A photovoltaic panel cleanliness detection system and method

Abstract

The application discloses a photovoltaic panel cleanliness detection system and method, and relates to the field of photovoltaic panel cleaning.The detection system comprises a cleaning robot, an infrared thermal imaging module, a dual-spectrum thermal imager, a control module and a software control platform.The infrared thermal imaging module is installed at one end of the cleaning robot, and the dual-spectrum thermal imager is installed at the other end of the cleaning robot.The control module is connected with the infrared thermal imaging module, the dual-spectrum thermal imager, the control bus of the cleaning robot and the software control platform, is used for receiving the data sent by the infrared thermal imaging module and the dual-spectrum thermal imager, generating an alarm strategy and a cleaning strategy after processing, and controlling the cleaning robot to clean the photovoltaic panel according to the cleaning strategy.Meanwhile, the control module sends the alarm strategy to the software control platform, and the software control platform sends the alarm strategy to the related operation and maintenance personnel.The hot spot detection and the dirt detection function are integrated on the cleaning robot, no additional inspection equipment is needed, the timeliness is strong, and the full-station photovoltaic panel physical examination is completed synchronously every time the photovoltaic panel cleaning operation is performed.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic equipment technology, specifically relating to a photovoltaic panel cleanliness detection system and method. Background Technology

[0002] With the large-scale construction of photovoltaic power plants, dirt such as dust, bird droppings, and leaves on the surface of photovoltaic panels, as well as the hot spot effect caused by internal defects, will significantly reduce power generation efficiency and may even cause the modules to burn out in severe cases.

[0003] Existing track-mounted photovoltaic cleaning robots typically only have a single wet cleaning or dry brushing function and lack the ability to assess cleaning effectiveness in real time. During operation and maintenance, the detection of hot spots or localized dirt on photovoltaic panels still relies on manual inspection with handheld thermal imaging cameras, which is inefficient, costly, and fails to meet the timeliness requirements for early fault identification. Summary of the Invention

[0004] To address the technical problem that existing technologies cannot simultaneously detect the degree of dirt and hot spot distribution on photovoltaic panels during the cleaning process, this invention provides a photovoltaic panel cleanliness detection system and method.

[0005] The technical problem to be solved by this invention is achieved through the following technical solution:

[0006] A photovoltaic panel cleanliness detection system includes: a cleaning robot, an infrared thermal imaging module, a dual-spectrum thermal imager, a control module, and a software control platform.

[0007] The infrared thermal imaging module is installed at one end of the cleaning robot's direction of travel and tilted downwards at an angle of 10° to 30° with the horizontal plane;

[0008] The dual-spectrum thermal imager is installed at the other end of the cleaning robot's travel direction and tilted downwards at an angle of 10° to 30° with the horizontal plane;

[0009] The control module is connected to the infrared thermal imaging module, the dual-spectrum thermal imager, the control bus of the cleaning robot, and the software control platform. It receives data sent by the infrared thermal imaging module and the dual-spectrum thermal imager, processes it to generate alarm and cleaning strategies, and controls the cleaning robot to clean the photovoltaic panels according to the cleaning strategies. At the same time, the control module sends the alarm strategies to the software control platform, which then sends them to the relevant maintenance personnel.

[0010] Another embodiment of the present invention discloses a method for detecting the cleanliness of photovoltaic panels, implemented using the aforementioned photovoltaic panel cleanliness detection system, comprising:

[0011] S1: Start the cleaning robot and move it along the photovoltaic panel array to perform cleaning operations on the photovoltaic panel surface. At the same time, use the infrared thermal imaging module to obtain the infrared temperature matrix of the current detection area; and use the dual-spectrum thermal imager to obtain the infrared thermal image and visible light image of the current detection area.

[0012] S2: Process the infrared temperature matrix to obtain apparent fault data of the photovoltaic panel area in the current detection area. The apparent fault data includes the area ratio of the abnormal area, the average temperature difference, and the location. The average temperature difference is obtained by subtracting the average temperature of the abnormal area from the average temperature of the photovoltaic panel area.

[0013] The infrared thermal image and the visible light image are processed to obtain the state feature data of the photovoltaic panel area in the current detection area. The state feature data includes the dust accumulation level, the anomaly type, and the corrected inter-panel temperature difference. The inter-panel temperature difference is obtained by subtracting the temperature value of a single pixel in the photovoltaic panel area from the average temperature value of the photovoltaic panel area.

[0014] S3: Determine the anomaly type and location based on the area ratio, average temperature difference, and location of the abnormal area, and determine the alarm strategy at the abnormal location based on the anomaly type;

[0015] The cleaning strategy is determined based on the dust accumulation level, the anomaly type, and the corrected inter-plate temperature difference.

[0016] Further, in step S2, the infrared temperature matrix is ​​processed to obtain apparent fault data of the photovoltaic panel area in the current detection area, including:

[0017] The infrared temperature matrix is ​​preprocessed to obtain an effective infrared temperature matrix;

[0018] The effective infrared temperature matrix is ​​subjected to threshold segmentation to generate a binary mask matrix, and the binary mask matrix is ​​superimposed with the effective infrared temperature matrix to obtain the infrared temperature matrix of the photovoltaic panel area in the current detection area.

[0019] Extract data from the infrared temperature matrix of the photovoltaic panel area to form an effective temperature list;

[0020] An adaptive threshold algorithm is used to calculate the data in the effective temperature list to obtain the optimal temperature threshold. The infrared temperature matrix of the photovoltaic panel area is then binarized based on the optimal temperature threshold to obtain the binary matrix of the abnormal area.

[0021] The data in the effective temperature list are calculated to obtain the average temperature and total effective pixel count of the photovoltaic panel area. Morphological denoising and connected component analysis are performed on the binary matrix of the abnormal area to finally calculate the area ratio, average temperature and center coordinates of each abnormal area.

[0022] The average temperature difference between the photovoltaic panel area and the average temperature of each abnormal area is obtained, and the location of each abnormal area is determined based on the center coordinates of each abnormal area.

[0023] Further, the infrared temperature matrix is ​​preprocessed to obtain an effective infrared temperature matrix, including:

[0024] Gaussian noise in the infrared temperature matrix is ​​removed using Gaussian filtering.

[0025] Then, median filtering is used to detect and replace bad pixels to obtain the effective infrared temperature matrix.

[0026] Further, the effective infrared temperature matrix is ​​subjected to threshold segmentation to generate a binary mask matrix, including:

[0027] The effective infrared temperature matrix is ​​normalized from 16 bits to an 8-bit map;

[0028] The effective infrared temperature matrix of the 8-bit image is thresholded using the big law method to generate the binary mask matrix.

[0029] Further, in step S3, the anomaly type and location are determined based on the area ratio, average temperature difference, and location of the abnormal region, and an alarm strategy at the abnormal location is determined based on the anomaly type; including:

[0030] The anomaly type and location are determined based on the area proportion, average temperature difference, and location of the abnormal region.

[0031] The control module generates an alarm strategy for the corresponding abnormal location based on the abnormality type, maps the infrared temperature matrix of the photovoltaic panel area into an 8-bit color image, and marks the abnormal location and abnormality type on the 8-bit color image; finally, the alarm strategy and the 8-bit color image are sent to the software control platform.

[0032] Furthermore, the alarm strategy includes:

[0033] When the average temperature difference ΔT ≥ 25℃ and the area of ​​the abnormal region ≥ 5%, the control module marks it as a severe hot spot, records the location and thermal imaging screenshot, and immediately reports it to the software control platform.

[0034] When the average temperature difference ΔT ≥ 15℃ and the area of ​​the abnormal region accounts for 2% to 5%, the control module marks it as a normal hot spot, records the location and thermal imaging screenshot, and controls the cleaning robot to complete the current cleaning task before reporting to the software control platform.

[0035] When the abnormal area is a localized low temperature and the edge of the area is clear, the control module marks it as an obstruction, records the location, and controls the cleaning robot to complete the cleaning task before reporting to the software control platform.

[0036] When the abnormal area is an irregular high-temperature zone with a large temperature difference, the control module marks it as damaged, records the location and thermal imaging screenshot, and controls the cleaning robot to complete the current cleaning task before reporting it to the software control platform first.

[0037] Further, the infrared thermal image and visible light image are processed to obtain state characteristic data of the photovoltaic panel area in the current detection area. The state characteristic data includes the dust accumulation level and the corrected inter-panel temperature difference, including:

[0038] The temperature matrix corresponding to the infrared thermal image is preprocessed to obtain the enhanced infrared thermal image.

[0039] A semantic segmentation model is used, and the enhanced infrared thermal image is segmented into photovoltaic panel regions using the visible light image to extract the infrared thermal image of the photovoltaic panel region.

[0040] Based on the infrared temperature matrix corresponding to the infrared thermal image of the photovoltaic panel area, multi-dimensional dust accumulation features are extracted and calculated. The multi-dimensional dust accumulation features include the temperature of a single pixel in the photovoltaic panel area, average temperature, temperature variance, temperature standard deviation, inter-panel temperature difference, temperature gradient features, equivalent infrared emissivity features, and texture features.

[0041] The infrared thermal image and multi-dimensional dust accumulation features of the photovoltaic panel area are input into a lightweight classification model for processing to obtain the dust accumulation level of the photovoltaic panel area.

[0042] The multi-dimensional dust accumulation features are processed to obtain the anomaly type of the photovoltaic panel area;

[0043] The environment in which the photovoltaic panels are located is compensated and corrected to eliminate the interference of weather and time of day on the temperature field, and the corrected temperature difference between the panels is calculated.

[0044] Furthermore, the anomaly types are as follows: when the average temperature of the abnormal area of ​​the photovoltaic panel minus the average temperature of the photovoltaic panel area is ≥15℃, and the abnormal area exceeding this temperature is ≥2%, the control module marks the location as having a hot spot and reports it to the software control platform; when there is a blocky high temperature area in the photovoltaic panel area with an irregular shape, the control module marks the location as having bird droppings / stains and reports it to the software control platform; when there is a large area of ​​low temperature in the photovoltaic panel area, the control module marks the location as having shading; when there is a clearly defined and low temperature area in the photovoltaic panel area, the control module marks the location as having shadows.

[0045] Furthermore, a cleaning strategy is determined based on the dust accumulation level, the corrected inter-plate temperature difference, and the anomaly type, including:

[0046] When the dust accumulation level of the photovoltaic panel is 0, no cleaning is required; when the dust accumulation level is 1, regular cleaning is performed; when the dust accumulation level is 2, the cleaning robot records the location and reduces its travel speed by 50% when cleaning that location; when the dust accumulation level is 3, the temperature difference between panels is ≥3℃, or the abnormality type is bird droppings / stains, the number of brushing cycles is increased by 2, and a fixed-point re-wash is performed and the data is verified until the dust accumulation level of the photovoltaic panel area drops below 1, at which point the cleaning process for the next area is started.

[0047] The beneficial effects of this invention are:

[0048] 1. This invention integrates hot spot detection and dirt detection functions into the existing track-mounted photovoltaic cleaning robot, eliminating the need for additional inspection equipment and providing high timeliness. During each photovoltaic panel cleaning operation, a full-station photovoltaic panel inspection is completed simultaneously, identifying potential hot spot hazards in the first instance, thereby improving production efficiency and reducing operation and maintenance costs, achieving "cleaning, detection, and precise operation and maintenance".

[0049] 2. In addition, the dual-spectrum thermal imager can be used to compare and distinguish between ordinary dust and stubborn stains, and the robot can be controlled to perform targeted rewashing or the software control platform can be alerted to prompt manual handling.

[0050] 3. By comparing data before and after cleaning, normal dust can be avoided from being misjudged as hot spots, thereby improving the accuracy of detection. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of the detection system of the present invention;

[0052] Figure 2 This is the overall flowchart of the present invention.

[0053] Explanation of reference numerals in the attached figures:

[0054] 1-Cleaning robot; 2-Infrared thermal imaging module; 3-Dual-spectrum thermal imager; 4-Control module; 5-Software control platform; 6-Robot control bus. Detailed Implementation

[0055] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0056] Example 1

[0057] This invention discloses a photovoltaic panel cleanliness detection system, specifically including a cleaning robot, an infrared thermal imaging module, a dual-spectrum thermal imager, a control module, and a software control platform.

[0058] The cleaning robot is equipped with a robot cover plate at its upper end, and the infrared thermal imaging module and the dual-spectrum thermal imager are mounted on the robot cover plate.

[0059] The infrared thermal imaging module is installed at one end of the cleaning robot's travel direction and tilted downwards at an angle of 10-30° to the horizontal plane, covering the entire surface of a single photovoltaic panel with a resolution of 32×24. It is used to collect the infrared thermal radiation temperature distribution on the surface of the photovoltaic panel, identify local shading or hot spot effects caused by damage to the photovoltaic panel, and send the specific location of the anomaly to the control module.

[0060] The dual-spectrum thermal imager is installed at the other end of the cleaning robot's travel direction and tilted downwards at an angle of 10-30° to the horizontal plane. It uses a microbolometer to collect the difference in reflectivity and emissivity of the photovoltaic panel surface after cleaning, in conjunction with an active light source or using ambient light, to identify residual stubborn dirt and send it to the control module.

[0061] The control module is connected to the infrared thermal imaging module, the dual-spectrum thermal imager, the control bus of the cleaning robot, and the software control platform. It receives data sent by the infrared thermal imaging module and the dual-spectrum thermal imager, processes it to generate alarm and cleaning strategies, and controls the cleaning robot to clean the photovoltaic panels according to the cleaning strategies. At the same time, the control module sends the alarm strategies to the software control platform, which then sends them to the relevant maintenance personnel.

[0062] In addition, the initial cleaning strategy for photovoltaic panels, including the number of cleaning cycles and cleaning time, is sent by the software control platform to the control bus of the cleaning robot through the control module, thereby setting the initial cleaning parameters of the cleaning robot.

[0063] Example 2

[0064] This invention discloses a method for detecting the cleanliness of photovoltaic panels, such as... Figure 1 As shown, the specific steps include:

[0065] S1: Start the cleaning robot and move it along the photovoltaic panel array to perform cleaning on the photovoltaic panel surface. At the same time, use the infrared thermal imaging module to obtain the infrared temperature matrix of the current detection area of ​​the photovoltaic panel. The infrared thermal imaging module outputs a 16-bit 32×24 two-dimensional floating-point temperature matrix in °C, with a sampling frequency of 10Hz and a data format of T_in

[32]

[24] . The dual-spectrum thermal imager outputs an infrared thermal image and a visible light image of the current detection area of ​​the photovoltaic panel with a diameter of 8~14μm.

[0066] Since the cleaning robot moves back and forth on the photovoltaic panel array to clean, when the cleaning robot moves forward for the first time, the infrared thermal imaging module collects the infrared temperature matrix of the photovoltaic panel before cleaning and inputs it into the STM32 main control unit, i.e., the control module, through the I2C communication protocol; the dual-spectrum thermal imager collects the infrared thermal image and visible light image of the photovoltaic panel after cleaning and inputs it into the control module through the I2C communication protocol.

[0067] When the control module receives data collected by the infrared thermal imaging module, it first determines the detection conditions: Condition 1: The average temperature of the photovoltaic panel area in the current detection area minus the ambient temperature is ≥3℃; Condition 2: The irradiance measured by the solar irradiance sensor on the cleaning robot is ≥200W / m². 2 If both conditions are met, the subsequent steps continue. If either condition is not met, hot spot detection is not performed in the current detection area, and the control module sends the status signal to the control bus of the cleaning robot to skip the current hot spot detection and continue to perform the cleaning action. The next sampling cycle will then determine the condition. When both detection conditions are met, the following data processing procedure will be executed.

[0068] It should be noted that the photovoltaic panel area in the following analysis and calculations refers to the photovoltaic panel area in the current detection area.

[0069] S2: Process the infrared temperature matrix to obtain apparent fault data of the photovoltaic panel area in the current detection area. The apparent fault data includes the area ratio of the abnormal area, the average temperature difference, and the location. The average temperature difference is obtained by subtracting the average temperature of the abnormal area of ​​the photovoltaic panel from the average temperature of the photovoltaic panel area.

[0070] The infrared thermal image and the visible light image are processed to obtain the state feature data of the photovoltaic panel area in the current detection area. The state feature data includes the dust accumulation level, the anomaly type, and the corrected inter-panel temperature difference. The inter-panel temperature difference refers to the difference between the temperature value of a single pixel in the photovoltaic panel area in the current detection area and the average temperature value of the photovoltaic panel area in the current detection area.

[0071] Specifically, the infrared temperature matrix is ​​processed to obtain apparent fault data of the photovoltaic panel area in the current detection area, including the following steps:

[0072] (1) The infrared temperature matrix is ​​preprocessed to obtain an effective infrared temperature matrix;

[0073] Furthermore, the preprocessing process includes: firstly, applying Gaussian filtering (kernel size 3×3, σ=0) to remove Gaussian noise in the infrared temperature matrix, then using median filtering to detect and replace bad pixels, and finally outputting the effective infrared temperature matrix after filtering and bad pixel correction.

[0074] (2) The effective infrared temperature matrix is ​​subjected to threshold segmentation to generate a binary mask matrix, and the binary mask matrix is ​​superimposed with the effective infrared temperature matrix to obtain the infrared temperature matrix of the photovoltaic panel area in the current detection area.

[0075] Specifically, the effective infrared temperature matrix is ​​transformed from a 16-bit normalized 8-bit image, and then the effective infrared temperature matrix of the 8-bit image is thresholded using the big rule method to generate a binary mask matrix.

[0076] (3) Extract the data from the infrared temperature matrix of the photovoltaic panel area to form an effective temperature list.

[0077] (4) The data in the effective temperature list are calculated using an adaptive threshold algorithm to obtain the optimal temperature threshold, and the infrared temperature matrix of the photovoltaic panel area is binarized according to the optimal temperature threshold to obtain the binary matrix of the abnormal area.

[0078] Specifically, the data in the effective temperature list are automatically calculated by an adaptive threshold algorithm to obtain the optimal temperature threshold. Then, the infrared temperature matrix of the photovoltaic panel area is binarized according to the optimal temperature threshold. Pixels higher than the optimal temperature threshold are marked as abnormal pixels and marked as 1; pixels lower than the optimal temperature threshold are marked as normal pixels and marked as 0. Finally, the binary matrix B_hot of the abnormal area is output

[32]

[24] .

[0079] (5) Calculate the data in the effective temperature list to obtain the average temperature and total number of effective pixels in the photovoltaic panel area, and perform morphological denoising and connected component analysis on the binary matrix of the abnormal area to finally calculate the area ratio, average temperature and center coordinates of each abnormal area.

[0080] (6) The average temperature difference between each abnormal area is obtained based on the average temperature of the photovoltaic panel area and the average temperature of each abnormal area, and the location of each abnormal area is determined based on the center coordinates of each abnormal area.

[0081] Specifically, the average temperature difference ΔT between each abnormal region is equal to the average temperature of each abnormal region minus the average temperature of the photovoltaic panel region.

[0082] Specifically, the infrared thermal image and the visible light image are processed to obtain state characteristic data of the photovoltaic panel area in the current detection area. The state characteristic data includes the dust accumulation level, anomaly type, and corrected inter-panel temperature difference, including the following steps:

[0083] (1) Preprocess the temperature matrix corresponding to the infrared thermal image to obtain the enhanced infrared thermal image;

[0084] Furthermore, the preprocessing includes temperature normalization, noise reduction, and histogram equalization of the temperature matrix corresponding to the infrared thermal image to obtain an enhanced infrared thermal image.

[0085] (2) A semantic segmentation model is adopted, and the enhanced infrared thermal image is segmented into photovoltaic panel regions using the visible light image to extract the infrared thermal image of the photovoltaic panel region.

[0086] The enhanced infrared thermal image and visible light image are input into an AI semantic model for segmentation. For example, U-Net, MobileNet-UNet, or SegFormer models perform pixel recognition on the enhanced infrared thermal image, automatically identifying and distinguishing interference areas such as photovoltaic panels, shadows, debris, and the ground in the image. The clear edges of the visible light image are used to more accurately segment the photovoltaic panel area. Only the photovoltaic panel area is retained in the output mask, while the other areas are masked. Finally, the infrared thermal image of the photovoltaic panel area is extracted.

[0087] (3) Based on the infrared temperature matrix corresponding to the infrared thermal image of the photovoltaic panel area, extract and calculate multi-dimensional dust accumulation features, including the temperature of a single pixel in the photovoltaic panel area, average temperature, temperature variance, temperature standard deviation, inter-panel temperature difference, temperature gradient features, equivalent infrared emissivity features, and texture features.

[0088] The temperature values ​​corresponding to the photovoltaic panel area are filtered out using array operations in Python / NumPy to form a one-dimensional array. Then, statistical calculations are performed on this array to obtain multi-dimensional dust accumulation features. Among them, the inter-panel temperature difference = the temperature value of a single pixel in the photovoltaic panel area of ​​the current detection area - the average temperature value of the photovoltaic panel area of ​​the current detection area; the infrared emissivity equivalent feature is indirectly calculated using the inter-panel temperature difference, which is a current technology; the texture feature is extracted using the LBP / GLCM algorithm; and the temperature gradient feature is the maximum rate of change of the temperature difference between adjacent pixels.

[0089] (4) Input the infrared thermal image and multi-dimensional dust accumulation features of the photovoltaic panel area into a lightweight classification model for processing to obtain the dust accumulation level of the photovoltaic panel area;

[0090] The lightweight classification model is the MobileNetV3 model; the resulting dust accumulation levels of the photovoltaic panel area include: 0: clean; 1: light dust accumulation; 2: moderate dust accumulation; 3: heavy dust accumulation.

[0091] (5) At the same time, the multi-dimensional dust accumulation features are processed to obtain the anomaly type of the photovoltaic panel area.

[0092] Specifically, there are four types of anomalies: When there is a temperature anomaly in the photovoltaic panel area, that is, the average temperature difference ΔT between the average temperature of the abnormal area and the average temperature of the photovoltaic panel area is ≥15℃, and the abnormal area is ≥2%, the control module marks the area as having a hot spot, records the location and thermal imaging screenshot, and reports it to the software control platform; When there is a block of high temperature in the photovoltaic panel area and the shape is irregular, the control module marks the area as having bird droppings / stains, records the location and thermal imaging screenshot, and reports it to the software control platform; When there is a large area of ​​low temperature in the photovoltaic panel area, the control module marks the area as having shading; When there is a clearly defined and low temperature area, the control module marks the area as having shadow.

[0093] Among them, the temperature of blocky high temperature refers to the average temperature difference ΔT between the average temperature of the temperature anomaly area of ​​the photovoltaic panel and the average temperature of the photovoltaic panel area, which is between 8℃ and 15℃. Whether it belongs to an irregular blocky structure can be distinguished by texture features or "shape factor". At the same time, since bird droppings / stains have obvious color and texture features on the visible light image, they can also be used to help distinguish whether they belong to a blocky structure.

[0094] (6) Compensate and correct the environment in which the photovoltaic panel is located, and correct the temperature of a single pixel point according to the corrected environmental parameters, and calculate the corrected inter-panel temperature difference; wherein, the environmental compensation and correction includes eliminating the interference of weather and time period on the temperature field.

[0095] Corrected interplate temperature difference = T 修正 - Average temperature of the photovoltaic panel area.

[0096] Among them, the corrected pixel temperature T 修正 The calculation formula is:

[0097]

[0098] in, Solar irradiance; Ambient temperature; Wind speed; , , These are compensation coefficients trained using experimental data. Generally speaking... Take 0.001-0.01℃ / (W / m²); Take a value between 0.1 and 0.4; Take 0.2-1.0℃ / (m / s).

[0099] It should be noted that steps (4), (5), and (6) in the process of processing the infrared thermal image and the visible light image are performed simultaneously without distinguishing the order.

[0100] S3: Determine the anomaly type and location based on the area ratio, average temperature difference, and location of the abnormal area, and determine the alarm strategy at the abnormal location based on the anomaly type; determine the cleaning strategy based on the dust accumulation level, anomaly type, and corrected inter-board temperature difference.

[0101] Specifically, the control module determines the anomaly type and location based on the area ratio, average temperature difference, and location of the abnormal region, and determines the alarm strategy at the abnormal location based on the anomaly type, including the following steps:

[0102] (1) The control module confirms the anomaly type and location based on the area ratio, average temperature difference and location of the anomaly area;

[0103] (2) The control module generates an alarm strategy for the corresponding abnormal location based on the abnormality type, and maps the infrared temperature matrix of the photovoltaic panel area into an 8-bit color image, and marks the abnormal location and abnormality type on the 8-bit color image; finally, the alarm strategy and the 8-bit color image are sent to the software control platform, and the software control platform notifies the operation and maintenance personnel to carry out specific processing.

[0104] Specifically, the alarm strategy is as follows: when the average temperature difference ΔT ≥ 25℃ and the area of ​​the abnormal region ≥ 5%, the control module marks it as a "severe hot spot," records its location and thermal imaging screenshot, and immediately reports it to the software control platform via 4G or WIFI, prompting manual handling; when the average temperature difference ΔT ≥ 15℃ and the area of ​​the abnormal region is 2%~5%, the control module marks it as a "normal hot spot," records its location and thermal imaging screenshot, and controls the cleaning robot to complete the current cleaning task before reporting it to the software control platform, prompting manual handling; when the abnormal area is a localized low temperature and the... If the edge of the abnormal area is clear, the area of ​​the abnormal area is not considered. The control module marks it as an obstruction, records its location, and controls the cleaning robot to complete the cleaning task before reporting it to the software control platform, prompting manual handling. Local low temperature means that the average temperature of the abnormal area minus the average temperature of the photovoltaic panel area is < 8℃. When the abnormal area is an irregular high temperature area with a large temperature difference, the control module marks it as damaged, records its location and thermal imaging screenshot, and controls the cleaning robot to complete the current cleaning task before reporting it to the software control platform first, prompting manual handling. Large temperature difference means that the average temperature difference ΔT of the abnormal area is ≥ 25℃.

[0105] It should be noted that the Canny edge detection method is used to determine whether the edges of abnormal regions are clear. First, pixels with larger gradient values ​​are more like boundaries. Second, the number of edge pixels detected by Canny on the boundary of the target region is counted. A high percentage equals a clear edge (above 60%: very clear edge; 40%-60%: relatively clear edge; below 40%: blurry edge).

[0106] Use the "shape factor" to determine if a shape is regular: calculate the area-to-perimeter ratio for high-temperature areas (circularity = 4π × area / perimeter). 2 If the value deviates significantly from 1 (1 for a circle), it indicates that the shape is irregular. Hot spots are usually round spots, while the high-temperature areas caused by physical damage (such as broken battery cells) are often elongated or irregular in shape.

[0107] In addition, the highest temperature, lowest temperature and standard deviation of the photovoltaic panel area can be calculated from the effective temperature list data in step (3) of the infrared temperature matrix processing process. The control module uses these three parameters to make a preliminary judgment on the type of abnormality. When the highest temperature of the photovoltaic panel area minus the average temperature of the photovoltaic panel area is less than 15°C, the control module determines that there are no hot spots in the photovoltaic panel area. When the lowest temperature of the photovoltaic panel area minus the average temperature of the photovoltaic panel area is greater than 8°C, the control module determines that there is no shading or shadow in the photovoltaic panel area. The control module can determine whether the temperature distribution difference of the photovoltaic panel area in the current test area is within the normal range by using the temperature standard deviation. When the temperature standard deviation is ≤3°C, the temperature of the photovoltaic panel area is determined to be within the normal range.

[0108] Furthermore, the control module determines the cleaning strategy based on the dust accumulation level of the photovoltaic panel, the anomaly type, and the corrected inter-panel temperature difference.

[0109] When the dust accumulation level of the photovoltaic panel is 0, no cleaning is required; when the dust accumulation level is 1, regular cleaning is performed; when the dust accumulation level is 2, the cleaning robot records the location and reduces its travel speed by 50% when cleaning the location next time; when the dust accumulation level is 3, the temperature difference between panels is ≥3℃, or the abnormality type is bird droppings / stains, the number of brushing operations is increased by 2, and a fixed-point re-wash is performed and the data is verified until the dust accumulation level of the area drops below 1, and the cleaning process for the next area is started.

[0110] In addition, after a complete cleaning operation is completed during the outbound phase, the photovoltaic cleaning robot will re-inspect the abnormal areas identified during the outbound phase during the return phase, and re-determine the abnormality type and dust accumulation level of each abnormal area. For areas marked as hot spot defects during the outbound phase, if the hot spot characteristics are not reproduced during the return phase inspection, it is determined that normal dust accumulation has been misidentified as hot spots. In this case, the coordinate information of the area and the corresponding thermal imaging screenshot are recorded and reported to the software control platform, and it is recommended to cancel the manual review process for the area.

[0111] In addition, the cost loss caused by photovoltaic panel contamination can be calculated by the average temperature and temperature standard deviation in the infrared temperature matrix corresponding to the infrared thermal image of the photovoltaic panel area.

[0112] First, the average temperature, temperature standard deviation, environmental irradiance, ambient temperature, and wind speed of the photovoltaic panel area are processed using a regression algorithm to obtain the dust accumulation surface density (g / m³). 2 ).

[0113] Among them, the ambient irradiance is measured by the solar irradiance sensor equipped on the cleaning robot, the ambient temperature is measured by the ambient temperature sensor on the cleaning robot, and the wind speed is measured by the wind speed sensor on the cleaning robot.

[0114] The five variables mentioned above are input into a pre-trained XGBoost regression model for calculation, and the output is the dust surface density. This regression model is pre-trained in the laboratory by manually spreading dust with different surface densities (0.5 / 1 / 3 / 5 / 10 / 15 / 20g / m²) and recording the corresponding infrared temperature data to train the mapping relationship between "temperature of dust-covered area → amount of dust". After applying it to the cleaning robot, the current amount of dust can be calculated by directly inputting real-time data.

[0115] Then, the percentage of power loss is calculated based on the ash surface density mapping algorithm, the power generation loss is calculated based on the percentage of power loss, and finally the cost loss is calculated based on the power generation loss and the unit electricity price.

[0116] Specifically, the mapping function is fitted from laboratory calibration data and pre-trained. , The percentage of power loss. Given the ash surface density, the percentage of power loss can be directly obtained through this mapping function. The power loss percentage can then be used to calculate the power generation loss per unit time, and finally, the cost loss per unit time.

[0117] During the training of this mapping function, the process involves spreading dust of different surface densities → power loss percentage → measuring the decrease in actual power generation → establishing a mapping curve, and finally fitting the mapping curve.

[0118] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A photovoltaic panel cleanliness detection system, characterized in that, It includes a cleaning robot, an infrared thermal imaging module, a dual-spectrum thermal imager, a control module, and a software control platform. The infrared thermal imaging module is installed at one end of the cleaning robot's direction of travel and tilted downwards at an angle of 10° to 30° with the horizontal plane; The dual-spectrum thermal imager is installed at the other end of the cleaning robot's travel direction and tilted downwards at an angle of 10° to 30° with the horizontal plane; The control module is connected to the infrared thermal imaging module, the dual-spectrum thermal imager, the control bus of the cleaning robot, and the software control platform. It receives data sent by the infrared thermal imaging module and the dual-spectrum thermal imager, processes it to generate alarm and cleaning strategies, and controls the cleaning robot to clean the photovoltaic panels according to the cleaning strategies. At the same time, the control module sends the alarm strategies to the software control platform, which then sends them to the relevant maintenance personnel.

2. A method for detecting the cleanliness of photovoltaic panels, implemented using the photovoltaic panel cleanliness detection system described in claim 1, characterized in that, include: S1: Start the cleaning robot and move it along the photovoltaic panel array to perform cleaning operations on the photovoltaic panel surface. At the same time, use the infrared thermal imaging module to obtain the infrared temperature matrix of the current detection area. The infrared thermal image and visible light image of the current detection area are obtained using a dual-spectrum thermal imager; S2: Process the infrared temperature matrix to obtain apparent fault data of the photovoltaic panel area in the current detection area. The apparent fault data includes the area ratio of the abnormal area, the average temperature difference, and the location. The average temperature difference is obtained by subtracting the average temperature of the abnormal area from the average temperature of the photovoltaic panel area. The infrared thermal image and the visible light image are processed to obtain the state feature data of the photovoltaic panel area in the current detection area. The state feature data includes the dust accumulation level, the anomaly type, and the corrected inter-panel temperature difference. The inter-panel temperature difference is obtained by subtracting the temperature value of a single pixel in the photovoltaic panel area from the average temperature value of the photovoltaic panel area. S3: Determine the anomaly type and location based on the area ratio, average temperature difference, and location of the abnormal area, and determine the alarm strategy at the abnormal location based on the anomaly type; The cleaning strategy is determined based on the dust accumulation level, the anomaly type, and the corrected inter-plate temperature difference.

3. The photovoltaic panel cleanliness detection method according to claim 2, characterized in that, In step S2, the infrared temperature matrix is ​​processed to obtain apparent fault data of the photovoltaic panel area in the current detection region, including: The infrared temperature matrix is ​​preprocessed to obtain an effective infrared temperature matrix; The effective infrared temperature matrix is ​​subjected to threshold segmentation to generate a binary mask matrix, and the binary mask matrix is ​​superimposed with the effective infrared temperature matrix to obtain the infrared temperature matrix of the photovoltaic panel area in the current detection area. Extract data from the infrared temperature matrix of the photovoltaic panel area to form an effective temperature list; An adaptive threshold algorithm is used to calculate the data in the effective temperature list to obtain the optimal temperature threshold. The infrared temperature matrix of the photovoltaic panel area is then binarized based on the optimal temperature threshold to obtain the binary matrix of the abnormal area. The data in the effective temperature list are calculated to obtain the average temperature and total number of effective pixels in the photovoltaic panel area. Morphological denoising and connected component analysis are performed on the binary matrix of the abnormal area to finally calculate the area ratio, average temperature and center coordinates of each abnormal area. The average temperature difference between the photovoltaic panel area and the average temperature of each abnormal area is obtained, and the location of each abnormal area is determined based on the center coordinates of each abnormal area.

4. The photovoltaic panel cleanliness detection method according to claim 3, characterized in that, The infrared temperature matrix is ​​preprocessed to obtain an effective infrared temperature matrix, including: Gaussian noise in the infrared temperature matrix is ​​removed using Gaussian filtering. Then, median filtering is used to detect and replace bad pixels to obtain the effective infrared temperature matrix.

5. The photovoltaic panel cleanliness detection method according to claim 3, characterized in that, The effective infrared temperature matrix is ​​subjected to threshold segmentation to generate a binary mask matrix, including: The effective infrared temperature matrix is ​​normalized from 16 bits to an 8-bit map; The effective infrared temperature matrix of the 8-bit image is thresholded using the big law method to generate the binary mask matrix.

6. The method for detecting the cleanliness of photovoltaic panels according to claim 2, characterized in that, In step S3, the anomaly type and location are determined based on the area ratio, average temperature difference, and location of the abnormal region, and an alarm strategy at the abnormal location is determined based on the anomaly type; including: The anomaly type and location are determined based on the area proportion, average temperature difference, and location of the abnormal region. The control module generates an alarm strategy for the corresponding abnormal location based on the abnormality type, maps the infrared temperature matrix of the photovoltaic panel area into an 8-bit color image, and marks the abnormal location and abnormality type on the 8-bit color image; finally, the alarm strategy and the 8-bit color image are sent to the software control platform.

7. The photovoltaic panel cleanliness detection method according to claim 6, characterized in that, The alarm strategy includes: When the average temperature difference ΔT ≥ 25℃ and the area of ​​the abnormal region ≥ 5%, the control module marks it as a severe hot spot, records the location and thermal imaging screenshot, and immediately reports it to the software control platform. When the average temperature difference ΔT ≥ 15℃ and the area of ​​the abnormal region accounts for 2% to 5%, the control module marks it as a normal hot spot, records the location and thermal imaging screenshot, and controls the cleaning robot to complete the current cleaning task before reporting to the software control platform. When the abnormal area is a localized low temperature and the edge of the area is clear, the control module marks it as an obstruction, records the location, and controls the cleaning robot to complete the cleaning task before reporting to the software control platform. When the abnormal area is an irregular high-temperature zone with a large temperature difference, the control module marks it as damaged, records the location and thermal imaging screenshot, and controls the cleaning robot to complete the current cleaning task before reporting it to the software control platform first.

8. The method for detecting the cleanliness of photovoltaic panels according to claim 2, characterized in that, The infrared thermal image and visible light image are processed to obtain state characteristic data of the photovoltaic panel area in the current detection area. The state characteristic data includes the dust accumulation level and the corrected inter-panel temperature difference, including: The temperature matrix corresponding to the infrared thermal image is preprocessed to obtain the enhanced infrared thermal image. A semantic segmentation model is used, and the enhanced infrared thermal image is segmented into photovoltaic panel regions using the visible light image to extract the infrared thermal image of the photovoltaic panel region. Based on the infrared temperature matrix corresponding to the infrared thermal image of the photovoltaic panel area, multi-dimensional dust accumulation features are extracted and calculated. The multi-dimensional dust accumulation features include the temperature of a single pixel in the photovoltaic panel area, average temperature, temperature variance, temperature standard deviation, inter-panel temperature difference, temperature gradient features, equivalent infrared emissivity features, and texture features. The infrared thermal image and multi-dimensional dust accumulation features of the photovoltaic panel area are input into a lightweight classification model for processing to obtain the dust accumulation level of the photovoltaic panel area. The multi-dimensional dust accumulation features are processed to obtain the anomaly type of the photovoltaic panel area; The environment in which the photovoltaic panels are located is compensated and corrected to eliminate the interference of weather and time of day on the temperature field, and the corrected temperature difference between the panels is calculated.

9. The method for detecting the cleanliness of photovoltaic panels according to claim 8, characterized in that, The anomaly types are as follows: when the average temperature of the abnormal area of ​​the photovoltaic panel minus the average temperature of the photovoltaic panel area is ≥15℃, and the abnormal area exceeding this temperature is ≥2%, the control module marks the location as having a hot spot and reports it to the software control platform; when there are blocky high temperatures in the photovoltaic panel area and the shape is irregular, the control module marks the location as having bird droppings / stains and reports it to the software control platform; when there are large areas of low temperature in the photovoltaic panel area, the control module marks the location as having shading; when there are areas of low temperature with clear boundaries in the photovoltaic panel area, the control module marks the location as having shadows.

10. The method for detecting the cleanliness of photovoltaic panels according to claim 9, characterized in that, The cleaning strategy is determined based on the dust accumulation level, the corrected inter-plate temperature difference, and the anomaly type, including: When the dust accumulation level of the photovoltaic panel is 0, no cleaning is required; when the dust accumulation level is 1, regular cleaning is performed; when the dust accumulation level is 2, the cleaning robot records the location and reduces its travel speed by 50% when cleaning that location; when the dust accumulation level is 3, the temperature difference between panels is ≥3℃, or the abnormality type is bird droppings / stains, the number of brushing cycles is increased by 2, and a fixed-point re-wash is performed and the data is verified until the dust accumulation level of the photovoltaic panel area drops below 1, at which point the cleaning process for the next area is started.

Images