under the current operating state, and is used for subsequent deviation analysis with the actual infrared thermal map.

[0092] By incorporating a digital twin model and a heat conduction simulation mechanism, this technical solution dynamically generates a temperature field prediction benchmark map tailored to the actual operating state, combining device classification information and real-time vibration spectrum data. This approach not only overcomes the limitations of traditional static templates, which cannot reflect the current thermal distribution characteristics of the device, but also accurately simulates the temperature conduction path through the finite element method, achieving highly accurate predictions of the device's temperature rise trends. This improves the scientific nature of temperature measurement analysis and the accuracy of fault prediction.

[0093] S500: performing pixel-by-pixel deviation analysis on the corrected infrared thermal image and the temperature field prediction reference image, and calculating the temperature deviation value between the real-time temperature field and the predicted temperature field;

[0094] Specifically, in order to ensure the consistency of the actual infrared thermal image and the predicted temperature field reference image in the spatial dimension, the two images are first accurately aligned, including rigid alignment based on image features (such as edges, corners or key structure contours), and then the affine transformation is applied to resample the corrected infrared image to ensure that the same pixel position corresponds to the same physical point of the device in the two images; after completing the image alignment, the temperature deviation value is calculated using the pixel-by-pixel difference operation method, assuming To predict the temperature of the corresponding pixel in the temperature field reference image, the temperature deviation value of the corresponding pixel is:

[0095] ;

[0096] in, The pixel in the corrected infrared thermal image obtained in step S200 The actual temperature, Indicates the temperature deviation value corresponding to the pixel.

[0097] S600: Input the temperature deviation value, image feature information, and vibration spectrum data as observation variables into the probability graphical model, and output the fault information of the target device.

[0098] Furthermore, the step of outputting current device fault information includes:

[0099] The temperature deviation value, image feature information and vibration spectrum data are used as observation variables to construct an observation vector and input it into a predefined probabilistic graphical model to perform joint probabilistic inference on the health status of the target device;

[0100] The maximum a posteriori estimation method is used to solve the occurrence probability of various typical failure modes and output the probability distribution vector of the current device under various failure types;

[0101] Fault information including a fault risk level and a most likely fault type is generated according to the probability distribution vector.

[0102] Furthermore, the three main observation variables are unified into a standard feature input format, including temperature deviation values, image feature information, and vibration spectrum data. The image feature information includes texture information, edge density, color gradient distribution, and other features from visible light images and rectified infrared thermal images. These features are obtained through the convolutional layer of the convolutional neural network. After standard normalization, the three main observation variables are used as input observation vectors for the next modeling process. A probabilistic graphical model with a Bayesian network as the core is used for fault causal reasoning. The model structure is pre-built based on expert knowledge and historical equipment failure data, and includes observation variable nodes (corresponding to the multimodal features of the input), implicit state nodes (representing equipment operating status parameters such as local overheating and insulation aging), and fault type nodes (outputting the most likely failure mode such as poor contact, overload operation, and abnormal heat dissipation). The model is solved using the maximum a posteriori estimation method to output the probability distribution vector of the equipment under various fault types. The probability distribution vector is converted into a fault risk level and the most likely fault type through post-processing and decision extraction. The fault type is determined by taking the fault label corresponding to the maximum index in the probability distribution vector. This step can be described as follows:

[0103] ;

[0104] in, Indicates the fault type , Indicates the fault type The probability of Represents the maximization function, and the fault risk level is determined by the maximum fault probability value and the threshold value division interval, such as , the risk level is extremely high.

[0105] By fusing temperature deviation values, image feature information, and vibration spectrum data into observation vectors and introducing a probabilistic graphical model for joint probabilistic inference, this approach enables comprehensive assessment of equipment operating status using multi-source information. Calculating the probability distribution of various failure modes using the maximum a posteriori estimation method helps accurately identify the most likely fault type and assess its risk level, providing a scientific basis for operational and maintenance decisions and improving both the accuracy of fault diagnosis and the intelligence of the system.

[0106] Example 2

[0107] The present invention also provides a temperature measurement system for substation equipment based on visible light guidance, such as Figure 2 As shown, including:

[0108] The data acquisition module is used to synchronously collect multimodal data of the target substation equipment, including visible light images, infrared thermal images, and vibration spectrum data;

[0109] The infrared thermal image correction module is used to analyze the light intensity distribution of the visible light image, construct a dynamic compensation coefficient matrix based on the analysis results, and apply the compensation coefficient matrix to the infrared thermal image to generate a corrected infrared thermal image;

[0110] An information acquisition module is used to input the corrected infrared thermal image and visible light image into a pre-trained heterogeneous image matching model, and output the category information of the target device and the three-dimensional spatial coordinates of the target device in the image space;

[0111] The simulation model establishment module is used to retrieve the corresponding standard temperature field template from the digital twin model library based on the category information of the target device, and establish a heat conduction path model in combination with the vibration spectrum data to generate a temperature field prediction benchmark map for the target device in its current state;

[0112] The temperature difference analysis module is used to perform pixel-by-pixel deviation analysis on the corrected infrared thermal image and the temperature field prediction reference image, and calculate the temperature deviation value between the real-time temperature field and the predicted temperature field;

[0113] The faulty equipment acquisition module is used to input the temperature deviation value, image feature information, and vibration spectrum data as observation variables into the probabilistic graphical model and output the fault information of the target equipment.

[0114] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A method for measuring temperature of substation equipment based on visible light guidance, characterized in that: include: Synchronously collect multimodal data of target substation equipment, including visible light images, infrared thermal images, and vibration spectrum data; Performing light intensity distribution analysis on the visible light image, constructing a dynamic compensation coefficient matrix based on the analysis results, and applying the compensation coefficient matrix to the infrared thermal image to generate a corrected infrared thermal image; Input the corrected infrared thermal image and visible light image into the pre-trained heterogeneous image matching model, and output the category information of the target device and the 3D spatial coordinates of the target device in the image space; Based on the target device's category information, the corresponding standard temperature field template is retrieved from the digital twin model library. Combined with the vibration spectrum data, a heat conduction path model is established to generate a temperature field prediction benchmark map for the target device in its current state. Perform pixel-by-pixel deviation analysis on the corrected infrared thermal image and the temperature field prediction reference image to calculate the temperature deviation between the real-time temperature field and the predicted temperature field; The temperature deviation value, image feature information, and vibration spectrum data are input into the probabilistic graphical model as observation variables, and the fault information of the target equipment is output.

2. The method for measuring temperature of substation equipment based on visible light guidance according to claim 1, characterized in that: The step of generating the corrected infrared thermal image comprises: Performing grayscale transformation and illumination mean filtering on the visible light image to obtain an image illumination intensity distribution map; Constructing a dynamic compensation coefficient matrix based on the light intensity distribution map, wherein the compensation coefficient of each pixel is determined according to the deviation between the light brightness at the corresponding position of the pixel and the set reference brightness; The dynamic compensation coefficient matrix is ​​applied to the original temperature pixel values ​​of the infrared thermal image, and light interference compensation is performed through pixel-level dot multiplication to obtain the corrected infrared thermal image.

3. The method for measuring temperature of substation equipment based on visible light guidance according to claim 1, characterized in that: The heterogeneous graph matching model includes: The device recognition unit is used to combine the visible light image and the rectified infrared thermal image to identify the category information of the target device through a convolutional neural classification model; Image feature extraction unit, used to extract key points and their corresponding descriptors from visible light images and rectified infrared thermal images through the SuperPoint model; Feature matching unit, used to perform feature matching on key point descriptors using Euclidean distance and eliminate false matches through cross-validation; The spatial geometry estimation unit is used to output the three-dimensional spatial coordinates of the target device in the image space using a geometric model fitting method based on a random sampling consistency algorithm.

4. The method for measuring temperature of substation equipment based on visible light guidance according to claim 1, characterized in that: The step of generating the temperature field prediction reference map includes: According to the category information of the target device, a standard temperature field template matching the structure, material and operating parameters of the target device is retrieved from the digital twin model library; Extracting the operating state parameters of the target device using the vibration spectrum data, including frequency peak, spectrum width, and spectral line energy distribution, and inputting the operating state parameters into the device heat conduction simulation module to calculate the temperature conduction path of the device in the current operating state using the finite element method; The temperature field is reconstructed and mapped based on the temperature conduction path and the standard temperature field template to generate a temperature field prediction reference map that is suitable for the current working state.

5. The method for measuring temperature of substation equipment based on visible light guidance according to claim 1, characterized in that: The steps for outputting current device fault information include: The temperature deviation value, image feature information and vibration spectrum data are used as observation variables to construct an observation vector and input it into a predefined probabilistic graphical model to perform joint probabilistic inference on the health status of the target device; The maximum a posteriori estimation method is used to solve the occurrence probability of various typical failure modes and output the probability distribution vector of the target equipment under various failure types; Fault information including a fault risk level and a most likely fault type is generated according to the probability distribution vector.

6. A temperature measurement system for substation equipment based on visible light guidance, characterized in that: include: The data acquisition module is used to synchronously collect multimodal data of the target substation equipment, including visible light images, infrared thermal images, and vibration spectrum data; The infrared thermal image correction module is used to analyze the light intensity distribution of the visible light image, construct a dynamic compensation coefficient matrix based on the analysis results, and apply the compensation coefficient matrix to the infrared thermal image to generate a corrected infrared thermal image; An information acquisition module is used to input the corrected infrared thermal image and visible light image into a pre-trained heterogeneous image matching model, and output the category information of the target device and the three-dimensional spatial coordinates of the target device in the image space; The simulation model building module is used to retrieve the corresponding standard temperature field template from the digital twin model library based on the category information of the target device, and combine it with the vibration spectrum data to establish a heat conduction path model to generate a temperature field prediction benchmark map for the target device in its current state; The temperature difference analysis module is used to perform pixel-by-pixel deviation analysis on the corrected infrared thermal image and the temperature field prediction reference image, and calculate the temperature deviation value between the real-time temperature field and the predicted temperature field; The faulty equipment acquisition module is used to input the temperature deviation value, image feature information, and vibration spectrum data as observation variables into the probabilistic graphical model and output the fault information of the target equipment.

7. The substation equipment temperature measurement system based on visible light guidance according to claim 6 is characterized in that: The step of generating the corrected infrared thermal image comprises: Performing grayscale transformation and illumination mean filtering on the visible light image to obtain an image illumination intensity distribution map; Constructing a dynamic compensation coefficient matrix based on the light intensity distribution map, wherein the compensation coefficient of each pixel is determined according to the deviation between the light brightness at the corresponding position of the pixel and the set reference brightness; The dynamic compensation coefficient matrix is ​​applied to the original temperature pixel values ​​of the infrared thermal image, and light interference compensation is performed through pixel-level dot multiplication to obtain the corrected infrared thermal image.

8. The substation equipment temperature measurement system based on visible light guidance according to claim 6 is characterized in that: The heterogeneous graph matching model includes: The device recognition unit is used to combine the visible light image and the rectified infrared thermal image to identify the category information of the target device through a convolutional neural classification model; Image feature extraction unit, used to extract key points and their corresponding descriptors from visible light images and rectified infrared thermal images through the SuperPoint model; Feature matching unit, used to perform feature matching on key point descriptors using Euclidean distance and eliminate false matches through cross-validation; The spatial geometry estimation unit is used to output the three-dimensional spatial coordinates of the target device in the image space using a geometric model fitting method based on a random sampling consistency algorithm.

9. The substation equipment temperature measurement system based on visible light guidance according to claim 6 is characterized in that: The step of generating the temperature field prediction reference map includes: According to the category information of the target device, a standard temperature field template matching the structure, material and operating parameters of the target device is retrieved from the digital twin model library; Extracting the operating state parameters of the target device using the vibration spectrum data, including frequency peak, spectrum width, and spectral line energy distribution, and inputting the operating state parameters into the device heat conduction simulation module to calculate the temperature conduction path of the device in the current operating state using the finite element method; The temperature field is reconstructed and mapped based on the temperature conduction path and the standard temperature field template to generate a temperature field prediction reference map that is suitable for the current working state.

10. The substation equipment temperature measurement system based on visible light guidance according to claim 6, characterized in that: The steps for outputting current device fault information include: The temperature deviation value, image feature information and vibration spectrum data are used as observation variables to construct an observation vector and input it into a predefined probabilistic graphical model to perform joint probabilistic inference on the health status of the target device; The maximum a posteriori estimation method is used to solve the occurrence probability of various typical failure modes and output the probability distribution vector of the target equipment under various failure types; Fault information including a fault risk level and a most likely fault type is generated according to the probability distribution vector.

Images