Unattended photovoltaic power station cloud edge collaborative intelligent inspection and operation method

Abstract

The present application relates to the technical field of intelligent inspection and operation and maintenance, and discloses a cloud-edge collaborative intelligent inspection and operation method for unattended photovoltaic power stations, comprising: acquiring a video stream of photovoltaic components and performing spatial coordinate registration; identifying a suspected abnormal heating region of interest and a healthy reference region, extracting the highest temperature and the reference temperature respectively, and calculating the temperature change rate; acquiring the real global GPS coordinates of the region of interest, and calculating the local real-time wind speed; constructing a heat flow decoupling model combining the differential heat capacity heat flow density and the differential convection heat flow density, calculating the severity index, and comparing the severity index with the fault determination threshold, and executing intelligent operation and maintenance decisions according to the comparison result. The present application removes the false signals of convection heat dissipation expansion, distinguishes the residual heat on the component surface due to thermal inertia from the internal real electric heat anomaly, thereby preventing the misjudgment of hot spots caused by environmental appearance interference, reducing the misjudgment rate and reducing invalid on-site personnel scheduling.

Description

Technical Field

[0001] This invention relates to the field of intelligent inspection and maintenance technology, specifically to a cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants. Background Technology

[0002] With the development of clean energy, unattended photovoltaic power stations located in remote areas and covering large areas are becoming increasingly common. In order to ensure the safety and power generation efficiency of the power stations, cloud-edge collaborative intelligent inspection and maintenance of unattended photovoltaic power stations has become an important part of this field.

[0003] In the daily operation and maintenance of photovoltaic power plants, for those covering large areas, drones equipped with infrared thermal imagers are typically used for large-scale inspections. This existing technology mainly relies on images collected by drones from the air to monitor the surface temperature of photovoltaic modules, thereby identifying potential hot spot faults such as microcracks in the cells.

[0004] However, existing drones equipped with infrared thermal imagers are often susceptible to interference from external environmental factors when identifying hot spot faults in photovoltaic modules. For example, uneven convection due to localized winds, cloud cover, bird droppings, or dust accumulation can cause localized high-temperature areas in the infrared image to appear similar to actual internal hot spots. The system easily misidentifies these false faults caused by environmental interference as genuine hot spot faults, leading to a high false alarm rate and ineffective on-site personnel dispatch, increasing the operational costs of unattended power plants. Therefore, a cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants is urgently needed to solve these problems. Summary of the Invention

[0005] To address the problems in related technologies, this invention provides a cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants, thereby overcoming the aforementioned technical problems in existing related technologies.

[0006] To solve the aforementioned technical problem, the present invention is achieved through the following technical solution: In a first aspect, embodiments of the present invention provide a cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants, specifically including: acquiring continuous multi-frame multimodal video streams of photovoltaic modules using inspection equipment, and performing spatial coordinate registration of the continuous multi-frame multimodal video streams at an edge computing node; identifying regions of interest suspected of abnormal heating in the continuous multi-frame multimodal video streams, selecting unobstructed healthy reference regions within the same photovoltaic string, extracting the highest temperature of the region of interest and the reference temperature of the healthy reference region, and calculating the temperature change rate; acquiring the real global GPS coordinates of the region of interest, and simultaneously calculating the local real-time wind speed of the global GPS coordinates; constructing a heat flow decoupling model by combining the highest temperature, reference temperature, temperature change rate, and local real-time wind speed, and calculating a severity index reflecting the actual electrothermal anomaly inside the module; acquiring the factory nominal parameters and initial commissioning time of the target photovoltaic module, calculating a dynamic fault judgment threshold by combining the current operating years, comparing the severity index with the dynamic fault judgment threshold, and executing intelligent operation and maintenance decisions based on the comparison results.

[0007] As a preferred embodiment of the cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants described in this invention, the method involves dynamically registering the spatial coordinates of the continuous multi-frame multimodal video stream at the edge computing node, specifically including: Key points of two consecutive frames of images are extracted using a feature extraction algorithm and preliminary feature matching is performed. A random sampling consensus algorithm is introduced to iteratively filter the initial matching points and calculate the global homography matrix representing the true rigid body geometric transformation relationship between two frames of images. The center pixel coordinates of the region of interest in the previous frame are converted into homogeneous coordinate vectors, and then combined with the global homography matrix to perform spatial linear transformation and perspective normalization processing to deduce the accurate two-dimensional center pixel coordinate vector of the region of interest in the current image frame.

[0008] As a preferred embodiment of the cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants described in this invention, the extraction of the highest temperature of the region of interest and the reference temperature of the healthy reference region, and the calculation of the temperature change rate, specifically includes: The original infrared radiation grayscale matrix of the region of interest is transformed into an absolute temperature matrix using a radiometric calibration function, and the maximum value is extracted as the highest temperature of the region of interest. The independent photovoltaic module set within the photovoltaic string is segmented using edge detection and line transformation algorithms. After removing photovoltaic modules containing abnormal areas, the infrared spatial temperature variance inside each remaining candidate module is calculated. The photovoltaic module corresponding to the global minimum value of the infrared spatial temperature variance is selected as the healthy reference area, and the spatial arithmetic mean of the pixel temperature within the healthy reference area is used as the reference temperature. Calculate the relative surface temperature difference between the highest temperature and the reference temperature in the current frame, and within a set sliding time window, use least squares linear regression to fit the surface temperature difference data of multiple consecutive frames to determine the rate of temperature change reflecting the thermodynamic transient process.

[0009] As a preferred embodiment of the cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants described in this invention, the method involves obtaining the actual global GPS coordinates of the region of interest and simultaneously calculating the local real-time wind speed at the global GPS coordinates, specifically including: Read the high-precision global positioning coordinates, absolute altitude, and camera gimbal attitude angle of the inspection equipment at the moment when abnormal features are extracted, and query the local terrain elevation corresponding to the current coordinates based on the preset digital elevation model; By combining the camera's nominal intrinsic parameters, the actual effective pitch angle and effective yaw angle pointing towards the region of interest can be calculated; By combining the nominal support height, installation tilt angle, and installation azimuth angle of the target photovoltaic module, the precise spatial slant distance from the camera optical center to the abnormal heat point on the surface of the photovoltaic module is calculated. Based on the precise spatial slant distance, the position offset of the target point relative to the vertical foot of the UAV is decomposed into east and north components, and the standard radius of curvature of the Earth is introduced and superimposed on the global positioning coordinates of the inspection equipment to determine the true global GPS coordinates of the abnormal heating point of the photovoltaic module. The cloud server uses an inverse distance spatial weighted interpolation algorithm, combined with real-time wind speeds collected by multiple micro-weather stations within the photovoltaic field and the straight-line physical spatial distance between the micro-weather stations and the global GPS coordinates, to calculate the local real-time wind speed of the region of interest.

[0010] As a preferred embodiment of the cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants described in this invention, the method involves constructing a heat flow decoupling model by combining the highest temperature, reference temperature, temperature change rate, and local real-time wind speed to calculate a severity index, specifically including: The differential heat flux density of the abnormal region compared to the healthy region due to thermal inertia is determined by multiplying the rate of temperature change with the surface equivalent dynamic heat capacity of the photovoltaic module. The differential convective heat flux density lost from the abnormal region to the outside compared to the healthy region is determined by multiplying the relative surface temperature difference between the highest temperature and the reference temperature by the dynamic surface heat transfer coefficient. The local abnormal electrothermal power density of the region of interest is determined by combining the differential heat capacity heat flux density absorbed by the abnormal region compared to the healthy region due to thermal inertia and the differential convective heat flux density lost by the abnormal region compared to the healthy region to the outside. The real-time effective solar irradiance projected onto the surface of the photovoltaic module is obtained, and the photoelectric conversion efficiency attenuation at the reference temperature is corrected using the peak power temperature coefficient. The nominal photoelectric conversion efficiency is obtained and a reference photothermal power density is constructed by combining the real-time effective solar irradiance. The ratio of the local abnormal electrothermal power density to the reference photothermal power density is calculated to construct a heat flow decoupling model and calculate the dynamic severity index.

[0011] As a preferred embodiment of the cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants described in this invention, the method for obtaining the surface equivalent dynamic heat capacity includes: Obtain the density, standard specific heat capacity, nominal thickness, and thermal conductivity of each physical encapsulation layer inside the photovoltaic module; The inherent static physical heat capacity background value of each physical encapsulation layer is calculated using the density, standard specific heat capacity and nominal thickness of each physical encapsulation layer. The cumulative normal thermal resistance from the geometric center of each layer to the observation surface is calculated based on the nominal thickness and thermal conductivity of each physical encapsulation layer. The dynamic convective heat transfer intensity on the surface of the photovoltaic module is evaluated by combining local real-time wind speed, and the cumulative normal thermal resistance is coupled with the dynamic convective heat transfer intensity to construct a transient thermal coupling attenuation weight. The static physical thermal capacity background value of each physical encapsulation layer is multiplied by its corresponding transient thermal coupling attenuation weight and then summed and aggregated to determine the surface equivalent dynamic thermal capacity.

[0012] As a preferred embodiment of the cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants described in this invention, the method for obtaining the dynamic surface heat transfer coefficient includes: Retrieve the nominal parameters of air properties and extract the installation tilt angle of the target photovoltaic module and the characteristic length characterizing the fluid flow span on its surface; The natural convection heat transfer coefficient dominated by temperature difference is calculated by combining the reference temperature, ambient absolute temperature and installation tilt angle. By combining local real-time wind speed, nominal air properties and the characteristic length, the forced convection heat transfer coefficient dominated by wind speed is calculated. The Churchill-Usagi equation is introduced to nonlinearly superimpose the natural convection heat transfer coefficient and the forced convection heat transfer coefficient to determine the dynamic surface heat transfer coefficient.

[0013] As a preferred embodiment of the cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants described in this invention, after calculating the severity index, an evaluation and elimination mechanism is also included: Combining the dynamic surface heat transfer coefficient, the equivalent temperature difference of the infrared thermal imager's hardware noise, and the fault judgment threshold, the minimum reference photoelectric power density threshold that just makes the infrared thermal imager reach the physical limit resolution is derived. When the real-time calculated reference photoelectric power density is lower than the minimum reference photoelectric power density threshold, it is determined that the current meteorological conditions do not have the thermodynamic conditions to effectively drive the component to generate heat, no evaluation calculation is performed, and the severity index is directly set to zero.

[0014] As a preferred embodiment of the cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants described in this invention, the severity index is compared with a fault determination threshold, and intelligent maintenance decisions are executed based on the comparison results, specifically including: If the calculated severity index is greater than the dynamic fault determination threshold, it is confirmed as an internal hot spot fault caused by abnormal physical defects and a maintenance work order is issued. If the severity index is less than or equal to the dynamic fault determination threshold, the heat generation is determined to be within the normal physical heat dissipation range allowed in the current natural aging stage, and is confirmed as normal.

[0015] Secondly, embodiments of the present invention provide a cloud-edge collaborative intelligent inspection and maintenance system for unattended photovoltaic power plants, comprising: a video stream acquisition module, used to acquire continuous multi-frame multimodal video streams of photovoltaic modules using inspection equipment, and perform spatial coordinate registration at an edge computing node; a temperature feature extraction module, used to identify regions of interest suspected of abnormal heating, and simultaneously select a healthy reference region, extract the highest temperature and reference temperature, and calculate the temperature change rate; a wind speed acquisition module, used to acquire the real global GPS coordinates of the region of interest and calculate the local real-time wind speed; a heat flow decoupling evaluation module, used to construct a heat flow decoupling model and calculate a severity index; and an intelligent maintenance decision module, used to compare the severity index with a fault judgment threshold and execute intelligent maintenance decisions.

[0016] The present invention has the following beneficial effects: 1. This invention constructs a heat flow decoupling model by combining real-time extracted local wind speed and temperature change rate. This model includes the differential heat capacity heat flux density of abnormal areas compared to healthy areas due to thermal inertia, and the differential convective heat flux density of abnormal areas compared to healthy areas. This model can distinguish between residual heat on the component surface due to thermal inertia and real internal electrothermal anomalies by removing false signals of convective heat dissipation expansion. This prevents misjudgment of hot spots caused by environmental appearance interference, thereby reducing the misjudgment rate and reducing ineffective on-site personnel scheduling.

[0017] 2. This invention retrieves the factory-specified attenuation parameters of the target component and combines them with its actual operating years to dynamically calculate the fault judgment threshold at the current life node. By comparing the severity index calculated in real time by the heat flow decoupling model with the dynamic fault judgment threshold, it adaptively distinguishes between normal physical aging and abnormal deep faults of the component. This reduces the need for ineffective on-site personnel scheduling and lowers the entry and exit maintenance costs of unattended photovoltaic power stations.

[0018] 3. This invention identifies regions of interest (ROIs) suspected of abnormal heating in continuous multi-frame multimodal video streams, and selects unobstructed healthy reference regions within the same photovoltaic string. The highest temperature of the ROI and the reference temperature of the healthy reference region are extracted, and the rate of temperature change is calculated. This method transforms isolated static temperature snapshots into continuous thermodynamic evolution trends, enabling the system to perceive whether suspected heating points are continuously generating heat driven by internal faults or are merely in a hysteretic cooling phase due to material thermal inertia. This helps to eliminate false signals from environmental appearances, ensuring the accuracy of identifying deep physical faults and reducing the false positive rate.

[0019] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, the drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 The present invention provides a flowchart of a cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants.

[0022] Figure 2 This is a schematic diagram of the S4 process provided by the present invention.

[0023] Figure 3 This is a schematic diagram of the S5 process provided by the present invention.

[0024] Figure 4 This invention provides a schematic diagram of a cloud-edge collaborative intelligent inspection and maintenance system for unattended photovoltaic power plants. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Example 1 In the daily operation and maintenance of photovoltaic power plants, especially for unattended photovoltaic power plants located in remote areas and covering large areas, drones equipped with infrared thermal imagers are usually used for inspection. However, existing drones equipped with infrared thermal imagers are often easily interfered with by external environmental factors when identifying hot spot faults in photovoltaic modules. For example, uneven convection due to localized winds, cloud cover, bird droppings, or dust accumulation can cause localized high-temperature areas on the infrared image to appear similar to actual internal structural hot spots (such as microcracks in the solar cells). This false fault caused by environmental interference can easily lead to a high false alarm rate, resulting in ineffective on-site personnel dispatch and increasing the operation and maintenance costs of unattended power plants.

[0027] To solve the above technical problems, such as Figure 1 As shown, Embodiment 1 of the present invention provides a cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power stations. Specifically, Embodiment 1 takes the inspection and renovation scenario of a large unattended photovoltaic power station in a desert in Northwest China as an example: an automated hangar for drones with edge computing capabilities and a photovoltaic micro-weather station are deployed on site. The maintenance system needs to identify deep physical faults of photovoltaic modules while minimizing invalid manual dispatching.

[0028] In the specific implementation of Example 1: First, a continuous multi-frame multimodal video stream of the photovoltaic module is acquired using inspection equipment. Spatial coordinate registration of the continuous multi-frame multimodal video stream is performed at the edge computing node. This prevents image jitter errors and achieves dynamic registration and alignment of spatial coordinates in the continuous multi-frame multimodal video stream, ensuring accurate locking of abnormal heating targets when disturbed by the external environment. Then, a region of interest suspected of abnormal heating is identified in the continuous multi-frame multimodal video stream. An unobstructed healthy reference region is selected within the same photovoltaic string. The highest temperature of the region of interest and the reference temperature of the healthy reference region are extracted, and the temperature change rate is calculated. This transforms the isolated static temperature snapshot into a continuous thermodynamic evolution trend, enabling the system to perceive whether the suspected heating point is continuously generating heat under the drive of an internal fault or is merely in a hysteresis cooling stage due to material thermal inertia. This provides parameter input for the subsequent construction of a heat flow decoupling model, helping to remove false signals from environmental appearance interference and ensuring the accuracy of deep physical fault identification. Next, the real global GPS coordinates of the region of interest are acquired, and the local real-time wind speed of the global GPS coordinates is calculated simultaneously. Subsequently, a heat flow decoupling model is constructed by combining the highest temperature, reference temperature, temperature change rate, and local real-time wind speed. A severity index is calculated, and by quantifying the differential heat capacity absorption term caused by thermal inertia, false signals of convective heat expansion can be removed, reflecting the true internal abnormal heating situation. This prevents false alarms caused by environmental interference and reduces the false alarm rate. Finally, the severity index is compared with the fault judgment threshold, and intelligent operation and maintenance decisions are executed based on the comparison results. This method can identify whether the heat generated in the heating area belongs to normal physical heat dissipation allowed by the natural decay of the module over long-term operation, or is a hot spot fault caused by abnormal deep physical structural damage. It avoids false fault alarms caused by reasonable natural aging of the module and, while ensuring the safe operation of the power station, minimizes ineffective on-site personnel scheduling and blind dispatching, reducing the entry and exit operation and maintenance costs of unattended photovoltaic power stations located in remote areas with large land areas.

[0029] Furthermore, to better introduce the technical solution of Embodiment 1 of the present invention, a detailed description is provided of the cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants, specifically including the following: S1. Acquire continuous multi-frame multimodal video streams of photovoltaic modules using inspection equipment, and dynamically register the spatial coordinates between the multi-frame images at the edge computing node to complete spatial coordinate alignment. This includes the following sub-steps: S11. The drone inspection equipment flies along a preset route, using its onboard infrared thermal imager and visible light camera to acquire continuous multi-frame, multi-modal video streams of the photovoltaic modules; the video sampling frame interval is set to... The on-site terminal locally caches the acquired raw video frames and performs Gaussian filtering to remove high-frequency image noise.

[0030] Specifically, the implementation steps of Gaussian filtering are as follows: S111. In order to filter out the inherent high-frequency spatial noise of the infrared sensor while preserving the gradient characteristics of the photovoltaic module edge to the greatest extent, the system sets the kernel radius of the Gaussian kernel to be [value missing]. And the relative coordinates of each point within the kernel are calculated using a two-dimensional Gaussian distribution function. initial weights : ; In the formula, The standard deviation of the Gaussian kernel determines the smoothing scale for high-frequency noise; and These represent the horizontal and vertical coordinate offsets of pixels within the smoothing kernel relative to the kernel center, with values ​​ranging from [value range missing]. Integers within.

[0031] S112. To ensure the conservation of total energy before and after image smoothing, i.e., to maintain the overall temperature baseline of the infrared image, the system applies the calculated positive-zero Gaussian initial weights. Normalization is performed to generate the final filter weights. : ; In the formula, The term is the sum of all initial weights within the kernel.

[0032] S113. The edge computing node uses a normalized two-dimensional Gaussian smoothing kernel to perform pixel-by-pixel sliding spatial convolution on the cached original multimodal video frames, obtaining smooth image frames after filtering out high-frequency noise. In the formula, The absolute pixel coordinates of the filtered image Pixel value at; This represents the pixel value of the original image within the corresponding local neighborhood.

[0033] For example, taking a frame of image acquired by an infrared thermal imager as an example, considering the ratio between the drone's inspection altitude and the target pixel size, the system sets the kernel radius. That is, construct a 3×3 Gaussian kernel and set the standard deviation. When processing coordinates as When dealing with a pixel, if that pixel happens to be affected by a sudden transient level pulse, its original image pixel value will be affected. An abnormally high value, indicating a high-frequency isolated noise extremum, is determined by substituting the coordinates of nine pixels within its local 3×3 neighborhood into the above formula, with the center point... The normalized weight is the largest, and the weight of the edge point decreases exponentially with the increase of Euclidean distance. After calculation by the formula, the energy of the abnormal isolated extreme value is forcibly dispersed to the surrounding neighborhood, reducing its amplitude.

[0034] S12. At the edge computing node, a feature point matching algorithm is used to dynamically register the spatial coordinates between multiple frames of images. The specific steps are as follows: S121. The edge computing node uses the ORB algorithm to extract global key points from the previous frame image and the current frame image respectively, and calculates the corresponding binary feature descriptors. The feature points are initially matched by Hamming distance.

[0035] S122. To eliminate the interference of erroneous matching points on motion estimation, the Random Sample Consensus (RANSAC) algorithm is introduced to iteratively filter the initial matching pairs and calculate the true rigid body geometric transformation relationship between the two frames. Global homography matrix .

[0036] S123. Extract the two-dimensional center pixel coordinates of the region of interest in the previous frame, convert them into homogeneous coordinate vectors, and correlate them with the homography matrix. Perform a spatial linear transformation and calculate the derived coordinates after dynamic registration. The specific expression is as follows: ; S124. Perform perspective normalization on the transformed homogeneous coordinate components to obtain... The precise two-dimensional center pixel coordinate vector of the region of interest in the current image frame at any given time. The specific expression is as follows: ; In the formula, and These are the x and y coordinates of the center point of the previous frame; to Elements of the homography matrix; and These are the scale coordinates after the derivation; This is the perspective distortion scale factor.

[0037] In this embodiment 1, a combination of ORB feature point matching and the RANSAC algorithm is used to calculate the global homography matrix that represents the true geometric transformation relationship between two consecutive image frames. This allows for real-time deduction and correction of the two-dimensional center pixel coordinates of the region of interest in the current image frame. Specifically, for example: set in At that moment, the system located the center coordinates of a suspected abnormal hotspot. At this moment, the drone experienced complex non-rigid motion, including translation, scaling, and slight yaw rotation, due to crosswinds. The edge nodes calculated the current... Global homography matrix Transform the coordinates of the previous frame into homogeneous coordinates and then input them into the formula to calculate the transformed homogeneous coordinate components: Perspective normalization processing (divided by scale factor) After that, the precise two-dimensional center pixel coordinate vector of the region of interest in the current image frame is calculated. This method overcomes the nonlinear image distortion and jitter errors caused by high-speed movement or large attitude changes of UAVs under complex weather conditions, and realizes dynamic registration and alignment of spatial coordinates in continuous multi-frame multimodal video streams, ensuring accurate locking of abnormally heated targets when subjected to strong external environmental interference.

[0038] S2. Identify regions of interest (ROIs) suspected of abnormal heating in a continuous multi-frame multimodal video stream, and select an unobstructed healthy reference region within the same photovoltaic string. Extract the highest temperature of the ROI and the reference temperature of the healthy reference region, and calculate the temperature change rate. This includes the following sub-steps: S21, Based on the locked coordinates The highest temperature of the region of interest is extracted using a target detection algorithm. Simultaneously, a healthy area within the same photovoltaic string, free from physical obstructions and with normal electrical operation, was selected as a reference area, and a reference temperature was extracted. Calculate the surface temperature difference of the current frame. The specific implementation steps are as follows: S211. The object detection algorithm outputs a two-dimensional bounding box matrix with absolute pixel coordinates in the video stream. Its range is strictly defined by the coordinates of the upper left and lower right corners. The system utilizes the radiometric calibration function that comes pre-installed on the infrared thermal imager. The original infrared radiation grayscale value matrix within the bounding box matrix. Pixel-by-pixel conversion to absolute temperature matrix .

[0039] Subsequently, the system traverses the bounding box matrix. By maximizing all valid coordinate points within the region, the highest temperature of the region of interest is extracted. ; .

[0040] S212. To ensure that the healthy reference region and the region of interest are in the same external macroscopic environment, i.e., belong to the same photovoltaic string, and that the reference region meets the conditions of no physical obstruction and normal electrical operation, the system uses the Canny edge detection and Hough line transform algorithm to segment all independent photovoltaic modules in the current physical array in the registered visible light image, forming a module set. And remove regions containing abnormalities. Components For any remaining component in the set Physical obstructions such as bird droppings or localized shadows on the surface can lead to localized low temperatures, while electrical anomalies such as minor hot spots can lead to localized high temperatures. Both of these conditions can significantly increase the spatial variance of the internal temperature matrix of the component. Therefore, the system calculates the infrared spatial temperature variance of each candidate component. : ; In the formula, For components The total number of pixels contained within; This is the arithmetic mean of the temperatures of all pixels within this component.

[0041] By solving the infrared space temperature variance The global minimum value is used to lock the photovoltaic module with the most uniform temperature distribution as the healthy reference area. : .

[0042] S213, Locking the health reference area Subsequently, to avoid interference from high-frequency thermal noise at a single point on the infrared sensor, the spatial arithmetic mean of the temperatures of all pixels within the matrix region is calculated and used as the reference temperature at that moment. : .

[0043] S22, within a set sliding time window (Include Within a series of consecutive sampling points, the rate of temperature change is calculated using linear regression based on the least squares method. The specific formula is as follows: ; In the formula, for The rate of temperature change over time; For the first in the sliding window Each discrete sampling time; This is the average value of all sampling times within the sliding window; for The surface temperature difference measured at any time; This represents the average temperature difference on the inner surface of the window; This represents the total number of frames within the sliding window.

[0044] For example: In actual inspection processes, sliding time windows The method for determining it is as follows: S221. Obtain the effective hovering observation time of the UAV inspection equipment on a single suspected faulty target, and the actual video sampling frame rate of the infrared camera. Based on these two physical parameters, determine the maximum number of consecutive image frames that the system can capture, and define the theoretical upper limit of the sliding time window.

[0045] S222. Synchronously retrieve the real-time wind speed change frequency and the nominal equivalent heat capacity of the target module from the micro-weather station in the field. When the external gusts suddenly change violently, the convective heat transfer state on the surface of the photovoltaic module switches very quickly. The system needs to adaptively shorten the time window to accurately capture the transient temperature drop process. Conversely, if the ambient wind speed is stable and the thermodynamic evolution is slow, the window can be appropriately extended to smooth and filter the random high-frequency thermal noise of the infrared sensor.

[0046] S223. Preset the minimum sample size requirement for linear regression calculation, i.e., the minimum number of consecutive effective frames, as a hard lower limit guarantee for the sliding time window to prevent the slope of the fitted rate of change from being distorted due to too few sampling points.

[0047] S224. Within the aforementioned time limit interval, the edge computing node adaptively performs dynamic optimization based on the fluctuation variance of the surface temperature difference data extracted from the current continuous image frames: selecting an effective duration that minimizes the linear fitting error and reflects a clear thermodynamic trend as the current sliding time window. .

[0048] In this embodiment 1, the infrared spatial temperature variance is dynamically searched within a photovoltaic string under the same physical environment. The smallest photovoltaic module as a health reference area This is used to calculate the relative surface temperature difference, offsetting the interference caused by fluctuations in overall illumination and ambient temperature; and by introducing a sliding time window Least squares linear regression was performed on surface temperature difference data from multiple consecutive frames to obtain the rate of temperature change characterizing the thermodynamic transient process. Specifically, for example: setting the sliding time window to include... There are sampling points, with a sampling interval of . Due to the strong cooling effect of gusts, the temperature difference extracted from 5 consecutive frames... They are respectively The rate of temperature change within the window can be calculated using the above formula. This method transforms isolated static temperature snapshots into continuous thermodynamic evolution trends, enabling the system to perceive whether suspected hot spots are continuously generating heat under the drive of internal faults or are merely in a hysteretic cooling phase due to material thermal inertia. This provides parameter input for the subsequent construction of a heat flow decoupling model, helps to remove false signals from environmental appearances, and ensures the accuracy of identifying deep physical faults.

[0049] S3. Map the spatial pose of the inspection equipment to the actual physical coordinates of the target abnormal heat point, and simultaneously retrieve and calculate the local real-time wind speed of the actual physical coordinates. This includes the following sub-steps: S31. Map the spatial pose of the inspection equipment to the actual global GPS coordinates of the target abnormal heat point. This includes the following steps: S311, the edge computing node synchronously reads data from the built-in RTK and IMU modules. High-precision global positioning coordinates of the constant inspection equipment Absolute altitude And the pitch angle of the camera gimbal relative to the horizontal plane. Yaw angle relative to true north Simultaneously, the system queries the corresponding local terrain elevation in the preset digital elevation model (DEM) based on the current coordinates. .

[0050] S312. Based on the camera's factory-specified intrinsic parameters, calculate the true effective pitch angle pointing towards the region of interest. With effective yaw angle The specific formula is as follows: ; ; In the formula, These are the coordinates of the optical principal point pixel of the photosensitive element; Nominal focal length; and The physical size of a single pixel.

[0051] S313. Extract the nominal support height of the target photovoltaic module from the cloud database. Installation tilt angle and installation azimuth angle Calculate the effective vertical relative height of the drone with respect to the bottom reference plane of the component. The specific formula is as follows: ; Subsequently, the precise spatial slant distance from the camera's optical center to the anomalous heating point on the tilted photovoltaic module surface was calculated. The specific formula is as follows: ; In the formula, This represents the slant distance in real space.

[0052] S314, Based on precise spatial slant distance Calculate the eastward offset of the target point relative to the vertical foot of the UAV. offset from true north The specific formula is as follows: ; ; By introducing the Earth's standard radius of curvature and superimposing the physical offset onto the RTK coordinates of the inspection equipment, the true global GPS coordinates of the abnormal heating points of the photovoltaic modules can be determined. The specific expression is as follows: ; ; In the formula, is the Earth's standard radius of curvature.

[0053] S32, The edge computing node will extract the highest temperature of the region of interest. Reference temperature Temperature change rate and the GPS global spatial coordinates of the inspection equipment Package and upload to the cloud server.

[0054] Cloud server receives GPS global spatial coordinates Then, using an inverse distance spatial weighted interpolation algorithm, the local real-time wind speed at the target coordinate point is simultaneously retrieved and calculated. : ; In the formula, This refers to the local real-time wind speed at the current inspection target coordinates. This represents the total number of micro-weather stations within the site. For the first Real-time wind speed collected by a micro-weather station; The current GPS coordinates of the drone and the first The straight-line physical spatial distance between micro-weather stations; To prevent The smallest positive number that causes the poles to diverge when it is zero, take .

[0055] S33. The cloud server extracts and distributes the thermodynamic and electrical nominal parameters of the photovoltaic module corresponding to the coordinate point from the database, and obtains the surface equivalent dynamic heat capacity. and peak power temperature coefficient .

[0056] Specifically: In this embodiment, the surface equivalent dynamic heat capacity The method for determining it is as follows: Based on the model of the target photovoltaic module, retrieve its Bill of Materials (BOM) from the cloud database. Sequentially obtain the density, standard specific heat capacity, nominal thickness, and thermal conductivity of each physical encapsulation layer that constitutes the module (such as high-transparency photovoltaic glass, upper EVA, silicon cell, lower EVA, insulating backsheet, etc.). Using the density, specific heat capacity, and thickness of each material layer, the inherent static physical heat capacity background value of each layer is calculated. Then, using the actual observation surface of the component by the infrared thermal imager as the reference surface, the cumulative normal thermal resistance from the geometric center of each internal physical layer to the observation surface is calculated by integrating layer by layer according to the thickness and thermal conductivity of each layer. The local real-time wind speed provided by the micro-weather station is read to assess the dynamic convective heat transfer intensity of the current component surface. The pre-calculated cumulative normal thermal resistance of each layer is coupled with the current dynamic convective heat transfer intensity of the surface to construct an exponentially varying attenuation weight factor: when the outside encounters strong winds and sudden cooling, and the convective heat transfer is extremely intense, the deep material is blocked by the internal thermal resistance, and its heat cannot respond in time and be conducted to the surface. Therefore, its corresponding attenuation weight will approach zero. The static physical heat capacity background value of each layer of the component is multiplied by its corresponding transient thermal coupling attenuation weight. All the calculated effective layered heat capacities are summed and aggregated to determine the surface equivalent dynamic heat capacity under the current meteorological and transient thermodynamic conditions. .

[0057] In this embodiment, the peak power temperature coefficient This is an inherent electrical degradation characteristic of semiconductor materials. The system directly retrieves the flash test matrix report generated at the factory for this model of photovoltaic module, based on the IEC 61215 certification standard, from a cloud database. This is based on standard test conditions (STC: Irradiance). (Atmospheric quality AM1.5), extract the maximum power point data for different controlled temperature gradients, and use mathematical partial derivatives to determine the peak power temperature coefficient. : ; In the formula, for The nominal maximum power under standard testing conditions.

[0058] In this embodiment 1, by fusing high-precision RTK positioning, IMU attitude data, and infrared camera nominal intrinsic parameters, the high-altitude patrol pose is projected and corrected to the true global GPS coordinates of the ground target heat point. Simultaneously, relying on the cloud-edge collaborative architecture, using this real global GPS coordinates As an index, the thermodynamic and electrical nominal parameters of the components are retrieved from the database. Combined with the gridded photovoltaic micro-weather stations within the site, the local real wind speed at the location of the heat source is obtained through an inverse distance spatial weighted interpolation algorithm. Specifically, for example: there are a total of A micro-weather station. The drone's coordinates are at a distance from station number 1. Measured wind speed Distance from Station 2 Measured wind speed Substituting into the formula, the local true wind speed can be calculated. This method eliminates the physical space mapping bias caused by the high-altitude observation perspective, ensuring that the target heat point corresponds to the actual wind load it bears and its own physical properties.

[0059] S4. Construct a heat flow decoupling model and calculate the dynamic severity index. To eliminate misjudgments of thermal hysteresis caused by sudden weather changes, such as Figure 2 As shown, the specific steps include: S41. For a healthy reference region within the same photovoltaic string, without physical obstruction and with normal electrical operation, the normal photothermal conversion energy received is equal to the sum of the convective heat dissipated outward from this region and the heat caused by temperature change due to the material's own thermal capacity absorption. For the region of interest, in addition to experiencing the same normal reference background heat flux as the healthy region, an abnormal electrothermal power density generated by internal high-impedance defects (such as microcracks or poor soldering) is additionally superimposed. Furthermore, since the abnormal heating region and the healthy reference region in the same string experience the same solar irradiance and environmental convection conditions, subtracting the transient thermal balance equations of the two will cancel out the normal photothermal conversion background energy, thus revealing the abnormal electrothermal power density purely generated by high-impedance defects (such as microcracks in the cells). The specific expression is as follows: ; In the formula, It is the differential heat capacity heat flux density that the abnormal region absorbs (or releases) more heat than the healthy region due to thermal inertia. This represents the differential convective heat flux density that is lost outward from the abnormal region compared to the healthy region.

[0060] S42, The system is based on the equivalent dynamic heat capacity of the target component surface. And combined with the rate of temperature change Calculate the differential heat capacity heat flux density: ; In the formula, It represents the differential heat capacity and heat flux density that the abnormal region absorbs (or releases) more heat than the healthy region due to thermal inertia.

[0061] S43. Utilize the highest temperature in the region of interest. Compared with reference temperature Calculate the differential convective heat flux density based on the surface temperature difference. : ; in, The dynamic surface heat transfer coefficient is given.

[0062] For example, in this embodiment, the dynamic surface heat transfer coefficient The method for determining it is as follows: The system synchronously retrieves nominal air properties at the current spatiotemporal point from cloud databases and meteorological platforms, including air thermal conductivity, kinematic viscosity, Prandtl number, and air volume expansion coefficient. Simultaneously, it extracts the installation tilt angle of the target photovoltaic module and the characteristic length used to characterize the fluid flow span on its surface. The real-time reference temperature and ambient absolute temperature of the healthy reference area are extracted, and the surface temperature difference between the two is used as the thermodynamic driving force. The gravitational acceleration is decomposed by combining the component installation tilt angle to evaluate the effective thermal buoyancy force. Subsequently, based on the theory of natural convection heat transfer on an inclined flat plate, the natural convection heat transfer coefficient dominated solely by temperature difference is calculated. The local real-time wind speed collected by the micro-weather station in the field is read and used as the core driving boundary of external forced cooling. Combined with the pre-extracted air kinematic viscosity and component characteristic length, the forced convection heat transfer coefficient dominated by wind speed is calculated. Given that natural convection and forced convection often act simultaneously and are nonlinearly coupled in real outdoor environments, the system introduces the Churchill-Usagi equation. The calculated natural convection heat transfer coefficient and the forced convection heat transfer coefficient are then nonlinearly superimposed using a third root-mean-square superposition to determine the current dynamic surface heat transfer coefficient. .

[0063] S44. The system extracts real-time and effective solar irradiance. Retrieve the nominal photoelectric conversion efficiency of the target component under standard test conditions from the database. And taking into account the process by which the component converts some light energy into electrical energy, the peak power temperature coefficient is utilized. Corrections are made for the photoelectric conversion efficiency degradation at the reference temperature to construct a baseline photothermal reference power density. : .

[0064] For example: In this embodiment, solar irradiance The determination steps include: S441, the cloud server synchronously reads data from the high-precision radiometer matrix of the field's micro-weather stations. Total irradiance of the horizontal plane at any time Horizontal diffuse irradiance and the ground reflected irradiance measured by the reflected radiation meter Simultaneously, the installation tilt angles of the target photovoltaic modules, recorded during factory manufacturing and construction acceptance, are extracted from the database. With component installation azimuth angle .

[0065] S442, Cloud server utilizes real global GPS coordinates in abnormal areas And precise UTC timestamps Substituting the values ​​into the Standard Ephemeris Algorithm (SPA), the solar space vector angle at that spatiotemporal node is calculated, including the solar zenith angle. With solar azimuth .

[0066] S443. The system performs a three-dimensional dot product operation between the spatial vector representing the normal direction of the photovoltaic module and the spatial vector representing the direct sunlight. Using spherical trigonometry formulas, it determines the true angle of incidence between the sunlight beam and the surface of the photovoltaic module. : .

[0067] S444. The system decomposes the total horizontal radiation into a normal direct component, and superimposes it with the scattered component and the ground reflection component according to the geometric projection relationship to calculate the real-time effective solar irradiance projected onto the photovoltaic module surface. : ; In the formula, The lower limit threshold of the zenith angle cosine is set for the system, for example, the cosine value corresponding to a zenith angle of 85°, to avoid overflow in the calculation of radiation values ​​under low solar altitude angle conditions.

[0068] S45, Local abnormal electrothermal power density Compared with the reference photothermal power density By performing ratio calculations, a heat-fluid decoupling model is constructed to calculate the dynamic severity index. : ; Substitute into the above formula: .

[0069] Specifically: under extreme gust wind sudden change conditions ( (A sharp increase), if the component temperature has not yet decreased due to thermal inertia, the temperature difference Maintaining a high level will lead to A false numerical expansion occurs; however, according to thermodynamic laws, strong convection will cause the component surface to begin cooling, meaning the rate of temperature change becomes negative. ).at this time, The number turned negative, thus canceling out the negative value. This leads to a false expansion. Therefore, the heat flow decoupling model can remain stable under meteorological changes by inversely transforming the delayed heat release into a mathematical compensation term.

[0070] S46. Set the minimum reference photothermal power density threshold. At night or under extreme cloud cover, when the baseline solar thermal power is too low and the thermodynamic driving conditions for photovoltaic power generation are not met, i.e. If the current weather conditions are deemed insufficient to effectively drive the components to generate heat, and no fault severity assessment is performed, then... .

[0071] in: ; In the formula, This is the noise equivalent temperature difference of an infrared thermal imager, representing the smallest effective temperature difference signal that the camera can resolve. For example, a certain industrial-grade thermal imager is labeled as 0.05K at the factory. This is the fault determination threshold.

[0072] In this embodiment 1, the abnormal electrothermal power generated by the high impedance defect in the target area is obtained. At the same time, utilizing real-time effective solar irradiance Temperature coefficient of peak power at the factory Determine the reference photothermal power density A heat flow decoupling model was constructed and the severity index was dynamically calculated. Specifically, for example: the system sets the nominal photoelectric conversion efficiency of the target component. Peak power temperature coefficient Current real-time effective solar irradiance Under a sudden gust of wind, the system calculates the dynamic surface heat transfer coefficient of the mixed convection. Combined with surface equivalent dynamic heat capacity The surface temperature difference between the region of interest and the healthy region was measured. Temperature change rate Reference temperature Substitute into the formula to calculate: Differential convection heat dissipation term. Differential heat capacity absorption term Pure abnormal electrothermal power density Reference power density The dynamic severity index, which characterizes the equivalent power loss ratio, is finally calculated. In the example above, under extreme sudden gusts of wind, the strong forced convection effect caused the differential convection heat dissipation term to increase by a significant amount. While it exhibits spurious expansion, this method captures the sudden drop in surface temperature of the component and utilizes dynamic heat capacity terms to reveal... The negative cancellation removes the false signal of heat flow expansion caused by meteorological changes, so that the calculation results truly reflect that the area is only the surface residual heat accumulated in the early stage. After being blown away by the wind, the actual abnormal heating inside is relatively weak, thus ensuring the accuracy of fault identification and reducing the false alarm rate.

[0073] S5. Severity Index With fault determination threshold Compare the data and make intelligent operation and maintenance decisions based on the comparison results, such as... Figure 3 As shown, the specific steps include: S511 compliant photovoltaic modules all include full-lifecycle power degradation commitment data in their factory BOM (Bill of Materials). The cloud server extracts factory degradation parameters from the database, including: the maximum permissible degradation rate in the first year. Annual maximum allowable linear attenuation rate and the initial commissioning timestamp of the component's grid connection for power generation. .

[0074] S512, Extract the current precise timestamp The actual service life of the computing components Since the factory-set degradation rate represents the overall uniform electrical aging of the module, while local hot spots represent non-uniform energy accumulation, the system introduces a local defect accumulation tolerance coefficient. This maps the macroscopic electrical attenuation benchmark to the microscopic thermally permissible limit of dissipation. : ; ; In the formula, for Threshold for determining micro-local faults at any given time; The physical number of seconds in a standard regression year; For mathematical truncation operators.

[0075] S52, Severity Index With fault determination threshold Compare: like This proves that the local heat accumulation in the area has exceeded the maximum microscopic non-uniform heat dissipation limit allowed by the current natural aging stage. The system confirms that the hot spot fault is caused by abnormal deep physical defects (such as severe microcracks or poor welding) and issues a maintenance work order. like The system determined that the heat generated in this area was normal due to the discrete heat generation caused by the internal resistance of the battery cells during the natural aging process, and confirmed it as normal.

[0076] For example, the tolerance coefficient for local defect aggregation The method for determining it is as follows: In a historical database of photovoltaic power plants or a laboratory environment, healthy photovoltaic modules of the same model as the target modules, but at different actual operating years, are selected as benchmark samples. These samples must undergo rigorous electroluminescence (EL) testing in advance to ensure that there are absolutely no abnormal deep physical defects such as microcracks, broken grids, or poor soldering, so as to ensure that their current performance degradation is purely caused by the natural aging of the materials. Under standard test conditions (STC), the actual macroscopic output power of each healthy sample component was measured using an IV curve tester, and the corresponding actual global power attenuation rate was calculated. At the same time, under outdoor steady-state illumination and natural convection conditions, the surface temperature distribution characteristics of the above sample components were collected using a high-resolution infrared thermal imager, and meteorological parameters at that time, including irradiance, wind speed, and ambient temperature, were recorded simultaneously. For all normal cell regions in the infrared images of the sample components, a dynamic severity index is calculated for each micro-pixel or cell level. Due to the microscopic inhomogeneities of semiconductor manufacturing tolerances and natural aging processes, the intrinsic resistance of each cell within a healthy component exhibits physical dispersion, manifesting as slight temperature differences. The system performs probability density statistics on the severity index of normal cells and introduces a normal distribution. The criteria are as follows: The maximum local heat dissipation extremum allowed by normal internal resistance dispersion under pure natural aging conditions is extracted. This maximum allowable local heat dissipation extremum is divided by the actual global power decay rate of the corresponding sample, and a least-squares linear regression is performed to fit and obtain a statistically significant local defect aggregation tolerance coefficient. .

[0077] In this embodiment 1, the factory nominal attenuation parameters and initial commissioning timestamp of the target component are retrieved from the database via a cloud server. The actual service life of the component is calculated by combining the inspection time. Calculate the dynamic fault determination threshold By using the severity index With fault determination threshold The comparison is used as the basis for the final decision on issuing maintenance work orders. This method can identify whether the heat generated in the heat-generating area is due to normal physical heat dissipation allowed by the natural degradation of the module over long-term operation, or a hot spot fault caused by abnormal deep physical structural damage. It avoids false fault alarms caused by reasonable natural aging of the module, and minimizes ineffective on-site personnel scheduling and blind work order dispatch while ensuring the safe operation of the power station. This reduces the entry and exit maintenance costs of unattended photovoltaic power stations located in remote areas with large land areas.

[0078] Example 2 As a second embodiment of the present invention, such as Figure 4 As shown in Example 1, this example also discloses a cloud-edge collaborative intelligent inspection and maintenance system for unattended photovoltaic power plants, specifically including: a video stream acquisition module, a temperature feature extraction module, a wind speed acquisition module, a heat flow decoupling evaluation module, and an intelligent maintenance decision-making module.

[0079] The system also includes drone inspection equipment equipped with infrared thermal imagers and visible light cameras, edge computing nodes and photovoltaic micro-weather stations deployed on-site, and remote cloud servers, which are used to collaboratively provide multimodal data acquisition and cross-terminal data interaction support.

[0080] The video stream acquisition module is used to acquire continuous multi-frame multimodal video streams of photovoltaic modules using inspection equipment, and to perform dynamic spatial coordinate registration at the edge computing node to obtain the two-dimensional center pixel coordinate vector of the region of interest in the current image frame. To overcome image jitter error and achieve precise locking of abnormally heated targets; The temperature feature extraction module is used to identify regions of interest suspected of abnormal heating in multiple consecutive frames of images. Within the same photovoltaic string, a healthy reference area with no physical obstruction and normal operation was selected. Extract the highest temperature respectively Compared with reference temperature And combined with sliding time window Calculate the rate of temperature change This is to transform isolated static temperature snapshots into continuous thermodynamic evolution trends; The wind speed acquisition module is used to map the spatial pose of the region of interest to real global GPS coordinates. The local real-time wind speed was calculated synchronously using an inverse distance spatial weighted interpolation algorithm combined with micro-weather station data. To eliminate high-altitude physical space mapping bias; The heat flux decoupling evaluation module is used to construct, based on the law of conservation of energy, the differential heat capacity and heat flux density that includes the additional absorption (or release) of heat in the abnormal region compared to the healthy region due to thermal inertia. The differential convective heat flux density lost outwards from the anomalous region compared to the healthy region. A heat flow decoupling model is used to dynamically calculate the severity index. In order to eliminate the misjudgment of false signals caused by sudden weather changes; The intelligent operation and maintenance decision-making module is used to combine the actual operating years of the target components. Dynamically calculate the fault determination threshold at the current lifespan node. Severity Index With fault determination threshold Compare and implement intelligent operation and maintenance decisions.

[0081] In the specific implementation of the above embodiment 2, firstly, the system processes the multimodal video stream acquired by the UAV at the edge computing node through the video stream acquisition module and obtains the two-dimensional center pixel coordinate vector. Registration overcomes image jitter errors caused by disturbances, ensuring accurate target locking. Secondly, the system dynamically extracts the highest temperature using a temperature feature extraction module. Compared with reference temperature The temperature change rate was calculated using least squares linear regression. By offsetting the interference of overall environmental fluctuations, the system can sense whether the internal heating driving force of components is continuous heat generation or delayed cooling. Then, through the collaborative calculation of the wind speed acquisition module and the heat flow decoupled evaluation module, the true global GPS coordinates are obtained. Local real-time wind speed The severity index is calculated by combining parameters with the thermodynamic properties of the components. By removing spurious signals from convective heat expansion, the system distinguishes between residual surface heat due to thermal inertia and genuine internal electrothermal anomalies. Finally, the intelligent operation and maintenance decision module uses the real-time calculated severity index. With fault determination threshold By comparing and adaptively distinguishing between normal physical aging and abnormal deep faults of components, the system minimizes ineffective on-site personnel scheduling and reduces the operation and maintenance costs of unattended photovoltaic power stations while ensuring safe operation of the power station.

[0082] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0083] The preferred embodiments of the invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants, characterized in that: include: The inspection equipment is used to acquire continuous multi-frame multimodal video streams of photovoltaic modules, and the continuous multi-frame multimodal video streams are spatially registered at the edge computing node; In the continuous multi-frame multimodal video stream, a region of interest suspected of abnormal heating is identified, and an unobstructed healthy reference region is selected within the same photovoltaic string. The highest temperature of the region of interest and the reference temperature of the healthy reference region are extracted respectively, and the temperature change rate is calculated. Obtain the actual global GPS coordinates of the region of interest, and simultaneously calculate the local real-time wind speed at the global GPS coordinates; A heat flow decoupling model is constructed by combining the highest temperature, reference temperature, temperature change rate, and local real-time wind speed, and a severity index reflecting the actual electrothermal anomaly inside the component is calculated. Obtain the factory nominal parameters and initial commissioning time of the target photovoltaic module, calculate the dynamic fault judgment threshold in combination with the current operating years, compare the severity index with the dynamic fault judgment threshold, and execute intelligent operation and maintenance decisions based on the comparison results.

2. The cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants according to claim 1, characterized in that, Dynamic spatial coordinate registration of the continuous multi-frame multimodal video stream at the edge computing node specifically includes: Key points of two consecutive frames of images are extracted using a feature extraction algorithm and preliminary feature matching is performed. A random sampling consensus algorithm is introduced to iteratively filter the initial matching points and calculate the global homography matrix representing the true rigid body geometric transformation relationship between two frames of images. The center pixel coordinates of the region of interest in the previous frame are converted into homogeneous coordinate vectors, and then combined with the global homography matrix to perform spatial linear transformation and perspective normalization processing to deduce the accurate two-dimensional center pixel coordinate vector of the region of interest in the current image frame.

3. The cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants according to claim 1, characterized in that, Extract the highest temperature of the region of interest and the reference temperature of the healthy reference region, and calculate the rate of temperature change, specifically including: The original infrared radiation grayscale matrix of the region of interest is transformed into an absolute temperature matrix using a radiometric calibration function, and the maximum value is extracted as the highest temperature of the region of interest. The independent photovoltaic module set within the photovoltaic string is segmented using edge detection and line transformation algorithms. After removing photovoltaic modules containing abnormal areas, the infrared spatial temperature variance inside each remaining candidate module is calculated. The photovoltaic module corresponding to the global minimum value of the infrared spatial temperature variance is selected as the healthy reference area, and the spatial arithmetic mean of the pixel temperature within the healthy reference area is used as the reference temperature. Calculate the relative surface temperature difference between the highest temperature and the reference temperature in the current frame, and within a set sliding time window, use least squares linear regression to fit the surface temperature difference data of multiple consecutive frames to determine the rate of temperature change reflecting the thermodynamic transient process.

4. The cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants according to claim 1, characterized in that, Obtain the actual global GPS coordinates of the region of interest, and simultaneously calculate the local real-time wind speed at the global GPS coordinates, specifically including: Read the high-precision global positioning coordinates, absolute altitude, and camera gimbal attitude angle of the inspection equipment at the moment when abnormal features are extracted, and query the local terrain elevation corresponding to the current coordinates based on the preset digital elevation model; By combining the camera's nominal intrinsic parameters, the actual effective pitch angle and effective yaw angle pointing towards the region of interest can be calculated; By combining the nominal support height, installation tilt angle, and installation azimuth angle of the target photovoltaic module, the precise spatial slant distance from the camera optical center to the abnormal heat point on the surface of the photovoltaic module is calculated. Based on the precise spatial slant distance, the position offset of the target point relative to the vertical foot of the UAV is decomposed into east and north components, and the standard radius of curvature of the Earth is introduced and superimposed on the global positioning coordinates of the inspection equipment to determine the true global GPS coordinates of the abnormal heating point of the photovoltaic module. The cloud server uses an inverse distance spatial weighted interpolation algorithm, combined with real-time wind speeds collected by multiple micro-weather stations within the photovoltaic field and the straight-line physical spatial distance between the micro-weather stations and the global GPS coordinates, to calculate the local real-time wind speed of the region of interest.

5. The cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants according to claim 1, characterized in that, A heat flux decoupling model is constructed by combining the highest temperature, reference temperature, temperature change rate, and local real-time wind speed to calculate the severity index, specifically including: The differential heat flux density of the abnormal region compared to the healthy region due to thermal inertia is determined by multiplying the rate of temperature change with the surface equivalent dynamic heat capacity of the photovoltaic module. The differential convective heat flux density lost from the abnormal region to the outside compared to the healthy region is determined by multiplying the relative surface temperature difference between the highest temperature and the reference temperature by the dynamic surface heat transfer coefficient. The local abnormal electrothermal power density of the region of interest is determined by combining the differential heat capacity heat flux density absorbed by the abnormal region compared to the healthy region due to thermal inertia and the differential convective heat flux density lost by the abnormal region compared to the healthy region to the outside. The real-time effective solar irradiance projected onto the surface of the photovoltaic module is obtained, and the photoelectric conversion efficiency attenuation at the reference temperature is corrected using the peak power temperature coefficient. The nominal photoelectric conversion efficiency is obtained and a reference photothermal power density is constructed by combining the real-time effective solar irradiance. The ratio of the local abnormal electrothermal power density to the reference photothermal power density is calculated to construct a heat flow decoupling model and calculate the dynamic severity index.

6. The cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants according to claim 5, characterized in that, The method for obtaining the surface equivalent dynamic heat capacity includes: Obtain the density, standard specific heat capacity, nominal thickness, and thermal conductivity of each physical encapsulation layer inside the photovoltaic module; The inherent static physical heat capacity background value of each physical encapsulation layer is calculated using the density, standard specific heat capacity and nominal thickness of each physical encapsulation layer. The cumulative normal thermal resistance from the geometric center of each layer to the observation surface is calculated based on the nominal thickness and thermal conductivity of each physical encapsulation layer. The dynamic convective heat transfer intensity on the surface of the photovoltaic module is evaluated by combining local real-time wind speed, and the cumulative normal thermal resistance is coupled with the dynamic convective heat transfer intensity to construct a transient thermal coupling attenuation weight. The static physical thermal capacity background value of each physical encapsulation layer is multiplied by its corresponding transient thermal coupling attenuation weight and then summed and aggregated to determine the surface equivalent dynamic thermal capacity.

7. The cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants according to claim 5, characterized in that, The method for obtaining the dynamic surface heat transfer coefficient includes: Retrieve the nominal parameters of air properties and extract the installation tilt angle of the target photovoltaic module and the characteristic length characterizing the fluid flow span on its surface; The natural convection heat transfer coefficient dominated by temperature difference is calculated by combining the reference temperature, ambient absolute temperature and installation tilt angle. By combining local real-time wind speed, nominal air properties and the characteristic length, the forced convection heat transfer coefficient dominated by wind speed is calculated. The Churchill-Usagi equation is introduced to nonlinearly superimpose the natural convection heat transfer coefficient and the forced convection heat transfer coefficient to determine the dynamic surface heat transfer coefficient.

8. The cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants according to claim 7, characterized in that, After calculating the severity index, the evaluation exclusion mechanism is also set up: Combining the dynamic surface heat transfer coefficient, the equivalent temperature difference of the infrared thermal imager's hardware noise, and the fault judgment threshold, the minimum reference photoelectric power density threshold that just makes the infrared thermal imager reach the physical limit resolution is derived. When the real-time calculated reference photoelectric power density is lower than the minimum reference photoelectric power density threshold, it is determined that the current meteorological conditions do not have the thermodynamic conditions to effectively drive the component to generate heat, no evaluation calculation is performed, and the severity index is directly set to zero.

9. The cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants according to claim 1, characterized in that, The severity index is compared with the fault determination threshold, and intelligent operation and maintenance decisions are executed based on the comparison results, specifically including: If the calculated severity index is greater than the dynamic fault determination threshold, it is confirmed as an internal hot spot fault caused by abnormal physical defects and a maintenance work order is issued. If the severity index is less than or equal to the dynamic fault determination threshold, the heat generation is determined to be within the normal physical heat dissipation range allowed in the current natural aging stage, and is confirmed as normal.

10. A cloud-edge collaborative intelligent inspection and maintenance system for unattended photovoltaic power plants, employing the cloud-edge collaborative intelligent inspection and maintenance method for unattended photovoltaic power plants as described in any one of claims 1 to 9, characterized in that... include: The video stream acquisition module is used to acquire continuous multi-frame multimodal video streams of photovoltaic modules using inspection equipment, and to perform spatial coordinate registration at the edge computing node; The temperature feature extraction module is used to identify regions of interest suspected of abnormal fever, select a healthy reference region, extract the highest temperature and reference temperature, and calculate the temperature change rate. The wind speed acquisition module is used to obtain the real global GPS coordinates of the area of ​​interest and calculate the local real-time wind speed. The thermal-fluid decoupling assessment module is used to build a thermal-fluid decoupling model and calculate the severity index. The intelligent operation and maintenance decision module is used to compare the severity index with the fault judgment threshold and execute intelligent operation and maintenance decisions.

Images