An ai-based distributed photovoltaic module management method

Abstract

The application provides an AI-based distributed photovoltaic module management method, and belongs to the technical field of management analysis, which comprises the following steps: obtaining rated parameters of photovoltaic modules and environmental characteristic parameters of a deployment area; planning a checkpoint matrix, and controlling a UAV equipped with a detection module to collect real scene images of the photovoltaic modules according to a preset flight route; calculating power generation differences in combination with theoretical power generation and actual power generation, and determining the causes of the differences; dividing power generation unit blocks according to a preset rule, evaluating power generation efficiency of each block based on the analysis results, and locking target blocks with abnormal efficiency; extracting photovoltaic panel working state characteristic parameters, locating faulty photovoltaic panels, and determining fault types and severity. The AI-based distributed photovoltaic module management method provided by the application effectively avoids invalid maintenance and fault omission, significantly improves photovoltaic module management efficiency, reduces operation and maintenance costs, and at the same time guarantees stable operation of the photovoltaic system and maximization of power generation efficiency.

Description

Technical Field

[0001] This invention belongs to the field of management analysis technology, and more specifically, relates to an AI-based distributed photovoltaic module management method. Background Technology

[0002] Distributed photovoltaic (PV) systems, as a core component of clean energy systems, have been widely applied in various scenarios such as industrial plants and residential rooftops due to their advantages of flexible installation, high resource utilization, and minimal environmental impact. Their power generation efficiency is affected by factors such as the condition of the modules themselves, their installation orientation, surface cleanliness, and external environmental conditions. The dispersed deployment and large number of modules make them susceptible to natural factors, leading to orientation shifts, surface contamination, and potential mechanical damage and hot spots, resulting in power degradation and energy loss. Therefore, constructing a precise and efficient module management system is crucial to ensuring the stable operation of PV systems.

[0003] Currently, the management of distributed photovoltaic modules mainly relies on manual inspection and traditional technical monitoring: manual inspection is labor-intensive, inefficient, and easily subject to environmental limitations, making it difficult to achieve comprehensive high-frequency monitoring; traditional technical monitoring is mostly limited to the preliminary identification of surface defects, lacks deep integration with power generation data, cannot distinguish the causes of power differences, and lacks a standardized block performance evaluation system, resulting in a lack of quantitative basis for fault judgment, which leads to insufficient targeted maintenance and makes it difficult to solve the core pain points of "inaccurate monitoring, superficial analysis, and indiscriminate maintenance". Summary of the Invention

[0004] The purpose of this invention is to provide an AI-based distributed photovoltaic module management method, which aims to solve the problems of insufficient integration with power generation data, making it impossible to distinguish the causes of power differences, and lack of quantitative basis for fault judgment, resulting in insufficient targeted maintenance.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is: to provide an AI-based distributed photovoltaic module management method, comprising: Obtain the rated parameters of the photovoltaic module and the environmental characteristic parameters of the deployment area, and calculate the theoretical power output of the photovoltaic module; Plan the checkpoint matrix and control drones equipped with detection modules to collect real-world images of photovoltaic modules along preset routes; Image analysis is used to extract posture change data and surface contamination data of photovoltaic panels. Combined with theoretical power generation and actual power generation, the difference in power generation is calculated and the cause of the difference is determined. The power generation unit blocks are divided according to preset rules. The power generation efficiency of each block is evaluated based on the above analysis results, and the target block with abnormal efficiency is locked. The drone is controlled to perform precise detection on the target block and extract the working status characteristic parameters of the photovoltaic panel. Based on the working status characteristic parameters and the detection results, the faulty photovoltaic panel is located and the fault type and severity are determined.

[0006] In one possible implementation, the rated parameters include the model and specifications of the photovoltaic modules, rated power, open-circuit voltage, short-circuit current, initial installation tilt angle, and module array spacing; the environmental characteristic parameters include the average daily solar irradiance, ambient temperature range, wind speed level, precipitation frequency, and atmospheric particulate matter concentration of the deployment area.

[0007] In one possible implementation, the planning checkpoint matrix includes: Based on the row and column distribution of the photovoltaic module array, several blind-spot-free monitoring sub-areas are divided. Within each monitoring sub-area, checkpoints are set according to a six-point layout of "four corners + center + edge midpoint". The shooting angle and height of each checkpoint are calibrated to ensure that the image completely covers all photovoltaic panels within the sub-area.

[0008] In one possible implementation, the theoretical power generation is expressed using a preset power generation calculation model:

[0009] In theory, This represents the theoretical power generation (kWh) of photovoltaic modules. The rated power (kW) of the photovoltaic module. This is the ratio of the actual light intensity to the light intensity under standard test conditions. This is the ambient temperature correction factor. This represents the power generation efficiency coefficient corresponding to the current tilt angle. The photoelectric conversion efficiency degradation coefficient of a photovoltaic module. The measurement period is (h). This represents the initial decay rate of the component.

[0010] In one possible implementation, the attitude change data includes tilt offset and planar displacement, and the surface contamination data includes the proportion of contaminant coverage area, estimated coverage thickness, and contaminant type feature vectors. The extraction process includes: using the SIFT algorithm to extract key feature points of the photovoltaic panel from the real-scene image and the reference calibration image; performing feature matching using the RANSAC algorithm; calculating the three-dimensional coordinate changes of the feature points to obtain attitude change data; and performing semantic segmentation of the photovoltaic panel surface based on a CNN image segmentation model to identify and quantify the contaminant coverage area, thereby forming a surface contamination dataset.

[0011] In one possible implementation, the determination of the cause of the difference includes: a preset power attenuation threshold. With environmental disturbance threshold Calculate the actual power attenuation. ; like This was determined to be a normal fluctuation in power generation. like Calculate the actual changes in environmental characteristic parameters ,like and and correlation coefficient This was determined to be an external environmental disturbance. like ,and or The problem was determined to be a component failure.

[0012] In one possible implementation, dividing the power generation unit blocks according to preset rules includes: Based on the electrical connection relationship and physical array distribution of photovoltaic modules, independent power generation unit blocks are divided using either a "single-panel partitioning" or "string unified partitioning" model. The evaluation of the power generation efficiency of each block includes constructing a block power generation efficiency evaluation model, selecting power attenuation rate, pollutant coverage ratio, tilt angle offset correction coefficient, and temperature influence coefficient as evaluation indicators, using the analytic hierarchy process to determine the indicator weights, calculating the comprehensive efficiency evaluation value through weighted summation, and marking blocks below a preset threshold as target blocks.

[0013] In one possible implementation, the detection module includes a high-definition visible light camera and an infrared thermal imaging module. The precise detection includes appearance detection and temperature detection. The extraction of photovoltaic panel operating status feature parameters includes: performing grayscale enhancement and noise filtering preprocessing on the high-definition visible light image; identifying appearance defects such as cracks, missing corners, aging of the encapsulation layer, and foreign matter adhesion on the photovoltaic panel surface using morphological algorithms; and outputting appearance feature parameters such as defect type, defect area, and defect location coordinates. The infrared thermal imaging data undergoes temperature calibration and pseudo-color processing; abnormally high-temperature areas are identified using hot spot detection algorithms; the temperature peak value, temperature gradient, and high-temperature area ratio of the high-temperature area are calculated; and temperature distribution feature parameters are output based on the photovoltaic panel circuit topology. The appearance feature parameters and temperature distribution feature parameters are fused to form a set of operating status feature parameters including defect severity level, hot spot risk coefficient, and performance degradation trend value.

[0014] In one possible implementation, locating the faulty photovoltaic panel and determining the fault type and severity includes: A photovoltaic module fault classification system is established, covering tilt offset faults, surface contamination faults, mechanical damage faults, hot spot faults, circuit connection faults, and module aging faults. The operating status characteristic parameters are input into a preset AI fault diagnosis model, which is constructed by fusing CNN and random forest algorithms to achieve fault type classification and identification. A fault severity grading standard is set, and faults are divided into three levels: minor faults, general faults, and severe faults based on the quantified value of defect parameters, the magnitude of temperature anomalies, and the proportion of power attenuation. A complete fault judgment result is output.

[0015] In one possible implementation, the following is included after outputting the fault determination result: The theoretical power generation, actual power generation, attitude change data, surface contamination data, fault judgment results, and detection image data are linked and stored in the cloud-based photovoltaic module management database to establish a full life-cycle health record for the modules. Based on the severity level of the fault, differentiated maintenance strategy recommendations are generated, including maintenance priority ranking, maintenance window planning, specific maintenance operation procedures, and spare parts replacement plans. A visual monitoring platform is built to display the efficiency evaluation value, fault distribution, and maintenance progress of each power generation unit block in real time in the form of heat maps and topology maps, triggering audible and visual warnings and multi-channel notifications for serious faults. Historical data in the cloud database is regularly statistically analyzed to iteratively optimize the parameters of the power generation calculation model and the AI ​​fault diagnosis model, achieving self-evolution and upgrading of management methods.

[0016] The beneficial effects of the AI-based distributed photovoltaic module management method provided by this invention are as follows: Compared with the prior art, the AI-based distributed photovoltaic module management method of this invention calculates the theoretical power generation by obtaining the rated parameters and environmental characteristic parameters of the photovoltaic module, plans the checkpoint matrix and uses drones to collect real-scene images, extracts key data such as attitude changes and surface contamination through image analysis, determines the cause by combining the difference between theoretical and actual power generation, locks the abnormal area through block-based performance evaluation, and finally uses drones to accurately detect and locate the faulty photovoltaic panel and the fault type, forming a complete management closed loop of "power generation calculation - status monitoring - cause determination - block evaluation - fault location". The system realizes precise management of distributed photovoltaic modules from the whole to the part and from the macro to the micro in all dimensions.

[0017] The method described in this application overcomes the inefficiencies of manual inspections and the limitations of traditional monitoring methods, achieving comprehensive coverage of module status, high-frequency monitoring, and quantitative analysis. It addresses the pain points of existing technologies, such as "inaccurate monitoring and superficial analysis." It enables precise fault location, type determination, and severity classification, providing a scientific basis for differentiated maintenance. This effectively avoids ineffective maintenance and fault omissions, significantly improving photovoltaic module management efficiency, reducing operation and maintenance costs, and simultaneously ensuring the stable operation of the photovoltaic system and maximizing power generation efficiency. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating an AI-based distributed photovoltaic module management method provided in this embodiment of the invention. Detailed Implementation

[0020] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0021] Please see Figure 1 The present invention will now describe an AI-based distributed photovoltaic module management method. An AI-based distributed photovoltaic module management method includes: Obtain the rated parameters of the photovoltaic module and the environmental characteristic parameters of the deployment area, and calculate the theoretical power output of the photovoltaic module; Plan the checkpoint matrix and control drones equipped with detection modules to collect real-world images of photovoltaic modules along preset routes; Image analysis is used to extract posture change data and surface contamination data of photovoltaic panels. Combined with theoretical power generation and actual power generation, the difference in power generation is calculated and the cause of the difference is determined. The power generation unit blocks are divided according to preset rules. The power generation efficiency of each block is evaluated based on the above analysis results, and the target block with abnormal efficiency is locked. The drone is controlled to perform precise detection on the target block and extract the working status characteristic parameters of the photovoltaic panel. Based on the working status characteristic parameters and the detection results, the faulty photovoltaic panel is located and the fault type and severity are determined.

[0022] The beneficial effects of the AI-based distributed photovoltaic module management method provided by this invention are as follows: Compared with the prior art, the AI-based distributed photovoltaic module management method of this invention calculates the theoretical power generation by obtaining the rated parameters and environmental characteristic parameters of the photovoltaic module, plans the checkpoint matrix and uses drones to collect real-scene images, extracts key data such as attitude changes and surface contamination through image analysis, determines the cause by combining the difference between theoretical and actual power generation, locks the abnormal area through block-based performance evaluation, and finally uses drones to accurately detect and locate the faulty photovoltaic panel and the fault type, forming a complete management closed loop of "power generation calculation - status monitoring - cause determination - block evaluation - fault location". The system realizes precise management of distributed photovoltaic modules from the whole to the part and from the macro to the micro in all dimensions.

[0023] The method described in this application overcomes the inefficiencies of manual inspections and the limitations of traditional monitoring methods, achieving comprehensive coverage of module status, high-frequency monitoring, and quantitative analysis. It addresses the pain points of existing technologies, such as "inaccurate monitoring and superficial analysis." It enables precise fault location, type determination, and severity classification, providing a scientific basis for differentiated maintenance. This effectively avoids ineffective maintenance and fault omissions, significantly improving photovoltaic module management efficiency, reducing operation and maintenance costs, and simultaneously ensuring the stable operation of the photovoltaic system and maximizing power generation efficiency.

[0024] In some embodiments of the AI-based distributed photovoltaic module management method provided in this application, the rated parameters include the model and specifications of the photovoltaic module, rated power, open-circuit voltage, short-circuit current, initial installation tilt angle, and module array spacing; the environmental characteristic parameters include the average daily solar irradiance, ambient temperature range, wind speed level, precipitation frequency, and atmospheric particulate matter concentration of the deployment area.

[0025] Defining the specific ranges of rated parameters for photovoltaic modules and environmental characteristics of the deployment area is crucial for providing comprehensive and accurate foundational data to support theoretical power generation calculations, power difference analysis, and cause determination. The selection logic for rated parameters closely revolves around the module's own performance and installation status. Model specifications and rated power are fundamental indicators of the module's power generation capacity; open-circuit voltage and short-circuit current reflect core electrical performance; initial installation tilt angle directly affects solar radiation reception efficiency; and module array spacing is related to shading effects. These parameters collectively constitute a complete description of module performance, ensuring the scientific validity of theoretical power generation calculations.

[0026] The selection of environmental characteristic parameters comprehensively covers key external factors affecting the power generation efficiency of the photovoltaic modules: average daily solar intensity is the core source of power generation energy; ambient temperature range affects photoelectric conversion efficiency; wind speed level is related to the attitude stability of the photovoltaic panels; and precipitation frequency and atmospheric particulate matter concentration respectively affect the natural cleaning effect and pollution level of surface pollutants. These parameters provide a reliable data foundation for subsequent quantification of environmental disturbances and differentiation of fault causes, ensuring the objectivity and accuracy of subsequent analysis results.

[0027] In some embodiments of the AI-based distributed photovoltaic module management method provided in this application, the planning checkpoint matrix includes: Based on the row and column distribution of the photovoltaic module array, several blind-spot-free monitoring sub-areas are divided. Within each monitoring sub-area, checkpoints are set according to a six-point layout of "four corners + center + edge midpoint". The shooting angle and height of each checkpoint are calibrated to ensure that the image completely covers all photovoltaic panels within the sub-area.

[0028] The core of refining the planning method for the checkpoint matrix is ​​to solve the problems of "coverage integrity" and "data comparability" in UAV image acquisition, laying the foundation for accurate extraction of subsequent feature parameters. The division of monitoring sub-regions is based on the row and column distribution of photovoltaic module arrays, ensuring that the boundaries of each sub-region are clear and there are no monitoring blind spots. This effectively adapts to distributed photovoltaic systems with different layouts and avoids image omissions caused by the dispersed distribution of modules.

[0029] The checkpoint layout employs a six-point design: four corners + center + edge midpoint. This is a highly efficient data acquisition solution optimized through practical experience. The four corner points cover the boundary components of the sub-region, the center point ensures image clarity for components in the core area, and the edge midpoints fill in acquisition gaps in the boundary transition areas. The six points work together to achieve complete coverage of all photovoltaic panels within the sub-region. Simultaneously, the shooting angle and height of each checkpoint are calibrated to ensure that images acquired at different times and from different checkpoints have a unified benchmark, avoiding feature extraction errors caused by differences in shooting perspectives and ensuring the accuracy of subsequent data comparison and analysis.

[0030] In some embodiments of the AI-based distributed photovoltaic module management method provided in this application, the theoretical power generation adopts a preset power generation calculation model expression as follows:

[0031] In theory, This represents the theoretical power generation (kWh) of photovoltaic modules. The rated power (kW) of the photovoltaic module. This is the ratio of the actual light intensity to the light intensity under standard test conditions. This is the ambient temperature correction factor. This represents the power generation efficiency coefficient corresponding to the current tilt angle. The photoelectric conversion efficiency degradation coefficient of a photovoltaic module. The measurement period is (h). This represents the initial decay rate of the component.

[0032] We provide a scientifically quantified theoretical power generation calculation model. The core of this model is to achieve a high degree of consistency between theoretical power generation and actual operating conditions through multi-factor correction, providing a reliable benchmark value for subsequent power difference analysis. The model parameters cover module rated power, light intensity correction, temperature correction, tilt efficiency coefficient, conversion efficiency degradation coefficient, calculation cycle, and initial degradation rate, comprehensively considering key influencing factors such as module performance, installation orientation, environmental conditions, and aging characteristics.

[0033] Compared to traditional, simplistic calculation methods that only consider illumination and rated power, this model's advantage lies in its multi-dimensional corrections and high degree of adaptability to actual operating conditions. The synergistic effect of various parameters: Correct for differences in lighting conditions. The effect of compensating for temperature on efficiency The effect of quantifying tilt angle changes, and The calculation method considers both long-term degradation and initial degradation separately. This multi-factor integrated approach makes the theoretical power generation more closely match the actual power generation scenario, providing an accurate benchmark for the "difference analysis between theoretical and actual power generation" and improving the reliability of subsequent fault diagnosis and cause determination.

[0034] In some embodiments of the AI-based distributed photovoltaic module management method provided in this application, the attitude change data includes tilt angle offset and planar displacement, and the surface contamination data includes the proportion of contaminant coverage area, estimated coverage thickness, and contaminant type feature vectors. The extraction process includes: using the SIFT algorithm to extract key feature points of the photovoltaic panel in the real-scene image and the reference calibration image, performing feature matching through the RANSAC algorithm, calculating the three-dimensional coordinate changes of the feature points to obtain attitude change data; performing semantic segmentation on the photovoltaic panel surface based on the CNN image segmentation model, identifying the contaminant coverage area and performing quantitative analysis to form a surface contamination data set.

[0035] This study clarifies the extraction methods for attitude change data and surface contamination data, and utilizes advanced image analysis technology to quantitatively extract key parameters, thus solving the problem of traditional methods that rely heavily on qualitative descriptions and lack quantitative analysis. The attitude change data extraction employs a combination of the SIFT and RANSAC algorithms: the SIFT algorithm possesses powerful feature point extraction capabilities, effectively identifying stable features such as photovoltaic panel corners and edges, and resisting image scaling and rotation interference; the RANSAC algorithm eliminates outliers, ensuring matching accuracy, and precisely quantifies tilt offset and planar displacement through changes in the three-dimensional coordinates of feature points.

[0036] Surface contamination data extraction is based on a CNN image segmentation model, which possesses powerful semantic recognition capabilities and can automatically distinguish between photovoltaic panel surfaces and contaminant areas. The contamination range is quantified by calculating the coverage area ratio, and the coverage thickness is estimated by combining grayscale value analysis. Texture feature extraction is used to form contaminant type feature vectors, constructing a complete set of contamination feature parameters. This parameter set not only quantifies the degree of contamination but also distinguishes contaminant types, providing a refined quantitative basis for subsequent performance assessment and maintenance priority determination.

[0037] In some embodiments of the AI-based distributed photovoltaic module management method provided in this application, the determination of the cause of the difference includes: a preset power attenuation threshold. With environmental disturbance threshold Calculate the actual power attenuation. ; like This was determined to be a normal fluctuation in power generation. like Calculate the actual changes in environmental characteristic parameters ,like and and correlation coefficient This was determined to be an external environmental disturbance. like ,and or The problem was determined to be a component failure.

[0038] Establishing scientific rules for determining the causes of power differences addresses the core pain point of existing technologies' inability to effectively distinguish between "external environmental disturbances" and "component failures," providing a decision-making basis for subsequent targeted treatment. The determination logic is centered on quantitative thresholds and correlation analysis: a preset power attenuation threshold is used. The threshold can be flexibly set according to component type and service life to distinguish between normal fluctuations and abnormal degradation; the actual changes in environmental characteristic parameters are calculated. With preset environmental disturbance threshold In comparison, the correlation coefficient between power attenuation and environmental changes was analyzed. Divide the causes.

[0039] This judgment logic realizes the transformation from "subjective experience judgment" to "objective data support": this quantitative judgment method avoids ineffective maintenance or fault omission caused by misjudgment, makes the cause judgment results more accurate and reliable, and provides a clear decision-making direction for subsequent differentiated processing.

[0040] In some embodiments of the AI-based distributed photovoltaic module management method provided in this application, the step of dividing the power generation unit blocks according to preset rules includes: Based on the electrical connection relationship and physical array distribution of photovoltaic modules, independent power generation unit blocks are divided using either a "single-panel partitioning" or "string unified partitioning" model. The evaluation of the power generation efficiency of each block includes constructing a block power generation efficiency evaluation model, selecting power attenuation rate, pollutant coverage ratio, tilt angle offset correction coefficient, and temperature influence coefficient as evaluation indicators, using the analytic hierarchy process to determine the indicator weights, calculating the comprehensive efficiency evaluation value through weighted summation, and marking blocks below a preset threshold as target blocks.

[0041] By clearly defining the division mode and performance evaluation model of power generation unit blocks, a shift from "overall evaluation" to "precise zonal evaluation" has been achieved, enabling rapid identification of areas with abnormal performance. The block division offers two modes: "single-panel zoning" and "string-unified zoning." "Single-panel zoning" is suitable for scenarios with a small number of modules and dispersed layouts, enabling precise evaluation of individual modules. "String-unified zoning" is suitable for scenarios with densely arranged modules, balancing evaluation accuracy and efficiency, and flexibly adapting to the layout characteristics of different distributed photovoltaic systems.

[0042] The performance evaluation model embodies comprehensiveness and objectivity: it selects power attenuation rate, pollutant coverage ratio, tilt offset correction coefficient, and temperature influence coefficient as core evaluation indicators, comprehensively covering key influencing factors; it employs the analytic hierarchy process (AHP) to determine the weights of each indicator, which can be dynamically adjusted according to actual scenarios, and obtains a comprehensive evaluation value through weighted summation. Marking blocks below a preset threshold as target blocks quickly focuses on problem areas, avoids invalid detection of normal blocks, and significantly improves the efficiency and targeting of subsequent accurate detection.

[0043] In some embodiments of the AI-based distributed photovoltaic module management method provided in this application, the detection module includes a high-definition visible light camera and an infrared thermal imaging module. The precise detection includes appearance detection and temperature detection. The extraction of photovoltaic panel operating status feature parameters includes: performing grayscale enhancement and noise filtering preprocessing on the high-definition visible light image; identifying appearance defects such as cracks, missing corners, encapsulation layer aging, and foreign matter adhesion on the photovoltaic panel surface through morphological algorithms; and outputting appearance feature parameters such as defect type, defect area, and defect location coordinates. The infrared thermal imaging data is then subjected to temperature calibration and pseudo-color processing; abnormally high-temperature areas are identified through hot spot detection algorithms; the temperature peak value, temperature gradient, and high-temperature area ratio of the high-temperature area are calculated; and temperature distribution feature parameters are output in conjunction with the photovoltaic panel circuit topology. Finally, the appearance feature parameters and temperature distribution feature parameters are fused to form a set of operating status feature parameters including defect severity level, hot spot risk coefficient, and performance degradation trend value.

[0044] The process of refining the content and feature parameter extraction of precise drone inspection has enabled comprehensive monitoring of photovoltaic panels' "appearance + temperature," providing rich quantitative data for fault diagnosis. The inspection module uses a combination of a high-definition visible light camera and an infrared thermal imaging module: the high-definition visible light camera is used to capture appearance defects such as surface cracks, missing corners, and aging of the encapsulation layer, while the infrared thermal imaging module is used to monitor temperature distribution and identify electrical faults such as hot spots. The combination of the two provides comprehensive coverage of both physical defects and internal faults in the components.

[0045] The feature parameter extraction process consists of three stages: preprocessing, identification, and fusion. Visible light images undergo grayscale enhancement and noise filtering preprocessing, and appearance defect parameters are quantified using morphological algorithms. Infrared thermal imaging data undergoes temperature calibration and pseudo-color processing, and temperature distribution features are extracted using hot spot detection algorithms. Finally, appearance and temperature parameters are fused to form a set of working state feature parameters that includes defect severity levels, hot spot risk coefficients, and performance degradation trend values. This parameter set provides a comprehensive and accurate data source for subsequent fault diagnosis, ensuring the comprehensiveness and accuracy of fault identification.

[0046] In some embodiments of the AI-based distributed photovoltaic module management method provided in this application, locating the faulty photovoltaic panel and determining the fault type and severity includes: A photovoltaic module fault classification system is established, covering tilt offset faults, surface contamination faults, mechanical damage faults, hot spot faults, circuit connection faults, and module aging faults. The operating status characteristic parameters are input into a preset AI fault diagnosis model, which is constructed by fusing CNN and random forest algorithms to achieve fault type classification and identification. A fault severity grading standard is set, and faults are divided into three levels: minor faults, general faults, and severe faults based on the quantified value of defect parameters, the magnitude of temperature anomalies, and the proportion of power attenuation. A complete fault judgment result is output.

[0047] A standardized fault classification system and severity grading method were established, enabling accurate fault identification and quantitative assessment, and providing a basis for differentiated maintenance. The fault classification system comprehensively covers common fault types of distributed photovoltaic modules, including tilt offset faults, surface contamination faults, mechanical damage faults, hot spot faults, circuit connection faults, and module aging faults, ensuring that no fault is missed and adapting to fault diagnosis needs in different scenarios.

[0048] The AI ​​fault diagnosis model is constructed by fusing CNN and random forest algorithms. CNN possesses powerful deep feature extraction capabilities, while random forest algorithms offer good classification stability and anti-interference capabilities. Compared to using a single algorithm, the fusion of the two results in higher diagnostic accuracy and stronger generalization ability. Fault severity is graded based on quantitative indicators, classifying faults into three levels: minor, moderate, and severe, according to the quantified values ​​of defect parameters, the magnitude of temperature anomalies, and the proportion of power attenuation. This avoids vague fault descriptions and provides a clear basis for subsequent maintenance priority determination and differentiated maintenance strategy formulation.

[0049] In some embodiments of the AI-based distributed photovoltaic module management method provided in this application, the method further includes the following after outputting the fault determination result: The theoretical power generation, actual power generation, attitude change data, surface contamination data, fault judgment results, and detection image data are linked and stored in the cloud-based photovoltaic module management database to establish a full life-cycle health record for the modules. Based on the severity level of the fault, differentiated maintenance strategy recommendations are generated, including maintenance priority ranking, maintenance window planning, specific maintenance operation procedures, and spare parts replacement plans. A visual monitoring platform is built to display the efficiency evaluation value, fault distribution, and maintenance progress of each power generation unit block in real time in the form of heat maps and topology maps, triggering audible and visual warnings and multi-channel notifications for serious faults. Historical data in the cloud database is regularly statistically analyzed to iteratively optimize the parameters of the power generation calculation model and the AI ​​fault diagnosis model, achieving self-evolution and upgrading of management methods.

[0050] The extended application of the method extends from "fault detection" to "full lifecycle management," enhancing its practicality and sustainability. The component's full lifecycle health record stores theoretical power generation, actual power generation, attitude change data, contamination data, fault determination results, and detection images in a cloud database, allowing for the tracing of component status changes at different stages and providing data support for component lifespan assessment and performance trend analysis.

[0051] Differentiated maintenance strategies are generated based on the severity level of faults. Corresponding maintenance priorities, time windows, operating procedures, and spare parts replacement recommendations are established for different fault levels, rationally allocating maintenance resources, improving maintenance efficiency, and reducing maintenance costs. The visualization monitoring platform displays data in real-time in the form of heatmaps and topology maps, triggering multi-channel early warnings for severe faults. Simultaneously, by regularly analyzing historical data, the power generation calculation model and AI fault diagnosis model parameters are iteratively optimized, allowing management methods to continuously improve performance with data accumulation and adapt to long-term application needs.

[0052] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An AI-based distributed photovoltaic module management method, characterized in that, include: Obtain the rated parameters of the photovoltaic module and the environmental characteristic parameters of the deployment area, and calculate the theoretical power output of the photovoltaic module; Plan the checkpoint matrix and control drones equipped with detection modules to collect real-world images of photovoltaic modules along preset routes; Image analysis is used to extract posture change data and surface contamination data of photovoltaic panels. Combined with theoretical power generation and actual power generation, the difference in power generation is calculated and the cause of the difference is determined. The power generation unit blocks are divided according to preset rules. The power generation efficiency of each block is evaluated based on the above analysis results, and the target block with abnormal efficiency is locked. The drone is controlled to perform precise detection on the target block and extract the working status characteristic parameters of the photovoltaic panel. Based on the working status characteristic parameters and the detection results, the faulty photovoltaic panel is located and the fault type and severity are determined.

2. The AI-based distributed photovoltaic module management method as described in claim 1, characterized in that, The rated parameters include the model and specifications of the photovoltaic modules, rated power, open-circuit voltage, short-circuit current, initial installation tilt angle, and module array spacing; the environmental characteristic parameters include the average daily solar irradiance, ambient temperature range, wind speed level, precipitation frequency, and atmospheric particulate matter concentration of the deployment area.

3. The AI-based distributed photovoltaic module management method as described in claim 1, characterized in that, The planning checkpoint matrix includes: Based on the row and column distribution of the photovoltaic module array, several blind-spot-free monitoring sub-areas are divided. Within each monitoring sub-area, checkpoints are set up in a six-point layout of "four corners + center + midpoint of the edge". The shooting angle and height of each checkpoint are calibrated to ensure that the image completely covers all photovoltaic panels within the sub-area.

4. The AI-based distributed photovoltaic module management method as described in claim 1, characterized in that, The theoretical power generation adopts a preset power generation calculation model expression as follows: In theory, This represents the theoretical power generation (kWh) of photovoltaic modules. The rated power (kW) of the photovoltaic module. This is the ratio of the actual light intensity to the light intensity under standard test conditions. This is the ambient temperature correction factor. This represents the power generation efficiency coefficient corresponding to the current tilt angle. The photoelectric conversion efficiency degradation coefficient of a photovoltaic module. The measurement period is (h). This represents the initial decay rate of the component.

5. The AI-based distributed photovoltaic module management method as described in claim 1, characterized in that, The attitude change data includes tilt angle offset and planar displacement. The surface contamination data includes the proportion of contaminant coverage area, estimated coverage thickness, and contaminant type feature vectors. The extraction process includes: using the SIFT algorithm to extract key feature points of the photovoltaic panel from the real-scene image and the reference calibration image; performing feature matching using the RANSAC algorithm; calculating the three-dimensional coordinate changes of the feature points to obtain attitude change data; and performing semantic segmentation of the photovoltaic panel surface based on the CNN image segmentation model to identify and quantify the contaminant coverage area, thereby forming a surface contamination dataset.

6. The AI-based distributed photovoltaic module management method as described in claim 1, characterized in that, The determination of the cause of the difference includes: a preset power attenuation threshold. With environmental disturbance threshold Calculate the actual power attenuation. ; like This was determined to be a normal fluctuation in power generation. like Calculate the actual changes in environmental characteristic parameters ,like and and correlation coefficient This was determined to be an external environmental disturbance. like ,and or The problem was determined to be a component failure.

7. The AI-based distributed photovoltaic module management method as described in claim 1, characterized in that, The division of power generation unit blocks according to preset rules includes: Based on the electrical connection relationship and physical array distribution of photovoltaic modules, independent power generation unit blocks are divided using either a "single-panel partitioning" or "string unified partitioning" model. The evaluation of the power generation efficiency of each block includes constructing a block power generation efficiency evaluation model, selecting power attenuation rate, pollutant coverage ratio, tilt angle offset correction coefficient, and temperature influence coefficient as evaluation indicators, using the analytic hierarchy process to determine the indicator weights, calculating the comprehensive efficiency evaluation value through weighted summation, and marking blocks below a preset threshold as target blocks.

8. The AI-based distributed photovoltaic module management method as described in claim 1, characterized in that, The detection module includes a high-definition visible light camera and an infrared thermal imaging module. The precise detection includes appearance inspection and temperature detection. The extraction of photovoltaic panel operating status feature parameters includes: performing grayscale enhancement and noise filtering preprocessing on the high-definition visible light image; identifying appearance defects such as cracks, missing corners, aging of the encapsulation layer, and foreign matter adhesion on the photovoltaic panel surface through morphological algorithms; and outputting appearance feature parameters such as defect type, defect area, and defect location coordinates. The infrared thermal imaging data is subjected to temperature calibration and pseudo-color processing; abnormally high-temperature areas are identified through hot spot detection algorithms; the temperature peak, temperature gradient, and high-temperature area ratio of the high-temperature area are calculated; and temperature distribution feature parameters are output in combination with the photovoltaic panel circuit topology. The appearance feature parameters and temperature distribution feature parameters are fused to form a set of operating status feature parameters that includes defect severity level, hot spot risk coefficient, and performance degradation trend value.

9. The AI-based distributed photovoltaic module management method as described in claim 8, characterized in that, The process of locating faulty photovoltaic panels and determining the type and severity of the fault includes: A photovoltaic module fault classification system is established, covering tilt offset faults, surface contamination faults, mechanical damage faults, hot spot faults, circuit connection faults, and module aging faults. The operating status characteristic parameters are input into a preset AI fault diagnosis model, which is constructed by fusing CNN and random forest algorithms to achieve fault type classification and identification. A fault severity grading standard is set, and faults are divided into three levels: minor faults, general faults, and severe faults based on the quantified value of defect parameters, the magnitude of temperature anomalies, and the proportion of power attenuation. A complete fault judgment result is output.

10. The AI-based distributed photovoltaic module management method as described in claim 9, characterized in that, The following is included after the output of the fault determination result: The theoretical power generation, actual power generation, attitude change data, surface contamination data, fault judgment results, and detection image data are linked and stored in the cloud-based photovoltaic module management database to establish a full life-cycle health record for the modules. Based on the severity level of the fault, differentiated maintenance strategy recommendations are generated, including maintenance priority ranking, maintenance window planning, specific maintenance operation procedures, and spare parts replacement plans. A visual monitoring platform is built to display the efficiency evaluation value, fault distribution, and maintenance progress of each power generation unit block in real time in the form of heat maps and topology maps, triggering audible and visual warnings and multi-channel notifications for serious faults. Historical data in the cloud database is regularly statistically analyzed to iteratively optimize the parameters of the power generation calculation model and the AI ​​fault diagnosis model, achieving self-evolution and upgrading of management methods.

Images