Photovoltaic power station abnormity monitoring method and system based on unmanned aerial vehicle image

Abstract

The invention discloses a photovoltaic power station abnormity monitoring method and system based on unmanned aerial vehicle images, and relates to the technical field of photovoltaic power station monitoring. The invention discloses a photovoltaic power station abnormity monitoring system based on unmanned aerial vehicle images. The system comprises an image acquisition module, a temperature rise propagation module, a collaborative analysis module, a cross-cycle registration module, a miniature variable analysis module and a fusion diagnosis module. Component-level, structure-level and sub-pixel-level joint registration is carried out on compensation registration images and night structure images of different inspection periods, the consistent corresponding relation of photovoltaic modules in multiple periods is constructed, sub-pixel quantization is carried out on frames, grid lines and surface geometric structures, and a multi-period micro-deformation quantity and structure health index time sequence is obtained. And the fine monitoring of the micro-deformation trend of the photovoltaic module is realized.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power plant monitoring technology, and in particular to a method and system for anomaly monitoring of photovoltaic power plants based on UAV imagery. Background Technology

[0002] The large-scale construction of photovoltaic (PV) power plants has led to a rapid increase in the demand for module condition inspection. Drones equipped with visible light and infrared imaging equipment have become a mainstream technology in PV power plant operation and maintenance. Drone imagery can cover large areas of PV arrays in a short time, collecting surface texture, structural information, and temperature distribution of the modules, providing basic data for identifying anomalies such as hot spots, microcracks, solder joint detachment, and shading. As the number of PV modules continues to increase and the array structure becomes increasingly complex, utilizing drone imagery for efficient and reliable automated anomaly detection has become an important direction for intelligent operation and maintenance.

[0003] Current photovoltaic module anomaly detection technologies based on UAV imagery still have several shortcomings. On the one hand, visible light and infrared images exhibit significant spatial discrepancies due to factors such as shooting angle, lighting conditions, and UAV attitude disturbances, making stable cross-modal registration difficult and resulting in a mismatch between temperature anomalies and visible structures. On the other hand, existing infrared detection is typically limited to single-moment analysis, making it difficult to distinguish between instantaneous temperature rises caused by shading and chain-like temperature rise propagation caused by electrical faults. It also fails to reveal the subtle deformation trends of the module structure over time, making it difficult to detect cross-module thermal crosstalk, long-term degradation, and potential structural risks in a timely manner. Furthermore, significant differences in environment and lighting between different inspection cycles make it difficult to align traditional cross-cycle comparison methods, thus limiting the ability to diagnose the long-term health status of modules.

[0004] Under conditions of attitude disturbances, illumination differences, and slight changes in component structure, achieving high-precision unified registration of visible light-infrared images from UAVs, and reliably distinguishing electrical anomalies, shading anomalies, and cross-component thermal crosstalk based on this, thereby realizing the time-series structural health monitoring of photovoltaic modules, is currently a key technical problem. Summary of the Invention

[0005] This invention proposes a photovoltaic power plant anomaly monitoring method based on UAV imagery. By constructing a unified reference board, multi-band compensation registration, nighttime structural stabilization processing, current and temperature conduction model, cross-cycle sub-pixel structure alignment, and multi-source evidence chain fusion diagnosis, it achieves comprehensive intelligent detection of photovoltaic module thermal anomalies, structural anomalies, and cross-module anomalies.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for anomaly monitoring of photovoltaic power plants based on UAV imagery includes: Visible light and infrared images of photovoltaic modules are acquired and projected onto a virtual reference board to obtain a reference image; multi-band registration is performed on the reference image, and local distortion correction under attitude disturbance is performed to obtain a compensated registration image; under nighttime inspection conditions, structural extraction and temperature reference calibration are performed on the infrared image to obtain a nighttime structural image. Based on the electrical series topology of photovoltaic modules, a current-temperature conduction model is constructed, and the temperature rise propagation characteristics are quantified according to the temperature spatial distribution extracted from the compensated registration image. Infrared and visible light co-analysis was performed on the compensated registration images to obtain co-anomaly features, and cross-component thermal crosstalk anomalies were identified by combining temperature rise propagation features. Cross-cycle registration is performed on compensated registration images and nighttime structure images from different inspection cycles to extract sub-pixel structural features of the frame and grid lines of photovoltaic modules. Micro-deformation of photovoltaic modules is detected based on sub-pixel structural features, structural health index is calculated, and micro-deformation trend results are obtained based on the time change of structural health index. Based on thermal crosstalk anomalies, micro-deformation trend results, and collaborative anomaly characteristics, a multi-source evidence chain fusion model is constructed to perform fusion diagnosis on the anomaly categories and levels of photovoltaic modules, output diagnostic confidence, generate operation and maintenance decisions based on the fusion diagnosis results, and adjust subsequent inspection strategies.

[0007] As a preferred technical solution of the present invention, obtaining the compensated registration image includes: constructing a virtual reference board based on the geographic coordinate system of the photovoltaic power station; mapping the real-time tracking angle of the photovoltaic module during shooting and the shooting pose parameters of the UAV onto the virtual reference board; performing viewpoint normalization processing on the visible light image and the infrared image to obtain a virtual reference image; adaptively selecting a combination of registration features from the module texture features, grid line features, and module edge contour features according to the module type, surface cleanliness, and shooting illumination conditions of the photovoltaic module; performing multi-band feature matching and geometric transformation on the reference image to obtain an initial registration image; constructing a distortion field to characterize local distortion based on the attitude disturbance information of the UAV during flight; and performing distortion correction on the local nonlinear geometric distortion caused by attitude changes in the initial registration image to obtain a compensated registration image; the compensated registration image includes a visible light compensated registration image and an infrared compensated registration image.

[0008] As a preferred technical solution of the present invention, obtaining the nighttime structure image includes: extracting the edge contour of the photovoltaic module, the temperature change line of the grid line, and the temperature feature points of the internal structure of the module based on the temperature gradient distribution of the nighttime infrared image to obtain the nighttime infrared structure features; constructing a nighttime temperature reference model based on the ambient temperature, the ground background temperature, and the average reference temperature of the photovoltaic module at the inspection time; calibrating the nighttime infrared structure features by temperature calibration; and performing structure stabilization processing on the temperature-calibrated nighttime infrared structure features, including structural continuity enhancement and local noise suppression, to obtain the nighttime structure image.

[0009] As a preferred embodiment of the present invention, the quantification of temperature rise propagation characteristics includes: constructing a current-temperature conduction model based on the connection sequence of photovoltaic modules in an electrical series branch, determining the current conduction path from upstream to downstream in the series branch and its impact on the module temperature distribution; extracting abnormal heat sources caused by local electrical faults based on the temperature spatial distribution in infrared compensated registration images, and calculating the propagation direction and temperature rise attenuation mode of temperature between upstream and downstream modules based on the current-temperature conduction model; and quantifying the temperature rise propagation characteristics of abnormal heat sources in the series branch based on the propagation direction, temperature rise amplitude, and propagation attenuation law.

[0010] As a preferred embodiment of the present invention, obtaining the collaborative anomaly features includes: extracting temperature anomaly regions on the surface of the photovoltaic module based on the temperature spatial distribution of the infrared compensated registration image, and obtaining temperature anomaly features including spatial distribution and temperature gradient; extracting spectral reflectance features, brightness variation features, and local shading features of the photovoltaic module surface from the visible light compensated registration image based on visible light reflectance characteristics to identify shading regions; performing collaborative matching between the temperature anomaly features and spectral reflectance features, and comparing the correspondence between temperature anomaly regions and shading regions to distinguish between shading-related temperature rise and electrical temperature rise; and generating collaborative anomaly features characterizing the thermal anomaly attributes of the photovoltaic module based on the collaborative matching results.

[0011] As a preferred embodiment of the present invention, the identification of cross-module thermal crosstalk anomalies includes: determining the propagation direction and propagation path of temperature rise in the series branch of photovoltaic modules based on temperature rise propagation characteristics; determining whether the temperature anomaly is independent of the surface shading area based on cooperative anomaly characteristics; identifying whether the temperature anomaly crosses the module boundary of adjacent photovoltaic modules in the propagation path of temperature rise, and determining whether the temperature rise change conforms to the propagation law of the current-temperature conduction model; when the temperature anomaly simultaneously satisfies cross-module propagation, continuous temperature rise propagation mode and independence from surface shading, determining that there is a cross-module thermal crosstalk anomaly, and outputting the corresponding anomaly influence range.

[0012] As a preferred embodiment of the present invention, the cross-cycle registration includes: performing component-level matching on the compensated registration images for different inspection cycles based on the component border features and grid line geometric features in the compensated registration images, and establishing an initial correspondence of the same photovoltaic component in different inspection cycles; performing structural consistency calibration on the initial correspondence based on the component outline and structural boundary features in the nighttime structural images to eliminate cross-cycle structural offsets caused by differences in illumination conditions, temperature conditions, and local shading; and performing sub-pixel-level fine alignment on the cross-cycle images based on the joint constraints of component border features, grid line geometric features, and component internal geometric features to obtain cross-cycle registration images for time-series analysis.

[0013] As a preferred technical solution of the present invention, obtaining the micro-deformation trend results includes: based on cross-period registered images, performing sub-pixel-level parameter quantization on the frame line segments, grid line straightness, and surface geometry of the photovoltaic module to obtain a multi-period structural feature sequence; based on the structural feature sequence, calculating the micro-deformation variables of the photovoltaic module in different inspection cycles, including frame offset, grid line curvature change, and local warping of the module surface; constructing a structural health index of the photovoltaic module based on the micro-deformation variables, and performing time series modeling on the structural health index to obtain the time series trend of the photovoltaic module structural health index; and based on the time series trend, determining the micro-deformation trend results of the photovoltaic module, including the development direction, rate of change, and trend stability of the micro-deformation.

[0014] As a preferred technical solution of the present invention, the fusion diagnosis includes: constructing a multi-source feature set containing temperature features, structural features, and spectral features based on thermal crosstalk anomalies, micro-deformation trend results, and collaborative anomaly features; inputting the multi-source feature set into a multi-source evidence chain fusion model, and obtaining the anomaly category and anomaly level of the photovoltaic module according to the credibility, correlation, and consistency of different feature sources; and calculating the confidence level of the anomaly diagnosis based on the fusion output of the multi-source evidence chain fusion model.

[0015] A photovoltaic power plant anomaly monitoring system based on UAV imagery includes: Image acquisition module: Acquires visible light and infrared images of photovoltaic modules and projects them onto a virtual reference board to obtain reference images; performs multi-band registration on the reference images and performs local distortion correction under attitude disturbance to obtain compensated registration images; under night inspection conditions, performs structure extraction and temperature reference calibration on infrared images to obtain nighttime structure images; Temperature rise propagation module: Based on the electrical series topology of photovoltaic modules in photovoltaic power plants, a current temperature conduction model is constructed, and the temperature rise propagation characteristics are quantified according to the temperature spatial distribution in the compensated registration image; Collaborative Analysis Module: Performs infrared and visible light collaborative analysis on the compensated registration images to obtain collaborative anomaly features, and identifies cross-component thermal crosstalk anomalies by combining temperature rise propagation features. Cross-cycle registration module: Performs cross-cycle registration on compensated registration images and nighttime structure images from different inspection cycles to extract sub-pixel structural features of the frame and grid lines of photovoltaic modules; Micro-deformation analysis module: Detects micro-deformation of photovoltaic modules based on sub-pixel structural features, calculates the structural health index, and obtains micro-deformation trend results based on the time change of the structural health index; Fusion Diagnosis Module: Based on thermal crosstalk anomalies, micro-deformation trend results, and collaborative anomaly characteristics, a multi-source evidence chain fusion model is constructed to perform fusion diagnosis on the anomaly categories and levels of photovoltaic modules, output diagnostic confidence, generate operation and maintenance decisions based on the fusion diagnosis results, and adjust subsequent inspection strategies.

[0016] The present invention has the following advantages: This invention constructs a virtual reference board based on the geographic coordinate system of a photovoltaic power station, and performs viewpoint normalization, multi-band feature matching, and local distortion correction under attitude disturbances on visible light and infrared images. This significantly improves the spatial registration accuracy of multi-source images and ensures accurate correspondence between temperature information and structural information in the same coordinate system. By performing temperature calibration and structural stabilization processing on nighttime infrared structural features under nighttime inspection conditions, nighttime structural images with good continuity and noise resistance are obtained, improving the ability to identify component geometry and potential anomalies in complex environments.

[0017] This invention constructs a current-temperature conduction model based on the electrical series topology of photovoltaic modules, extracts abnormal heat sources caused by local electrical faults by utilizing the temperature spatial distribution in infrared compensated registration images, and quantifies the propagation direction and attenuation law of temperature rise between upstream and downstream modules, thus achieving a quantitative characterization of temperature rise propagation characteristics. By co-matching temperature anomaly characteristics with shading characteristics, it achieves an effective distinction between shading-related temperature rise and electrical temperature rise.

[0018] This invention combines temperature rise propagation characteristics with cooperative anomaly characteristics to determine whether a temperature anomaly crosses the boundary of adjacent components and whether the propagation mode conforms to the current-temperature conduction law. Under the premise of excluding surface shading interference, it identifies cross-component thermal crosstalk anomalies and accurately determines the scope of the anomaly and the associated components.

[0019] By jointly registering compensated registration images from different inspection cycles with nighttime structural images at the component, structural, and sub-pixel levels, a consistent correspondence between photovoltaic modules in multiple cycles is constructed. Sub-pixel quantization is performed on the frame, grid lines, and surface geometry to obtain time series of micro-deformation variables and structural health indices in multiple cycles, thus achieving refined monitoring of micro-deformation trends of photovoltaic modules.

[0020] By constructing a multi-source feature set including temperature, structural, and spectral features, and inputting thermal crosstalk anomaly results, micro-deformation trend results, and collaborative anomaly features into a multi-source evidence chain fusion model, and performing fusion analysis based on the credibility, correlation, and consistency of feature sources, the abnormality category, anomaly level, and diagnostic confidence of photovoltaic modules are obtained, realizing comprehensive intelligent diagnosis and hierarchical assessment of multiple types of anomalies. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are only schematic diagrams of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort. Figure 1 This is a schematic diagram of the structure of a photovoltaic power station anomaly monitoring system based on UAV imagery, used in an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0023] Example 1: A method for anomaly monitoring of photovoltaic power plants based on UAV imagery, comprising the following steps: Step S1: Acquire visible light and infrared images of photovoltaic modules and project them onto a virtual reference board to obtain a reference image; perform multi-band registration on the reference image and perform local distortion correction under attitude disturbance to obtain a compensated registration image; under nighttime inspection conditions, perform structure extraction and temperature reference calibration on the infrared image to obtain a nighttime structure image. In this embodiment, the geographical location information of each photovoltaic module within the photovoltaic power station is uniformly calibrated based on the power station design drawings and on-site survey data to obtain the planar coordinates and installation height of each module in the power station's geographic coordinate system. The UAV is equipped with a visible light camera and an infrared camera, both of which are jointly calibrated to obtain their respective intrinsic and relative extrinsic parameters. When performing inspection tasks, the UAV collects three-dimensional position and attitude angle information through a global satellite positioning module and an inertial measurement unit, and records a timestamp and corresponding pose parameters for each frame of visible light and infrared imagery. Visible light images are stored in color image format with a resolution, for example, set to 4000×3000 pixels; infrared images are stored in temperature matrix format, with each pixel corresponding to an apparent temperature value obtained through camera calibration. By ensuring that both types of images are acquired synchronously under the same inspection route and altitude conditions, the data source in subsequent projection and registration steps has a consistent spatial reference basis.

[0024] The process of obtaining the compensated registration image includes: constructing a virtual reference board based on the geographic coordinate system of the photovoltaic power station; mapping the real-time tracking angle of the photovoltaic module during shooting to the shooting pose parameters of the UAV onto the virtual reference board; performing viewpoint normalization processing on the visible light image and the infrared image to obtain a virtual reference image; adaptively selecting a combination of registration features from the module texture features, grid line features, and module edge contour features according to the module type, surface cleanliness, and shooting lighting conditions; performing multi-band feature matching and geometric transformation on the reference image to obtain an initial registration image; constructing a distortion field to characterize local distortion based on the attitude disturbance information of the UAV during flight; and performing distortion correction on the local nonlinear geometric distortion caused by attitude changes in the initial registration image to obtain a compensated registration image; the compensated registration image includes a visible light compensated registration image and an infrared compensated registration image.

[0025] In this embodiment, the "virtual reference plane" refers to a virtual reference plane constructed in the geographic coordinate system of the photovoltaic power station. This reference plane is parallel to or substantially coincides with the average installation plane of the photovoltaic modules, and is used to uniformly map images from different shooting perspectives to the same plane coordinate system. The virtual reference plane is divided into regular grids with a fixed resolution. Each grid cell corresponds to an area cell in the geographic coordinate system of the power station, and thus corresponds to a pixel position in the compensated registration image. By utilizing the UAV pose parameters and camera imaging model, pixels in the visible light image and infrared image are back-projected one by one into three-dimensional space, and then projected onto the virtual reference plane to achieve the construction of a reference image after perspective normalization.

[0026] In the adaptive selection of registration feature combinations, this embodiment quantitatively evaluates the image quality of candidate feature regions to ensure that the features used for registration have sufficient discriminativeness and stability. Specifically, in component texture features, regions with rich gray-level variations and high detail information are selected as texture feature point sets by calculating the gray-level variance and local information entropy of candidate regions; in grid line features, grid line segments with obvious edges and prominent linear structures are selected by calculating the contrast and edge intensity along and perpendicular to the grid line direction; in component edge contour features, component bounding boxes are extracted using edge detection operators, and their continuity and straightness are evaluated. For different component types, surface cleanliness, and lighting conditions, a set of quality evaluation thresholds and weights are pre-set, and indicators such as image entropy, contrast, and edge intensity are combined into a feature scoring function. Based on the score, the optimal or suboptimal combination among texture features, grid line features, and edge contour features is selected, so that the feature point set used for multi-band feature matching has high robustness under the current inspection conditions.

[0027] When performing multi-band feature matching and geometric transformation based on the aforementioned adaptive feature combination, this embodiment uses the virtual reference plate coordinate system as a unified reference. It estimates the geometric mapping relationship between visible light and infrared images through feature point correspondence, including affine transformation or higher-order homography transformation, transforming both types of images onto the virtual reference plate to obtain initial registered images. Because the UAV is affected by factors such as wind disturbances and motor speed changes during flight, its attitude still undergoes short-term, small-amplitude changes, resulting in nonlinear deformation in local areas of the initial registered images. Therefore, this embodiment constructs a "torsion field" based on the high-frequency attitude disturbance sequence recorded by the UAV flight. This torsion field describes the local geometric offset caused by attitude changes at various positions on the virtual reference plate within a short timescale. By interpolating the torsion field on the virtual reference plate grid, the displacement vector of each pixel position is obtained, and inverse compensation is performed on the corresponding pixels in the initial registered images, thereby performing distortion correction on the local nonlinear geometric distortion, ultimately obtaining visible light compensated registered images and infrared compensated registered images.

[0028] The process of obtaining the nighttime structure image includes: extracting the edge contours of the photovoltaic module, the temperature abrupt change lines of the grid lines, and the temperature feature points of the internal structure of the module based on the temperature gradient distribution of the nighttime infrared image to obtain the nighttime infrared structure features; constructing a nighttime temperature reference model based on the ambient temperature, the ground background temperature, and the average reference temperature of the photovoltaic module at the time of inspection; calibrating the nighttime infrared structure features by temperature calibration; and performing structure stabilization processing on the temperature-calibrated nighttime infrared structure features, including enhancing structural continuity and suppressing local noise, to obtain the nighttime structure image.

[0029] In this embodiment, nighttime infrared images are acquired under conditions of minimal light impact. The surface temperature distribution of the components is mainly affected by factors such as ambient temperature, component self-heating, and sky radiation. Temperature gradient analysis of the nighttime infrared images reveals abrupt or step-like temperature change rates at component edges and along the grid lines. This embodiment searches for sets of pixels with large gradient amplitudes in the image according to a preset gradient threshold. The gradient direction is combined with the geometric location to identify the component edge contours and grid line temperature abrupt change lines. Within the component's internal region, temperature feature points representing internal structural zones are extracted by detecting local temperature extreme points and their neighborhood distribution, forming nighttime infrared structural features.

[0030] The nighttime temperature baseline model is constructed based on three types of data: ambient temperature, surface background temperature, and average baseline temperature of photovoltaic modules. Ambient temperature is measured by on-site meteorological sensors; surface background temperature is obtained from temperature statistics of the area surrounding the modules in infrared imagery; and the average baseline temperature of the photovoltaic modules is calculated from the average or weighted average temperature within the corresponding area for each module. This model performs offset and normalization processing on the temperature values ​​in the nighttime infrared structural features, ensuring the comparability of nighttime infrared data obtained at different inspection times and under different environmental conditions.

[0031] In the structural stabilization process, this embodiment addresses the low signal-to-noise ratio problem in nighttime infrared images by introducing geometric continuity constraints on the edge contours and grating structures to suppress isolated noise points. For edge breaks, edge connectivity is restored through interpolation and connection along the main structural direction, thereby enhancing the integrity of the component contours and grating structures. Simultaneously, within the component's internal region, neighborhood consistency is used for temperature feature points, eliminating isolated points significantly inconsistent with the surrounding structure and retaining stable feature points reflecting the boundaries of the component's internal structure. After temperature calibration and structural stabilization, the resulting nighttime structural image exhibits high geometric stability and temporal consistency in terms of component contours, grating distribution, and internal structural hierarchy, providing a reliable structural benchmark for subsequent cross-period registration and micro-deformation analysis.

[0032] Step S2: Based on the electrical series topology of photovoltaic modules, construct a current-temperature conduction model, and quantify the temperature rise propagation characteristics according to the temperature spatial distribution extracted from the compensated registration image; In this embodiment, the electrical connection relationships of each component within the photovoltaic power station are provided by the power station design documents and combiner box wiring records to obtain the component sequence information in each series branch. For each series branch, the components are numbered from upstream to downstream according to their physical arrangement and electrical connection, and these numbers are mapped to the component positions in the virtual reference board coordinate system, establishing a one-to-one correspondence between the electrical topology and the image space. The temperature spatial distribution in the compensated registration image comes from the infrared compensated registration image in step S1. Each pixel position represents the temperature value on the surface of the photovoltaic module. The temperature matrix undergoes viewpoint normalization and distortion correction to ensure that the temperature data is aligned in the component coordinate system.

[0033] The quantification of temperature rise propagation characteristics includes: constructing a current-temperature conduction model based on the connection sequence of photovoltaic modules in the electrical series branch, determining the current conduction path from upstream to downstream in the series branch and its impact on the module temperature distribution; extracting abnormal heat sources caused by local electrical faults based on the temperature spatial distribution in the infrared compensated registration image, and calculating the propagation direction and temperature rise attenuation mode of temperature between the upstream and downstream modules based on the current-temperature conduction model; and quantifying the temperature rise propagation characteristics of abnormal heat sources in the series branch based on the propagation direction, temperature rise amplitude, and propagation attenuation law.

[0034] In this embodiment, the "current-temperature conduction model" refers to a simplified physical model used to describe the relationship between power loss and temperature rise caused by current conduction from upstream to downstream components in a series branch. The current-temperature conduction model is based on the characteristic of current consistency in a series circuit, meaning that all components in the same branch flow with the same current during operation. In this embodiment, the thermal effect generated by the current is characterized as an equivalent parameter related to the component's series resistance and power loss. Specifically, for component i, its temperature rise is positively correlated with the branch current and the component's own equivalent series resistance. The resistance value can be obtained by combining factors such as cell material, electrode structure, and aging degree, and estimated using existing periodic IV characteristic test records and component rated parameters from the power plant. This model does not require the construction of complex differential equations but uses a sequential propagation method to describe the trend of temperature rise. That is, if an anomaly occurs in component k causing a local temperature rise, this temperature rise gradually decreases in the direction of propagation among electrically adjacent components. The degree of attenuation is determined by the difference in equivalent resistance, heat dissipation conditions, and the average temperature of the neighborhood.

[0035] In the extraction of abnormal heat sources, this embodiment identifies clusters of temperatures significantly higher than the average reference temperature of the components by performing local peak detection on the temperature matrix in the infrared compensated registration image. Each abnormal region is considered a potential "abnormal heat source" by calculating its maximum temperature rise, area, and gradient shape. The location of the abnormal source is recorded in a virtual reference board coordinate system and mapped back to the component number, giving it a clear upstream and downstream relationship in the electrical topology.

[0036] Based on the current-temperature conduction model, this embodiment calculates the propagation characteristics of abnormal heat sources along the series branch direction. For the component containing the abnormal source, numbered m, the temperature rise difference sequence is calculated both upstream (numbered m-1, m-2…) and downstream (numbered m+1, m+2…) to obtain the direction of temperature rise propagation. If the temperature rise continuously decreases downstream, it indicates that the temperature rise mainly propagates downstream; if attenuation occurs both upstream and downstream, the attenuation rate and the location of the abnormal source are combined to determine that the heat source originates from a local fault in the middle. This method avoids complex temperature distribution calculations and constructs a temperature rise propagation mode through the difference sequence and attenuation gradient.

[0037] In quantifying the characteristics of temperature rise propagation, this embodiment records three key indicators: propagation direction, propagation path, and propagation attenuation law. The propagation direction is represented as "upstream → downstream" or "downstream → upstream"; the propagation path is recorded as a sequence of component numbers, for example (a group of three components, numbered 12→13→14) indicating that the temperature rise propagates along this path; the propagation attenuation law is obtained based on the temperature rise difference between each component in the sequence and the average temperature difference of the neighborhood, used to describe the trend of temperature rise during propagation, such as a linear decrease, stepwise attenuation, or a local plateau-like change. These three types of indicators together constitute the characteristics of temperature rise propagation, providing data for cross-component thermal crosstalk identification in the subsequent step S3.

[0038] Step S3: Perform infrared and visible light co-analysis on the compensated registration image to obtain co-anomaly features, and identify cross-component thermal crosstalk anomalies by combining temperature rise propagation features. In this embodiment, both the infrared-compensated registration image and the visible light-compensated registration image originate from the registration result of step S1. The two types of images have been spatially aligned in the virtual reference plate coordinate system, so that each pixel simultaneously possesses temperature information and visible light reflection information. Through this multimodal alignment, the temperature anomalies, occlusion conditions, and spectral reflectance characteristics of the component surface are compared and analyzed based on the same coordinate system, providing a reliable data foundation for cross-component anomaly identification.

[0039] The obtained collaborative anomaly features include: extracting temperature anomaly regions on the surface of photovoltaic modules based on the temperature spatial distribution of infrared compensated registration images, and obtaining temperature anomaly features including spatial distribution and temperature gradient; extracting spectral reflectance features, brightness change features, and local shading features of the photovoltaic module surface from visible light compensated registration images based on visible light reflectance characteristics to identify shading regions; performing collaborative matching between temperature anomaly features and spectral reflectance features, and comparing the correspondence between temperature anomaly regions and shading regions to distinguish between shading-related temperature rise and electrical temperature rise; and generating collaborative anomaly features characterizing the thermal anomaly attributes of photovoltaic modules based on the collaborative matching results.

[0040] In the extraction of temperature anomaly regions, this embodiment utilizes a temperature matrix to perform local extremum search on the compensated registration image to identify temperature rise regions exceeding the component's average reference temperature. For each potential anomaly region, its area, maximum temperature rise, temperature gradient direction, and gradient magnitude are calculated to form a corresponding temperature anomaly feature description. The temperature gradient direction is represented by information pointing towards the temperature rise center, used to identify the concentration of local hotspots; the gradient magnitude reflects the intensity of the temperature rise. The locations of all temperature anomaly regions are recorded using a virtual reference plate coordinate system and mapped to specific component numbers.

[0041] In the visible light compensated registration image, this embodiment extracts the spectral reflectance features of the photovoltaic module surface based on the reflectance characteristics and brightness structure of the visible light channel. The spectral reflectance features are obtained by analyzing the light reflection, color changes, and material consistency of the module surface. Among these, reflectance changes can reflect the presence of dust deposits, bird droppings, or localized shading on the module surface. To identify shading patterns, this embodiment further analyzes the brightness change features, calculating the average brightness, brightness gradient, and texture consistency of candidate regions. When a region exhibits high brightness differences and weakened texture information, that region is marked as a shading or contaminated area. Furthermore, large-area dust shading or spot contamination can be identified through anomalous changes in color distribution. The coordinates of all shading areas are also recorded based on a virtual reference board for spatial comparison with areas of temperature anomalies.

[0042] "Shading-induced temperature rise" refers to the temperature increase in the shaded area caused by uneven local illumination due to external shading (such as dust, bird droppings, or debris). "Electrical temperature rise" refers to the temperature rise caused by abnormal current due to internal defects in the module (such as microcracks, solder strip detachment, or cell damage). The former is usually accompanied by shading or brightness abnormalities in visible light images, while the latter usually occurs in areas without obvious signs of shading and is related to the temperature rise propagation characteristics in the electrical topology. This embodiment performs spatial overlap analysis on the temperature anomaly area and the shading area: when the temperature anomaly area and the shading area highly overlap in space and the temperature rise gradient is weak, it is labeled as shading-induced temperature rise; when the temperature anomaly area and the shading area do not overlap or the gradient is significant, and the abnormal behavior is consistent with the texture of the module material, it is labeled as electrical temperature rise. The result of the collaborative matching consists of a set of labels and parameters used to record the type, morphology, and corresponding visible light characteristics of the temperature anomaly, thereby forming collaborative anomaly characteristics.

[0043] The identification of cross-module thermal crosstalk anomalies includes: determining the propagation direction and propagation path of temperature rise in the series branch of photovoltaic modules based on temperature rise propagation characteristics; determining whether the temperature anomaly is independent of the surface shading area based on cooperative anomaly characteristics; identifying whether the temperature anomaly crosses the module boundary of adjacent photovoltaic modules in the temperature rise propagation path, and determining whether the temperature rise change conforms to the propagation law of the current-temperature conduction model; when the temperature anomaly simultaneously satisfies cross-module propagation, continuous temperature rise propagation mode and independence from surface shading, it is determined that there is a cross-module thermal crosstalk anomaly, and the corresponding anomaly impact range is output.

[0044] In this embodiment, "cross-component thermal crosstalk anomaly" refers to the temperature rise caused by an electrical hot spot inside a component, which propagates continuously along the current path of the series circuit between adjacent components, resulting in a series-type temperature rise characteristic in multiple components. This type of anomaly is usually consistent with the current-temperature conduction model, that is, the temperature rise gradually decreases along the electrical topology direction, while its propagation mode shows continuity in space along the component arrangement direction.

[0045] During the identification process, the temperature rise propagation path obtained in step S2 is first used to determine whether the temperature rise propagates along a component sequence with the same direction. For example, the propagation path can be represented as "component 12 → component 13 → component 14". Secondly, combined with cooperative anomaly characteristics, it is confirmed that the temperature anomaly on the propagation path is not due to obstruction, i.e., no obvious visible light obstruction features appear in the propagation area, thus eliminating false judgments of temperature anomalies caused by surface contamination. Subsequently, a stability analysis is performed on the temperature rise value in the propagation path. When the temperature rise change exhibits a monotonically decaying or slowly decaying mode consistent with the current-temperature conduction model and crosses at least two adjacent component boundaries, the propagation path is determined to be a valid cross-component propagation path.

[0046] "Continuous temperature rise propagation pattern" refers to the temperature rise caused by internal electrical anomalies in photovoltaic modules, which propagates sequentially between adjacent modules along the electrical connections within the electrical series circuit of the photovoltaic modules. During propagation, it exhibits consistent direction, spatial adjacency, and consistent temperature rise variation patterns. In practical implementation, the continuity of the temperature rise propagation pattern is determined by jointly judging the directional consistency, spatial adjacency, and changing trend of the temperature rise difference sequence between adjacent modules.

[0047] This embodiment records the impact range of cross-component thermal crosstalk anomalies, outputs the component numbers involved in the propagation as the impact range of the anomalies, and provides evidence signals of electrical anomalies for the subsequent fusion diagnosis in step S6.

[0048] Step S4: Perform cross-cycle registration on the compensated registration images and nighttime structure images of different inspection cycles to extract the sub-pixel structure features of the frame and grid lines of the photovoltaic modules. In this embodiment, different inspection cycles correspond to different dates and weather conditions. Each cycle includes visible light compensated registration images, infrared compensated registration images, and nighttime structural images. Spatial alignment between images within the same cycle is achieved through viewpoint normalization and distortion correction in step S1. However, due to significant differences in illumination, temperature background, and UAV micro-attitude between different cycles, directly comparing images from different periods will result in structural offsets. Therefore, this embodiment uses cross-cycle registration to align the structural features of the same photovoltaic module in different cycles to the same coordinate reference, supporting subsequent micro-deformation analysis and structural health index construction. All cross-cycle registration results use a virtual reference board coordinate system as a unified reference, thereby ensuring the comparability and consistency of images from different cycles.

[0049] The cross-cycle registration process includes: performing component-level matching on compensated registration images for different inspection cycles based on component border features and grid line geometric features in the compensated registration images to establish an initial correspondence for the same photovoltaic component in different inspection cycles; performing structural consistency calibration on the initial correspondence based on component outline and structural boundary features in nighttime structural images to eliminate cross-cycle structural offsets caused by differences in illumination conditions, temperature conditions, and local shading; and performing sub-pixel-level fine alignment on the cross-cycle images based on the joint constraints of component border features, grid line geometric features, and component internal geometric features to obtain cross-cycle registration images for time-series analysis.

[0050] In component-level matching, this embodiment first performs coarse matching of component positions in two-period images based on the global position of the components in the virtual reference plate coordinate system. For each component region, this embodiment constructs a component structure description vector based on the bounding box segments, the four corner positions of the bounding box, and the distribution of the grid lines extracted from the compensated registration image. This vector includes the component's bounding box length, width, bounding box orientation angle, number of grid lines, and the relative positions of each grid line. Subsequently, these structural description vectors are matched between different periods, and the regions with the highest structural description similarity are marked as the same component, thereby establishing an initial correspondence. This process relies on the stability of the geometric features of the bounding box and grid lines, and the compensated registration image ensures the geometric consistency of these features through projection and distortion correction in step S1.

[0051] In structural consistency calibration, this embodiment utilizes the high stability of nighttime structural images to correct the initial correspondence. Nighttime structural images, after temperature reference calibration and structural stabilization, are least affected by illumination differences between different inspection cycles, making them suitable for compensating for structural shifts between visible and infrared images across cycles. Specifically, this embodiment extracts component outlines and structural boundaries from nighttime structural images. By assessing the straightness of the component outlines, boundary closure, and structural morphological stability, structural consistency measurements are performed on the same component region from different cycles. When a shift in structural features is detected between two cycles, this embodiment applies the shift as a calibration term to the spatial adjustment of the compensated registration image, thereby eliminating cross-cycle shifts at the component level and ensuring more accurate structural correspondences.

[0052] "Subpixel-level fine alignment" refers to fine-tuning the position of the component structure within the pixel size to obtain cross-cycle images with higher geometric accuracy. This embodiment performs fine alignment under the joint constraints of three types of data: component border features, gate line geometry features, and component internal geometry features. First, by detecting minute positional differences in border line segments across different cycles, the border contour is adjusted through subpixel-level translation and rotation. Second, for gate line features, subpixel-level compensation is performed on gate line position offsets through local correlation analysis along the gate line direction and the vertical direction. Finally, within the component's internal region, local alignment is performed on the geometric consistency of cell boundaries, solder ribbon texture, and temperature gradient changes, ensuring high consistency of all structural features in the virtual reference plate coordinate system. The resulting cross-cycle registered image, by incorporating multi-level structural information from the border, gate lines, and internal structures, achieves a high-precision representation of the component structure for subsequent time-series analysis.

[0053] When extracting subpixel structural features, this embodiment performs structural parameter quantization on the cross-cycle registration image for each cycle. The structural parameters include: subpixel position coordinates of the border line segments to record the small offset of the border over time; straightness parameters of the grid lines, including the position offset of the grid line centerline and the local bending amplitude; and local geometric structural features of the component surface, which record whether there is warping or slight deformation on the component surface by analyzing the consistency of local texture direction and temperature gradient changes.

[0054] Step S5: Detect micro-deformation of photovoltaic modules based on sub-pixel structural features, calculate the structural health index, and obtain the micro-deformation trend results based on the time change of the structural health index; In this embodiment, step S4 has performed cross-cycle registration on images from different inspection cycles and extracted sub-pixel information of the frame, grid lines, and internal geometric features related to the module structure. Because photovoltaic modules are affected by factors such as thermal expansion and contraction, support deformation, load changes, and material fatigue during long-term operation, their geometric structure will experience small but continuously accumulating deformations. This embodiment quantifies and analyzes the sub-pixel structural features extracted across cycles to identify the micro-deformations of the photovoltaic module and assess its health status, providing a basis for identifying hidden structural risks.

[0055] The process of obtaining micro-deformation trend results includes: quantizing the parameters of the photovoltaic module's frame segments, grid line straightness, and module surface geometry at the sub-pixel level based on cross-period registered images to obtain a multi-period structural feature sequence; calculating the micro-deformation variables of the photovoltaic module in different inspection cycles based on the structural feature sequence, including frame offset, grid line curvature change, and local warping of the module surface; constructing a structural health index of the photovoltaic module based on the micro-deformation variables, and performing time series modeling on the structural health index to obtain the time series trend of the photovoltaic module's structural health index; and determining the micro-deformation trend results of the photovoltaic module based on the time series trend, including the development direction, rate of change, and trend stability of the micro-deformation.

[0056] In the sub-pixel quantization of the border line segments, this embodiment precisely parameterizes the position of the component border in each inspection cycle. The border is represented by four straight line segments, each described by its endpoint coordinates and orientation angle. By comparing the positional changes of the corresponding border line segments in different cycles, the border offset is calculated, and its unit can be a pixel or a specific length unit in the virtual reference board coordinate system. The border offset reflects the overall translational, tilting, or twisting tendency of the component and is an important indicator of structural changes.

[0057] In the quantization of grid line straightness, this embodiment extracts the centerline of each grid line from the cross-cycle registration image and calculates its deviation from the straight line. For each grid line, the change in grid line curvature is obtained by analyzing the minute changes in its centerline in the principal and vertical directions. The change in curvature reflects the structural deformation of the area where the grid line is located, such as bending caused by local stress on the cell or offset caused by changes in solder ribbon stress. Since the grid lines are uniformly distributed in space, their straightness change reflects the local deformation of the internal structure of the module.

[0058] In the analysis of the surface geometry of the component, this embodiment identifies the local warpage of the component surface by comparing the consistency of local texture direction and the temperature gradient change. The local warpage is obtained by comparing the texture direction, local boundary morphology, and temperature gradient characteristics of the component surface in different periods. When slight warpage occurs on the surface, the local geometry in the compensated registration image will show changes in orientation consistency or temperature gradient distortion, and this change can be quantified as the local warpage.

[0059] "Micro-variables" are a comprehensive data set across the three dimensions mentioned above, used to represent subtle structural changes in components throughout their lifecycle. Micro-variables possess directional, amplitude, and spatial distribution characteristics, and can reflect potential risks such as component structural aging, frame loosening, or installation loosening.

[0060] The structural health index is a comprehensive health metric built upon micro-variables. This index integrates border offset, changes in grid line curvature, and local warpage in a weighted manner, generating a scalar value based on preset structural feature importance weights. This scalar represents the overall structural health of the component in the current inspection cycle. A higher index value indicates greater structural instability or deformation. A structural health index is generated for each cycle.

[0061] In time series modeling, this embodiment constructs a structural health index sequence by arranging the structural health indices obtained from multiple inspection cycles in chronological order. This sequence is used to observe the temporal variation of structural health. This embodiment uses trend slope, short-term volatility, and long-term stability as key features for time series modeling. The trend slope reflects whether structural deformation is accelerating, slowing down, or remaining stable; short-term volatility reflects whether the component is affected by abnormal loads or local environmental changes; and long-term stability reflects whether the component structure remains within a healthy range. By analyzing the trend of the structural health index sequence, the time series trend of the structural health index is obtained.

[0062] The calculation of the structural health index specifically includes: A1. Extract sub-pixel structural features (for each photovoltaic module). Extract the following feature vectors from the cross-period registration image: Boundary feature vector: ,in These are the subpixel coordinates of the four corner points of the component in the virtual reference plate coordinate system.

[0063] Grid line feature vector: Each grid line It is represented by a set of points along its center line, and its linear equation or curvature characteristics are obtained by fitting.

[0064] Surface geometric eigenvectors: The texture orientation consistency (variance of the principal gradient direction) and the local temperature gradient magnitude are calculated through a local window (5×5 pixels) to form a surface feature map.

[0065] A2. Calculate the subtle variables by comparing the eigenvectors of the current period t with those of the previous period t-1: Border offset: ,in For component number The sub-pixel coordinates of each angle in the virtual reference plate coordinate system at time t. Euclidean distance, in pixels or millimeters (converted according to virtual reference board resolution).

[0066] Change in grid line curvature: ,in For the first The average curvature of the grid lines at time t is calculated by fitting a curve. This represents the total number of grid lines.

[0067] Local warping of the surface: ,in For the first The main direction of the texture of each local window (calculated using the gradient structure tensor). This represents the total number of windows.

[0068] A3. Normalization: To avoid dimensional differences, the minor variables are normalized to the [0,1] interval. These correspond to the normalized border offset, the change in grid line curvature, and the local surface warping, respectively. For example, the calculation is as follows: ,in The maximum allowable change threshold for the border offset is set based on historical data or engineering experience. The other two parameters are obtained in the same way.

[0069] A4. Calculate the single-cycle structural health index: ,in The corresponding configuration weights are set through expert experience or historical data fitting, and are combined to 1, with the initial weight allocation being 5:3:2.

[0070] A5. Construct a time series for n consecutive inspection cycles to obtain the sequence: Each cycle corresponds to a timestamp. (Inspection date).

[0071] A6. Trend analysis of the time series of the structural health index: The trend slope is calculated using linear regression, short-term volatility is calculated using the sliding window standard deviation, and long-term stability is calculated using the coefficient of variation.

[0072] The long-term stability is calculated as follows: ,in The function representing the calculation of standard deviation. The function that represents the calculation of the average value. The smaller the value, the more stable it is in the long run.

[0073] This embodiment interprets the above trends and obtains the micro-deformation trend results. The micro-deformation trend results include three key aspects: the direction of micro-deformation development, i.e., whether the structure is gradually deteriorating, remaining stable, or showing signs of recovery; the rate of change, which measures the speed of micro-deformation accumulation through the trend slope; and the trend stability, which measures whether the component's structural state is stable and reliable through long-term volatility and structural consistency. The micro-deformation trend results will serve as an important basis for judging the long-term health of the component.

[0074] Step S6: Based on the results of thermal crosstalk anomalies, micro-deformation trends, and collaborative anomaly characteristics, construct a multi-source evidence chain fusion model to perform fusion diagnosis on the anomaly categories and levels of photovoltaic modules, output the diagnostic confidence level, generate operation and maintenance decisions based on the fusion diagnosis results, and adjust subsequent inspection strategies.

[0075] In this embodiment, step S3 has identified cross-module thermal crosstalk anomalies, step S5 has obtained the micro-deformation trend results of the photovoltaic module, and the multimodal analysis in step S3 has also generated collaborative anomaly features. These three types of information reflect electrical anomalies, structural anomalies, and surface reflection / shading anomalies of the module, respectively, and are the three most critical indicators of the photovoltaic module's operating status. In order to comprehensively analyze anomaly information from different sources, this embodiment constructs a multi-source evidence chain fusion model, and classifies, rates, and outputs diagnostic confidence levels for the operating status of the module using the module as the basic unit.

[0076] The fusion diagnosis includes: constructing a multi-source feature set containing temperature features, structural features, and spectral features based on thermal crosstalk anomalies, micro-deformation trend results, and collaborative anomaly characteristics; inputting the multi-source feature set into a multi-source evidence chain fusion model, and obtaining the anomaly category and anomaly level of the photovoltaic module according to the credibility, correlation, and consistency of different feature sources; and calculating the confidence level of the anomaly diagnosis based on the fusion output of the multi-source evidence chain fusion model.

[0077] In constructing the multi-source feature set, this embodiment unifies temperature features, structural features, and spectral features into three sub-feature sets. Temperature features include the temperature rise amplitude, temperature gradient direction, and temperature rise propagation characteristics of temperature anomaly regions; structural features include the trend slope obtained from the structural health index sequence, structural stability indices, and micro-variables; spectral features include brightness variations, reflectance variations, and the spatial distribution of occluded areas. Each type of feature uses a virtual reference plate coordinate system as a benchmark to ensure spatial alignment and comparability between multi-source features.

[0078] The "multi-source evidence chain fusion model" refers to a statistical model used to jointly reason about information from three different physical sources. It can form an overall diagnostic result for a component based on the relationships and credibility between different features. In this embodiment, the multi-source evidence chain fusion model is implemented based on DS evidence theory, Bayesian networks, or a trained machine learning model. If DS evidence theory is used, a basic credibility function is assigned to different features, and then the final confidence interval is obtained using evidence combination rules. If a Bayesian network is used, the input features are converted into probability outputs of anomaly categories by establishing conditional probability relationships between temperature anomalies, structural deformations, and occlusion features. If a machine learning model is used, feature patterns are learned through pre-constructed training samples to identify component anomaly types.

[0079] During anomaly classification, the model categorizes component anomalies into multiple types, including but not limited to: electrical hotspots, cross-component thermal crosstalk, shading-related contamination anomalies, structural anomalies caused by component surface warping, latent cell damage, solder ribbon anomalies, or internal current path anomalies. Anomaly levels are assigned based on the severity and potential risk, such as Level 1 (mild anomaly), Level 2 (moderate anomaly), and Level 3 (severe anomaly). The level classification is based on factors such as temperature rise amplitude, propagation path length, the slope of the micro-deformation trend, and the spatial coverage of spectral reflectance anomalies.

[0080] Diagnostic confidence is a score assessing the model's reliability in classifying anomalies. For example, when temperature anomalies and shading features are spatially independent and the temperature rise propagation path is stable, the model has a high confidence level in electrical hotspots; when microdeformation trends are accelerating and structural health indices deviate from the normal range for an extended period, the model has a high confidence level in structural anomalies; and when temperature anomalies are entirely within shading areas and exhibit significant spectral reflectance characteristics, the model has a high confidence level in shading anomalies. Diagnostic confidence is output as a 0–1 or percentage, quantifying the reliability of anomaly classification.

[0081] When generating operation and maintenance decisions, this embodiment provides corresponding measures to power plant operation and maintenance personnel based on the anomaly category and level. For example, for a Level 3 severe electrical hotspot or cross-component thermal crosstalk anomaly, the output suggestion is "immediately arrange shutdown for inspection or component replacement"; for a Level 2 structural anomaly with a deteriorating trend, the output suggestion is "planned maintenance and intensive retesting"; for a Level 1 anomaly caused by shading and contamination, the output suggestion is "perform cleaning treatment". In addition, if multiple components in a certain area have the same type of anomaly, this embodiment will add an area risk marker to the operation and maintenance suggestion to indicate the existence of array-level problems or support structure problems.

[0082] In adjusting the inspection strategy, this embodiment optimizes future UAV inspections based on anomaly levels and diagnostic confidence. For high-risk components, the inspection frequency is increased or nighttime inspections are added; for areas with obvious structural anomaly trends, high-resolution compensated registration image acquisition is increased; for branches with cross-component thermal crosstalk, the flight path is adjusted to increase sampling density in the electrical path direction. All inspection strategy adjustments aim to improve component monitoring accuracy and reduce potential risks, and the final strategy is recorded as a reference for subsequent inspection cycles.

[0083] Example 2: A photovoltaic power plant anomaly monitoring system based on UAV imagery, see [link / reference] Figure 1 As shown, it includes the following modules: Image acquisition module: Acquires visible light and infrared images of photovoltaic modules and projects them onto a virtual reference board to obtain reference images; performs multi-band registration on the reference images and performs local distortion correction under attitude disturbance to obtain compensated registration images; under night inspection conditions, performs structure extraction and temperature reference calibration on infrared images to obtain nighttime structure images; Temperature rise propagation module: Based on the electrical series topology of photovoltaic modules in photovoltaic power plants, a current temperature conduction model is constructed, and the temperature rise propagation characteristics are quantified according to the temperature spatial distribution in the compensated registration image; Collaborative Analysis Module: Performs infrared and visible light collaborative analysis on the compensated registration images to obtain collaborative anomaly features, and identifies cross-component thermal crosstalk anomalies by combining temperature rise propagation features. Cross-cycle registration module: Performs cross-cycle registration on compensated registration images and nighttime structure images from different inspection cycles to extract sub-pixel structural features of the frame and grid lines of photovoltaic modules; Micro-deformation analysis module: Detects micro-deformation of photovoltaic modules based on sub-pixel structural features, calculates the structural health index, and obtains micro-deformation trend results based on the time change of the structural health index; Fusion Diagnosis Module: Based on thermal crosstalk anomalies, micro-deformation trend results, and collaborative anomaly characteristics, a multi-source evidence chain fusion model is constructed to perform fusion diagnosis on the anomaly categories and levels of photovoltaic modules, output diagnostic confidence, generate operation and maintenance decisions based on the fusion diagnosis results, and adjust subsequent inspection strategies.

[0084] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for anomaly monitoring of photovoltaic power plants based on UAV imagery, characterized in that, include: The visible light and infrared images of the photovoltaic module are acquired and projected onto a virtual reference board to obtain the reference image; Multi-band registration is performed on the reference image, and local distortion correction under attitude disturbance is performed to obtain the compensated registration image; under night inspection conditions, structure extraction and temperature reference calibration are performed on the infrared image to obtain the night structure image. Based on the electrical series topology of photovoltaic modules, a current-temperature conduction model is constructed, and the temperature rise propagation characteristics are quantified according to the temperature spatial distribution extracted from the compensated registration image. Infrared and visible light co-analysis was performed on the compensated registration images to obtain co-anomaly features, and cross-component thermal crosstalk anomalies were identified by combining temperature rise propagation features. Cross-cycle registration is performed on compensated registration images and nighttime structure images from different inspection cycles to extract sub-pixel structural features of the frame and grid lines of photovoltaic modules. Micro-deformation of photovoltaic modules is detected based on sub-pixel structural features, structural health index is calculated, and micro-deformation trend results are obtained based on the time change of structural health index. Based on thermal crosstalk anomalies, micro-deformation trend results, and collaborative anomaly characteristics, a multi-source evidence chain fusion model is constructed to perform fusion diagnosis on the anomaly categories and levels of photovoltaic modules, output diagnostic confidence, generate operation and maintenance decisions based on the fusion diagnosis results, and adjust subsequent inspection strategies.

2. The method for anomaly monitoring of photovoltaic power plants based on UAV imagery according to claim 1, characterized in that, The process of obtaining the compensated registration image includes: constructing a virtual reference board based on the geographic coordinate system of the photovoltaic power station; mapping the real-time tracking angle of the photovoltaic module during shooting to the shooting pose parameters of the UAV onto the virtual reference board; performing viewpoint normalization processing on the visible light image and the infrared image to obtain a virtual reference image; adaptively selecting a combination of registration features from the module texture features, grid line features, and module edge contour features according to the module type, surface cleanliness, and shooting lighting conditions; performing multi-band feature matching and geometric transformation on the reference image to obtain an initial registration image; constructing a distortion field to characterize local distortion based on the attitude disturbance information of the UAV during flight; and performing distortion correction on the local nonlinear geometric distortion caused by attitude changes in the initial registration image to obtain a compensated registration image; the compensated registration image includes a visible light compensated registration image and an infrared compensated registration image.

3. The method for anomaly monitoring of photovoltaic power plants based on UAV imagery according to claim 1, characterized in that, The process of obtaining the nighttime structure image includes: extracting the edge contours of the photovoltaic module, the temperature abrupt change lines of the grid lines, and the temperature feature points of the internal structure of the module based on the temperature gradient distribution of the nighttime infrared image to obtain the nighttime infrared structure features; constructing a nighttime temperature reference model based on the ambient temperature, the ground background temperature, and the average reference temperature of the photovoltaic module at the time of inspection; calibrating the nighttime infrared structure features by temperature calibration; and performing structure stabilization processing on the temperature-calibrated nighttime infrared structure features, including enhancing structural continuity and suppressing local noise, to obtain the nighttime structure image.

4. The method for anomaly monitoring of photovoltaic power plants based on UAV imagery according to claim 2, characterized in that, The quantification of temperature rise propagation characteristics includes: constructing a current-temperature conduction model based on the connection sequence of photovoltaic modules in the electrical series branch, determining the current conduction path from upstream to downstream in the series branch and its impact on the module temperature distribution; extracting abnormal heat sources caused by local electrical faults based on the temperature spatial distribution in the infrared compensated registration image, and calculating the propagation direction and temperature rise attenuation mode of temperature between the upstream and downstream modules based on the current-temperature conduction model; and quantifying the temperature rise propagation characteristics of abnormal heat sources in the series branch based on the propagation direction, temperature rise amplitude, and propagation attenuation law.

5. The method for anomaly monitoring of photovoltaic power plants based on UAV imagery according to claim 2, characterized in that, The obtained collaborative anomaly features include: extracting temperature anomaly regions on the surface of photovoltaic modules based on the temperature spatial distribution of infrared compensated registration images, and obtaining temperature anomaly features including spatial distribution and temperature gradient; extracting spectral reflectance features, brightness change features, and local shading features of the photovoltaic module surface from visible light compensated registration images based on visible light reflectance characteristics to identify shading regions; performing collaborative matching between temperature anomaly features and spectral reflectance features, and comparing the correspondence between temperature anomaly regions and shading regions to distinguish between shading-related temperature rise and electrical temperature rise; and generating collaborative anomaly features characterizing the thermal anomaly attributes of photovoltaic modules based on the collaborative matching results.

6. The method for anomaly monitoring of photovoltaic power plants based on UAV imagery according to claim 1, characterized in that, The identification of cross-module thermal crosstalk anomalies includes: determining the propagation direction and propagation path of temperature rise in the series branch of photovoltaic modules based on temperature rise propagation characteristics; determining whether the temperature anomaly is independent of the surface shading area based on cooperative anomaly characteristics; identifying whether the temperature anomaly crosses the module boundary of adjacent photovoltaic modules in the temperature rise propagation path, and determining whether the temperature rise change conforms to the propagation law of the current-temperature conduction model; when the temperature anomaly simultaneously satisfies cross-module propagation, continuous temperature rise propagation mode and independence from surface shading, it is determined that there is a cross-module thermal crosstalk anomaly, and the corresponding anomaly impact range is output.

7. The method for anomaly monitoring of photovoltaic power plants based on UAV imagery according to claim 1, characterized in that, The cross-cycle registration process includes: performing component-level matching on compensated registration images for different inspection cycles based on component border features and grid line geometric features in the compensated registration images to establish an initial correspondence for the same photovoltaic component in different inspection cycles; performing structural consistency calibration on the initial correspondence based on component outline and structural boundary features in nighttime structural images to eliminate cross-cycle structural offsets caused by differences in illumination conditions, temperature conditions, and local shading; and performing sub-pixel level alignment on the cross-cycle images based on the joint constraints of component border features, grid line geometric features, and component internal geometric features to obtain cross-cycle registration images for time-series analysis.

8. The method for anomaly monitoring of photovoltaic power plants based on UAV imagery according to claim 1, characterized in that, The process of obtaining micro-deformation trend results includes: quantizing the parameters of the photovoltaic module's frame segments, grid line straightness, and module surface geometry at the sub-pixel level based on cross-period registered images to obtain a multi-period structural feature sequence; calculating the micro-deformation variables of the photovoltaic module in different inspection cycles based on the structural feature sequence, including frame offset, grid line curvature change, and local warping of the module surface; constructing a structural health index of the photovoltaic module based on the micro-deformation variables, and performing time series modeling on the structural health index to obtain the time series trend of the photovoltaic module's structural health index; and determining the micro-deformation trend results of the photovoltaic module based on the time series trend, including the development direction, rate of change, and trend stability of the micro-deformation.

9. The method for anomaly monitoring of photovoltaic power plants based on UAV imagery according to claim 1, characterized in that, The fusion diagnosis includes: constructing a multi-source feature set containing temperature features, structural features, and spectral features based on thermal crosstalk anomalies, micro-deformation trend results, and collaborative anomaly characteristics; inputting the multi-source feature set into a multi-source evidence chain fusion model, and obtaining the anomaly category and anomaly level of the photovoltaic module according to the credibility, correlation, and consistency of different feature sources; and calculating the diagnostic confidence based on the fusion output of the multi-source evidence chain fusion model.

10. A photovoltaic power plant anomaly monitoring system based on UAV imagery, characterized in that, The system employs a photovoltaic power plant anomaly monitoring method based on UAV imagery as described in any one of claims 1 to 9, comprising: Image acquisition module: Acquires visible light and infrared images of photovoltaic modules and projects them onto a virtual reference board to obtain reference images; performs multi-band registration on the reference images and performs local distortion correction under attitude disturbance to obtain compensated registration images; under night inspection conditions, performs structure extraction and temperature reference calibration on infrared images to obtain nighttime structure images; Temperature rise propagation module: Based on the electrical series topology of photovoltaic modules in photovoltaic power plants, a current temperature conduction model is constructed, and the temperature rise propagation characteristics are quantified according to the temperature spatial distribution in the compensated registration image; Collaborative Analysis Module: Performs infrared and visible light collaborative analysis on the compensated registration images to obtain collaborative anomaly features, and identifies cross-component thermal crosstalk anomalies by combining temperature rise propagation features. Cross-cycle registration module: Performs cross-cycle registration on compensated registration images and nighttime structure images from different inspection cycles to extract sub-pixel structural features of the frame and grid lines of photovoltaic modules; Micro-deformation analysis module: Detects micro-deformation of photovoltaic modules based on sub-pixel structural features, calculates the structural health index, and obtains micro-deformation trend results based on the time change of the structural health index; Fusion Diagnosis Module: Based on thermal crosstalk anomalies, micro-deformation trend results, and collaborative anomaly characteristics, a multi-source evidence chain fusion model is constructed to perform fusion diagnosis on the anomaly categories and levels of photovoltaic modules, output diagnostic confidence, generate operation and maintenance decisions based on the fusion diagnosis results, and adjust subsequent inspection strategies.

Images