Photovoltaic power station equipment operation state automatic inspection method and system

Abstract

The application provides a photovoltaic power station equipment operation state automatic inspection method and system, relates to the technical field of data processing, and comprises the following steps: performing spatial attribute segmentation on a photovoltaic station area image to obtain a spatial attribute partition map; determining the attribution partition type of each high-temperature candidate region according to the spatial position overlap ratio between each high-temperature candidate region and the spatial attribute partition map in the photovoltaic station area image, performing background interference elimination to obtain an abnormal candidate region; comparing the area proportion of the abnormal candidate region with the attribution partition type with preset abnormality judgment rules to generate abnormality alarm information, generating a key inspection path of a control ball and a differential treatment instruction according to the regional alarm density and the attribution partition type of the abnormality alarm information, and outputting an inspection scheduling message. Through the spatial attribute segmentation on the photovoltaic station area image, the high-temperature region attribution determination, the abnormality alarm generation and the inspection path planning, the application realizes automatic identification and accurate scheduling treatment of the abnormal state of the photovoltaic power station equipment.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method and system for automatic inspection of the operating status of photovoltaic power plant equipment. Background Technology

[0002] During long-term operation, photovoltaic power plants are susceptible to abnormal temperature rises in certain areas of the modules due to factors such as dust accumulation, localized damage to the solar cells, or loose solder joints. This can lead to decreased power generation efficiency or safety hazards. To achieve automatic detection of these anomalies, photovoltaic power plants commonly deploy surveillance cameras to acquire thermal imaging images of the photovoltaic array area. Image processing algorithms are then used to automatically identify high-temperature areas in the acquired images and generate inspection results.

[0003] Existing automated inspection methods typically mark areas in thermal images that exceed temperature thresholds as anomaly candidate areas, then uniformly output alarm information and plan inspection paths according to fixed rules. However, since photovoltaic array thermal images contain multiple sources of heat radiation, including the module panels, module gaps, frame structures, and the ground, the temperature response characteristics of different areas vary. The aforementioned methods treat all high-temperature areas uniformly, failing to differentiate between the spatial location characteristics of each type of area. This results in background heat radiation interference areas being treated the same as genuine equipment anomaly areas.

[0004] Therefore, how to effectively distinguish between background thermal radiation interference and actual equipment anomalies in the thermal images of photovoltaic arrays during automatic inspection, thereby improving the accuracy of automatic inspection results, has become an urgent problem to be solved. Summary of the Invention

[0005] This application provides an automatic inspection method and system for the operating status of photovoltaic power station equipment, in order to solve the problems of existing automatic inspection methods that uniformly process all high-temperature areas in thermal images, fail to effectively distinguish between background thermal radiation interference and actual equipment abnormalities, and thus result in insufficient accuracy of inspection results.

[0006] The first aspect of this application provides an automatic inspection method for the operating status of photovoltaic power station equipment, including: performing spatial attribute segmentation on a photovoltaic power station area image based on the structural arrangement characteristics of the photovoltaic array to obtain a spatial attribute partition map; Based on the spatial overlap ratio between each high-temperature candidate region in the photovoltaic station area image and the spatial attribute zoning map, the zoning type of each high-temperature candidate region is determined, and background interference removal is performed based on the zoning type to obtain abnormal candidate regions. Based on the area ratio of the abnormal candidate regions and the partition type to which they belong, an abnormal alarm message is generated by comparing the information with the preset abnormal judgment rules. Based on the regional alarm density of the abnormal alarm information and the type of the assigned partition, a key inspection path and differentiated handling instructions are generated for the deployment ball, and an inspection scheduling message is output.

[0007] Optionally, in one possible implementation of the first aspect, the spatial attribute partitioning map includes a splicing gap cold zone, an array body zone, an edge-prone zone, and a background interference zone; The method of spatial attribute segmentation of the photovoltaic station area image based on the structural arrangement characteristics of the photovoltaic array to obtain a spatial attribute partition map includes: The image of the photovoltaic station area is projected in grayscale along the component arrangement direction, and continuous grayscale low-level areas are extracted and identified as splicing gap cold zones. Based on the average gray value of the cold zone in the splicing gap, the array body area is defined by extending outwards. Using a fixed ratio of the average spacing between adjacent splicing gap cold zones as the extension width, the extension width is extended inward along the lower boundary of the array body area and marked as the edge-prone area. The area outside the splicing gap cold zone, the array body area, and the edge-prone area is marked as the background interference area.

[0008] Optionally, in one possible implementation of the first aspect, the step of extending the array body region outward based on the average grayscale value of the cold zone in the splicing gap includes: Extract the boundary column coordinates of each adjacent splicing gap cold zone in the photovoltaic station area image facing the side of the panel; The column coordinate segment between the boundary column coordinates of two adjacent splicing gap cold zones facing one side of the plate is taken as the candidate body range; Extract the average grayscale value of each column in the column direction within the candidate ontology range; The longest continuous column segment formed by the pixel column whose cumulative grayscale average value in the column direction is higher than the grayscale average value of the cold zone in the splicing gap is defined as the array body area.

[0009] Optionally, in one possible implementation of the first aspect, the step of performing grayscale projection on the photovoltaic station area image along the component arrangement direction and extracting continuous grayscale low-level regions to determine the splicing gap cold zone includes: The image of the photovoltaic station area is processed by accumulating pixel gray levels along the column direction to generate a horizontal gray level distribution curve; Extract all consecutive pixel column segments in the horizontal grayscale distribution curve whose cumulative grayscale value is lower than the mean of the horizontal grayscale distribution curve, calculate the spacing between adjacent pixel column segments, and obtain the spacing sequence; Based on the spacing sequence, the upper and lower limits of the spacing are determined. Pixel column segments in the spacing sequence whose spacing values ​​fall outside the range of the lower to upper limits of the spacing are removed, and the remaining pixel column segments are determined as the cold zone of the splicing gap.

[0010] Optionally, in one possible implementation of the first aspect, determining the partition type of each high-temperature candidate region based on the spatial overlap ratio between each high-temperature candidate region in the photovoltaic station area image and the spatial attribute partition map, and performing background interference removal based on the partition type to obtain abnormal candidate regions, includes: The partition type to which the high-temperature candidate region belongs is determined based on the proportion of the overlapping area between each high-temperature candidate region and each partition in the spatial attribute partitioning map. The edge grayscale change sequence is determined based on the edge pixel grayscale difference sequence within the assigned partition type; High-temperature candidate regions whose partition type is background interference region and whose edge gray-level change sequence does not contain a continuous unidirectional gray-level increasing subsequence are removed from the candidate set, while high-temperature candidate regions whose partition type is array body region or edge-prone region are retained to obtain the abnormal candidate regions.

[0011] Optionally, in one possible implementation of the first aspect, determining the partition type of the high-temperature candidate region based on the proportion of the overlapping area between each high-temperature candidate region and each partition in the spatial attribute partitioning map includes: Based on the actual pixel area ratio of each partition in the spatial attribute partitioning map, the threshold range of the overlapping area ratio of the background interference area, the edge-prone area and the array body area is determined. When the overlapping area ratio falls within the threshold range corresponding to the background interference area, the partition type of the high temperature candidate region is determined to be the background interference area. When the overlapping area ratio falls within the threshold range corresponding to the edge-prone zone, the partition type of the high-temperature candidate region is determined to be the edge-prone zone. When the overlapping area ratio falls within the threshold range corresponding to the array body region, the partition type of the high-temperature candidate region is determined to be the array body region.

[0012] Optionally, in one possible implementation of the first aspect, after determining that the overlapping area ratio falls within the threshold range corresponding to the array body region, and after determining that the partition type of the high-temperature candidate region is the array body region, the method further includes: Extract the geometric center coordinates of each high-temperature candidate region whose partition type is the array body region; Based on the offset distance between the geometric center coordinates and the midpoint of the long side of the array body area, and the pixel width of the long side of the array body area, the upper limit of the offset distance is determined. The upper limit of the number of outline pixels is determined based on the average pixel area of ​​a single component in the array body region. Based on the comparison results of the offset distance and the upper limit of the offset distance determination, the number of pixels enclosed by the closed contour line of the high temperature candidate region and the upper limit of the number of contour pixels, the high temperature candidate region is marked as a junction box interference region and removed from the candidate set, or the high temperature candidate region is retained in the abnormal candidate region.

[0013] Optionally, in one possible implementation of the first aspect, the step of generating an anomaly alarm information based on the area ratio of the anomaly candidate region and the partition type, in accordance with a preset anomaly determination rule, includes: The area ratio is determined based on the number of pixels enclosed by the closed contour lines of each candidate region within the abnormal candidate region and the total number of pixels in the array body region to which it belongs. Based on the gray-level difference sequence between the boundary pixels of each candidate region and the adjacent non-candidate region pixels within the abnormal candidate region, edge gray-level gradient features are obtained. The area ratio, the edge grayscale gradient features, and the partition type are input into a preset anomaly determination rule for comparison, and anomaly type labels are output. Based on the maximum gray-level difference in the edge gray-level gradient features, the emergency level of the candidate region is divided, and the abnormal alarm information is generated based on the abnormality type label and the emergency level.

[0014] Optionally, in one possible implementation of the first aspect, the preset anomaly determination rule includes: When the partition type is determined to be the array body area, the area ratio is lower than the average pixel area ratio of the single component in the array body area, and the maximum gray-level difference in the edge gray-level gradient feature is higher than the median of the maximum gray-level difference of all abnormal candidate regions, the abnormal type label is output as hot spot abnormal. When the partition type is determined to be the array body area, the area ratio is higher than the average pixel area ratio of the single component in the array body area, and the ratio of the maximum gray level difference to the average gray level difference in the edge gray level gradient features is lower than the median of the corresponding ratios of all abnormal candidate regions, the abnormal type label is output as the whole string of failure abnormalities. When the partition type is determined to be an edge-prone area and the area ratio is lower than the average pixel area ratio of a single component in the array body area, the abnormality type label is output as edge hot spot abnormality. When the partition type is determined to be a splicing gap cold area and the maximum gray level difference in the edge gray level gradient features is higher than the difference between the gray level average of the splicing gap cold area and the gray level average of the array body area, the output anomaly type label is gap anomaly.

[0015] Optionally, in one possible implementation of the first aspect, the step of dividing the emergency level of candidate regions based on the maximum gray-level difference in the edge gray-level gradient features, and generating the anomaly alarm information based on the anomaly type label and the emergency level, includes: When the maximum grayscale difference is determined to be greater than the first preset comparison value, the emergency level is classified as a high emergency level. When the maximum grayscale difference is determined to be less than a first preset comparison value and greater than a second preset comparison value, the emergency level is classified as a medium emergency level. When the maximum grayscale difference is determined to be less than the second preset comparison value, the emergency level is classified as a low emergency level. The anomaly type label, the emergency level, and the center coordinates of the candidate area are concatenated to generate the anomaly alarm information.

[0016] Optionally, in one possible implementation of the first aspect, the step of generating a key inspection path and differentiated handling instructions for the surveillance system based on the regional alarm density of the abnormal alarm information and the type of the assigned partition, and outputting an inspection scheduling message, includes: The number of abnormal alarm messages within each assigned partition type is counted, and the regional alarm density of each assigned partition type is determined based on the area of ​​each assigned partition type and the corresponding number of alarms. Alarms are sorted in descending order of regional alarm density according to each partition type. Alarms within the same partition type are aggregated into inspection clusters. Alarms within each inspection cluster are connected to the center coordinates in descending order of emergency level. Inspection clusters are connected sequentially in descending order of regional alarm density to generate key inspection paths for the control ball. Based on the environmental operation characteristics corresponding to the partition type to which each abnormal alarm information belongs, retrieve the corresponding safety equipment list and handling time limit requirements, and generate differentiated handling instructions; The key inspection path of the control ball and the differentiated handling instruction are encapsulated into a message, and the inspection scheduling message is output.

[0017] Optionally, in one possible implementation of the first aspect, the step of retrieving the corresponding safety equipment list and handling time limit requirements based on the environmental operation characteristics corresponding to the partition type to which each abnormal alarm information belongs, and generating differentiated handling instructions, includes: When the assigned zone type is determined to be an edge-prone area or a cold zone in the splicing gap, the environmental operation characteristics are identified as a ground operation environment, and the corresponding ground operation safety equipment list and short-term response time limit are matched according to the emergency level. When the assigned partition type is determined to be the array body area, the environmental operation characteristics are identified as a high-altitude operation environment, and the corresponding high-altitude operation safety equipment list and standard handling time limit requirements are matched according to the emergency level. When the assigned partition type is determined to be a background interference zone, the environmental operation characteristics are identified as a remote observation environment, and the corresponding remote verification operation instructions and long-term handling time limits are matched according to the emergency level. The matching list of safety equipment, time limit requirements for handling, or remote verification operation instructions are bound to the corresponding alarm coordinates to generate the differentiated handling instructions.

[0018] A second aspect of this application provides an automatic inspection system for the operating status of photovoltaic power station equipment, including: a spatial segmentation module, used to perform spatial attribute segmentation on the image of the photovoltaic station area based on the structural arrangement characteristics of the photovoltaic array, to obtain a spatial attribute partition map; The overlap determination module is used to determine the partition type of each high temperature candidate region based on the spatial overlap ratio between each high temperature candidate region in the photovoltaic station area image and the spatial attribute partition map, and to perform background interference removal based on the partition type to obtain abnormal candidate regions. The rule matching module is used to generate abnormal alarm information based on the area ratio of the abnormal candidate region and the partition type to which it belongs, by comparing with preset abnormal judgment rules. The strategy generation module is used to generate key inspection paths and differentiated handling instructions for the deployment ball based on the regional alarm density of the abnormal alarm information and the type of the assigned partition, and output inspection scheduling messages.

[0019] The automatic inspection method for the operating status of photovoltaic power plant equipment provided in this application has the following beneficial effects: 1. This application addresses the highly regular arrangement of photovoltaic power station arrays by performing spatial attribute segmentation on thermal images collected by the control ball, combining high-temperature candidate areas with their spatial location features for anomaly determination, thereby enabling differentiated processing of high-temperature areas from different sources in the periodic arrangement structure, and accurately distinguishing between actual equipment anomalies and background thermal radiation interference. 2. Due to the periodic spatial structure of the photovoltaic array, the location distribution of various areas is predictable. Therefore, this application utilizes this predictability to eliminate interference in the anomaly identification process, thereby improving the identification accuracy. This ensures that each stage, including alarm classification, inspection path planning, and handling instruction generation, is based on accurate anomaly identification results, avoiding the step-by-step propagation of errors throughout the entire process. 3. Different spatial locations of abnormal areas naturally correspond to different on-site operating environments. Therefore, this application generates differentiated handling instructions based on the spatial location of the abnormal area, so that inspection personnel can know the safety equipment and handling time limit requirements that match the operating environment before arriving at the site. This realizes the fully automatic closed loop from anomaly identification, alarm classification to on-site handling scheduling. In the actual operation and maintenance scenario of large-scale photovoltaic power plants, it reduces the proportion of invalid inspection operations caused by interference false alarms and improves the response time and handling of power plant equipment anomalies. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the automatic inspection method for the operating status of photovoltaic power station equipment provided in this application embodiment; Figure 2 This is an application environment diagram of the automatic inspection method for the operating status of photovoltaic power station equipment provided in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of the automatic inspection system for the operating status of photovoltaic power station equipment provided in the embodiments of this application; Figure 4 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0022] The technical solutions of this application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0023] See Figure 1 This is a flowchart illustrating the automatic inspection method for the operating status of photovoltaic power station equipment provided in this application embodiment. Figure 1The execution entity of the method shown can be a software and / or hardware device. The execution entity of this application can include, but is not limited to, at least one of the following: user equipment, network equipment, etc. User equipment can include, but is not limited to, computers, smartphones, personal digital assistants (PDAs), and the aforementioned electronic devices. Network equipment can include, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of computers or network servers. Cloud computing is a type of distributed computing, consisting of a super virtual computer composed of a group of loosely coupled computers. This embodiment does not impose any limitations on this.

[0024] The automatic inspection method for the operating status of photovoltaic power station equipment provided in this application embodiment can be applied to, for example... Figure 2 In the application environment shown, surveillance cameras are deployed within the photovoltaic power station area. These cameras acquire thermal images of the photovoltaic array area, and the acquired images are transmitted to the inspection and processing server via wired or wireless network. The inspection and processing server performs spatial attribute segmentation, high-temperature candidate area attribution determination, background interference removal, anomaly type determination, and alarm density statistics on the photovoltaic station area images, generating anomaly alarm information and inspection dispatch messages. The inspection dispatch messages are sent to the surveillance camera control terminal via the network. The surveillance camera control terminal drives the surveillance camera to perform fixed-point inspections according to the key inspection path in the message, and simultaneously pushes differentiated handling instructions to the mobile terminals of maintenance personnel. Maintenance personnel prepare equipment according to the differentiated handling instructions and proceed to the corresponding alarm coordinates to perform on-site handling. The data storage system is used to store historical inspection cycle grayscale mean difference records, preset judgment rules for various anomalies, and safety equipment lists, etc. The data storage system can be integrated locally on the inspection and processing server or deployed in the cloud or other network storage nodes.

[0025] The automatic inspection method for the operating status of photovoltaic power station equipment provided in this application includes steps S1 to S4, as detailed below: Step S1: Based on the structural arrangement characteristics of the photovoltaic array, perform spatial attribute segmentation on the photovoltaic station area image to obtain a spatial attribute partition map.

[0026] It should be noted that the high-temperature areas from different sources in the photovoltaic (PV) station area images are not randomly distributed in space, but rather highly correspond to the periodic structure of the PV array. If spatial attribute segmentation is not performed on the images beforehand, and high-temperature areas are directly extracted uniformly from the thermal imaging images, then real anomalies on the module panels, structural thermal radiation at the splicing gaps, and background thermal radiation from the ground will be treated the same, making it impossible to distinguish the differences in the sources when identifying anomalies. This step utilizes the regular arrangement of the PV array, using spatial attribute segmentation to provide spatial location basis for the subsequent differentiated processing of various high-temperature areas, thereby freeing the anomaly identification results from dependence on human experience-based interpretation.

[0027] The photovoltaic (PV) power station area image is a thermal imaging image acquired by a satellite imager over the PV power station array area. It includes various regions such as the PV module panels, the gaps between modules, the array edge structure, and the ground background. Each region exhibits a different grayscale response in the thermal imaging image. The structural arrangement characteristics of the PV array refer to the periodic spatial structure formed by the regular arrangement of PV modules at fixed intervals. Specifically, this is manifested as the alternating appearance of module panels and gaps between modules in the horizontal direction of the image.

[0028] The spatial attribute partitioning map includes four types of partitions: cold zone at the splicing gap, array body area, edge-prone area, and background interference area.

[0029] It is easy to understand that the periodic arrangement of the photovoltaic array makes the position of each partition in the image predictable, which is a prerequisite for accurate spatial attribute segmentation. The generation of the spatial attribute partition map distinguishes the originally mixed regions in the image at the spatial level, providing a basis for the subsequent determination of the attribution of high-temperature candidate regions and avoiding false alarms caused by confusion of region origins.

[0030] In some embodiments, step S1 includes steps S11 to S14: Step S11: Perform grayscale projection on the photovoltaic station area image along the component arrangement direction, extract continuous grayscale low-level areas, and determine them as splicing gap cold zones.

[0031] The component arrangement direction refers to the direction in which photovoltaic modules are arranged horizontally side-by-side in the array, corresponding to the column direction in the photovoltaic station area image. Gray-scale projection refers to summing the gray-scale values ​​of each pixel column in the photovoltaic station area image along the column direction, compressing the two-dimensional image into a one-dimensional curve reflecting the horizontal gray-scale distribution pattern. Continuous gray-scale trough regions are pixel column segments whose cumulative gray-scale values ​​are consistently lower than the overall curve level, extracted from the gray-scale projection curve. Since there is no photovoltaic panel coverage at the splicing gaps between components, the thermal radiation intensity is lower than that of the component panel area, which appears as vertical stripes with low gray-scale values ​​in the thermal imaging image. After gray-scale projection, periodic troughs are formed, and the cold area at the splicing gaps is determined from the continuous gray-scale trough regions based on the above characteristics.

[0032] It is understandable that the gaps between modules appear as grayscale valleys in thermal imaging images, which is determined by the physical structure of the photovoltaic array. The gaps lack the thermal radiation contribution from the module panels, and the valley locations correspond one-to-one with the gap locations after grayscale projection. Compressing the lateral structural information of a two-dimensional image into a one-dimensional distribution curve through grayscale projection, and then extracting the valley segments from the curve, is an effective way to automatically locate the gaps using the periodic arrangement characteristics of the photovoltaic array.

[0033] It should be noted that step S11 extracts the cold zone of the splicing gap using grayscale projection, allowing the position of the gap between modules to be automatically determined from the thermal imaging image without relying on pre-entered module layout drawings. In the actual operation and maintenance of photovoltaic power plants, the module arrangement varies due to site conditions. Relying on fixed drawings would result in poor adaptability. The grayscale projection method directly locates the gap from the grayscale distribution of the image itself, ensuring that the delineation result of the cold zone of the splicing gap is consistent with the current actual image.

[0034] In some embodiments, step S11 includes steps S111 to S113: Step S111: Accumulate pixel grayscale values ​​in the column direction of the photovoltaic station area image to generate a horizontal grayscale distribution curve.

[0035] The horizontal grayscale distribution curve is a one-dimensional curve with the pixel column number as the horizontal axis and the cumulative grayscale value of the corresponding column as the vertical axis. The peak section of the curve corresponds to the component board area, and the trough section corresponds to the splicing gap area between components.

[0036] It is easy to understand that the gray values ​​of the same pixel column in a thermal imaging image reflect the total thermal radiation intensity of all rows in that column. After accumulating along the column direction, the gray value difference between the component plate column and the gap column is amplified on the horizontal gray value distribution curve, making the identification of the trough section more stable and less susceptible to interference from single-row noise.

[0037] It should be noted that step S111 compresses the horizontal grayscale distribution pattern of the two-dimensional image into a one-dimensional curve. Steps S112 and S113 are both based on the horizontal grayscale distribution curve. The amount of computation is greatly reduced compared to directly operating on the two-dimensional image pixel by pixel, while retaining all the horizontal structural information required to identify the cold area of ​​the splicing gap.

[0038] Step S112: Extract all consecutive pixel column segments in the horizontal gray-level distribution curve whose cumulative gray-level value is lower than the mean of the horizontal gray-level distribution curve, calculate the spacing between adjacent pixel column segments, and obtain the spacing sequence.

[0039] Among them, a continuous pixel column segment whose cumulative grayscale value is lower than the mean of the horizontal grayscale distribution curve refers to a continuous segment on the horizontal grayscale distribution curve whose cumulative grayscale value is consistently lower than the mean of the horizontal grayscale distribution curve. Each continuous segment corresponds to a candidate gap position. The spacing between adjacent pixel column segments refers to the difference between the center column coordinates of two adjacent candidate gap segments, or the difference between the boundary column coordinates of two adjacent candidate gap segments. Arranging the spacing values ​​of all adjacent candidate gap segments in order yields the spacing sequence.

[0040] It should be noted that candidate segments extracted solely based on the condition that the cumulative grayscale value is lower than the mean may include segments with low grayscale values ​​caused by local noise, ground shadows, or other non-gap factors. If all candidate segments are directly identified as cold areas of splicing gaps, noisy segments will be mistakenly labeled as gaps, interfering with the accuracy of the subsequent spatial attribute zoning map. Step S113, which filters candidate segments based on the spacing sequence, is precisely to eliminate the aforementioned non-gap interference segments.

[0041] Step S113: Determine the upper and lower limits of the spacing based on the spacing sequence, remove pixel column segments in the spacing sequence whose spacing values ​​fall outside the range of the lower and upper limits of the spacing, and determine the remaining pixel column segments as cold areas for splicing gaps.

[0042] The upper and lower limits of the spacing are screening boundaries determined based on the statistical distribution characteristics of the spacing sequence. Specifically, the interquartile range (ICM) of the spacing sequence can be calculated. The upper limit is the sum of the median of the spacing sequence and twice the ICM, and the lower limit is the difference between the median and twice the ICM. The ICM reflects the dispersion of the middle 50% of the data in the spacing sequence. Extending the ICM by twice the ICM from the median value can cover most of the spacing values ​​in the spacing sequence that conform to the actual arrangement of the photovoltaic array, while excluding abnormal spacing values ​​that deviate from the arrangement pattern. Pixel column segments with spacing values ​​falling outside the range of the lower and upper limits indicate that the corresponding candidate segment does not conform to the periodic spacing pattern of the photovoltaic array and belongs to noise or non-gap interference segments, which are then eliminated. The remaining pixel column segments are determined as cold zones for splicing gaps.

[0043] It should be noted that the determination of the upper and lower limits of the spacing does not require pre-input of component layout parameters; it is calculated entirely based on the statistical distribution of the spacing sequence itself. This method allows the screening of cold zones in the splicing gaps to adapt to photovoltaic power plants with different component spacing specifications. Those skilled in the art can also select different statistical interval widths based on the consistency of component spacing in the actual power plant. The higher the spacing consistency, the narrower the screening interval can be to improve accuracy.

[0044] It should be noted that step S113 retains candidate segments that conform to the periodic arrangement pattern as splicing gap cold zones through spacing statistics screening, and eliminates noise and non-gap interference segments, so that the positioning result of the splicing gap cold zone is consistent with the actual gap position of the photovoltaic array.

[0045] Step S12: Based on the average gray value of the cold area in the splicing gap, extend the array body area outward to define the main body area.

[0046] The outward expansion refers to extending the panel area between two adjacent cold zones along the splicing gaps, defining the pixel column range of the module panel in the image. The array body area refers to the continuous segment formed by pixel columns whose cumulative gray-level average in the column direction is higher than the average gray-level average of the cold zones along the splicing gaps, corresponding to the main coverage area of ​​the photovoltaic module panel.

[0047] It is understandable that the grayscale value of the component board area is generally higher than that of the gap area. Using the average grayscale value of the cold area of ​​the splicing gap as a reference benchmark to distinguish between the board and non-board pixel columns, the array body area can be delineated by using the grayscale statistical information of the image itself without introducing external parameters.

[0048] It should be noted that, based on the cold zone of the splicing gap determined in step S11, step S12 further precisely delineates the plate area between adjacent gaps, so that the boundary of the array body area in the spatial attribute partitioning diagram matches the actual component plate boundary, providing an accurate spatial range for the anomaly determination of the high temperature candidate area belonging to the array body area.

[0049] In some embodiments, step S12 includes steps S121 to S124: Step S121: Extract the boundary column coordinates of each adjacent splicing gap cold zone in the photovoltaic station area image facing the side of the panel.

[0050] Here, the boundary column coordinates facing the panel side refer to the column number corresponding to the boundary of each splicing gap cold zone on the side closest to the component panel in the column direction. For two adjacent splicing gap cold zones, the boundary column coordinates facing the panel side are located inside the two gaps, that is, on the side where the two gaps face each other. From the splicing gap cold zones determined in step S11, the column range of each splicing gap cold zone is extracted one by one, and the column number facing the adjacent panel area in the column range of each splicing gap cold zone is taken to obtain the boundary column coordinates facing the panel side of each adjacent splicing gap cold zone.

[0051] Step S122: The column coordinate segment between the boundary column coordinates of the cold zones of two adjacent splicing gaps facing one side of the plate is taken as the candidate body range.

[0052] The candidate body range is a column coordinate segment enclosed by the boundary column coordinates of the cold areas of two adjacent splicing gaps facing the board body, covering all pixel columns between the two adjacent gaps. The candidate body range is the initial search space of the array body area. Further filtering based on grayscale statistics within the candidate body range yields the pixel column range that actually belongs to the component board body.

[0053] Step S123: Extract the cumulative mean of grayscale values ​​in the column direction of each column within the candidate ontology range.

[0054] Step S124: The longest continuous column segment formed by the pixel column whose cumulative gray value in the column direction is higher than the gray value of the cold area in the splicing gap is defined as the array body area.

[0055] Within the candidate body area, pixel columns with a column-direction cumulative grayscale mean higher than the splicing gap cold zone mean grayscale value correspond to the component panel coverage area. Pixel columns with a column-direction cumulative grayscale mean lower than or equal to the splicing gap cold zone mean grayscale value correspond to non-panel areas outside the panel edge, such as the transition area between the component frame and the ground. The portion forming the longest continuous column segment among the pixel columns within the candidate body area whose column-direction cumulative grayscale mean is higher than the splicing gap cold zone mean grayscale value is designated as the array body area. Choosing the longest continuous column segment, rather than all pixel columns meeting the condition, is to eliminate interference from isolated high-grayscale pixel columns formed by local noise within the candidate body area on the boundary delineation of the array body area.

[0056] It should be noted that if the entire candidate body area is directly identified as the array body area, the component border area and the transition area of ​​the board edge will also be included in the array body area. This results in an overly large range for judging the anomalies of high-temperature candidate areas within the array body area, and the thermal radiation of non-board areas will be mistakenly included in the board anomaly analysis. Step S124 uses grayscale mean filtering and optimization of the longest continuous column segment to precisely limit the array body area to the actual coverage column range of the component board, improving the degree of conformity between the boundary of the array body area and the actual board boundary.

[0057] Step S13: Using a fixed ratio of the average spacing of the cold zones between adjacent splicing gaps as the extension width, extend the extension width inward along the lower boundary of the array body area, and mark it as the edge-prone area.

[0058] The average spacing of the cold zones between adjacent splicing gaps is a statistical measure obtained by averaging the spacing sequence obtained in step S112, reflecting the average level of spacing between adjacent modules in the photovoltaic array. The extension width is the number of pixel columns obtained by multiplying the average spacing of the cold zones between adjacent splicing gaps by a fixed ratio. The specific value of the fixed ratio can be selected by those skilled in the art based on the proportional relationship between the actual width of the hot spot-prone area at the lower edge of the photovoltaic module and the module spacing. Typically, the proportion of the edge area to the total width of the module can be referenced in the module specifications. The edge-prone area is the pixel column segment covered by the extension width extending inwards from the lower boundary of the array body area, corresponding to the lower edge area of ​​the photovoltaic module. Due to factors such as the module installation tilt angle and uneven dust accumulation, the probability of hot spots occurring in this area is higher than in the middle area of ​​the module panel.

[0059] It should be noted that the lower edge area of ​​photovoltaic modules is a high-incidence location for hot spot anomalies in actual operation and maintenance. If the lower edge area is categorized together with the central area of ​​the module panel into the array body area without distinction, the high-temperature candidate areas in the lower edge area will be treated according to the same anomaly judgment rules as the central area, which may lead to missed or false alarms for edge hot spots. Marking the lower edge area separately as an edge-prone area allows subsequent anomaly judgment to use different judgment rules for the edge-prone area than for the array body area, improving the accuracy of edge hot spot anomaly identification.

[0060] It should be noted that step S13 uses the average spacing of the cold zones between adjacent splicing gaps as the calculation benchmark for the extension width, so that the width of the edge-prone area is proportional to the actual component size in the current image. In photovoltaic power plants with different component specifications, the edge-prone area delineation result that matches the actual edge width of the component can be obtained.

[0061] Step S14: Mark the areas outside the splicing gap cold zone, the array body area, and the edge-prone area as background interference areas.

[0062] Background interference zone refers to all remaining areas in the photovoltaic station area image except for the cold zone at the splicing gap, the array body area, and the edge-prone area. This includes non-component areas such as the ground, support structure, and image edges. The heat radiation source of these areas is unrelated to the photovoltaic modules, and their high temperature performance is not considered an abnormality of the photovoltaic equipment.

[0063] It is easy to understand that by uniformly marking the areas outside the three types of partitions mentioned above as background interference areas, the spatial attribute partition map can completely cover all pixel areas in the photovoltaic station area image, with no unassigned pixels, thus ensuring that each subsequent high-temperature candidate area can obtain a clear partition type.

[0064] It should be noted that the background interference area is determined by using a complement method after the cold area in the splicing gap, the array body area, and the edge-prone area are identified, without the need for additional calculations, while ensuring the integrity of the spatial attribute partition map. High-temperature candidate areas belonging to the background interference area will be verified by the edge grayscale change sequence in subsequent steps to determine whether to remove them, rather than retaining them all directly. This design takes into account the need to identify occasional real anomalies in the background area.

[0065] Preferably, step S1 divides the photovoltaic station area image into four categories—cold areas between splicing gaps, array body areas, edge-prone areas, and background interference areas—through spatial attribute segmentation, ensuring that each pixel in the image has a clear spatial classification. The regular gap distribution formed by the periodic arrangement of the photovoltaic array is the fundamental basis for step S1 to complete spatial attribute segmentation using pure image statistics. The entire segmentation process does not require external layout drawings; it directly extracts structural information from the grayscale distribution of the photovoltaic station area image itself. The generation of the spatial attribute partition map provides a structured spatial benchmark for subsequent high-temperature candidate area classification, enabling spatial differentiation of high-temperature areas from different sources before entering the anomaly judgment process. This avoids the problem of treating background thermal radiation interference and actual equipment anomalies equally, ensuring the accuracy of the input information in each stage of alarm classification and inspection scheduling in steps S2 to S4.

[0066] Step S2: Based on the spatial overlap ratio between each high-temperature candidate region in the photovoltaic station area image and the spatial attribute zoning map, determine the zoning type of each high-temperature candidate region, and perform background interference removal based on the zoning type to obtain abnormal candidate regions.

[0067] It should be noted that the thermal imaging images acquired by the surveillance sphere in the photovoltaic station area image, after high-temperature region extraction, may result in high-temperature candidate regions distributed across various areas such as the module panels, module gaps, frame structures, or ground background. If all high-temperature candidate regions are uniformly subjected to anomaly judgment, ground background thermal radiation, gap structure thermal radiation, and genuine anomalies in the module panels will be treated the same, leading to a persistently high false alarm rate. This step performs spatial overlap calculation between the high-temperature candidate regions and the spatial attribute partition map generated in step S1 to determine the partition type to which each high-temperature candidate region belongs. Then, background interference removal is performed based on the partition type, ensuring that all candidate regions entering the subsequent anomaly judgment process have completed spatial source differentiation.

[0068] In some embodiments, step S2 includes steps S21 to S23: Step S21: Determine the partition type to which the high temperature candidate region belongs based on the proportion of the overlapping area between each high temperature candidate region and each partition in the spatial attribute partitioning map.

[0069] High-temperature candidate regions are continuous pixel regions with grayscale values ​​exceeding the extraction reference grayscale value extracted from the photovoltaic power station area image. The extraction reference grayscale value can be selected by those skilled in the art based on the calibration parameters of the actual power station thermal imaging equipment and ambient temperature conditions. The overlap area ratio refers to the proportion of pixels in each high-temperature candidate region that spatially overlaps with a certain partition in the spatial attribute partitioning map, out of the total number of pixels in that high-temperature candidate region. This reflects the degree to which the high-temperature candidate region spatially belongs to each partition. The partition type is determined based on the overlap area ratio falling within the corresponding threshold range of each partition, indicating which type of partition the high-temperature candidate region primarily belongs to spatially.

[0070] The high-temperature candidate region is a continuous pixel area extracted from the photovoltaic station area image whose grayscale value exceeds the extraction reference grayscale value. The extraction reference grayscale value can be selected based on the calibration parameters of the actual power station thermal imaging equipment and the ambient temperature conditions. The calibration parameters of the thermal imaging equipment are a set of parameters recorded during the equipment's factory inspection or on-site calibration. This set includes a complete table of temperature and grayscale correspondences for the equipment under its current operating configuration. Based on this, the grayscale reading corresponding to the normal operating temperature of the modules under the current ambient temperature can be directly obtained, as well as the grayscale reading corresponding to the minimum identifiable temperature rise (usually about 10°C) required for hot spot anomaly detection, superimposed on this. The sum of these two values ​​is used as the extraction reference grayscale value. It should be noted that if the extraction reference grayscale value is lower than the calculated value, normally operating module areas will also be included in the high-temperature candidate region, leading to an artificially inflated number of candidate regions and an increased computational burden on subsequent background interference removal steps. If the extraction reference grayscale value is higher than the calculated value, hot spot areas whose temperature rise does not reach the minimum identifiable temperature rise will be missed, resulting in missed detections. In this embodiment, the extraction reference grayscale value can be preferably set to 190. In practical applications, it can be adjusted within the range of 160 to 220. When the ambient temperature rises, the corresponding grayscale value is adjusted upwards, and when the ambient temperature falls, the corresponding grayscale value is adjusted downwards.

[0071] It is understandable that a high-temperature candidate region may span multiple partitions in the spatial attribute partitioning map. The partition with the largest overlap area reflects the main spatial affiliation of the high-temperature candidate region. This is used as the basis for determining the partition type, so that each high-temperature candidate region can obtain a unique partition type.

[0072] It should be noted that step S21 introduces spatial location information into the attribution determination of high temperature candidate regions, so that subsequent background interference elimination and anomaly determination can adopt differentiated rules based on the attribution partition type, rather than applying a uniform judgment to all high temperature candidate regions, thus solving the problem of high temperature regions from different sources being treated equally from the source.

[0073] In some embodiments, step S21 includes steps S211 to S214: Step S211: Based on the actual pixel area ratio of each partition in the spatial attribute partitioning map, determine the threshold range of the overlapping area ratio corresponding to the background interference area, the edge-prone area and the array body area.

[0074] The actual pixel area ratio refers to the proportion of the number of pixels in each partition of the spatial attribute partitioning map to the total number of pixels in the photovoltaic station area image, reflecting the spatial proportion of each partition in the image. The overlap area ratio threshold range is the boundary for determining the attribution based on the actual pixel area ratio of each partition. Specifically, the array body area has the largest proportion in the image, so the upper limit of the corresponding overlap area ratio threshold range is set to 1, and the lower limit can be selected by those skilled in the art based on the difference between the pixel area ratio of the array body area and the edge-prone area; the edge-prone area has the second largest proportion, and the upper and lower limits of the corresponding threshold range are determined based on the distribution relationship between the pixel area ratio of the edge-prone area and the background interference area; the upper limit of the threshold range corresponding to the background interference area is connected to the lower limit of the threshold range of the edge-prone area, and the lower limit is 0, ensuring that the three threshold ranges completely cover the entire value range from 0 to 1 and do not overlap.

[0075] It should be noted that if a fixed threshold for the overlap area ratio is set for each partition without considering the actual pixel area ratio of each partition, the same fixed threshold may lead to discrepancies between the attribution determination results and the actual spatial attribution in photovoltaic power plants of different sizes and component densities. Determining the threshold range for each partition based on the actual pixel area ratio allows the threshold to adapt to the actual partition area distribution in the power plant image, ensuring the accuracy of the attribution determination results.

[0076] It should be noted that step S211 determines the threshold range by the actual pixel area ratio of each partition, which is the premise for the partition classification judgment in steps S212 to S214. The complete coverage of the three threshold ranges ensures that each high temperature candidate area can fall into and only fall into one partition type, avoiding blanks or overlaps in the classification judgment.

[0077] Step S212: When the overlapping area ratio falls within the threshold range corresponding to the background interference area, determine that the high temperature candidate area belongs to the background interference area.

[0078] The background interference area corresponds to the high-temperature candidate region within the threshold range, indicating that the high-temperature candidate region is mainly distributed in non-component areas such as the ground and support structures. The high-temperature candidate region belonging to the background interference area is not directly eliminated, but enters the edge grayscale change sequence verification in step S22. After verification, it is decided whether to retain it in the abnormal candidate region, so as to take into account the need to identify occasional real anomalies in the background area.

[0079] It should be noted that step S212 marks the high-temperature candidate regions belonging to the background interference area separately, so that the subsequent step S23 can apply additional edge grayscale verification to the candidate regions of the background interference area, instead of directly retaining or removing them all. This reduces false alarms from background interference while retaining the ability to identify real anomalies.

[0080] Step S213: When the overlapping area ratio falls within the threshold range corresponding to the edge-prone area, determine that the high-temperature candidate area belongs to the edge-prone area partition type.

[0081] The edge-prone area corresponds to the high-temperature candidate region within the threshold range, indicating that the high-temperature candidate region is mainly distributed in the lower edge area of ​​the module. Due to the influence of the photovoltaic module installation structure and dust accumulation distribution, the probability of hot spots occurring in the edge-prone area is higher than that in the middle of the module panel. The high-temperature candidate region belonging to the edge-prone area is directly retained as an abnormal candidate region in the subsequent step S23 and enters the abnormality type determination in step S3.

[0082] It should be noted that step S213 marks the candidate areas of the edge-prone area separately, so that edge hot spot anomalies can be processed in the subsequent step S3 according to the judgment rules that are different from those of hot spots in the array body area, thus avoiding the omission of edge hot spots due to their different area characteristics from those of hot spots in the middle of the plate.

[0083] Step S214: When the overlap area ratio falls within the threshold range corresponding to the array body area, determine that the partition type of the high temperature candidate region belongs to the array body area.

[0084] The high-temperature candidate region within the threshold range corresponding to the array body area indicates that this high-temperature candidate region is mainly distributed in the main body area of ​​the component board, and is the main location where hot spot anomalies and string failure anomalies occur. The high-temperature candidate region belonging to the array body area is retained as an abnormal candidate region for subsequent judgment after the junction box interference elimination in steps A1 to A4 is completed.

[0085] It should be noted that step S214 marks the high-temperature candidate area belonging to the array body area separately, providing a clear operational range for the fine elimination of the junction box interference area in steps A1 to A4, ensuring that the array body area candidate area entering the anomaly judgment has eliminated the interference of junction box thermal radiation.

[0086] In some embodiments, after step S214, steps A1 to A4 are further included: Step A1: Extract the geometric center coordinates of each high-temperature candidate region whose partition type is array body region.

[0087] The geometric center coordinates are obtained by averaging the pixel coordinates of the area enclosed by the closed contour line of the high-temperature candidate region belonging to the array body region. These coordinates reflect the spatial center position of the high-temperature candidate region in the photovoltaic station area image. The geometric center coordinates are extracted one by one for each high-temperature candidate region belonging to the array body region, resulting in a set of geometric center coordinates for each high-temperature candidate region.

[0088] It should be noted that the geometric center coordinates extracted in step A1 are the input basis for calculating the offset distance in step A2. The accuracy of the geometric center coordinates directly affects the identification result of the interference area of ​​the junction box in the subsequent step.

[0089] Step A2: Determine the upper limit of the offset distance based on the offset distance between the geometric center coordinates and the midpoint of the long side of the array body area, as well as the pixel width of the long side of the array body area.

[0090] The midpoint of the long side of the array body area is the coordinate of the midpoint of the array body area defined in step S1 along the long side, reflecting the geometric center of the component board in the column direction. The offset distance is the pixel distance between the geometric center coordinates of the high-temperature candidate region extracted in step A1 and the midpoint of the long side of the corresponding array body area, reflecting the degree to which the high-temperature candidate region deviates from the center of the component board. The pixel width of the long side of the array body area is the number of pixel columns of the array body area determined in step S1 in the column direction, reflecting the actual pixel scale of the component board.

[0091] In the specific calculation, the process expands pixel by pixel from the midpoint of the long side of the array body area to the left and right. The distance between each pixel column and the midpoint of the long side is taken as the offset distance to be tested. This offset distance is then compared with the pixel width of the long side of the array body area to obtain its proportion of the long side pixel width. Pixel columns whose proportion does not exceed the preset central region proportion are retained, and the offset distance corresponding to the pixel column farthest from the midpoint of the long side is recorded as the upper limit of the offset distance judgment. In this embodiment, the preset central region proportion is 15%. For example, when the pixel width of the long side of the array body area is 200 pixels, when expanding from the midpoint of the long side to either side to the 30th pixel, the offset distance accounts for 15% of the pixel width of the long side. Therefore, 30 pixels is taken as the upper limit of the offset distance judgment. The actual offset distance of the high-temperature candidate area is compared with this upper limit.

[0092] It should be noted that the junction box installed on the back of the photovoltaic module will appear as a localized high-temperature area within the array body area in the thermal imaging image. The heat radiation from the junction box is not indicative of module abnormality, and failure to eliminate it will result in false alarms. The junction box's installation position on the module panel is relatively fixed, typically located in the center of the panel. The offset distance between the geometric center coordinates of the corresponding high-temperature candidate area and the midpoint of the long side of the array body area is small, and the junction box area is much smaller than the area of ​​a single module. Steps A2 and A3 provide the basis for identifying the junction box interference area from the two dimensions of positional offset and area, respectively.

[0093] It should be noted that the upper limit of the offset distance is determined based on the pixel width of the long side of the array body area, so that the upper limit of the judgment can be adapted to the actual pixel scale of different component specifications, and can accurately delineate the possible location range of the junction box under different image resolutions and component specifications.

[0094] Step A3: Determine the upper limit of the number of outline pixels based on the average pixel area of ​​a single component in the array body region.

[0095] The average pixel area of ​​a single component in the array body area is the ratio of the total number of pixels in the array body area determined in calculation step S1 to the number of components covered by the array body area. The resulting pixel count reflects the average pixel area of ​​a single component in the current thermal imaging image. Specifically, the actual dimensions of the junction box and the single component are read from the product drawings of the same model of component. The ratio of the actual area of ​​the junction box to the actual area of ​​the single component is calculated, and this ratio is used as the pixel area conversion ratio. Then, the average pixel area of ​​a single component is multiplied by the pixel area conversion ratio, and the resulting pixel count is rounded up. The rounded pixel count is the upper limit of the outline pixel count.

[0096] When product drawings are unavailable, pre-annotated thermal imaging samples of the same model can be used to determine the maximum proportion of high-temperature contour pixels in the junction box relative to the pixel area of ​​a single component. This maximum proportion can then be used as the pixel area conversion ratio. For example, if the average pixel area of ​​a single component is 20,000 pixels and the junction box area accounts for 2%, the upper limit for the number of contour pixels is 400 pixels. If the number of pixels within the closed contour of the high-temperature candidate region does not exceed this upper limit, it indicates that its area size conforms to the area characteristics of the junction box's thermal radiation area.

[0097] It should be noted that step A3 provides an upper limit constraint for the identification of the interference area of ​​the junction box from the area dimension. Together with the position offset dimension of step A2, it constitutes a dual judgment condition for junction box identification, reducing the risk of false rejection that may occur from a single dimension judgment.

[0098] Step A4: Based on the comparison results of the offset distance and the upper limit of the offset distance determination, the number of pixels enclosed by the closed contour of the high temperature candidate region and the upper limit of the number of contour pixels, mark the high temperature candidate region as the junction box interference region and remove it from the candidate set, or retain the high temperature candidate region in the abnormal candidate region.

[0099] Specifically, when the offset distance is determined to be less than the upper limit of the offset distance judgment, and the number of pixels surrounded by the closed contour line of the high temperature candidate region is less than the upper limit of the number of contour pixels, the corresponding high temperature candidate region is marked as a junction box interference region and removed from the candidate set; when the offset distance is determined to be greater than or equal to the upper limit of the offset distance judgment, or the number of pixels surrounded by the closed contour line of the high temperature candidate region is greater than or equal to the upper limit of the number of contour pixels, the corresponding high temperature candidate region is retained in the abnormal candidate region.

[0100] It should be noted that the determination of junction box interference areas uses a logic of eliminating areas only if both positional offset and area conditions are met simultaneously, rather than eliminating areas based on either one condition. This is because real hot spots may exist in the central area of ​​the component board. If a region is eliminated solely based on a small offset distance, real hot spots in the central area of ​​the board would be mistakenly eliminated. Similarly, if a region is eliminated solely based on a small area, small real hot spots would be mistakenly eliminated. Only when both conditions are met can a region be reliably identified as a junction box interference area. If the offset distance is greater than or equal to the upper limit of the offset distance determination, or if the number of pixels enclosed by the closed contour line is greater than or equal to the upper limit of the contour pixel count, then the failure to meet either condition indicates that the high-temperature candidate area does not meet the typical characteristics of junction box thermal radiation and is therefore retained.

[0101] It should be noted that step A4 achieves a balance between eliminating junction box interference and retaining real hot spots through a dual-condition joint judgment, so that the interference of junction box thermal radiation is effectively controlled in the abnormal candidate area of ​​the array body area entering the subsequent step S3, while avoiding the underreporting of real hot spots due to excessive elimination.

[0102] Step S22: Determine the edge grayscale change sequence based on the edge pixel grayscale difference sequence within the assigned partition type.

[0103] The edge pixel grayscale difference sequence is obtained by calculating the grayscale difference between each boundary pixel of a high-temperature candidate region and its adjacent non-candidate region pixels within the same partition type, and arranging them in spatial order according to the boundary pixels. This sequence reflects the grayscale transition at the edge of the high-temperature candidate region. The edge grayscale change sequence is the grayscale change trend information extracted from the edge pixel grayscale difference sequence, specifically the direction and continuity of the grayscale difference changes in the sequence.

[0104] Understandably, in real-world equipment anomalies, thermal imaging images show a concentrated increase in localized temperature, with grayscale values ​​at the edges exhibiting a continuous decrease from the anomaly area to the surrounding region. This corresponds to a continuous, unidirectional increasing subsequence in the edge pixel grayscale difference sequence. In contrast, the grayscale distribution in the background thermal radiation area is typically more uniform or disordered, and there is no stable, continuous, unidirectional increasing subsequence in the edge grayscale difference sequence. The edge grayscale change sequence is precisely the feature quantity used to describe these differences.

[0105] It should be noted that step S22 extracts the edge grayscale change sequence, which provides a grayscale verification basis for the retention and elimination of candidate regions of background interference area in step S23, so that the high temperature candidate regions belonging to the background interference area can be further distinguished from the real anomaly and the background thermal radiation through grayscale features.

[0106] Step S23: Remove high-temperature candidate regions that belong to the background interference region and do not have continuous unidirectional gray-level increasing subsequences in the edge gray-level change sequence from the candidate set, and retain high-temperature candidate regions that belong to the array body region or edge-prone region to obtain abnormal candidate regions.

[0107] For high-temperature candidate regions belonging to the background interference zone, if there is no continuous unidirectional gray-level increasing subsequence in the edge gray-level change sequence, it indicates that the edge of the high-temperature candidate region lacks the concentrated temperature rise transition characteristics that a true anomaly should have, and it is judged as background thermal radiation interference and removed from the candidate set; if there is a continuous unidirectional gray-level increasing subsequence in the edge gray-level change sequence, it is retained in the anomaly candidate region. For high-temperature candidate regions belonging to the array body area or edge-prone area, after completing the junction box interference removal, they are directly retained in the anomaly candidate region.

[0108] It should be noted that candidate regions in the background interference area are not all eliminated. Instead, the decision to retain them is made after verification using the edge grayscale change sequence. This is because real anomalies such as occasional localized overheating of the support structure or obstruction by foreign objects exist within the background interference area. Removing all such anomalies would lead to missed detections. The selection of continuous, unidirectional grayscale increasing subsequences as the retention criterion is based on the fact that real abnormal thermal radiation exhibits a clear grayscale gradient concentration characteristic in thermal imaging images, while background thermal radiation typically does not possess this characteristic. The two can be distinguished by their edge grayscale change sequences.

[0109] It should be noted that the abnormal candidate regions obtained after background interference removal in step S23 have completed source differentiation in two dimensions: spatial affiliation and grayscale features. The anomaly type determination in the subsequent step S3 can adopt differentiated rules for candidate regions of different affiliation and partition types to ensure the accuracy of the determination results.

[0110] In some embodiments, step S23 further includes steps B1 to B3: Step B1: If the response anomaly candidate region is an empty set, extract the average gray value of the cold area in the splicing gap and the average gray value of the array body region from the spatial attribute partition map, and determine the difference in average gray value for the current time based on the average gray value of the cold area in the splicing gap and the average gray value of the array body region.

[0111] An empty set of abnormal candidate regions means that after elimination in step S23, no high-temperature candidate regions are retained in the candidate set. In this case, it is necessary to further determine whether the empty set result is due to the absence of equipment abnormality in the current image or due to the extraction baseline gray value being set too high, resulting in the real abnormal region not being extracted as a high-temperature candidate region. The average gray value of the cold area in the splicing gap and the average gray value of the array body area are both extracted from the spatial attribute partition map generated in step S1. The gray value difference is the difference between the average gray value of the cold area in the splicing gap and the average gray value of the array body area, reflecting the overall gray value contrast level between the component plate and the gap at the current image acquisition time.

[0112] It is understandable that the mean difference in grayscale reflects the overall radiation intensity of the current thermal imaging image. By comparing the mean difference in grayscale accumulated in historical inspection cycles, it can be determined whether the radiation intensity of the current image is within the normal range.

[0113] It should be noted that step B1 establishes a remedial judgment mechanism when the abnormal candidate region is an empty set by the difference of the mean gray value of the current extraction, which avoids the failure to report real anomalies due to the problem of setting the baseline gray value in a single extraction, and improves the reliability of the inspection results.

[0114] Step B2: When the mean difference of grayscale falls within the range of mean difference of grayscale in the historical inspection cycle, the extraction benchmark grayscale value of the high-temperature candidate region is recalibrated based on the mean grayscale value of the cold area in the splicing gap. Based on the calibrated extraction benchmark grayscale value, the extraction of the high-temperature candidate region and the removal of background interference are re-executed to obtain the updated abnormal candidate region.

[0115] The historical inspection cycle gray-scale mean difference range is a reasonable range of values ​​obtained by statistically analyzing the gray-scale mean difference records of multiple historical inspections. Those skilled in the art can determine the statistical method for this range based on the accumulated historical inspection data of the actual power plant, such as using the interquartile range of historical records. When the current gray-scale mean difference falls within the historical inspection cycle gray-scale mean difference range, it indicates that the overall radiation intensity of the current image is at a historically normal level. An empty set of abnormal candidate regions is likely due to an excessively high extraction reference gray-scale value. In this case, the extraction reference gray-scale value is recalibrated using the gray-scale mean of the cold area in the splicing gap as the reference, ensuring that the extraction reference gray-scale value matches the actual gray-scale distribution of the current image. Then, the high-temperature candidate region extraction and background interference removal are re-executed to obtain an updated abnormal candidate region.

[0116] It should be noted that the average gray value of the cold area in the stitching gap is used as the calibration benchmark because the cold area in the stitching gap always appears as a low gray value region in the thermal imaging image. Its average gray value can stably reflect the overall gray value benchmark level of the current image. Using this as a reference point, the benchmark gray value is extracted for calibration, so that the calibration result is adapted to the actual radiation intensity level of the current image.

[0117] It should be noted that the calibration and re-extraction mechanism in step B2 enables the extraction reference gray value to be adjusted according to the actual radiation intensity level of the current image, which solves the problem of missing extraction of real anomalies caused by fixing the extraction reference gray value under different time periods and temperature conditions, and improves the stability of inspection results under different environmental conditions.

[0118] Step B3: When the mean difference of grayscale in the current inspection is lower than the lower limit of the mean difference of grayscale in the historical inspection cycle, it is determined that the overall radiation intensity of the current photovoltaic station area image has not reached the hot spot display condition, the current inspection result is marked as valid, and the current inspection process is terminated.

[0119] When the mean grayscale difference is lower than the lower limit of the mean grayscale difference range for the historical inspection cycle, it indicates that the overall grayscale contrast between the module panel and the gap is insufficient at the time of image acquisition. This suggests that the overall radiation intensity of the photovoltaic module is low under the current environmental conditions, and temperature anomalies such as hot spots have not yet formed identifiable grayscale differences in the thermal imaging image. Even if the baseline grayscale value is recalibrated and extracted, no meaningful anomaly candidate region can be obtained. In this case, it is determined that the current image does not meet the conditions for hot spot manifestation, a valid marker for the current inspection result is output, the reason why the anomaly was not detected due to insufficient radiation intensity is recorded, and the current inspection process is terminated to avoid continuing to perform meaningless judgment operations on invalid images.

[0120] It is worth mentioning that steps B2 and B3 cover two cases: the current grayscale mean difference falls within the range of grayscale mean differences in the historical inspection period, and the current grayscale mean difference falls below the lower limit of the range. If the current grayscale mean difference is higher than the upper limit of the range of grayscale mean differences in the historical inspection period, it indicates that the overall radiation intensity of the current image is higher than the historical normal level, and the result that the abnormal candidate region is an empty set is reliable. The conclusion that there is no abnormality in the current inspection can be directly output.

[0121] It should be noted that steps B1 to B3 constitute the complete processing link when the anomaly candidate region is an empty set. By comparing the gray-scale mean difference of the current time with the gray-scale mean difference range of the historical inspection cycle, the empty set results are divided into three cases: high extraction benchmark, insufficient radiation intensity, and no real anomaly, which are processed separately to ensure the interpretability and traceability of the inspection conclusion.

[0122] Preferably, step S2 uses two levels of screening—spatial attribution determination and edge grayscale verification—to distinguish high-temperature candidate regions from different sources in the photovoltaic station area image into abnormal candidate regions and interference regions. This ensures that the anomaly type determination in subsequent step S3 is based on candidate regions with clear sources and eliminated interference. The empty set remediation mechanism in steps B1 to B3 further guarantees the integrity of inspection results under different environmental conditions. The two levels work together to ensure that the output of step S2 accurately reflects the actual distribution of equipment anomalies in the current image while avoiding false alarms and missed alarms caused by extraction benchmark deviations or background interference.

[0123] Step S3: Based on the area ratio and partition type of the abnormal candidate region, generate abnormal alarm information by referring to the preset abnormal judgment rules.

[0124] It should be noted that while the spatial source of the candidate anomalies obtained in step S2 has been distinguished, the anomaly type of each candidate region has not yet been determined. Different types of equipment anomalies exhibit different area and edge grayscale features in thermal imaging images. For example, hot spot anomalies typically present as small local areas with concentrated edge grayscale gradients, while a series of failure anomalies present as large areas with relatively dispersed edge grayscale gradients. If a uniform anomaly type determination criterion is applied to all candidate regions, the feature differences between different anomaly types will be ignored, leading to inaccurate anomaly type labeling. This step combines area proportion, edge grayscale gradient features, and the type of zone to which the region belongs, and uses preset anomaly determination rules to label the anomaly type of each candidate region, classifying the urgency level according to the edge grayscale gradient intensity, and finally generating anomaly alarm information.

[0125] In some embodiments, step S3 includes steps S31 to S34: Step S31: Determine the area ratio based on the number of pixels enclosed by the closed contour lines of each candidate region within the abnormal candidate region and the total number of pixels in the array body region to which they belong.

[0126] The number of pixels enclosed by the closed contour line is a value obtained by counting each pixel within the area enclosed by the boundary contour line of each anomaly candidate region, reflecting the actual pixel area of ​​the candidate region in the thermal imaging image. The total number of pixels in the array body area to which the candidate region belongs is the total number of pixels within the array body area to which the candidate region belongs, reflecting the total pixel area of ​​the corresponding single-column component board. The area ratio is the ratio of the number of pixels enclosed by the closed contour line to the total number of pixels in the array body area, reflecting the area ratio of the candidate region relative to the component board.

[0127] Understandably, the area ratio normalizes the absolute pixel area of ​​the candidate region to a proportional value relative to the board of its component, decoupling the area feature from the image resolution and component specifications, and ensuring the comparability of area features under different resolution conditions.

[0128] It should be noted that the area ratio determined in step S31 is one of the core feature quantities that distinguishes hot spot anomalies from whole-string failure anomalies in the preset anomaly judgment rule in step S33. The comparison result of the area ratio and the average pixel area ratio of a single component in the array body area directly determines the output direction of the anomaly type label in subsequent steps C1 and C2.

[0129] Step S32: Obtain edge gray-level gradient features based on the gray-level difference sequence between the boundary pixels of each candidate region and the adjacent non-candidate region pixels within the abnormal candidate region.

[0130] Boundary pixels are pixels on the closed contour line of the anomaly candidate region, while adjacent non-candidate region pixels are pixels spatially adjacent to the boundary pixels but not belonging to the candidate region. The gray-level difference sequence is obtained by calculating the gray-level difference between each boundary pixel and its adjacent non-candidate region pixels, and arranging them in spatial order according to the boundary pixels. Edge gray-level gradient features are statistical features extracted from the gray-level difference sequence, including the maximum gray-level difference and the average gray-level difference. The maximum gray-level difference reflects the strongest gray-level contrast at the edge of the candidate region, while the average gray-level difference reflects the overall level of gray-level contrast at the edge.

[0131] It is understandable that hot spot anomalies are usually caused by concentrated temperature rise due to localized cell damage, with strong gray-scale contrast at the edges and a high maximum gray-scale difference. In contrast, string failure anomalies are characterized by large-area uniform temperature rise, with a more uniform distribution of gray-scale differences at the edges, and the ratio of the maximum gray-scale difference to the average gray-scale difference is close to 1. Edge gray-scale gradient features are precisely the characteristic quantities that capture the above differences.

[0132] It should be noted that the edge grayscale gradient features extracted in step S32, together with the area ratio and the type of partition to which they belong, participate in the anomaly type determination in step S33, and serve as the basis for emergency level classification in step S34. They are the core feature quantities that run through the entire process of step S3.

[0133] Step S33: Input the area ratio, edge grayscale gradient features and the type of partition to the preset anomaly judgment rule for comparison, and output the anomaly type label.

[0134] The preset anomaly judgment rules are judgment logics pre-defined based on actual operation and maintenance experience of photovoltaic power plants and thermal imaging characteristics of various equipment anomalies. They use area ratio, edge grayscale gradient characteristics and the type of the assigned partition as joint judgment conditions to output anomaly type labels.

[0135] It should be noted that step S33 inputs the three types of features into the preset anomaly judgment rule. Compared with the judgment method based on a single feature, it can distinguish candidate regions that are similar in appearance on a single feature but have different actual anomaly types. For example, hot spots with similar areas but different gray-scale gradient distributions and entire failures, thus improving the accuracy of anomaly type labeling.

[0136] In some embodiments, the preset anomaly determination rule in step S33 includes steps C1 to C4: Step C1: When the partition type is array body area, the area ratio is lower than the average pixel area ratio of the single component in the array body area, and the maximum gray level difference in the edge gray level gradient feature is higher than the median of the maximum gray level difference of all abnormal candidate regions, the abnormal type label is output as hot spot abnormal.

[0137] The average pixel area ratio of a single component in the array body area is the ratio of the total number of pixels in the array body area to the number of components covered by the array body area in step S31, and then normalized relative to the total number of pixels in the array body area. This ratio reflects the average pixel area ratio occupied by a single component in the array body area. The median of the maximum grayscale difference of all abnormal candidate regions in this inspection is a statistic obtained by taking the median of the maximum grayscale difference of the edge grayscale gradient features of all abnormal candidate regions in this inspection. This reflects the overall level of grayscale contrast at the edges of abnormal candidate regions in this inspection.

[0138] It should be noted that the typical characteristics of hot spot anomalies are small-area concentrated temperature rises caused by localized cell damage, with an area lower than the average level of a single module and strong gray-scale contrast at the edges. The median of the maximum gray-scale difference across all candidate anomaly regions in a given inspection, rather than a fixed threshold, is used as a reference because ambient temperature and solar radiation intensity differ at different inspection times, causing the absolute gray-scale difference to vary. Using the median as a relative reference eliminates the influence of environmental factors on the threshold, ensuring the stability of the judgment results under different conditions.

[0139] It should be noted that step C1 combines the two conditions of an area ratio lower than that of a single component and a grayscale contrast higher than the median as the criteria for judging hot spot anomalies. This ensures that the labeling of hot spot anomalies requires both the area characteristics to conform to the scale characteristics of local damage and the grayscale gradient characteristics to conform to the intensity characteristics of concentrated temperature rise. Both are indispensable, thus avoiding the mislabeling of small but weakly contrasted background residual areas as hot spot anomalies.

[0140] Step C2: When the partition type is array body area, the area ratio is higher than the average pixel area ratio of the single component in the array body area, and the ratio of the maximum gray level difference to the average gray level difference in the edge gray level gradient feature is lower than the median of the corresponding ratios of all abnormal candidate regions, the output abnormal type label is "entire string failure abnormal".

[0141] The ratio of the maximum grayscale difference to the average grayscale difference reflects the concentration of grayscale contrast at the edge. A ratio closer to 1 indicates a more uniform grayscale difference across the edge, corresponding to the grayscale distribution characteristics of a uniform temperature rise across the entire spectrum. A larger ratio indicates a more concentrated grayscale contrast at the edge, corresponding to the grayscale distribution characteristics of localized temperature rises in hot spots. The median of the ratios corresponding to all anomaly candidate regions in a given inspection is a statistic obtained by taking the median of the ratios of the maximum grayscale difference to the average grayscale difference for each of the anomaly candidate regions in that inspection.

[0142] It should be noted that the typical characteristics of a string failure anomaly are a large-area uniform temperature rise caused by circuit breakage or severe attenuation of the entire array or string of components. The area exceeds the average level of a single component, and the grayscale contrast at the edges is uniformly dispersed. The area exceeding the average level of a single component is the basis for distinguishing the area of ​​a string failure from that of a single hot spot. The ratio of the maximum grayscale difference to the average grayscale difference being lower than the median is the basis for distinguishing the grayscale distribution of a uniform temperature rise across the entire string from a localized concentrated temperature rise. The combination of these two factors ensures the accuracy of the labeling of string failure anomalies.

[0143] It should be noted that steps C1 and C2 together cover all candidate regions belonging to the array body region partition type. Hotspot anomalies are those where the area percentage is lower than the average level for a single component and the maximum grayscale difference is higher than the median; string failure anomalies are those where the area percentage is higher than the average level for a single component and the ratio of the maximum grayscale difference to the average grayscale difference is lower than the median. Candidate regions with other combinations will be retained as anomalies for further verification in subsequent steps and will not be forcibly categorized into the above two labels.

[0144] Step C3: When the partition type is determined to be an edge-prone area and the area ratio is lower than the average pixel area ratio of the single component of the array body area, the abnormality type label is output as edge hot spot abnormality.

[0145] Candidate areas belonging to the zoning type of edge-prone area appear in the lower edge area of ​​the module and have an area lower than the average level of a single module. This corresponds to the local temperature rise caused by uneven dust accumulation or damage to the edge cells at the lower edge of the module, and is marked as edge hot spot anomaly.

[0146] It should be noted that edge hot spot anomalies and array body area hot spot anomalies differ in their causes and handling methods. Edge hot spots are usually related to the module installation tilt angle and dust accumulation and distribution, and the main treatment is cleaning; array body area hot spots are usually related to cell damage or loose solder joints, and the main treatment is replacement or repair. Distinguishing between these two types of anomalies by different labels allows for the generation of differentiated handling instructions in step S4.

[0147] It should be noted that in step C3, only the area ratio condition is set for the candidate areas belonging to the edge-prone area, and the median comparison of the maximum gray-scale difference is not introduced. This is because the high-temperature candidate areas of the edge-prone area have already been screened by spatial belonging determination in step S2, and the reliability of the source is guaranteed when entering step S3, so there is no need to add gray-scale intensity conditions.

[0148] Step C4: When the partition type is determined to be a cold area in the splicing gap and the maximum gray level difference in the edge gray level gradient features is higher than the difference between the gray level average of the cold area in the splicing gap and the gray level average of the array body area, the output anomaly type label is gap anomaly.

[0149] The difference between the average grayscale value of the cold zone in the splicing gap and the average grayscale value of the array body area is the difference between the average grayscale values ​​of the two types of partitions extracted from the spatial attribute partitioning map generated in step S1. It reflects the grayscale comparison benchmark between the gap area and the plate area under normal conditions. When the maximum grayscale difference of the candidate area belonging to the cold zone of the splicing gap is higher than the above benchmark difference, it indicates that the thermal radiation intensity of the gap area is abnormally high, exceeding the range of normal grayscale comparison between the gap and the plate. This indicates that there is a gap anomaly caused by foreign objects, short circuits, etc. at the corresponding gap location.

[0150] It should be noted that the cold zone in the splicing gap should normally be a low-grayscale area in the thermal imaging image. If a high-temperature candidate area appears in the gap area with a grayscale value higher than the grayscale comparison benchmark between the normal plate and the gap, it indicates that the thermal radiation at the gap has exceeded the normal range of structural thermal radiation, and there is an anomaly that needs to be investigated. The difference between the average grayscale value of the cold zone in the splicing gap and the average grayscale value of the array body area is used as the judgment benchmark, so that the judgment threshold is directly derived from the grayscale statistics of the current image, without the need to pre-set a fixed threshold.

[0151] It should be noted that the four anomaly type determination rules in steps C1 to C4 cover common thermal imaging anomaly types in photovoltaic power plants. The corresponding partition types of each rule do not overlap, ensuring that each candidate area can only trigger one type of anomaly label and avoiding duplicate labeling.

[0152] Step S34: Based on the maximum gray-level difference in the edge gray-level gradient features, classify the emergency level of the candidate region, and generate anomaly alarm information based on the anomaly type label and emergency level.

[0153] The maximum grayscale difference, extracted from the edge grayscale gradient features in step S32, reflects the strongest grayscale contrast at the edge of the candidate region and is positively correlated with the local temperature rise intensity of the candidate region, serving as a quantitative basis for emergency level classification. The anomaly alarm information is structured information generated by concatenating the anomaly type label, emergency level, and candidate region center coordinates. It is used in step S4 for generating key inspection paths for the control ball and retrieving differentiated handling instructions.

[0154] It should be noted that step S34 uses the maximum grayscale difference as the basis for classifying the emergency level, and integrates the spatial location, type and urgency of the anomaly into structured anomaly alarm information, so that step S4 can directly complete path planning and disposal scheduling based on the anomaly alarm information without having to re-analyze the original candidate area.

[0155] In some embodiments, step S34 includes steps D1 to D4: Step D1: When the maximum grayscale difference is greater than the first preset comparison value, the emergency level is classified as high emergency level.

[0156] The first preset comparison value is the grayscale difference boundary used to distinguish between high and medium emergency levels. Specifically, before the inspection, calibration samples are selected where the component temperature rises to a state requiring immediate action. Thermal imaging images are acquired under the same emissivity, shooting distance, and imaging parameters as the actual inspection. The maximum grayscale difference between the edge pixels of the calibration area and the pixels of the adjacent normal area is calculated, and the average of multiple calculation results is set as the first preset comparison value. When the maximum grayscale difference is greater than the first preset comparison value, it indicates that the temperature rise of the candidate area has reached a level requiring immediate action, the emergency level is classified as high emergency, and action is prioritized.

[0157] Step D2: When the maximum grayscale difference is less than the first preset comparison value and greater than the second preset comparison value, the emergency level is classified as medium emergency level.

[0158] The second preset comparison value is used to distinguish between medium and low urgency levels based on the grayscale difference, and its value is less than the first preset comparison value. Specifically, calibration samples where the component temperature rises to a state requiring tracking and observation are selected. Using the same imaging conditions and calculation method as the first preset comparison value, the maximum grayscale difference between the edge pixels of the calibration area and the pixels of the adjacent normal area is obtained, and the average value of multiple calculations is set as the second preset comparison value. When the maximum grayscale difference is greater than the second preset comparison value but less than the first preset comparison value, it indicates that the candidate area needs to be tracked and observed but has not yet reached the level requiring immediate action, and the urgency level is classified as medium urgency.

[0159] Step D3: When the maximum grayscale difference is less than the second preset comparison value, the emergency level is classified as a low emergency level.

[0160] When the maximum grayscale difference is less than the second preset comparison value, the temperature rise intensity of the corresponding candidate area is at a relatively low level, which is judged as a low emergency level and can be included in the routine inspection plan.

[0161] It should be noted that steps D1 to D3 divide the range of the maximum grayscale difference into three intervals: greater than the first preset comparison value, between the second preset comparison value and the first preset comparison value, and less than the second preset comparison value. These intervals correspond to high, medium, and low urgency levels, respectively. These three intervals completely cover the entire range of the maximum grayscale difference. When the maximum grayscale difference at the endpoint is exactly equal to the first preset comparison value, it is classified as medium urgency; when it is exactly equal to the second preset comparison value, it is classified as low urgency. This ensures that each candidate area receives a uniquely determined urgency level. The three-level urgency classification allows the connection order of alarms within the key inspection path of the monitoring ball in step S4 to be arranged in descending order of urgency, ensuring that inspection resources prioritize coverage of the candidate area with the highest temperature rise intensity.

[0162] Step D4: Concatenate the anomaly type label, emergency level, and candidate area center coordinates to generate anomaly alarm information.

[0163] The center coordinates of the candidate region are the average pixel coordinates of the area enclosed by the closed contour line of the candidate region, reflecting the spatial location of the candidate region in the photovoltaic station area image. Field concatenation combines the anomaly type label, urgency level, and candidate region center coordinates into a structured string according to a predetermined format to obtain the anomaly alarm information. The anomaly alarm information contains three types of information: anomaly location, type, and urgency level, and is the complete input for step S4 to generate the key inspection path of the surveillance camera and differentiated handling instructions.

[0164] It should be noted that step D4 integrates location, type, and urgency into structured abnormal alarm information, enabling step S4 to directly complete inspection path planning and handling scheduling based on abnormal alarm information. The structured form of the alarm information also facilitates message encapsulation of subsequent inspection scheduling messages.

[0165] Preferably, step S3 uses a combined determination of area proportion, edge grayscale gradient features, and assigned partition type to classify anomaly candidate regions into four categories: hot spot anomalies, entire failure anomalies, edge hot spot anomalies, and gap anomalies. It then classifies these regions into three urgency levels based on the maximum grayscale difference, ensuring that each anomaly candidate region receives a clear type and urgency label. Each determination rule uses the statistics of all anomaly candidate regions in the current iteration as a relative reference benchmark, eliminating the interference of environmental factors on the absolute grayscale difference and ensuring that the anomaly type labeling results remain stable under different inspection time conditions. The output of step S3 provides a complete structured input for the differentiated scheduling in step S4, enabling targeted allocation of inspection resources according to anomaly type and urgency.

[0166] Step S4: Based on the regional alarm density and the type of the abnormal alarm information, generate the key inspection path and differentiated handling instructions for the deployment ball, and output the inspection scheduling message.

[0167] It should be noted that the abnormal alarm information generated in step S3 includes the spatial location, anomaly type, and urgency level of each candidate area, but it has not yet formed an inspection path that can be executed by the control sphere or a handling instruction that can be directly used by maintenance personnel. Abnormal alarm information in a photovoltaic power station may be distributed across different zoning types, such as the array body area, edge-prone areas, and splicing gap cold areas. The on-site working environment varies depending on the zoning type. For example, the maintenance of the module panels corresponding to the array body area requires high-altitude work, while the handling of ground-based locations corresponding to edge-prone areas and splicing gap cold areas only requires ground work. If a unified handling instruction is generated for all alarms, maintenance personnel can only determine the required equipment and time limits after arriving on-site, potentially leading to duplicate deployments due to insufficient preparation. This step, based on the regional alarm density and zoning type, simultaneously generates key inspection paths for the control sphere and differentiated handling instructions, ensuring that the inspection dispatch message includes both path and handling information at once.

[0168] In some embodiments, step S4 includes steps S41 to S44: Step S41: Count the number of abnormal alarm messages in each home partition type, and determine the regional alarm density of each home partition type based on the area of ​​each home partition type and the corresponding number of alarms.

[0169] The number of abnormal alarm messages within each assigned partition type is the number of alarms in each group obtained by grouping all abnormal alarm messages generated in step S3 according to their assigned partition type. The area of ​​each assigned partition type is the pixel area of ​​each partition in the spatial attribute partition map generated in step S1, which is converted into actual geographic area and used as the area benchmark for density calculation. The regional alarm density is the ratio of the number of alarms corresponding to each assigned partition type to the area of ​​each assigned partition type, reflecting the degree of concentration of abnormal alarms per unit area.

[0170] It is understandable that the regional alarm density of the assigned partition type corresponds to the area with concentrated anomalies. The deployment ball inspection resources should prioritize covering the high-density partition, so that the limited inspection time and path length are concentrated in the area with concentrated anomalies.

[0171] It should be noted that step S41 uses the area alarm density rather than the absolute number of alarms as the priority basis for path planning because the area of ​​different partition types is different. The degree of abnormal concentration per unit area of ​​a partition with a large absolute number but a large area may be lower than that of a partition with a small absolute number but a small area. Using density as the basis can more accurately reflect the degree of abnormal concentration of each partition and ensure that the inspection path covers the truly dense areas of abnormality.

[0172] Step S42: Sort alarms in descending order of regional alarm density according to each partition type, aggregate alarms within the same partition type into inspection clusters, connect the center coordinates of alarms in each inspection cluster in descending order of emergency level, and connect the inspection clusters in descending order of regional alarm density to generate the key inspection path of the deployment ball.

[0173] An inspection cluster is a spatial collection of all abnormal alarm information within the same assigned partition type, with one inspection cluster corresponding to each partition type. Alarms within an inspection cluster are connected by their center coordinates in descending order of urgency level; that is, the center coordinates of high-urgency alarms are connected before medium-urgency and low-urgency alarms, forming a local path within the inspection cluster. Inspection clusters are then connected sequentially in descending order of regional alarm density; that is, the inspection cluster with the highest regional alarm density is at the forefront of the path, followed by the next highest density cluster, and so on. All inspection clusters are then connected in series to form the key inspection path for the monitoring sphere.

[0174] It should be noted that the key inspection path of the surveillance camera adopts a two-level sorting logic: between clusters, sorted by density in descending order; within clusters, sorted by urgency level in descending order, rather than sorting all alarms uniformly by urgency level. The basis for this design is that the overall processing efficiency of areas with concentrated anomalies is higher than that of dispersed areas. Completing all alarm inspections in high-density areas first and then moving to low-density areas results in a shorter path and lower travel consumption for the surveillance camera compared to frequently switching between single high-urgency alarms across areas. Within the same area, alarms with higher urgency are still processed first according to their urgency level, ensuring that high-risk anomalies are not delayed due to density sorting.

[0175] It should be noted that the key inspection path of the control ball generated in step S42 uses the center coordinates of the abnormal alarm information as the path node. The control ball covers all abnormal candidate areas in the order of the path nodes, so that the inspection resources of the control ball are tilted towards the areas where real abnormalities are concentrated, reducing the proportion of invalid movement in areas without abnormalities.

[0176] Step S43: Based on the environmental operation characteristics corresponding to the partition type to which each abnormal alarm information belongs, retrieve the corresponding safety equipment list and handling time limit requirements, and generate differentiated handling instructions.

[0177] Environmental operation characteristics are job types determined based on the physical environment characteristics corresponding to the assigned zone type. Different assigned zone types correspond to different on-site operation environments, and thus different safety equipment and response time limits. The safety equipment list is a pre-defined list of safety protection equipment that maintenance personnel should carry for a specific operation environment. Response time limits are pre-defined standards for completing anomaly response within a specific operation environment and urgency level.

[0178] It should be noted that step S43 establishes a mapping between the assigned partition type and the environmental operation characteristics, so that the generation of differentiated handling instructions does not require maintenance personnel to go to the site to judge the operation environment. The operation environment information is included in the inspection and scheduling message in advance, so that maintenance personnel can complete the equipment preparation before departure.

[0179] In some embodiments, step S43 includes steps E1 to E4: Step E1: When the assigned zone type is determined to be an edge-prone area or a cold zone between splicing gaps, the environmental operation characteristics are identified as a ground operation environment. Based on the emergency level, the corresponding list of ground operation safety equipment and short-term response time limit requirements are matched.

[0180] The edge-prone area corresponds to the lower edge of the component near the ground, and the cold zone in the splicing gap corresponds to the ground layer area at the gap between components. Both types of zones are located at ground level, requiring no aerial work equipment from personnel, and are thus identified as ground-based work environments. The short-time response limit is a time limit set for completing responses in ground-based work environments. Those skilled in the art can select this limit based on the actual operation time and safety procedures required for ground-based work. Ground-based work has shorter preparation and operation times compared to aerial work, hence the corresponding short-time response limit.

[0181] Step E2: When the assigned partition type is determined to be the array body area, the environmental operation characteristics are identified as a high-altitude operation environment. Based on the emergency level, the corresponding list of high-altitude operation safety equipment and standard handling time limit requirements are matched.

[0182] The array body area corresponds to the main body of the photovoltaic module panel. The module panels are mounted on a support frame, requiring workers to use climbing equipment or scaffolding for maintenance operations, thus constituting a high-altitude work environment. The list of safety equipment for high-altitude work includes essential protective equipment such as safety belts, safety helmets, and non-slip shoes. Those skilled in the art can configure these according to the actual power plant's operational safety procedures. The standard handling time limit is a time limit set specifically for handling situations in high-altitude work environments. Compared to ground-based operations, these require longer preparation and operation time, hence the corresponding standard handling time limit.

[0183] Step E3: When the assigned partition type is determined to be a background interference zone, the environmental operation characteristics are identified as a remote observation environment, and the corresponding remote verification operation instructions and long-term handling time limits are matched according to the emergency level.

[0184] The background interference area corresponds to areas such as the ground background and support structures. Although step S23 has performed edge grayscale verification on the candidate areas of the background interference area, the candidate areas of the background interference area retained in the anomaly candidate area have a certain probability of being real anomalies, but further remote verification is still required. The remote verification operation instruction is a remote confirmation operation instruction for anomalies in the background interference area, including operations such as retrieving historical images for comparison and adjusting the observation angle of the surveillance sphere, without requiring maintenance personnel to arrive on-site immediately. The long-term handling time limit requirement is a handling completion time limit set for the remote observation environment, which is longer than the time limit for ground and high-altitude operations, allowing remote verification to be completed first before deciding whether on-site handling is required.

[0185] It should be noted that steps E1 to E3 cover the determination of environmental operation characteristics for all four types of zoning categories. Background interference areas, having already been screened through edge grayscale verification in step S23, are retained as candidate areas for anomalies and are considered questionable and awaiting verification. The identification of remote observation environments and the long-term processing time limits correspond to this, linking the investment of processing resources to the verification results and avoiding unnecessary on-site operations triggered by questionable alarms. Cold areas in splicing gaps may output gap anomaly labels in step C4. Since the gap location is at ground level, it belongs to the same ground operation environment as edge-prone areas and is handled uniformly in step E1.

[0186] It should be noted that steps E1 to E3, through the mapping between the class partition type and the environmental operation characteristics, bring the on-site operation environment judgment forward to the inspection and scheduling stage. This allows the differentiated handling instructions to include the equipment list and time limit requirements in the inspection and scheduling message. After receiving the message, the operation and maintenance personnel can directly prepare according to the instructions without having to go to the site to assess the operation conditions.

[0187] Step E4: Bind the matching list of safety equipment, time limit requirements for handling or remote verification operation instructions to the corresponding alarm coordinates to generate differentiated handling instructions.

[0188] The alarm coordinates are the center coordinates of the candidate area contained in the abnormal alarm information in step D4, reflecting the spatial location of the corresponding abnormality. Step E4 associates the safety equipment list, handling time limit requirements, or remote verification operation instructions with the corresponding alarm coordinates into a record, so that each differentiated handling instruction contains two types of information: handling location and handling requirements.

[0189] Step S44: Encapsulate the key inspection paths and differentiated handling instructions of the surveillance ball into messages and output the inspection scheduling message.

[0190] The inspection and dispatch message contains the path node sequence of the ball and the differentiated handling instructions corresponding to each node. It is a complete dispatch output for the ball control system and maintenance personnel.

[0191] It should be noted that step S44 integrates path information and handling information into an inspection and dispatch message, realizing a fully automated closed loop from anomaly identification to inspection and dispatch. This enables the deployment ball control system to directly perform path inspections based on the message, and maintenance personnel to directly complete on-site handling preparations based on the message, reducing the need for manual secondary analysis.

[0192] Preferably, step S4 transforms the structured anomaly alarm information output in step S3 into executable key inspection paths and differentiated handling instructions by jointly applying regional alarm density and attribution partition type. Regional alarm density ensures that inspection paths cover areas with concentrated anomalies, while attribution partition type ensures that handling instructions match the on-site operating environment. The combination of these two factors enables targeted allocation of inspection dispatch messages in both spatial coverage and handling adaptation dimensions, completing a full closure from the accuracy of anomaly identification to the effectiveness of inspection dispatch.

[0193] Based on the above steps, this application also includes the following embodiments: See Figure 3 This is a schematic diagram of the structure of the automatic inspection system for the operating status of photovoltaic power station equipment provided in this application embodiment, including: The spatial segmentation module is used to perform spatial attribute segmentation on the photovoltaic station area image based on the structural arrangement characteristics of the photovoltaic array, and obtain a spatial attribute partition map. The overlap determination module is used to determine the partition type of each high temperature candidate region based on the spatial overlap ratio between each high temperature candidate region in the photovoltaic station area image and the spatial attribute partition map, and to perform background interference removal based on the partition type to obtain abnormal candidate regions. The rule matching module is used to generate abnormal alarm information based on the area ratio and partition type of the abnormal candidate region and by comparing it with the preset abnormal judgment rules. The strategy generation module is used to generate key inspection paths and differentiated handling instructions for the deployment ball based on the regional alarm density and the type of the abnormal alarm information, and output inspection scheduling messages.

[0194] Figure 3 The system of the illustrated embodiment can be used to perform corresponding operations. Figure 1 The steps in the method embodiments shown are implemented in a similar manner and have similar technical effects, and will not be repeated here.

[0195] See Figure 4 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. The electronic device 40 includes: a processor 41, a memory 42, and a computer program; wherein, The memory 42 is used to store the computer program, and the memory may also be flash memory. The computer program is, for example, an application program or functional module that implements the above method.

[0196] The processor 41 is configured to execute the computer program stored in the memory to implement the various steps performed by the device in the above method. For details, please refer to the relevant descriptions in the preceding method embodiments.

[0197] Alternatively, the memory 42 can be either standalone or integrated with the processor 41.

[0198] When the memory 42 is a device independent of the processor 41, the device may further include: Bus 43 is used to connect the memory 42 and the processor 41.

[0199] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. An automatic inspection method for the operating status of photovoltaic power station equipment, characterized in that, include: Based on the structural arrangement characteristics of photovoltaic arrays, spatial attribute segmentation is performed on the image of the photovoltaic station area to obtain a spatial attribute partition map. Based on the spatial overlap ratio between each high-temperature candidate region in the photovoltaic station area image and the spatial attribute zoning map, the zoning type of each high-temperature candidate region is determined, and background interference removal is performed based on the zoning type to obtain abnormal candidate regions. Based on the area ratio of the abnormal candidate regions and the partition type to which they belong, an abnormal alarm message is generated by comparing the information with the preset abnormal judgment rules. Based on the regional alarm density of the abnormal alarm information and the type of the assigned partition, a key inspection path and differentiated handling instructions are generated for the deployment ball, and an inspection scheduling message is output.

2. The method according to claim 1, characterized in that, The spatial attribute partitioning map includes a cold zone at the splicing gap, an array body area, an edge-prone area, and a background interference area; The method of spatial attribute segmentation of the photovoltaic station area image based on the structural arrangement characteristics of the photovoltaic array to obtain a spatial attribute partition map includes: The image of the photovoltaic station area is projected in grayscale along the component arrangement direction, and continuous grayscale low-level areas are extracted and identified as splicing gap cold zones. Based on the average gray value of the cold zone in the splicing gap, the array body area is defined by extending outwards. Using a fixed ratio of the average spacing between adjacent splicing gap cold zones as the extension width, the extension width is extended inward along the lower boundary of the array body area and marked as the edge-prone area. The area outside the splicing gap cold zone, the array body area, and the edge-prone area is marked as the background interference area.

3. The method according to claim 2, characterized in that, The process of defining the array body area by extending it outward based on the average grayscale value of the cold zone in the splicing gap includes: Extract the boundary column coordinates of each adjacent splicing gap cold zone in the photovoltaic station area image facing the side of the panel; The column coordinate segment between the boundary column coordinates of two adjacent splicing gap cold zones facing one side of the plate is taken as the candidate body range; Extract the average grayscale value of each column in the column direction within the candidate ontology range; The longest continuous column segment formed by the pixel column whose cumulative grayscale average value in the column direction is higher than the grayscale average value of the cold zone in the splicing gap is defined as the array body area.

4. The method according to claim 2, characterized in that, The process of performing grayscale projection on the photovoltaic station area image along the component arrangement direction, extracting continuous grayscale low-level areas, and determining them as splicing gap cold zones includes: The image of the photovoltaic station area is processed by accumulating pixel gray levels along the column direction to generate a horizontal gray level distribution curve; Extract all consecutive pixel column segments in the horizontal grayscale distribution curve whose cumulative grayscale value is lower than the mean of the horizontal grayscale distribution curve, calculate the spacing between adjacent pixel column segments, and obtain the spacing sequence; Based on the spacing sequence, the upper and lower limits of the spacing are determined. Pixel column segments in the spacing sequence whose spacing values ​​fall outside the range of the lower to upper limits of the spacing are removed, and the remaining pixel column segments are determined as the cold zone of the splicing gap.

5. The method according to claim 1, characterized in that, The process involves determining the partition type of each high-temperature candidate region based on the spatial overlap ratio between the candidate regions in the photovoltaic station area image and the spatial attribute partition map, and performing background interference removal based on the partition type to obtain abnormal candidate regions, including: The partition type to which the high-temperature candidate region belongs is determined based on the proportion of the overlapping area between each high-temperature candidate region and each partition in the spatial attribute partitioning map. The edge grayscale change sequence is determined based on the edge pixel grayscale difference sequence within the assigned partition type; High-temperature candidate regions whose partition type is background interference region and whose edge gray-level change sequence does not contain a continuous unidirectional gray-level increasing subsequence are removed from the candidate set, while high-temperature candidate regions whose partition type is array body region or edge-prone region are retained to obtain the abnormal candidate regions.

6. The method according to claim 5, characterized in that, The step of determining the partition type of a high-temperature candidate region based on the proportion of overlap between each high-temperature candidate region and each partition in the spatial attribute partitioning map includes: Based on the actual pixel area ratio of each partition in the spatial attribute partitioning map, the threshold range of the overlapping area ratio of the background interference area, the edge-prone area and the array body area is determined. When the overlapping area ratio falls within the threshold range corresponding to the background interference area, the partition type of the high temperature candidate region is determined to be the background interference area. When the overlapping area ratio falls within the threshold range corresponding to the edge-prone zone, the partition type of the high-temperature candidate region is determined to be the edge-prone zone. When the overlapping area ratio falls within the threshold range corresponding to the array body region, the partition type of the high-temperature candidate region is determined to be the array body region.

7. The method according to claim 6, characterized in that, When determining that the overlap area ratio falls within the threshold range corresponding to the array body region, after determining that the high-temperature candidate region belongs to the array body region, the method further includes: Extract the geometric center coordinates of each high-temperature candidate region whose partition type is the array body region; Based on the offset distance between the geometric center coordinates and the midpoint of the long side of the array body area, and the pixel width of the long side of the array body area, the upper limit of the offset distance is determined. The upper limit of the number of outline pixels is determined based on the average pixel area of ​​a single component in the array body region. Based on the comparison results of the offset distance and the upper limit of the offset distance determination, the number of pixels enclosed by the closed contour line of the high temperature candidate region and the upper limit of the number of contour pixels, the high temperature candidate region is marked as a junction box interference region and removed from the candidate set, or the high temperature candidate region is retained in the abnormal candidate region.

8. The method according to claim 1, characterized in that, The step of generating anomaly alarm information based on the area ratio of the abnormal candidate regions and the partition type to which they belong, in accordance with preset anomaly determination rules, includes: The area ratio is determined based on the number of pixels enclosed by the closed contour lines of each candidate region within the abnormal candidate region and the total number of pixels in the array body region to which it belongs. Based on the gray-level difference sequence between the boundary pixels of each candidate region and the adjacent non-candidate region pixels within the abnormal candidate region, edge gray-level gradient features are obtained. The area ratio, the edge grayscale gradient features, and the partition type are input into a preset anomaly determination rule for comparison, and anomaly type labels are output. Based on the maximum gray-level difference in the edge gray-level gradient features, the emergency level of the candidate region is divided, and the abnormal alarm information is generated based on the abnormality type label and the emergency level.

9. The method according to claim 8, characterized in that, The preset anomaly determination rules include: When the partition type is determined to be the array body area, the area ratio is lower than the average pixel area ratio of the single component in the array body area, and the maximum gray-level difference in the edge gray-level gradient feature is higher than the median of the maximum gray-level difference of all abnormal candidate regions, the abnormal type label is output as hot spot abnormal. When the partition type is determined to be the array body area, the area ratio is higher than the average pixel area ratio of the single component in the array body area, and the ratio of the maximum gray level difference to the average gray level difference in the edge gray level gradient features is lower than the median of the corresponding ratios of all abnormal candidate regions, the abnormal type label is output as the whole string of failure abnormalities. When the partition type is determined to be an edge-prone area and the area ratio is lower than the average pixel area ratio of a single component in the array body area, the abnormality type label is output as edge hot spot abnormality. When the partition type is determined to be a splicing gap cold area and the maximum gray level difference in the edge gray level gradient features is higher than the difference between the gray level average of the splicing gap cold area and the gray level average of the array body area, the output anomaly type label is gap anomaly.

10. The method according to claim 8, characterized in that, The step of classifying the emergency level of candidate regions based on the maximum gray-level difference in the edge gray-level gradient features, and generating the anomaly alarm information based on the anomaly type label and the emergency level, includes: When the maximum grayscale difference is determined to be greater than the first preset comparison value, the emergency level is classified as a high emergency level. When the maximum grayscale difference is determined to be less than a first preset comparison value and greater than a second preset comparison value, the emergency level is classified as a medium emergency level. When the maximum grayscale difference is determined to be less than the second preset comparison value, the emergency level is classified as a low emergency level. The anomaly type label, the emergency level, and the center coordinates of the candidate area are concatenated to generate the anomaly alarm information.

11. The method according to claim 1, characterized in that, Based on the regional alarm density and the assigned partition type of the abnormal alarm information, the system generates a key inspection path and differentiated handling instructions for the surveillance system, and outputs an inspection scheduling message, including: The number of abnormal alarm messages within each assigned partition type is counted, and the regional alarm density of each assigned partition type is determined based on the area of ​​each assigned partition type and the corresponding number of alarms. Alarms are sorted in descending order of regional alarm density according to each partition type. Alarms within the same partition type are aggregated into inspection clusters. Alarms within each inspection cluster are connected to the center coordinates in descending order of emergency level. Inspection clusters are connected sequentially in descending order of regional alarm density to generate key inspection paths for the control ball. Based on the environmental operation characteristics corresponding to the partition type to which each abnormal alarm information belongs, retrieve the corresponding safety equipment list and handling time limit requirements, and generate differentiated handling instructions; The key inspection path of the control ball and the differentiated handling instruction are encapsulated into a message, and the inspection scheduling message is output.

12. The method according to claim 11, characterized in that, The process involves retrieving the corresponding safety equipment list and handling time limits based on the environmental operation characteristics corresponding to the partition type to which each abnormal alarm information belongs, and generating differentiated handling instructions, including: When the assigned zone type is determined to be an edge-prone area or a cold zone in the splicing gap, the environmental operation characteristics are identified as a ground operation environment, and the corresponding ground operation safety equipment list and short-term response time limit are matched according to the emergency level. When the assigned partition type is determined to be the array body area, the environmental operation characteristics are identified as a high-altitude operation environment, and the corresponding high-altitude operation safety equipment list and standard handling time limit requirements are matched according to the emergency level. When the assigned partition type is determined to be a background interference zone, the environmental operation characteristics are identified as a remote observation environment, and the corresponding remote verification operation instructions and long-term handling time limits are matched according to the emergency level. The matching list of safety equipment, time limit requirements for handling, or remote verification operation instructions are bound to the corresponding alarm coordinates to generate the differentiated handling instructions.

13. An automatic inspection system for the operating status of photovoltaic power station equipment, employing the automatic inspection method for the operating status of photovoltaic power station equipment as described in any one of claims 1 to 12, characterized in that, include: The spatial segmentation module is used to perform spatial attribute segmentation on the photovoltaic station area image based on the structural arrangement characteristics of the photovoltaic array, and obtain a spatial attribute partition map. The overlap determination module is used to determine the partition type of each high temperature candidate region based on the spatial overlap ratio between each high temperature candidate region in the photovoltaic station area image and the spatial attribute partition map, and to perform background interference removal based on the partition type to obtain abnormal candidate regions. The rule matching module is used to generate abnormal alarm information based on the area ratio of the abnormal candidate region and the partition type to which it belongs, by comparing with preset abnormal judgment rules. The strategy generation module is used to generate key inspection paths and differentiated handling instructions for the deployment ball based on the regional alarm density of the abnormal alarm information and the type of the assigned partition, and output inspection scheduling messages.

Images