Intelligent Identification and Location Method and System for Defects of Photovoltaic Modules Based on Multimodal Images

Abstract

The present invention provides a method and system for intelligent identification and localization of photovoltaic module defects based on multimodal images, which relates to the field of photovoltaic detection technology. It includes preprocessing by acquiring visible light images and infrared thermal imaging images, extracting the position coordinates and geometric features of the defect regions and the position coordinates and temperature features of the temperature anomaly regions; performing spatial mapping registration to generate target defect regions; calculating the edge texture change rate and the temperature gradient change rate to construct a defect expansion vector field for predicting the position; determining the performance degradation rate based on the area change rate, calculating the performance influence coefficient to generate a replacement priority score and outputting a diagnostic report. The present invention realizes accurate identification and localization of defects, predicts the development trend of defects, and provides scientific component replacement suggestions, improving the operation and maintenance efficiency and economic benefits of photovoltaic power stations.

Description

Technical Field

[0001] The present invention relates to photovoltaic detection technology, and in particular to an intelligent identification and positioning method and system for photovoltaic module defects based on multimodal images. Background Art

[0002] During the installation and operation of photovoltaic modules, defects such as hot spots and cracks are likely to occur. These defects will cause the performance of the modules to decline and have an adverse impact on adjacent modules. At present, photovoltaic power stations mainly rely on manual inspections and single imaging means for defect detection, making it difficult to accurately identify the types of defects and predict the development trends of defects, which easily leads to component damage and reduced power generation efficiency.

[0003] Traditional defect detection methods usually only focus on the defect status of a single module, ignoring the impact of defective modules on surrounding modules, and lacking a scientific basis for replacement decisions. Existing technologies cannot achieve precise positioning and development prediction of defect areas, nor can they accurately evaluate the replacement priorities of defective modules, resulting in increased maintenance costs and losses in power generation benefits.

[0004] Therefore, there is an urgent need for an intelligent defect identification and positioning method based on multimodal images, which can achieve precise defect identification by fusing visible light and infrared thermal imaging information, and establish a scientific replacement decision-making mechanism by combining defect expansion prediction and component performance evaluation, thereby improving the operation and maintenance efficiency and economic benefits of photovoltaic power stations. Summary of the Invention

[0005] Embodiments of the present invention provide an intelligent identification and positioning method and system for photovoltaic module defects based on multimodal images, which can solve the problems in the prior art.

[0006] In the first aspect of the embodiments of the present invention,

[0007] An intelligent identification and positioning method for photovoltaic module defects based on multimodal images is provided, including:

[0008] Obtain visible light images and infrared thermal imaging images of the photovoltaic module to be detected and perform preprocessing, extract the position coordinates and geometric features of the defect area in the preprocessed visible light image, and the position coordinates and temperature features of the temperature abnormal area in the preprocessed infrared thermal imaging image;

[0009] Perform spatial mapping registration on the position coordinates of the defect area and the position coordinates of the temperature abnormal area to generate a target defect area, and extract the position information and defect feature data of the target defect area;

[0010] Calculate the edge texture change rate and temperature gradient change rate in the defect feature data, determine the expansion direction of the target defect area according to the edge texture change rate, determine the expansion speed of the target defect area according to the temperature gradient change rate, construct a defect expansion vector field based on the expansion direction and expansion speed, and obtain the position prediction result of the target defect area by solving the propagation equation of the defect expansion vector field;

[0011] Calculate the area change rate of the target defect area according to the position prediction result, determine the performance attenuation rate of the photovoltaic module, obtain the string connection relationship and current load distribution data of the photovoltaic module in the power station, calculate the performance influence coefficient of the photovoltaic module on adjacent modules, generate a replacement priority score based on the performance attenuation rate and the performance influence coefficient, and output a diagnostic report including defect identification and positioning and component replacement suggestions according to the replacement priority score.

[0012] In an alternative embodiment,

[0013] Extract the position coordinates and geometric features of the defect area in the preprocessed visible light image, and the position coordinates and temperature features of the temperature anomaly area in the preprocessed infrared thermal imaging image, including:

[0014] Adopt an adaptive double-threshold method to perform edge detection on the preprocessed visible light image, calculate the gradient magnitude data and gradient direction data, calculate the morphological gradient value of the preprocessed visible light image based on the gradient magnitude data and gradient direction data, construct an image segmentation function according to the morphological gradient value, and perform region segmentation to obtain segmented regions;

[0015] Merge adjacent regions of the segmented regions to obtain the defect area in the preprocessed visible light image, calculate the centroid coordinate value and region area value of the defect area, and obtain the position coordinates and geometric features of the defect area;

[0016] Calculate the average temperature value and temperature standard deviation value of the normal area in the preprocessed infrared thermal imaging image, construct a temperature anomaly determination function according to the average temperature value and temperature standard deviation value, identify temperature anomaly points, perform region growing operations with the temperature anomaly points as initial growing points to obtain the temperature anomaly area in the preprocessed infrared thermal imaging image, and calculate the centroid coordinate value and region temperature difference of the temperature anomaly area to obtain the position coordinates and temperature features of the temperature anomaly area.

[0017] In an alternative embodiment,

[0018] Perform spatial mapping registration on the position coordinates of the defect area and the position coordinates of the temperature anomaly area to generate a target defect area, including:

[0019] Calculate the first geometric center of the position coordinates of the defect area and the second geometric center of the position coordinates of the temperature anomaly area, and use the coordinate difference between the first geometric center and the second geometric center as the translation reference;

[0020] Determine the translation transformation component of the spatial transformation matrix according to the translation reference, determine the rotation transformation component of the spatial transformation matrix based on the distribution characteristics of the position coordinates of the defect area and the position coordinates of the temperature anomaly area, and generate a spatial transformation matrix including the translation transformation component and the rotation transformation component;

[0021] Construct the position deviation between the position coordinates of the defect area and the position coordinates of the temperature anomaly area as an optimization objective function, which includes the translation transformation component and the rotation transformation component of the spatial transformation matrix, and solve the optimization objective function by an iterative optimization method to obtain the optimal spatial transformation matrix that minimizes the position deviation;

[0022] Apply the optimal spatial transformation matrix to the position coordinates of the defect area to obtain the position coordinates of the defect area after spatial mapping, calculate the overlap area ratio between the position coordinates of the defect area after spatial mapping and the position coordinates of the temperature anomaly area, and when the overlap area ratio is greater than the preset overlap threshold, determine the overlapping area as the target defect area.

[0023] In an alternative embodiment,

[0024] Calculate the edge texture change rate and the temperature gradient change rate in the defect feature data, determine the expansion direction of the target defect area according to the edge texture change rate, and determine the expansion speed of the target defect area according to the temperature gradient change rate, including:

[0025] Establish a polar coordinate system with the defect center as the origin, perform spline fitting on the defect boundary point sequence to obtain a boundary curve function, and establish a local coordinate system including a normal vector and a tangent vector at each boundary point;

[0026] Extract the texture feature parameters in the defect feature data, project the texture feature parameters in the local coordinate system to obtain the normal texture gradient value and the tangent texture gradient value, and linearly combine the normal texture gradient value and the tangent texture gradient value based on a preset texture weight to obtain the edge texture change rate;

[0027] Extract the temperature field matrix in the defect feature data, calculate the spatial gradient of the temperature field matrix, project the spatial gradient in the local coordinate system to obtain the temperature normal gradient and the temperature tangent gradient, and linearly combine the temperature normal gradient and the temperature tangent gradient based on a preset temperature weight to obtain the temperature gradient change rate;

[0028] Calculate the components of the edge texture change rate in the tangential and normal directions, calculate the expansion angle based on the ratio of the obtained tangential component and normal component, and combine the expansion angle with the normal vector of the boundary curve function to determine the expansion direction of the target defect area;

[0029] Project the temperature gradient change rate onto the expansion direction to obtain a projection value, and determine the expansion speed of the target defect area based on the magnitude of the projection value in combination with the material properties of the target defect area.

[0030] In an alternative embodiment,

[0031] Construct a defect expansion vector field based on the expansion direction and expansion speed, and obtain the position prediction result of the target defect area by solving the propagation equation of the defect expansion vector field, including:

[0032] Use the expansion direction as the direction component and the expansion speed as the amplitude component to construct a defect expansion vector field, and perform spatial weight smoothing on the defect expansion vector field to obtain a smoothed defect expansion vector field;

[0033] Construct an initial level set function representing the boundary of the target defect area, make the boundary of the target defect area correspond to the zero value position of the initial level set function, and the initial level set function takes negative values inside the target defect area and positive values outside the target defect area;

[0034] Perform an inner product operation on the smoothed defect expansion vector field and the gradient of the initial level set function to construct the driving term of the propagation equation, and establish the propagation equation of the initial level set function;

[0035] Perform time discretization on the propagation equation to obtain a time iteration expression, and perform spatial discretization on the time iteration expression according to the local direction of the smoothed defect expansion vector field to construct an iterative equation set;

[0036] Solve the iterative equation set to obtain the level set function at the prediction moment, extract the zero value position of the level set function at the prediction moment to obtain the boundary of the target defect area at the prediction moment, and obtain the position prediction result of the target defect area according to the boundary of the target defect area at the prediction moment.

[0037] In an alternative embodiment,

[0038] Calculate the area change rate of the target defect area according to the position prediction result, and determine the performance attenuation rate of the photovoltaic module, including:

[0039] Obtain defect boundary images at different times based on the position prediction results, construct the defect boundary images into time series data, perform forward difference and backward difference calculations on the time series data respectively using the central difference method, and obtain the area change rate of the target defect area through numerical recursive iteration optimization;

[0040] Collect the temperature distribution data and stress distribution data of the target defect area, judge the hot spot characteristics of the target defect area according to the temperature gradient and temperature extreme values in the temperature distribution data, judge the crack characteristics of the target defect area according to the stress concentration degree and stress extreme values in the stress distribution data, combine the area change rate with the hot spot characteristics and crack characteristics respectively to construct a non-linear mapping relationship of power loss, calculate the power loss value based on the non-linear mapping relationship of power loss, and perform continuous time derivative operation on the power loss value to obtain the performance degradation rate of the photovoltaic module.

[0041] In an alternative embodiment,

[0042] Obtain the string connection relationship and current load distribution data of the photovoltaic module in the power station, calculate the performance influence coefficient of the photovoltaic module on adjacent modules, and generate a replacement priority score based on the performance degradation rate and the performance influence coefficient, including:

[0043] Use a string topology detector to collect the string connection relationship data of the photovoltaic module in the power station, use a current collector to collect the current load distribution data in the string, establish a series-parallel connection matrix of the photovoltaic module and adjacent modules according to the string connection relationship data, and calculate the current difference between the photovoltaic module and adjacent modules according to the current load distribution data;

[0044] Collect the physical distance data and component surface temperature data between adjacent components, calculate the heat conduction coefficient between components according to the physical distance data, calculate the temperature gradient between components according to the component surface temperature data, calculate the initial value of the hot spot influence of adjacent components according to the current difference, heat conduction coefficient and temperature gradient, establish a hot spot diffusion path tree according to the series-parallel connection matrix, and perform forward recurrence and backward recurrence iteration calculations on the initial value of the hot spot influence along the diffusion path tree respectively to obtain the performance influence coefficient of the photovoltaic module on adjacent modules;

[0045] Multiply the performance degradation rate by the degradation weight calibrated based on historical data to obtain the component degradation score, multiply the performance influence coefficient by the influence weight calibrated based on experimental data to obtain the component interaction score, collect the installation position data of the photovoltaic module, and multiply the installation position data by the position weight calibrated based on maintenance experience to obtain the installation position score;

[0046] The deterioration score of the component, the component interaction score, and the installation location score are weighted and combined and optimized to obtain a replacement priority score.

[0047] In a second aspect of the embodiments of the present invention,

[0048] A photovoltaic component defect intelligent identification and positioning system based on multi-modal images is provided, including:

[0049] A first unit configured to acquire visible light images and infrared thermal imaging images of a photovoltaic component to be detected and perform preprocessing, extract the position coordinates and geometric features of the defect area in the preprocessed visible light image, and the position coordinates and temperature features of the temperature abnormal area in the preprocessed infrared thermal imaging image;

[0050] A second unit configured to perform spatial mapping registration on the position coordinates of the defect area and the position coordinates of the temperature abnormal area to generate a target defect area, and extract the position information and defect feature data of the target defect area;

[0051] A third unit configured to calculate the edge texture change rate and the temperature gradient change rate in the defect feature data, determine the expansion direction of the target defect area according to the edge texture change rate, determine the expansion speed of the target defect area according to the temperature gradient change rate, construct a defect expansion vector field based on the expansion direction and the expansion speed, and obtain the position prediction result of the target defect area by solving the propagation equation of the defect expansion vector field;

[0052] A fourth unit configured to calculate the area change rate of the target defect area according to the position prediction result, determine the performance attenuation rate of the photovoltaic component, obtain the string connection relationship and current load distribution data of the photovoltaic component in the power station, calculate the performance influence coefficient of the photovoltaic component on adjacent components, generate a replacement priority score based on the performance attenuation rate and the performance influence coefficient, and output a diagnostic report including defect identification and positioning and component replacement suggestions according to the replacement priority score.

[0053] In a third aspect of the embodiments of the present invention,

[0054] An electronic device is provided, including:

[0055] A processor;

[0056] A memory for storing instructions executable by the processor;

[0057] Wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0058] In a fourth aspect of the embodiments of the present invention,

[0059] Provided is a computer-readable storage medium having computer program instructions stored thereon, and when the computer program instructions are executed by a processor, the foregoing method is implemented.

[0060] In this embodiment, through the multi-modal analysis by combining visible light images and infrared thermal imaging images, the accurate identification and precise positioning of photovoltaic module defects are achieved, overcoming the identification limitations of single imaging technologies, improving the comprehensiveness and reliability of defect detection, making the acquisition of defect features more comprehensive and three-dimensional, and providing rich data support for the determination of defect types and the assessment of severity. Based on the edge texture change rate and temperature gradient change rate in the defect feature data, a defect expansion vector field and propagation equation are constructed to realize the dynamic prediction of the defect expansion trend. It can not only identify the current defect state but also predict the development trend and potential risks of defects, providing a scientific basis for the preventive maintenance of photovoltaic power stations and effectively avoiding component damage and power station performance degradation caused by defect expansion. By calculating the performance degradation rate of photovoltaic modules and the performance influence coefficient on adjacent modules, a component replacement priority score is generated, realizing the transformation from single-component defect analysis to the optimization of the entire power station system. Considering system factors such as string connection relationships and current load distributions, it provides quantitative indicators and priority suggestions for power station operation and maintenance decisions, improving maintenance efficiency and optimizing maintenance costs, and extending the service life and power generation efficiency of photovoltaic power stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a schematic flowchart of the method for intelligent identification and positioning of photovoltaic module defects based on multi-modal images according to an embodiment of the present invention;

[0062] Figure 2 It is a heat map for calculating the overlapping area ratio and determining the target defect area according to an embodiment of the present invention;

[0063] Figure 3 It is a comparison chart of prediction accuracy according to an embodiment of the present invention;

[0064] Figure 4 It is a bar chart for comparing the power losses of different defect types according to an embodiment of the present invention;

[0065] Figure 5 It is a schematic structural diagram of the system for intelligent identification and positioning of photovoltaic module defects based on multi-modal images according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0066] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0067] The technical solutions of the present invention will be described in detail below with specific embodiments. These specific embodiments may be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0068] Figure 1 The flowchart of the intelligent identification and positioning method for photovoltaic module defects based on multi-modal images according to the embodiments of the present invention is as Figure 1 shown, and the method includes:

[0069] Obtain the visible light image and infrared thermal imaging image of the photovoltaic module to be detected and perform preprocessing, extract the position coordinates and geometric features of the defect area in the preprocessed visible light image, and the position coordinates and temperature features of the temperature anomaly area in the preprocessed infrared thermal imaging image;

[0070] Perform spatial mapping registration on the position coordinates of the defect area and the position coordinates of the temperature anomaly area to generate a target defect area, and extract the position information and defect feature data of the target defect area;

[0071] Calculate the edge texture change rate and temperature gradient change rate in the defect feature data, determine the expansion direction of the target defect area according to the edge texture change rate, determine the expansion speed of the target defect area according to the temperature gradient change rate, construct a defect expansion vector field based on the expansion direction and expansion speed, and obtain the position prediction result of the target defect area by solving the propagation equation of the defect expansion vector field;

[0072] Calculate the area change rate of the target defect area according to the position prediction result, determine the performance attenuation rate of the photovoltaic module, obtain the string connection relationship and current load distribution data of the photovoltaic module in the power station, calculate the performance influence coefficient of the photovoltaic module on adjacent modules, generate a replacement priority score based on the performance attenuation rate and performance influence coefficient, and output a diagnostic report including defect identification and positioning and component replacement suggestions according to the replacement priority score.

[0073] Among them, visible light images are obtained by a drone equipped with a high-resolution camera, with a shooting resolution of 4000×3000 pixels, and each image covers 4 - 6 photovoltaic modules; infrared thermal imaging images are obtained by a drone equipped with a thermal imager, with a temperature accuracy of ±0.5°C and a shooting resolution of 640×480 pixels. Geometric correction, brightness equalization, and image enhancement processing are performed on the visible light images to filter out noise and improve image quality; temperature calibration, color mapping, and spatial filtering processing are performed on the infrared thermal imaging images to ensure the accuracy of temperature data.

[0074] In an optional implementation manner, extracting the position coordinates and geometric features of the defect region in the preprocessed visible light image, and the position coordinates and temperature features of the temperature anomaly region in the preprocessed infrared thermal imaging image includes:

[0075] Adaptive double-threshold method is used to perform edge detection on the preprocessed visible light image, calculate the gradient magnitude data and gradient direction data, calculate the morphological gradient value of the preprocessed visible light image based on the gradient magnitude data and gradient direction data, construct an image segmentation function according to the morphological gradient value, and perform region segmentation to obtain the segmentation region;

[0076] Adjacent region merging is performed on the segmentation region to obtain the defect region in the preprocessed visible light image, calculate the centroid coordinate value and region area value of the defect region, and obtain the position coordinates and geometric features of the defect region;

[0077] Calculate the average temperature value and temperature standard deviation value of the normal region in the preprocessed infrared thermal imaging image, construct a temperature anomaly determination function according to the average temperature value and temperature standard deviation value, identify the temperature anomaly points, use the temperature anomaly points as the initial growth points to perform region growth operations, obtain the temperature anomaly region in the preprocessed infrared thermal imaging image, calculate the centroid coordinate value and region temperature difference of the temperature anomaly region, and obtain the position coordinates and temperature features of the temperature anomaly region.

[0078] Exemplarily, adaptive double-threshold edge detection is performed on the preprocessed visible light image. Gaussian filtering is performed on the visible light image using a 5×5 Gaussian kernel with a standard deviation of 1.4 to eliminate the influence of noise. After filtering, the Sobel operator is used to calculate the gradient magnitude and gradient direction of the image. The Sobel operator convolves the image in the x-direction and y-direction respectively, and then synthesizes the gradient magnitude and gradient direction data according to the calculation results. For each pixel point (i, j) in the image, its gradient magnitude is calculated by taking the square root of the sum of the squares of the x-direction gradient and the y-direction gradient. The gradient direction is calculated by taking the arctangent of the y-direction gradient to the x-direction gradient. Next, the morphological gradient value is calculated based on the gradient magnitude and gradient direction data. The morphological gradient is obtained by performing dilation operation and erosion operation on the image and then calculating the difference between the two. The dilation operation uses a 3×3 rectangular structuring element, and the erosion operation uses the same structuring element. The dilation operation takes the maximum value within the area covered by the structuring element, and the erosion operation takes the minimum value.

[0079] An image segmentation function is constructed based on the morphological gradient value. The segmentation function sets a high threshold and a low threshold. The high threshold is 70% of the maximum value of the morphological gradient, and the low threshold is 40% of the high threshold. Pixel points with gradient values greater than the high threshold are directly marked as edge points; pixel points with gradient values less than the low threshold are marked as non-edge points; for pixel points with gradient values between the two, if they are adjacent to edge points, they are marked as edge points, otherwise they are marked as non-edge points.

[0080] When performing region segmentation, the connected component analysis method is used to group adjacent edge points into the same region. Specifically, the 8-neighborhood connectivity rule is adopted, and the labeling of the connected regions is realized through the depth-first search or breadth-first search algorithm. For example, in an image with a resolution of 640×480, 15 - 20 initial segmentation regions may be obtained after edge detection.

[0081] Adjacent region merging operation is performed on the segmented regions using the regional similarity measurement criterion. If the difference in the average gray values of two adjacent regions is less than a threshold (e.g., 10 gray levels) and the boundary strength is lower than a threshold (e.g., 30% of the average value of the morphological gradient), then these two regions are merged into one region. The merging process is iterated until there are no adjacent regions that meet the merging conditions. This process may merge the initial segmentation regions into 5 - 8 larger regions.

[0082] Obtain the defective regions in the preprocessed visible light image and judge by regional features. Defective regions usually have obvious edge features and gray - level discontinuities. Set the criteria for judging defective regions: the average gradient value of the region is greater than 1.5 times the average gradient value of the overall image; the gray - level difference between the region and the surrounding region is greater than 20 gray - level grades; the area of the region is greater than 50 pixels and less than 10% of the image area. Regions that meet these conditions are marked as defective regions.

[0083] Calculate the centroid coordinate values and the area values of the defective regions. For each connected region marked as a defective region, count all the pixel coordinates (x i , y i ) it contains. The centroid coordinates (x c , y c ) of the defective region are the average values of the x - coordinates and y - coordinates of all pixel points. The area value is equal to the total number of pixel points within the region. For example, a defective region may have centroid coordinates (320, 240) and an area value of 1200 pixel points.

[0084] Calculate the average temperature value and the temperature standard deviation value of the normal region for the preprocessed infrared thermal imaging image. Select the four - corner regions of the image (each corner is 1 / 8 of the image size) as the reference normal regions, calculate the average value of the temperature values of all pixel points within these regions to obtain the average temperature value μ. The temperature standard deviation value σ is obtained by calculating the square root of the average value of the sum of the squares of the differences between the temperature values within these regions and the average temperature value. For example, for an infrared thermal imaging image with a resolution of 320×240, the average temperature value of the normal region may be 25.5°C and the temperature standard deviation value may be 0.8°C.

[0085] Construct a temperature anomaly judgment function based on the average temperature value and the temperature standard deviation value. The 3σ criterion is used for temperature anomaly judgment, that is, pixel points whose temperature values exceed the range of 3 times the standard deviation above and below the average temperature value are judged as temperature anomaly points. That is, pixel points with temperature values greater than μ + 3σ or less than μ - 3σ are marked as temperature anomaly points. In the above example, pixel points with temperatures higher than 27.9°C or lower than 23.1°C will be judged as temperature anomaly points.

[0086] Perform region - growing operations with the temperature anomaly points as the initial growing points. The criterion for region - growing is: if a pixel point is adjacent to a pixel point within the already marked temperature anomaly region and the difference between its temperature value and the temperature value of the initial seed point is less than a set threshold (for example, 1.5 times the temperature standard deviation value), then this pixel point is incorporated into the temperature anomaly region. The region - growing process continues until no new pixel points meet the growing conditions.

[0087] During the region growing process, 8-neighborhood connectivity checking is used to ensure the connectivity of the temperature anomaly regions. For each initial growing point, a temperature anomaly region is grown. If two temperature anomaly regions are adjacent and have similar temperature characteristics (the temperature difference is less than the standard deviation of the temperature), they are merged into one region. After obtaining the temperature anomaly regions in the preprocessed infrared thermal image, the centroid coordinate value and the regional temperature difference of the region are calculated. The calculation method of the centroid coordinate value is the same as that of the visible light image, which is the average value of the coordinates of all pixel points in the region. The regional temperature difference is defined as the difference between the average temperature value in the temperature anomaly region and the average temperature value in the normal region. For example, the centroid coordinates of a temperature anomaly region may be (160, 120), and the regional temperature difference is 5.8 °C (i.e., the average temperature of this region is 31.3 °C, 5.8 °C higher than the average temperature of the normal region).

[0088] In this embodiment, by fusing the information of the visible light image and the infrared thermal image, high-precision recognition and positioning of the defect region and the temperature anomaly region are achieved. The adaptive double-threshold method and morphology gradient are used to enhance the image edge extraction effect and improve the accuracy of region segmentation. The adjacent region merging strategy effectively reduces the false detection rate and ensures the integrity and accuracy of defect region recognition. By introducing the temperature anomaly determination function and the region growing algorithm, the temperature anomaly regions in the infrared image can be accurately extracted, and combined with feature information such as the centroid and temperature difference, the joint analysis of multi-source images is realized. The overall scheme improves the robustness of image defect recognition and the accuracy of multi-dimensional feature extraction, and is applicable to intelligent detection tasks under complex working conditions.

[0089] In an alternative embodiment, spatial mapping registration is performed on the position coordinates of the defect region and the position coordinates of the temperature anomaly region to generate the target defect region, including:

[0090] Calculate the first geometric center of the position coordinates of the defect region and the second geometric center of the position coordinates of the temperature anomaly region, and use the coordinate difference between the first geometric center and the second geometric center as the translation reference;

[0091] Determine the translation transformation component of the spatial transformation matrix according to the translation reference, and determine the rotation transformation component of the spatial transformation matrix based on the distribution characteristics of the position coordinates of the defect region and the position coordinates of the temperature anomaly region, and generate a spatial transformation matrix including the translation transformation component and the rotation transformation component;

[0092] Construct the position deviation between the position coordinates of the defect region and the position coordinates of the temperature anomaly region as an optimization objective function, which includes the translation transformation component and the rotation transformation component of the spatial transformation matrix, and solve the optimization objective function by an iterative optimization method to obtain the optimal spatial transformation matrix that minimizes the position deviation;

[0093] Apply the optimal spatial transformation matrix to the position coordinates of the defect region to obtain the position coordinates of the defect region after spatial mapping. Calculate the overlapping area ratio between the position coordinates of the defect region after spatial mapping and the position coordinates of the temperature anomaly region. When the overlapping area ratio is greater than the preset overlapping threshold, determine the overlapping region as the target defect region.

[0094] During the intelligent identification and positioning of photovoltaic module defects, due to differences in imaging principles, perspectives, and resolutions between visible light images and infrared thermal imaging images, direct superposition will result in position deviations. To solve this problem, this method first calculates the first geometric center of the position coordinates of the defect region and the second geometric center of the position coordinates of the temperature anomaly region as the registration reference points. For the defect region extracted from the visible light image, the first geometric center is obtained by calculating the average value of the boundary point coordinates. For example, the boundary point coordinate set of a certain hot spot defect region contains 32 points, and the average values of the abscissa and ordinate of all points are calculated to obtain the first geometric center coordinates as (245, 367) pixels. Similarly, for the temperature anomaly region in the infrared thermal imaging image, assuming there are 18 boundary points, the second geometric center coordinates are calculated as (125, 198) pixels.

[0095] Use the coordinate difference between the first geometric center and the second geometric center as the translation reference. In the above example, the translation reference is a horizontal difference of 120 pixels and a vertical difference of 169 pixels. This translation reference reflects the position deviation between the two imaging systems and will be used as the translation transformation component of the spatial transformation matrix.

[0096] Next, determine the rotation transformation component of the spatial transformation matrix based on the distribution characteristics of the defect region and the temperature anomaly region. The distribution characteristics are obtained by analyzing the spatial distribution pattern of the region boundary points, including the main axis direction, shape characteristics, etc. For example, for linear crack - type defects, by calculating the main axis direction of the spatial distribution of the region boundary points, the main axis direction of the crack in the visible light image is 57 degrees, and the main axis direction of the corresponding region in the infrared thermal imaging is 63 degrees. The angle difference between the two is 6 degrees, and this angle difference is the rotation transformation component.

[0097] Combine the translation transformation component and the rotation transformation component to generate an initial spatial transformation matrix. In the above example, the initial spatial transformation matrix includes a horizontal translation of 120 pixels, a vertical translation of 169 pixels, and a transformation parameter of a 6 - degree clockwise rotation. This matrix provides a preliminary mapping relationship between the two coordinate systems.

[0098] To further optimize the spatial transformation matrix, the positional deviation between the position coordinates of the defect region and the position coordinates of the temperature anomaly region is constructed as the optimization objective function. The positional deviation is defined as the average Euclidean distance between the boundary points of the defect region and the boundary points of the temperature anomaly region after spatial transformation. For example, the average Euclidean distance after the initial transformation is 8.7 pixels, indicating that the registration accuracy needs to be improved.

[0099] The optimization objective function is solved by an iterative optimization method to obtain the optimal spatial transformation matrix that minimizes the positional deviation. The iterative optimization uses the gradient descent algorithm, which fine-tunes the translation and rotation components of the transformation matrix in each iteration, calculates the new positional deviation, and adjusts in the direction of decreasing deviation. The maximum number of iterations in the optimization process is set to 200, and the convergence threshold is 0.1 pixel. In actual operation, the optimization iteration usually converges within 50 - 100 times, and the finally obtained optimal spatial transformation matrix includes a horizontal translation of 118.5 pixels, a vertical translation of 172.3 pixels, and a rotation angle of 5.8 degrees.

[0100] The optimal spatial transformation matrix is applied to the position coordinates of the defect region to obtain the position coordinates of the defect region after spatial mapping. Taking the hot spot defect as an example, after the transformation of the original defect region coordinate set, its center point changes from (245, 367) to (126.5, 197.2), and the deviation from the center point (125, 198) of the temperature anomaly region is reduced to 1.9 pixels, indicating a significant improvement in registration accuracy. After completing the spatial mapping, the overlapping area ratio between the position coordinates of the transformed defect region and the position coordinates of the temperature anomaly region is calculated. The overlapping area ratio is defined as the ratio of the intersection area of the two regions to the union area of the two regions, which reflects the registration quality. In the above example, the calculated overlapping area ratio is 0.87, that is, 87% of the regions achieve effective overlap.

[0101] According to the actual application requirements, a preset overlapping threshold is set, and the typical value is 0.75. When the calculated overlapping area ratio is greater than the preset overlapping threshold, the overlapping region is determined as the target defect region. In this example, the overlapping area ratio of 0.87 is greater than the preset threshold of 0.75, so the overlapping region is confirmed as the effective target defect region. For cases that do not meet the overlapping threshold, the system will mark them as uncertain regions, and secondary registration can be performed by adding more feature points or using non-rigid registration methods. In actual applications, about 92% of the defect regions can reach the preset overlapping threshold in the first registration.

[0102] After completing the spatial mapping registration, the target defect region contains both the characteristic information of visible light and infrared thermal imaging, with both structural characteristics and temperature distribution characteristics. For example, the registered hot spot defect region not only contains visual characteristics such as surface discoloration and texture changes in the visible light image, but also contains thermal characteristics such as temperature distribution and thermal gradient in the infrared thermal imaging, providing comprehensive characteristic data for subsequent defect classification, severity assessment, and trend prediction.

[0103] Figure 2 This is the heat map for calculating the overlapping area ratio and determining the target defect area in the embodiment of the present invention. As Figure 2 shown, this figure intuitively shows the spatial distribution and overlapping situation of the defect area and the temperature anomaly area. The figure clearly shows the positional relationship between the temperature anomaly area (dashed ellipse, center coordinates (300, 180)) and the defect area (solid rectangle, center coordinates (310, 200)). The area ratio of the overlapping area (gray filled area) is as high as 87.6%, far exceeding the preset overlapping threshold, so it is determined as the target defect area. The transformation parameters shown below the figure give the specific values for achieving precise registration in detail: the translation reference formed by the difference in the geometric centers of the defect area and the temperature anomaly area is (10, 20), and the rotation angle is 2.7°. These parameters are obtained by optimizing the objective function, so that the defect area and the temperature anomaly area achieve the best match in space. The gray-scale distribution in the heat map intuitively reflects the temperature gradient, so that the temperature characteristic distribution of potential defect areas can be accurately identified, providing an important basis for defect diagnosis.

[0104] The prior art mainly uses simple superposition or fixed-parameter transformation for image registration, often ignoring the non-linear differences between different imaging modalities, resulting in low registration accuracy, especially significant errors in the case of uneven illumination and large viewing angle changes. In this application, the difference in geometric centers is used as the translation reference, and the rotation component is determined in combination with the distribution characteristics, constructing a spatial transformation matrix including translation and rotation transformations. By using the iterative optimization method to minimize the position deviation, and then using the overlapping area ratio as the quality evaluation index, high-precision multi-modal image registration is achieved. This feature-based adaptive registration method effectively solves the problems of scale, viewing angle, and resolution differences between visible light and infrared images, significantly improving the accuracy of the spatial correspondence between the defect area and the temperature anomaly area. The starting point for improving this method is to solve the position deviation problem in multi-modal image fusion, ultimately achieving precise positioning of the defect position, improving the integrity and accuracy of defect feature extraction, enhancing the reliability of defect expansion prediction, providing more reliable technical support for photovoltaic module defect diagnosis and predictive maintenance, and effectively reducing the operation and maintenance costs of photovoltaic power plants.

[0105] In an alternative embodiment, calculating the edge texture change rate and the temperature gradient change rate in the defect feature data, determining the expansion direction of the target defect area according to the edge texture change rate, and determining the expansion speed of the target defect area according to the temperature gradient change rate includes:

[0106] Establish a polar coordinate system with the defect center as the origin, perform spline fitting on the defect boundary point sequence to obtain the boundary curve function, and establish a local coordinate system including the normal vector and the tangent vector at each boundary point;

[0107] Extract the texture feature parameters from the defect feature data, project the texture feature parameters in the local coordinate system to obtain the normal texture gradient value and the tangential texture gradient value, and linearly combine the normal texture gradient value and the tangential texture gradient value based on a preset texture weight to obtain the edge texture change rate;

[0108] Extract the temperature field matrix from the defect feature data, calculate the spatial gradient of the temperature field matrix, project the spatial gradient in the local coordinate system to obtain the temperature normal gradient and the temperature tangential gradient, and linearly combine the temperature normal gradient and the temperature tangential gradient based on a preset temperature weight to obtain the temperature gradient change rate;

[0109] Calculate the components of the edge texture change rate in the tangential and normal directions, calculate the expansion angle based on the ratio of the obtained tangential component and normal component, and combine the expansion angle with the normal vector of the boundary curve function to determine the expansion direction of the target defect area;

[0110] Project the temperature gradient change rate onto the expansion direction to obtain a projection value, and determine the expansion speed of the target defect area based on the magnitude of the projection value in combination with the material properties of the target defect area.

[0111] Exemplarily, first establish a polar coordinate system with the defect center as the origin, which is convenient for describing the radial expansion characteristics of the defect. The specific approach is to arrange the boundary points of the target defect area in a clockwise or counterclockwise order, calculate the distance and angle of these boundary points from the geometric center of the defect, and form a polar coordinate representation. For example, if the defect center coordinates are (256, 342) pixels and the coordinates of a certain point on the boundary are (289, 365) pixels, then its polar coordinate representation is a distance of 42.15 pixels and an angle of 36.9 degrees.

[0112] Perform spline fitting on the defect boundary point sequence to obtain the boundary curve function. This step connects the discrete boundary points into a smooth and continuous closed curve through a cubic spline interpolation algorithm. Set an appropriate smoothing factor during spline fitting to ensure both the smoothness of the curve and the preservation of the local characteristics of the boundary. The obtained boundary curve function describes the mapping relationship from the polar coordinate angle to the radial distance, enabling the acquisition of the corresponding boundary point positions at any angle.

[0113] Establish a local coordinate system containing the normal vector and the tangent vector at each boundary point. The normal vector points to the external vertical direction of the defect boundary, and the tangent vector is along the tangent direction of the defect boundary. Determine the tangent direction by calculating the derivative of the boundary curve at this point, and then rotate it by 90 degrees to obtain the normal direction. For example, if the tangent vector at a certain point on the boundary is (0.707, 0.707), then its normal vector is (-0.707, 0.707). This local coordinate system describes the possible expansion direction characteristics of the defect at this point.

[0114] Extract texture feature parameters from the defect feature data, including statistical indicators such as the contrast, uniformity, and entropy of the gray-level co-occurrence matrix, as well as texture descriptors such as wavelet transform coefficients and local binary patterns. Set an annular region around the defect boundary, with a width usually of 10 - 20 pixels, extract the texture features within this region, and analyze the changing trend of the texture.

[0115] Project the texture feature parameters in the local coordinate system to obtain the normal texture gradient value and the tangential texture gradient value. The normal texture gradient represents the intensity of texture change perpendicular to the boundary, and the tangential texture gradient represents the intensity of texture change along the boundary. These two gradient values are obtained by calculating the spatial difference of the texture features in the normal and tangential directions. For example, at a certain boundary point, the normal texture gradient value is 8.3, and the tangential texture gradient value is 2.1, indicating that the texture change perpendicular to the boundary at this point is much greater than the texture change along the boundary.

[0116] Based on the preset texture weights, linearly combine the normal texture gradient value and the tangential texture gradient value to obtain the edge texture change rate. The preset texture weights reflect the influence degree of texture changes in different directions on the defect expansion. Usually, the normal weight is greater than the tangential weight. For example, set the normal weight to 0.7 and the tangential weight to 0.3, then the edge texture change rate is the weighted average of the two.

[0117] Extract the temperature field matrix from the defect feature data. This matrix comes from the infrared thermal imaging image and records the temperature distribution in the defect region and its surroundings. Calculate the spatial gradient of the temperature field matrix, that is, the change rate of temperature in the horizontal and vertical directions. The spatial gradient is obtained by calculating the temperature difference between adjacent pixel points and reflects the changing trend of the temperature field.

[0118] Project the temperature spatial gradient in the local coordinate system to obtain the temperature normal gradient and the temperature tangential gradient. The temperature normal gradient represents the change rate of temperature perpendicular to the boundary, and the temperature tangential gradient represents the change rate of temperature along the boundary. For example, at a certain boundary point, the temperature normal gradient is 1.8 °C / pixel, and the temperature tangential gradient is 0.3 °C / pixel, indicating that the temperature at this point mainly changes in the normal direction.

[0119] Based on the preset temperature weights, linearly combine the temperature normal gradient and the temperature tangential gradient to obtain the temperature gradient change rate. The preset temperature weights reflect the influence degree of temperature changes in different directions on the defect expansion and are determined according to the thermodynamic principle. For example, set the temperature normal weight to 0.8 and the temperature tangential weight to 0.2, then the temperature gradient change rate is the weighted average of the two.

[0120] Calculate the tangential and normal components of the edge texture change rate, and calculate the expansion angle based on the ratio of the obtained tangential component and normal component. The expansion angle reflects the angle between the direction in which the defect is most likely to expand and the normal direction. When the tangential component is much smaller than the normal component, the expansion direction is close to the normal direction; when the tangential component is comparable to the normal component, the expansion direction deviates from the normal direction to a certain extent. For example, when the ratio of the tangential component to the normal component is 0.25, the expansion angle is about 14 degrees.

[0121] Combine the expansion angle with the normal vector of the boundary curve function to determine the expansion direction of the target defect area. The specific method is to rotate the normal vector by the expansion angle to obtain the expansion direction vector. This combination takes into account the comprehensive effects of boundary morphology and material properties on defect expansion and is more in line with the actual situation.

[0122] Project the temperature gradient change rate onto the expansion direction to obtain a projection value, which represents the temperature change intensity in the determined expansion direction. The larger the projection value, the greater the temperature gradient in this direction and the stronger the driving force for defect expansion. Based on the magnitude of the projection value and combined with the material properties of the target defect area, determine the expansion speed of the target defect area. Material properties include parameters such as the coefficient of thermal expansion, elastic modulus, and ultimate strength, which determine the deformation and failure rate of the material under the action of a temperature gradient.

[0123] For example, in the detection of hot spot defects in a certain photovoltaic module, a target defect area located in the upper right corner of the module is identified, and its center coordinates are (876, 234) pixels. Perform spline fitting on 36 points on the boundary of the defect area to obtain a closed boundary curve function. At a point at a boundary angle of 45 degrees, the normal vector in its local coordinate system is (0.64, 0.77), and the tangent vector is (-0.77, 0.64). Extract the texture features of the 20-pixel-wide annular area around the boundary, and calculate that the normal texture gradient value at this point is 7.6, and the tangential texture gradient value is 1.9. Set the normal texture weight to 0.75 and the tangential texture weight to 0.25, and calculate that the edge texture change rate at this point is 6.18. Extract the temperature field matrix from the infrared thermal imaging image. The normal temperature gradient at this point is 2.1 °C / pixel, and the tangential temperature gradient is 0.4 °C / pixel. Set the normal temperature weight to 0.8 and the tangential temperature weight to 0.2, and obtain a temperature gradient change rate of 1.76 °C / pixel. The normal component of the edge texture change rate is 4.63, the tangential component is 1.55, the ratio is 0.33, and the corresponding expansion angle is 18.3 degrees. Rotate the normal vector by 18.3 degrees to obtain the expansion direction vector (0.32, 0.95). The projection value of the temperature gradient change rate in this direction is 1.59 °C / pixel. This defect area is a silicon solar cell, and the coefficient of thermal expansion of the material is 2.6×10 -6 / °C, the ultimate strength is 200 MPa. According to the empirical model, the expansion speed at this projection value is calculated to be 0.46% / month, that is, the monthly defect area increase rate is 0.46%.

[0124] In this embodiment, by constructing a polar coordinate system for the defect area and combining boundary geometric features with multi-source data analysis, high-precision prediction of the defect expansion direction and speed is achieved. Through the joint analysis of the edge texture change rate and the temperature gradient change rate, the change trend of the defect area in terms of texture and thermal response is accurately reflected, improving the accuracy of defect development trend recognition. By establishing a local coordinate system and introducing directional gradient analysis, the judgment of the expansion trend not only has the ability to distinguish directions but also has the ability to quantitatively evaluate, effectively overcoming the problem of fuzzy estimation of the defect expansion direction by traditional methods. In addition, by combining the thermal and mechanical properties of the material, the quantitative evaluation of the defect expansion speed is realized, providing a scientific basis for subsequent maintenance decisions and life prediction, and significantly improving the intelligence and reliability levels of photovoltaic module defect diagnosis.

[0125] In an alternative embodiment, a defect expansion vector field is constructed based on the expansion direction and expansion speed. The position prediction result of the target defect area is obtained by solving the propagation equation of the defect expansion vector field, including:

[0126] Taking the expansion direction as the direction component and the expansion speed as the amplitude component, a defect expansion vector field is constructed, and the spatial weight smoothing process is performed on the defect expansion vector field to obtain the smoothed defect expansion vector field;

[0127] An initial level set function representing the boundary of the target defect area is constructed, so that the boundary of the target defect area corresponds to the zero value position of the initial level set function. The initial level set function takes a negative value inside the target defect area and a positive value outside the target defect area;

[0128] The inner product operation of the smoothed defect expansion vector field and the gradient of the initial level set function is used to construct the driving term of the propagation equation, and the propagation equation of the initial level set function is established;

[0129] The time discretization process is performed on the propagation equation to obtain a time iteration expression, and the spatial discretization process is performed on the time iteration expression according to the local direction of the smoothed defect expansion vector field to construct an iterative equation set;

[0130] The iterative equation set is solved to obtain the level set function at the prediction moment, the zero value position of the level set function at the prediction moment is extracted to obtain the boundary of the target defect area at the prediction moment, and the position prediction result of the target defect area is obtained according to the boundary of the target defect area at the prediction moment.

[0131] Exemplarily, first, obtain the historical monitoring data of the target defect area, which includes the boundary contour, expansion direction, and expansion speed information of the defect area. The expansion direction and expansion speed of each point on the defect boundary can be calculated by comparing the boundary positions of the defect area at different time points. Taking the expansion direction as the direction component and the expansion speed as the amplitude component, construct the defect expansion vector field. Specifically, for each point on the defect boundary, record the expansion direction vector and expansion speed value of this point to form a vector. For example, for a point P on the crack boundary, if its expansion direction is 45 degrees and the expansion speed is 2 mm / day, then the expansion vector of this point is (1.414 mm / day). Combine the expansion vectors of all boundary points to form the vector field on the boundary.

[0132] Perform spatial weight smoothing on the defect expansion vector field. By taking the weighted average of the expansion vector of each point on the boundary and the expansion vectors of its surrounding points, obtain the smoothed vector field. The weight can be set based on the distance between points, and the closer the distance, the greater the weight. For example, taking a certain point on the boundary as the center, take all boundary points within a range of 5 cm, and perform weighted average on their expansion vectors according to the inverse of the distance to obtain the smoothed expansion vector of the center point. This processing can reduce the influence of local noise and make the vector field smoother and more continuous.

[0133] Construct an initial level set function that characterizes the boundary of the target defect area. This function makes the boundary of the target defect area correspond to the zero value position of the level set function, takes negative values inside the target defect area, and positive values outside. In specific implementation, a signed distance function can be used, that is, the shortest distance from any point to the defect boundary, and take negative values inside the defect. For example, for a known crack area, calculate the shortest distance from any point in space to the crack boundary, take the negative distance value for points inside the crack, and positive for points outside to form the initial level set function. Perform an inner product operation on the smoothed defect expansion vector field and the gradient of the initial level set function to construct the driving term of the propagation equation. The gradient of the level set function represents the rate of change and direction of the level set function at each point in space. The result of the inner product operation represents the degree of consistency between the defect expansion direction and the level set gradient direction, and this value is used as the driving force of the propagation equation to promote the evolution of the level set function along the defect expansion direction. For example, at each node of the computational grid, calculate the inner product of the level set function gradient vector and the expansion vector at that place as the driving term value at that point.

[0134] Establish the propagation equation of the initial level set function and perform time discretization on the propagation equation to obtain the time iteration expression. Adopt the forward difference format to discretize continuous time into multiple time steps, and update the level set function through iterative calculation within each time step. For example, set the time step to 1 day, then calculate the level set function value for the next day in each iteration. The time iteration expression represents the relationship between the level set function value at the current moment and the level set function value at the next moment.

[0135] Perform spatial discretization on the time iteration expression according to the local direction of the smoothed defect expansion vector field to construct an iterative equation set. In the spatial domain, a format suitable for the defect expansion direction is used for discretization. For example, for the problem of crack propagation, the area around the crack region is divided into 100×100 nodes on the computational grid. At each node, according to the expansion direction at that place, an upwind format or a central difference format is selected for spatial discretization to construct an iterative equation set including all nodes. Finally, solve the iterative equation set to obtain the level set function at the predicted time. For example, for predicting the crack propagation situation after 7 days, 7 iterative calculations are performed, and each iteration represents one day's expansion. Extract the zero-value position from the level set function at the predicted time, that is, obtain the boundary of the target defect region at the predicted time. In specific implementation, a linear interpolation method can be used to determine the position where the level set function value is zero. According to this boundary information, the position prediction result of the target defect region can be obtained.

[0136] Figure 3 This is the comparison chart of the prediction accuracy of the embodiments of the present invention. As Figure 3 shown, the horizontal axis represents the grid size of the defect region, and the vertical axis represents the calculation time (seconds). The black triangular marker column represents the technical solution of the present invention, the black square marker column represents the traditional standard level set method, and the black circular marker column represents the existing fast marching method. At the 50×50 grid scale, the calculation time of the technical solution of the present invention is 0.82 seconds, the traditional level set method is 1.37 seconds, and the fast marching method is 1.05 seconds; at the 100×100 grid scale, the calculation times of the three methods are 3.45 seconds, 6.82 seconds, and 4.71 seconds respectively; at the 200×200 grid scale, the calculation times are 12.58 seconds, 28.64 seconds, and 18.92 seconds respectively; at the 300×300 grid scale, the calculation times are 25.73 seconds, 63.45 seconds, and 42.18 seconds respectively. The data shows that as the grid scale increases, the calculation efficiency advantage of the technical solution of the present invention becomes more obvious. At the 300×300 grid scale, it saves 59.4% of the calculation time compared with the traditional level set method and 39.0% of the calculation time compared with the fast marching method. Further analysis finds that the time complexity of the technical solution of the present invention is significantly better than that of the traditional level set method and the fast marching method. This improvement in efficiency is mainly due to the fact that the direction adaptive discretization strategy adopted by the technical solution of the present invention reduces the calculation amount, and at the same time, the global re-initialization is avoided through the local update technology, and the peak value of the calculation memory is also reduced by about 43%, which has significant advantages compared with the traditional Osher-Sethian algorithm and the Tsitsiklis fast marching method.

[0137] In the prior art, defect location prediction often relies on static image analysis or simple extrapolation methods, which cannot fully consider the directionality and speed changes during the defect expansion process, resulting in insufficient prediction accuracy, especially when dealing with defects with complex shapes or dynamic evolution. This application enhances the continuity and stability of the defect evolution trend by introducing the means of defect expansion vector field modeling, constructing a vector field by combining the expansion direction and speed of the defect, and performing spatial smoothing on this basis. Further, by introducing this vector field into the level set evolution framework, using the inner product of the vector field driving term and the gradient of the level set function as the core of the propagation equation, the dynamic change process of the defect boundary is accurately simulated. The time and space discretization strategy is adopted to solve the propagation equation, ensuring that the evolution process of the predicted boundary not only follows the physical expansion law but also has numerical stability. Compared with the prior art's static fitting prediction method based only on boundary shape changes, this solution starts from the directional mechanism of defect expansion, integrates the velocity amplitude information, and realizes the dynamic evolution prediction of the future shape of the defect boundary, significantly improving the accuracy and adaptability of defect location prediction.

[0138] In an alternative embodiment, calculating the area change rate of the target defect region based on the location prediction result and determining the performance degradation rate of the photovoltaic module includes:

[0139] Obtaining defect boundary images at different times based on the location prediction result, constructing the defect boundary images into time series data, performing forward difference and backward difference calculations on the time series data respectively using the central difference method, and obtaining the area change rate of the target defect region through numerical recursive iteration optimization;

[0140] Collecting the temperature distribution data and stress distribution data of the target defect region, judging the hot spot characteristics of the target defect region according to the temperature gradient and temperature extreme value in the temperature distribution data, judging the crack characteristics of the target defect region according to the stress concentration and stress extreme value in the stress distribution data, combining the area change rate with the hot spot characteristics and crack characteristics respectively to construct a non-linear power loss mapping relationship, calculating the power loss value based on the non-linear power loss mapping relationship, and performing continuous time derivative operation on the power loss value to obtain the performance degradation rate of the photovoltaic module.

[0141] Exemplarily, defect boundary images at different times are obtained based on the location prediction result, and these boundary images reflect the expansion state of the defect at different future time points. The location prediction result comes from the solution of the aforementioned defect expansion vector field and usually includes the predicted boundaries at the current time, 30 days later, 90 days later, and 180 days later.

[0142] Construct the defect boundary image as time - series data, that is, a set of defect boundaries arranged in chronological order. The defect boundary at each time point is represented by a closed curve, and the set of points on the curve describes the shape and size of the defect area. The time - series data reflects the dynamic change process of the defect over time and provides a basis for subsequent calculations.

[0143] Use the central - difference method to perform forward - difference and backward - difference calculations on the time - series data respectively. The central - difference method is a numerical differentiation technique used to estimate the derivative value of a function at a certain point. Forward - difference calculates the derivative using the data of the current point and the next - moment point, and backward - difference calculates the derivative using the data of the current point and the previous - moment point. By calculating the forward - difference and backward - difference simultaneously, the numerical error can be reduced and the calculation accuracy can be improved. In the difference calculation, calculate the defect area at each time point, and then calculate the ratio of the area change between adjacent time points to the time interval to obtain a preliminary estimate of the area change rate. Forward - difference calculates the change rate from the current moment to the next moment, and backward - difference calculates the change rate from the previous moment to the current moment. The area change rate of the target defect area is obtained through numerical recursive iteration optimization. Numerical recursive iteration is a method that gradually approaches the true value through repeated calculations. The specific approach is to take the weighted average of the results of the forward - difference and backward - difference as the initial estimate, then introduce a smoothing constraint condition, and perform iterative calculations by minimizing the error function until the change in the area change rate is less than the preset threshold or the maximum number of iterations is reached. This method can effectively suppress the influence of noise and obtain a smooth and continuous area change rate curve.

[0144] Collect the temperature - distribution data and stress - distribution data of the target defect area. The temperature - distribution data comes from infrared thermal - imaging images and records the temperature field of the defect area and its surroundings. The stress - distribution data is obtained through photoelastic stress analysis or finite - element simulation and describes the mechanical stress state of the defect area. Judge the hot - spot characteristics of the target defect area according to the temperature gradient and temperature extreme values in the temperature - distribution data. The temperature gradient is the spatial change rate of temperature, reflecting the non - uniformity of the temperature field; the temperature extreme value is the highest temperature in the area, reflecting the severity of the hot spot. The hot - spot characteristics are usually represented by the hot - spot intensity index, which comprehensively considers the temperature gradient and temperature extreme values. The larger the value, the more severe the hot spot.

[0145] Judge the crack characteristics of the target defect area according to the stress concentration degree and stress extreme values in the stress - distribution data. The stress concentration degree describes the non - uniformity of the stress distribution. The larger the value, the more concentrated the stress; the stress extreme value is the maximum stress value in the area, reflecting the degree of danger that may cause material failure. The crack characteristics are usually represented by the crack - risk index, which comprehensively considers the stress concentration degree and stress extreme values. The larger the value, the higher the crack risk.

[0146] The rate of area change is combined with hot spot characteristics and crack characteristics respectively to construct a non-linear mapping relationship of power loss. This mapping relationship describes how the defect area, hot spot characteristics, and crack characteristics jointly affect the power output of the photovoltaic module. The non-linear mapping adopts a piecewise function form, and different mapping parameters are set according to different defect types and severities. For example, for defects of the same area, the area with more significant hot spot characteristics has a greater impact on power; for the same hot spot characteristics, the area accompanied by crack characteristics has more serious power loss.

[0147] The power loss value is calculated based on the non-linear mapping relationship of power loss. The power loss value represents the percentage reduction in the output power of the module due to defects and is a direct indicator for evaluating the impact of defects. The calculation process takes into account the combined effects of the defect area, hot spot characteristics, and crack characteristics, and obtains a predicted curve of power loss varying with time. The derivative operation of the power loss value with respect to continuous time is performed to obtain the performance degradation rate of the photovoltaic module. The performance degradation rate describes the speed at which the output power of the module decreases over time and is usually expressed as a percentage per month or a percentage per year. The derivative operation uses a numerical differentiation method to calculate the change rate of the power loss value at adjacent time points, and then performs smoothing processing to obtain a continuous degradation rate curve.

[0148] Exemplarily, for the hot spot defect of a component found in a certain photovoltaic power station, defect boundary images at the current time, 30 days later, 90 days later, and 180 days later are obtained based on the position prediction results. The current defect area is 12 square centimeters, the predicted area 30 days later is 13.2 square centimeters, the predicted area 90 days later is 16.8 square centimeters, and the predicted area 180 days later is 24 square centimeters. The defect boundary images at these four time points are constructed into time series data. The central difference method is used to calculate the rate of area change at each time point. Initially, the forward difference value at the 30-day point is 3% per month, and the backward difference value is 4% per month; the forward difference value at the 90-day point is 4.5% per month, and the backward difference value is 4% per month; the forward difference value at the 180-day point is 8% per month, and the backward difference value is 4.5% per month. Through numerical recursive iteration optimization, with a smoothing factor of 0.3, a maximum number of iterations of 50, and a convergence threshold of 0.01%, after 28 iterations of convergence, the rate of area change at each time point is obtained: 3.5% per month at the 30-day point, 4.2% per month at the 90-day point, and 5.8% per month at the 180-day point.

[0149] Temperature distribution data of the defect area is collected from the infrared thermal imaging image. The highest temperature at the defect center is 72 °C, which is 28 °C higher than the surrounding normal area, and the temperature gradient reaches 2.3 °C / cm. Stress distribution data is obtained through photoelastic stress analysis. The stress concentration factor at the defect edge is 3.2, and the maximum stress value is 65% of the nominal strength of the component. The hot spot intensity index is calculated to be 0.78 based on the temperature gradient and temperature extreme value, and the crack risk index is calculated to be 0.42 based on the stress concentration factor and stress extreme value.

[0150] Substitute the area change rate, hot spot intensity index, and crack risk index into the pre-established non-linear mapping model of power loss, which is obtained by fitting a large amount of experimental data. The calculated current power loss is 5.3%, the power loss after 30 days is 6.1%, the power loss after 90 days is 8.7%, and the power loss after 180 days is 14.2%. Perform a continuous-time derivative operation on the power loss values to obtain the performance decay rate at each time point: currently 0.27% per month, 0.32% per month after 30 days, 0.46% per month after 90 days, and 0.61% per month after 180 days. This indicates that as the defect expands, the component performance decay rate shows an accelerating growth trend, and maintenance measures should be taken as early as possible.

[0151] Figure 4 This is the bar chart for comparing the power losses of different defect types in the embodiments of the present invention, as Figure 4 shown. This figure shows the percentage of power loss caused by six common photovoltaic module defect types, and compares the prediction results of the present technical solution (solid rectangles) with those of the linear regression model (hollow rectangles). The present technical solution constructs a non-linear mapping relationship of power loss by combining the area change rate with the hot spot characteristics and crack characteristics respectively, and obtains a more accurate power loss prediction. From the data in the figure, it can be seen that the power loss caused by glass breakage is the most serious, reaching 32.6%, followed by hidden cracks (21.5%) and poor soldering (18.9%). In contrast, the power loss caused by hot spot defects is relatively low (8.7%). The predicted values of the linear regression model for all defect types are lower than those of the present technical solution, with an average difference of 3.5 percentage points, which indicates that the linear model may underestimate the impact of defects on the component performance. Especially for serious defects (such as glass breakage), the present technical solution predicts 32.6%, while the linear regression model is only 28.1%, with a difference of 4.5 percentage points. This difference is particularly important when predicting the long-term performance degradation of components, because the cumulative effect will cause the actual power loss to far exceed the predicted value of the linear model. The non-linear mapping relationship of the present technical solution can more comprehensively consider the complex relationship between defect characteristics and power loss.

[0152] In this embodiment, it is possible to accurately model and continuously track the dynamic expansion process of the defective area of the photovoltaic module, and based on the fusion analysis of the defect expansion trend and its thermodynamic characteristics, accurately evaluate the degree of influence of the defect on the power output of the module, so as to deduce the continuous change of the performance degradation rate. By introducing the time series of defect boundary images and the numerical difference optimization method, the noise interference is effectively suppressed, and the calculation stability of the area change rate is improved; a non-linear power loss mapping relationship is constructed by combining the hot spot and crack characteristic indexes, making the evaluation process more physically meaningful and adaptable. The finally formed performance degradation rate result not only has high timeliness and continuity, but also significantly improves the accuracy and response efficiency of defect early warning and operation and maintenance decision-making.

[0153] In an alternative embodiment, the string connection relationship and current load distribution data of the photovoltaic modules in the power station are obtained, the performance influence coefficient of the photovoltaic module on adjacent modules is calculated, and the replacement priority score is generated based on the performance degradation rate and the performance influence coefficient, including:

[0154] A string topology detector is used to collect the string connection relationship data of the photovoltaic modules in the power station, a current collector is used to collect the current load distribution data in the string, a series-parallel connection matrix of the photovoltaic module and adjacent modules is established according to the string connection relationship data, and the current difference between the photovoltaic module and adjacent modules is calculated according to the current load distribution data;

[0155] The physical distance data and the component surface temperature data between adjacent components are collected, the heat conduction coefficient between the components is calculated according to the physical distance data, the temperature gradient between the components is calculated according to the component surface temperature data, the initial value of the hot spot influence of the adjacent components is calculated according to the current difference, the heat conduction coefficient and the temperature gradient, and a hot spot diffusion path tree is established according to the series-parallel connection matrix, and the initial value of the hot spot influence is respectively subjected to forward recursion and backward recursion iterative calculation along the diffusion path tree to obtain the performance influence coefficient of the photovoltaic module on adjacent modules;

[0156] The component degradation score is obtained by multiplying the performance degradation rate by the degradation weight calibrated based on historical data, the component interaction score is obtained by multiplying the performance influence coefficient by the influence weight calibrated based on experimental data, the installation position data of the photovoltaic module is collected, and the installation position score is obtained by multiplying the installation position data by the position weight calibrated based on maintenance experience;

[0157] The component degradation score, the component interaction score and the installation position score are weighted and combined and optimized to obtain the replacement priority score.

[0158] This embodiment provides a photovoltaic module replacement priority scoring system, which can accurately evaluate the replacement priority of modules based on the module connection relationship, current load distribution, performance degradation rate, and the performance influence coefficient of adjacent modules. First, a string topology detector is used to collect the module connection relationship in a photovoltaic power station. The string topology detector determines the series-parallel relationship between modules by sending detection signals with specific frequencies to each string and recording the signal return time and path information. For example, in a 100MW photovoltaic power station, there are 5000 strings in total, and each string contains 20 modules. A connection relationship database is established based on the data collected by the topology detector, recording the unique identification code of each module and the identification codes of its upstream and downstream connected modules.

[0159] At the same time, a current collector is used to collect the current load distribution in the string. The current collector is installed at the output end of each module and records the working current value of the module at 1-minute intervals. Under standard test conditions (irradiance intensity 1000W / m², temperature 25°C), the current of a normally working photovoltaic module is about 8.5A, while the current of a module with degraded performance may drop below 7.0A. Based on the collected string connection relationship data, a series-parallel connection matrix of photovoltaic modules and adjacent modules is established. This matrix is stored in the form of a two-dimensional table, where both the row and column indices are module identification codes, and the matrix element values are 1 (series relationship), 2 (parallel relationship), or 0 (no direct connection relationship).

[0160] For the current load distribution data, calculate the current difference between modules. For example, for series-connected modules A and B, if the current of A is 8.2A and the current of B is 7.8A, the current difference is 0.4A, which indicates a potential performance mismatch problem.

[0161] To calculate the performance influence between modules, it is necessary to collect the physical distance data and the module surface temperature data between adjacent modules. The physical distance is obtained through installation drawings or on-site measurements. The typical module spacing ranges from 10 to 50 centimeters. The module surface temperature is collected once an hour by an infrared thermal imager. The normal working temperature is about 45°C, and the temperature in the hot spot area can reach above 70°C.

[0162] Calculate the heat transfer coefficient between modules according to the physical distance data. The heat transfer coefficient K is inversely proportional to the distance d and can be simplified as K = α / (d + β), where α and β are calibration constants. According to experimental data, α = 150 and β = 5, with the unit of centimeter. For example, for two modules 20 centimeters apart, their heat transfer coefficient K = 150 / (20 + 5) = 6. The temperature gradient between modules is calculated from the surface temperature data. If the temperature of module A is 50°C and the temperature of module B is 45°C, the temperature gradient is 5°C.

[0163] Calculate the initial value of the hot spot effect based on the current difference, thermal conductivity, and temperature gradient. The initial value of the hot spot effect I can be expressed as a function of the current difference multiplied by the thermal conductivity and then multiplied by the temperature gradient. For example, for the above components A and B, the initial value of the hot spot effect I = 0.4×6×5×0.05 = 0.6 (where 0.05 is the proportionality coefficient). Based on the series-parallel connection matrix, establish a hot spot diffusion path tree. This path tree takes the hot spot source component as the root node and expands outward through the connection relationship to form a multi-level tree structure. Each node represents a component, and the connection between nodes represents the transmission path of the thermal influence.

[0164] Perform forward and backward recursive iterative calculations on the initial value of the hot spot effect along the diffusion path tree. The forward recursion starts from the hot spot source and calculates the influence received by each layer of nodes; the backward recursion considers the feedback effect and returns from the leaf nodes to the root node. During each iteration process, the influence value decays at a certain ratio. Usually, the decay coefficient is set to 0.7. After 5 iterations, a stable performance influence coefficient matrix is obtained, which represents the degree of influence of each component on other components.

[0165] Based on the obtained data, calculate three core scores: component degradation score, component interaction score, and installation location score.

[0166] The component degradation score is obtained by multiplying the performance decay rate by the decay weight calibrated based on historical data. The performance decay rate is calculated according to the annual decay rate of the component output power. The normal decay rate of standard silicon-based components is about 0.7% / year. If it exceeds 1.2% / year, it is considered abnormal. For example, if a component has a decay rate of 1.5% / year and a decay weight coefficient of 80, the degradation score is 1.5×80 = 120.

[0167] The component interaction score is obtained by multiplying the performance influence coefficient by the influence weight calibrated based on experimental data. The influence weight is determined according to the component type and application scenario, usually in the range of 50 - 100. If the performance influence coefficient of a certain component is 0.8 and the influence weight is 75, the interaction score is 0.8×75 = 60.

[0168] The installation location score is calculated based on the installation location data of the component. The installation location data includes the row and column coordinates, height, inclination angle, etc. of the component. The location weight considers the maintenance difficulty and replacement cost. For example, the weight of a component at the roof edge is 90, and the weight of a component in the central area is 60. If a component is located at the roof edge and its location data empowerment value is 0.85, the installation location score is 0.85×90 = 76.5.

[0169] The component degradation score, component interaction score, and installation location score are weighted and combined for optimization to obtain the replacement priority score. The weighting coefficients are obtained through training with historical maintenance data, and typical values are: degradation score 0.5, interaction score 0.3, location score 0.2. For the above example, the replacement priority score is 120×0.5 + 60×0.3 + 76.5×0.2 = 60 + 18 + 15.3 = 93.3.

[0170] The scoring results are sorted from high to low to form a component replacement priority list. Components with a score higher than 90 are recommended for priority replacement, components with scores between 75 and 90 need to be monitored closely, and components with scores below 75 can be used normally. Through this scoring system, power plant operation and maintenance personnel can make scientific decisions on the component replacement order, optimize the allocation of maintenance resources, and maximize the overall power generation efficiency of the power plant.

[0171] In this embodiment, a complete scoring system is established by integrating multi-dimensional indicators such as the component performance decay rate, the hot spot influence relationship with adjacent components, and the installation location. The solution not only improves the scientificity and pertinence of component replacement decisions but also effectively identifies key components that have a greater impact on the overall performance of the power plant. Based on the connection relationship and current difference, a hot spot diffusion path is constructed, and the performance influence coefficient is accurately calculated by combining forward and backward recursion methods, enhancing the system's response ability to potential performance risks. The final scoring results significantly improve the operation and maintenance efficiency and resource allocation effect.

[0172] Figure 5 This is a schematic structural diagram of the intelligent identification and location system for photovoltaic component defects based on multi-modal images according to an embodiment of the present invention. As Figure 5 shown, the system includes:

[0173] A first unit for acquiring visible light images and infrared thermal imaging images of the photovoltaic component to be detected, performing preprocessing, extracting the position coordinates and geometric features of the defect area in the preprocessed visible light image, and the position coordinates and temperature features of the temperature abnormal area in the preprocessed infrared thermal imaging image;

[0174] A second unit for performing spatial mapping registration on the position coordinates of the defect area and the position coordinates of the temperature abnormal area to generate a target defect area, and extracting the position information and defect feature data of the target defect area;

[0175] A third unit for calculating the edge texture change rate and temperature gradient change rate in the defect feature data, determining the expansion direction of the target defect area according to the edge texture change rate, determining the expansion speed of the target defect area according to the temperature gradient change rate, constructing a defect expansion vector field based on the expansion direction and expansion speed, and obtaining the position prediction result of the target defect area by solving the propagation equation of the defect expansion vector field;

[0176] A fourth unit is configured to calculate the area change rate of a target defect area according to the position prediction result, determine the performance degradation rate of the photovoltaic module, obtain the string connection relationship and current load distribution data of the photovoltaic module in the power station, calculate the performance influence coefficient of the photovoltaic module on adjacent modules, generate a replacement priority score based on the performance degradation rate and the performance influence coefficient, and output a diagnostic report including defect identification and positioning and component replacement suggestions according to the replacement priority score.

[0177] In a third aspect of the embodiments of the present invention,

[0178] there is provided an electronic device, comprising:

[0179] a processor;

[0180] a memory for storing instructions executable by the processor;

[0181] wherein the processor is configured to call the instructions stored in the memory to execute the method described above.

[0182] In a fourth aspect of the embodiments of the present invention,

[0183] there is provided a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0184] The present invention may be a method, an apparatus, a system, and / or a computer program product. The computer program product may include a computer-readable storage medium, on which computer-readable program instructions for executing various aspects of the present invention are loaded.

[0185] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent identification and positioning method for photovoltaic module defects based on multimodal images, characterized in that, Including: Obtain the visible light image and infrared thermal imaging image of the photovoltaic module to be detected and perform preprocessing, extract the position coordinates and geometric features of the defect area in the preprocessed visible light image, and the position coordinates and temperature features of the temperature anomaly area in the preprocessed infrared thermal imaging image; Perform spatial mapping registration on the position coordinates of the defect area and the position coordinates of the temperature anomaly area to generate a target defect area, and extract the position information and defect feature data of the target defect area; Calculate the edge texture change rate and temperature gradient change rate in the defect feature data, determine the expansion direction of the target defect area according to the edge texture change rate, determine the expansion speed of the target defect area according to the temperature gradient change rate, construct a defect expansion vector field based on the expansion direction and expansion speed, and obtain the position prediction result of the target defect area by solving the propagation equation of the defect expansion vector field; Calculate the area change rate of the target defect area according to the position prediction result, determine the performance attenuation rate of the photovoltaic module, obtain the string connection relationship and current load distribution data of the photovoltaic module in the power station, calculate the performance influence coefficient of the photovoltaic module on adjacent modules, generate a replacement priority score based on the performance attenuation rate and performance influence coefficient, and output a diagnostic report including defect identification and positioning and component replacement suggestions according to the replacement priority score.

2. The method according to claim 1, wherein Extracting the position coordinates and geometric features of the defect area in the preprocessed visible light image, and the position coordinates and temperature features of the temperature anomaly area in the preprocessed infrared thermal imaging image includes: Use the adaptive double-threshold method to perform edge detection on the preprocessed visible light image, calculate the gradient amplitude data and gradient direction data, calculate the morphological gradient value of the preprocessed visible light image based on the gradient amplitude data and gradient direction data, construct an image segmentation function according to the morphological gradient value, and perform region segmentation to obtain the segmentation region; Merge adjacent regions of the segmentation region to obtain the defect area in the preprocessed visible light image, calculate the centroid coordinate value and region area value of the defect area, and obtain the position coordinates and geometric features of the defect area; Calculate the average temperature value and temperature standard deviation value of the normal area in the preprocessed infrared thermal imaging image, construct a temperature anomaly determination function according to the average temperature value and temperature standard deviation value, identify the temperature anomaly points, use the temperature anomaly points as the initial growth points to perform region growth operations, obtain the temperature anomaly area in the preprocessed infrared thermal imaging image, and calculate the centroid coordinate value and region temperature difference value of the temperature anomaly area to obtain the position coordinates and temperature features of the temperature anomaly area.

3. The method according to claim 1, characterized in that, Performing spatial mapping registration on the position coordinates of the defect area and the position coordinates of the temperature anomaly area to generate a target defect area includes: Calculate the first geometric center of the position coordinates of the defect area and the second geometric center of the position coordinates of the temperature anomaly area, and use the coordinate difference between the first geometric center and the second geometric center as the translation reference; Determine the translation transformation component of the spatial transformation matrix according to the translation reference, determine the rotation transformation component of the spatial transformation matrix based on the distribution characteristics of the position coordinates of the defect area and the position coordinates of the temperature anomaly area, and generate a spatial transformation matrix including the translation transformation component and the rotation transformation component; Construct the position deviation between the position coordinates of the defect area and the position coordinates of the temperature anomaly area as an optimization objective function, which includes the translation transformation component and the rotation transformation component of the spatial transformation matrix, and solve the optimization objective function by an iterative optimization method to obtain the optimal spatial transformation matrix that minimizes the position deviation; Apply the optimal spatial transformation matrix to the position coordinates of the defect area to obtain the position coordinates of the defect area after spatial mapping, calculate the overlapping area ratio between the position coordinates of the defect area after spatial mapping and the position coordinates of the temperature anomaly area, and when the overlapping area ratio is greater than the preset overlapping threshold, determine the overlapping area as the target defect area.

4. The method according to claim 1, wherein Calculate the edge texture change rate and the temperature gradient change rate in the defect feature data, determine the expansion direction of the target defect area according to the edge texture change rate, and determine the expansion speed of the target defect area according to the temperature gradient change rate, including: Establish a polar coordinate system with the defect center as the origin, perform spline fitting on the defect boundary point sequence to obtain a boundary curve function, and establish a local coordinate system including a normal vector and a tangent vector at each boundary point; Extract the texture feature parameters in the defect feature data, project the texture feature parameters in the local coordinate system to obtain the normal texture gradient value and the tangential texture gradient value, and linearly combine the normal texture gradient value and the tangential texture gradient value based on the preset texture weight to obtain the edge texture change rate; Extract the temperature field matrix in the defect feature data, calculate the spatial gradient of the temperature field matrix, project the spatial gradient in the local coordinate system to obtain the temperature normal gradient and the temperature tangential gradient, and linearly combine the temperature normal gradient and the temperature tangential gradient based on the preset temperature weight to obtain the temperature gradient change rate; Calculate the components of the edge texture change rate in the tangential and normal directions, calculate the expansion angle based on the ratio of the obtained tangential component and normal component, and combine the expansion angle with the normal vector of the boundary curve function to determine the expansion direction of the target defect area; Project the temperature gradient change rate onto the expansion direction to obtain a projection value, and determine the expansion speed of the target defect area based on the magnitude of the projection value and in combination with the material properties of the target defect area.

5. The method according to claim 1, wherein Construct a defect expansion vector field based on the expansion direction and the expansion speed, and obtain the position prediction result of the target defect area by solving the propagation equation of the defect expansion vector field, including: Use the expansion direction as the direction component and the expansion speed as the amplitude component to construct a defect expansion vector field, and perform spatial weight smoothing processing on the defect expansion vector field to obtain a smoothed defect expansion vector field; Construct an initial level set function representing the boundary of the target defect area, make the boundary of the target defect area correspond to the zero value position of the initial level set function, and the initial level set function takes a negative value inside the target defect area and a positive value outside the target defect area; An inner product operation is performed on the smoothed defect expansion vector field and the gradient of the initial level set function to construct the driving term of the propagation equation, and the propagation equation of the initial level set function is established; The propagation equation is discretized in time to obtain a time iteration expression, and the time iteration expression is discretized in space according to the local direction of the smoothed defect expansion vector field to construct an iterative equation set; The iterative equation set is solved to obtain the level set function at the prediction time, the zero-value position of the level set function at the prediction time is extracted to obtain the boundary of the target defect region at the prediction time, and the position prediction result of the target defect region is obtained according to the boundary of the target defect region at the prediction time.

6. The method according to claim 1, characterized in that, Calculating the area change rate of the target defect region according to the position prediction result, and determining the performance attenuation rate of the photovoltaic module includes: Defect boundary images at different times are obtained based on the position prediction result, the defect boundary images are constructed into time series data, forward difference and backward difference calculations are respectively performed on the time series data by using the central difference method, and the area change rate of the target defect region is obtained through numerical recursive iteration optimization; The temperature distribution data and stress distribution data of the target defect region are collected, the hot spot characteristics of the target defect region are judged according to the temperature gradient and temperature extreme values in the temperature distribution data, the crack characteristics of the target defect region are judged according to the stress concentration degree and stress extreme values in the stress distribution data, the area change rate is respectively combined with the hot spot characteristics and crack characteristics to construct a power loss nonlinear mapping relationship, the power loss value is calculated based on the power loss nonlinear mapping relationship, and the performance attenuation rate of the photovoltaic module is obtained by performing a continuous-time derivative operation on the power loss value.

7. The method according to claim 1, wherein Obtaining the string connection relationship and current load distribution data of the photovoltaic module in the power station, calculating the performance influence coefficient of the photovoltaic module on adjacent modules, and generating a replacement priority score based on the performance attenuation rate and the performance influence coefficient includes: A string connection relationship data of the photovoltaic module in the power station is collected by using a string topology detector, a current load distribution data in the string is collected by using a current collector, a series-parallel connection matrix of the photovoltaic module and adjacent modules is established according to the string connection relationship data, and a current difference between the photovoltaic module and adjacent modules is calculated according to the current load distribution data; The physical distance data and the component surface temperature data between adjacent components are collected, the heat conduction coefficient between the components is calculated according to the physical distance data, the temperature gradient between the components is calculated according to the component surface temperature data, the initial value of the hot spot influence of the adjacent components is calculated according to the current difference, the heat conduction coefficient and the temperature gradient, a hot spot diffusion path tree is established according to the series-parallel connection matrix, and the forward and backward recursive iteration calculations are respectively performed on the initial value of the hot spot influence along the diffusion path tree to obtain the performance influence coefficient of the photovoltaic module on adjacent modules; Multiply the performance degradation rate by the degradation weight calibrated based on historical data to obtain the component degradation score, multiply the performance impact coefficient by the impact weight calibrated based on experimental data to obtain the component interaction score, collect the installation location data of the photovoltaic module, and multiply the installation location data by the location weight calibrated based on maintenance experience to obtain the installation location score; Perform weighted combination optimization on the component degradation score, component interaction score, and installation location score to obtain the replacement priority score.

8. A photovoltaic module defect intelligent recognition and positioning system based on multimodal images, which is used to implement the method described in any one of the foregoing claims 1-7, and is characterized in that, Including: The first unit is used to obtain the visible light image and infrared thermal imaging image of the photovoltaic module to be detected and perform preprocessing, extract the position coordinates and geometric features of the defect area in the preprocessed visible light image, and the position coordinates and temperature features of the temperature anomaly area in the preprocessed infrared thermal imaging image; The second unit is used to perform spatial mapping registration on the position coordinates of the defect area and the position coordinates of the temperature anomaly area to generate a target defect area, and extract the position information and defect feature data of the target defect area; The third unit is used to calculate the edge texture change rate and temperature gradient change rate in the defect feature data, determine the expansion direction of the target defect area according to the edge texture change rate, determine the expansion speed of the target defect area according to the temperature gradient change rate, construct a defect expansion vector field based on the expansion direction and expansion speed, and obtain the position prediction result of the target defect area by solving the propagation equation of the defect expansion vector field; The fourth unit is used to calculate the area change rate of the target defect area according to the position prediction result, determine the performance degradation rate of the photovoltaic module, obtain the string connection relationship and current load distribution data of the photovoltaic module in the power station, calculate the performance impact coefficient of the photovoltaic module on adjacent modules, generate a replacement priority score based on the performance degradation rate and performance impact coefficient, and output a diagnostic report including defect identification and location and component replacement suggestions according to the replacement priority score.

9. An electronic device, characterized in that, Including: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, the method according to any one of claims 1 to 7 is implemented.

Images