Unmanned aerial vehicle inspection method and device for photovoltaic power station

Abstract

The application relates to the technical field of image processing, and discloses a kind of unmanned aerial vehicle inspection method and device of photovoltaic power station.The method comprises: carrying out multimodal imaging scanning to photovoltaic module, and carrying out time correlation processing to scanning data and electrical parameter data by IEEE1588 clock synchronization protocol, to obtain synchronous detection data package;Synchronous detection data package is processed by feature recognition through multidimensional feature extraction algorithm, and is processed by mapping correlation with electrical parameter, to obtain component health assessment model;Component health assessment model is processed by abnormal feature recognition through historical data comparison algorithm, to obtain component diagnosis data package;Component diagnosis data package is processed by area weight calculation to carry out inspection parameter adjustment, to obtain inspection strategy data package.The application improves the efficiency and accuracy of unmanned aerial vehicle inspection of photovoltaic power station.

Description

Technical Field

[0001] This application relates to the field of image processing, and in particular to a method and apparatus for unmanned aerial vehicle (UAV) inspection of a photovoltaic power station. Background Technology

[0002] Currently, the inspection of photovoltaic (PV) power plants mainly relies on manual inspections or traditional equipment monitoring systems. These methods typically depend on staff periodically checking the physical condition of PV modules and assessing their operational status through basic electrical parameters. Traditional inspection methods have many problems, such as low efficiency, high labor costs, limited accuracy, and the inability to obtain detailed real-time status information of the PV power plant. To improve inspection efficiency and accuracy, several automated inspection technologies have been proposed, among which the use of drones for PV power plant inspection is receiving increasing attention. Drone inspection, combined with high-definition cameras, infrared thermal imaging, and other multimodal imaging technologies, can achieve efficient and accurate module inspection, with its advantages being particularly prominent in large-scale PV power plants. Furthermore, drone inspection can utilize real-time data analysis and digital twin technology to dynamically model the PV power plant, further improving the intelligence level of the inspection process.

[0003] However, existing drone inspection systems still face several challenges, especially in the complex environments of photovoltaic power plants. For example, accurately synchronizing data from different types of sensors (such as electrical parameters and imaging data), efficiently processing massive amounts of data collected from various sensors, and conducting health assessments of components based on this data remain pressing technical challenges. Traditional inspection systems struggle to achieve real-time correlation of multimodal data, hindering the effective improvement of the intelligence and accuracy of inspections. Summary of the Invention

[0004] This application provides a method and apparatus for drone inspection of photovoltaic power plants, which improves the efficiency and accuracy of drone inspection of photovoltaic power plants.

[0005] Firstly, this application provides a method for unmanned aerial vehicle (UAV) inspection of a photovoltaic power station, the method comprising:

[0006] The location, structure, power, voltage, and current data of each photovoltaic module in the photovoltaic power station are synchronously acquired and processed using the POWERBUS bus to obtain a digital twin model data package of the photovoltaic module.

[0007] The component layout features, spacing features, obstacle features, and performance features in the digital twin model data package are processed by regional optimization calculation to obtain the inspection route data package;

[0008] Multimodal imaging scanning of photovoltaic modules is performed, and the scanning data and electrical parameter data are time-correlated processed using the IEEE 1588 clock synchronization protocol to obtain a synchronous detection data packet;

[0009] The synchronous detection data packets are processed for feature recognition using a multi-dimensional feature extraction algorithm, and then mapped and associated with electrical parameters to obtain a component health assessment model.

[0010] The component health assessment model is processed using a historical data comparison algorithm to identify abnormal features, resulting in a component diagnostic data package;

[0011] The inspection strategy data packet is obtained by adjusting the inspection parameters of the component diagnostic data packet through regional weight calculation.

[0012] Secondly, this application provides a drone inspection device for photovoltaic power plants, the drone inspection device for photovoltaic power plants comprising:

[0013] The data acquisition module is used to synchronously acquire and process the location data, structural data, power data, voltage data, and current data of each photovoltaic module in the photovoltaic power station via the POWERBUS bus, and obtain a digital twin model data package of the photovoltaic module.

[0014] The calculation module is used to perform regional optimization calculations on the component layout features, spacing features, obstacle features, and performance features in the digital twin model data package to obtain the inspection route data package;

[0015] The correlation module is used to perform multimodal imaging scanning of photovoltaic modules and to perform time correlation processing on the scanning data and electrical parameter data through the IEEE1588 clock synchronization protocol to obtain a synchronous detection data packet;

[0016] The mapping module is used to perform feature recognition processing on the synchronous detection data packet using a multi-dimensional feature extraction algorithm, and to perform mapping and association processing with electrical parameters to obtain a component health assessment model;

[0017] The identification module is used to perform abnormal feature identification processing on the component health assessment model through historical data comparison algorithm to obtain component diagnostic data package;

[0018] The processing module is used to adjust the inspection parameters of the component diagnostic data packet by calculating the regional weight, so as to obtain the inspection strategy data packet.

[0019] The technical solution provided in this application effectively identifies the performance change trend of components and predicts potential failures by performing time-series analysis on the state characteristic data in the component health assessment model, thus providing a scientific basis for subsequent inspections and maintenance. Through the hierarchical weighting of failure level data, it is possible not only to accurately assess the failure risk in different areas but also to rationally allocate inspection resources, prioritizing inspections of high-risk areas and effectively reducing maintenance time and power plant downtime after a failure. Compared with traditional periodic inspection methods, the dynamic inspection strategy based on failure level and regional importance assessment can more effectively improve inspection coverage and efficiency, ensuring that key areas are inspected more frequently and promptly. This data-driven intelligent inspection method significantly improves the accuracy and flexibility of inspections, avoiding the inefficient practice of applying the same inspection frequency to all areas. Secondly, by combining performance impact factor data with inspection cycle data and configuring the inspection cycle based on regional weight coefficients, a more scientific and reasonable inspection plan can be achieved. The adjustment of the inspection cycle is not only based on historical failure data of components but also takes into account the specific needs and importance of each area, thus avoiding unnecessary resource waste. For example, for areas showing signs of fault or exhibiting long-term anomalies, the system automatically adjusts the inspection frequency to ensure these areas receive more inspections and data collection, effectively reducing the risk of power plant failures. Furthermore, based on sampling requirements calculated from fault type characteristic data, the system can automatically adjust sampling parameters according to the characteristics of different fault types, ensuring sufficiently accurate and comprehensive data support during fault diagnosis, avoiding the omission of key data, and thus improving the accuracy and timeliness of fault diagnosis. Through the application of strategy generation algorithms, personalized inspection plans can be formulated based on sampling parameters and inspection cycle data, avoiding the limitations of fixed plans in traditional inspection methods and further enhancing the flexibility and adaptability of inspections. Depending on the fault type, the inspection plan can automatically adjust the inspection content and frequency, making the inspection plan not only more efficient but also capable of accurately identifying and repairing different types of faults. Attached Figure Description

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

[0021] Figure 1 This is a schematic diagram of one embodiment of the drone inspection method for photovoltaic power plants in this application.

[0022] Figure 2This is a schematic diagram of one embodiment of the drone inspection device for a photovoltaic power station in this application. Detailed Implementation

[0023] This application provides a method and apparatus for unmanned aerial vehicle (UAV) inspection of a photovoltaic power station. The terms "first," "second," "third," "fourth," etc. (if present)," in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0024] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the unmanned aerial vehicle (UAV) inspection method for photovoltaic power plants in this application includes:

[0025] Step S101: The location data, structural data, power data, voltage data, and current data of each photovoltaic module in the photovoltaic power station are synchronously acquired and processed by POWERBUS bus to obtain a digital twin model data package of the photovoltaic module.

[0026] Step S102: Perform regional optimization calculations on the component layout features, spacing features, obstacle features, and performance features in the digital twin model data packet to obtain the inspection route data packet;

[0027] Step S103: Perform multimodal imaging scanning on the photovoltaic module, and perform time correlation processing on the scanning data and electrical parameter data through the IEEE1588 clock synchronization protocol to obtain a synchronization detection data packet;

[0028] Step S104: The synchronous detection data packet is processed by a multi-dimensional feature extraction algorithm for feature recognition, and then mapped and associated with electrical parameters to obtain the component health assessment model;

[0029] Step S105: The component health assessment model is processed by anomaly feature identification using a historical data comparison algorithm to obtain a component diagnostic data package;

[0030] Step S106: Adjust the inspection parameters of the component diagnostic data packet by calculating the regional weight to obtain the inspection strategy data packet.

[0031] It is understood that the executing entity of this application can be a drone inspection device for a photovoltaic power station, or it can be a terminal or a server; no specific limitation is made here. This application's embodiment uses a server as the executing entity for illustration.

[0032] Specifically, multiple data points from each photovoltaic (PV) module are synchronously acquired and processed via the POWERBUS bus, including location, structural, power, voltage, and current data, to generate a digital twin model data package for the PV module. The digital twin model is a virtual representation of the real physical module, with data sourced from multiple sensors and measuring instruments, encompassing various characteristics of the PV module. Specifically, location data undergoes UTM projection coordinate transformation to ensure data standardization; tilt and orientation parameters in the structural data are geometrically standardized to provide accurate spatial positioning of the module; power, voltage, and current data are acquired through ADC sampling circuits and shunt monitors, respectively, and encapsulated into MODBUS protocol data frames. All these data points, after time-synchronized processing, form a complete digital twin data package for the module, available for subsequent analysis. Multiple features in the digital twin model data package, such as module layout, spacing, obstacles, and performance, are processed through area optimization calculations to generate an inspection route data package. Component layout features are processed using a grid segmentation algorithm to obtain the basic inspection area; spacing features are analyzed by calculating the minimum spacing to assess the passage space between components; obstacle features are assessed using a 3D grid modeling algorithm to evaluate potential collision risks; and performance features are identified using a weighted allocation algorithm to identify important areas. Based on these optimization results, a waypoint generation algorithm is used to calculate candidate waypoints, which are further optimized using safe flight path constraints, and the density of the inspection path is adjusted to ensure that the UAV can conduct a comprehensive inspection without affecting inspection efficiency.

[0033] Building upon this foundation, drones scan photovoltaic modules using multimodal imaging devices (such as spectral cameras and infrared thermal imagers) to acquire surface temperature and spectral reflectance data. This data undergoes time-correlation processing via the IEEE 1588 clock synchronization protocol to ensure synchronization between image and electrical parameter data. Image data typically includes timestamps; this synchronization protocol ensures accurate alignment between electrical parameter and image data, forming a synchronized detection data packet. This packet serves as a crucial basis for health assessment in subsequent processing. The synchronized detection data packet employs multidimensional feature extraction algorithms to identify key features from both image and electrical data. For example, spectral data is analyzed using spectral analysis algorithms to extract defect features, while infrared thermal imaging data is analyzed using temperature gradients to extract hotspot features. Subsequently, the features of the image data are mapped and correlated with the performance parameters of the electrical data to form a module health assessment model. This assessment model encompasses not only the physical state of the module but also electrical performance parameters, such as power output and current / voltage stability, providing a comprehensive module health score.

[0034] The component health assessment model compares and analyzes historical data, using historical data comparison algorithms to identify abnormal features and generate component diagnostic data packages. Time-series analysis of historical data reveals long-term trends and short-term changes. Based on threshold judgments using this data, the potential for component failures or performance degradation can be assessed. For example, deviation analysis of current and voltage from historical data can identify problems such as power attenuation or current instability. These issues can be further analyzed using fault mode recognition algorithms to determine the fault type and location. Finally, the component diagnostic data package adjusts inspection parameters through regional weight calculations, generating an inspection strategy data package. Specifically, by weighting fault level data, different importance assessment weights can be assigned to different regions, thereby optimizing inspection frequency and cycle. For example, if abnormal temperatures or electrical performance degradation occur in certain regions, the inspection frequency will increase, and vice versa. The configuration of sampling requirements and inspection cycles is achieved through a strategy generation algorithm, ultimately forming an inspection plan suitable for the current photovoltaic power plant.

[0035] For example, suppose a region in a photovoltaic power plant experiences voltage and current data deviating from normal values ​​by 5%. Comparison with historical data reveals this trend has persisted for approximately two months, and temperature data also indicates overheating in this area. Using a feature fusion algorithm, after generating a diagnostic data package for this region, the system will increase the inspection frequency of this region based on a weighted algorithm, adjusting the inspection strategy and increasing the inspection time and frequency to ensure timely detection of potential problems and prevent the spread of photovoltaic module failures.

[0036] In this embodiment, by performing time-series analysis on the state characteristic data in the component health assessment model, the changing trends of component performance can be effectively identified, and potential failures can be predicted in advance, thus providing a scientific basis for subsequent inspections and maintenance. Through the hierarchical weighting of failure level data, not only can the failure risk of different areas be accurately assessed, but inspection resources can also be rationally allocated, prioritizing inspections of high-risk areas, thereby effectively reducing maintenance time and power plant downtime after a failure. Compared with traditional periodic inspection methods, the dynamic inspection strategy based on failure level and regional importance assessment can more effectively improve inspection coverage and efficiency, ensuring that key areas are inspected more frequently and promptly. This data-driven intelligent inspection method significantly improves the accuracy and flexibility of inspections, avoiding the inefficient practice of applying the same inspection frequency to all areas. Secondly, by combining performance impact factor data with inspection cycle data and configuring the inspection cycle based on regional weight coefficients, a more scientific and reasonable inspection plan can be achieved. The adjustment of the inspection cycle is not only based on historical failure data of components but also takes into account the specific needs and importance of each region, thus avoiding unnecessary resource waste. For example, for areas showing signs of fault or exhibiting long-term anomalies, the system automatically adjusts the inspection frequency to ensure these areas receive more inspections and data collection, effectively reducing the risk of power plant failures. Furthermore, based on sampling requirements calculated from fault type characteristic data, the system can automatically adjust sampling parameters according to the characteristics of different fault types, ensuring sufficiently accurate and comprehensive data support during fault diagnosis, avoiding the omission of key data, and thus improving the accuracy and timeliness of fault diagnosis. Through the application of strategy generation algorithms, personalized inspection plans can be formulated based on sampling parameters and inspection cycle data, avoiding the limitations of fixed plans in traditional inspection methods and further enhancing the flexibility and adaptability of inspections. Depending on the fault type, the inspection plan can automatically adjust the inspection content and frequency, making the inspection plan not only more efficient but also capable of accurately identifying and repairing different types of faults.

[0037] In one specific embodiment, the process of performing step S101 may specifically include the following steps:

[0038] (1) The location data is processed by UTM projection coordinate transformation to obtain component standardized coordinate data, and the tilt angle parameter and orientation parameter in the structural data are processed by geometric parameter standardization to obtain structural feature data;

[0039] (2) The power data is acquired and processed by a high-precision ADC sampling circuit to obtain power sampling data, and the power sampling data is encapsulated into data frames by the MODBUS protocol to obtain power parameter data.

[0040] (3) The voltage and current data are sampled and processed by the INA210 shunt monitor to obtain electrical sampling data, and the electrical sampling data is encapsulated into data frames by the MODBUS protocol to obtain voltage and current parameter data.

[0041] (4) The standardized coordinate data and structural feature data of the components are processed by spatial relationship modeling through a three-dimensional spatial mapping algorithm to obtain the spatial distribution data of the components. The spatial distribution data of the components are then processed by a regional division algorithm to obtain the regional grouping data of the components.

[0042] (5) The power parameter data and voltage and current parameter data are acquired and processed through the POWERBUS bus to obtain the component operating status data, and the component operating status data is processed through the STM32F334 digital power supply to obtain the synchronous acquisition data.

[0043] (6) The component area grouping data and synchronously collected data are mapped through a data association algorithm to obtain the digital twin model data package of the photovoltaic module.

[0044] Specifically, the location data of the photovoltaic modules is processed using UTM (Universal Transverse Mercator) projection coordinate transformation to convert the geographic coordinates of the modules from a geographic coordinate system (such as latitude and longitude) to a planar coordinate system for subsequent data processing and spatial analysis. This process involves converting the original geographic coordinates (lat, lon) to planar coordinates (x, y), where the transformation uses the mathematical formula of UTM projection to ensure high-precision geographic positioning. After obtaining the standardized module coordinates, the tilt angle and orientation parameters in the structural data are also geometrically standardized. This step is to unify the structural data of all photovoltaic modules, enabling them to be compared and analyzed under the same standard. Tilt angle and orientation parameters are usually expressed in angles. Through a three-dimensional geometric transformation algorithm, the tilt angle and orientation of each module are converted into geometric parameters in a standard coordinate system, obtaining the structural feature data of the modules. Next, power data acquisition and processing are performed through a high-precision ADC sampling circuit. The ADC (Analog-to-Digital Converter) can convert analog signals into digital signals, accurately acquiring the power output data of each photovoltaic module. Power data is encapsulated into data frames using the MODBUS protocol, a commonly used industrial communication protocol for data acquisition from remote devices. During encapsulation, the power data is divided into different data frames and packaged according to the MODBUS protocol format. This ensures the integrity and consistency of the data during network transmission, thus yielding the power parameter data.

[0045] Voltage and current data are sampled using the INA210 shunt monitor. The INA210 is an integrated circuit for current and voltage monitoring. It converts current signals into corresponding voltage signals, which are then processed by an ADC sampling circuit to obtain electrical sampling data. The voltage and current data are also encapsulated using the MODBUS protocol to generate voltage and current parameter data. This data provides crucial information for subsequent operational status assessments and component health assessments. All this data—including standardized coordinate data, structural feature data, power data, and electrical parameter data—needs to be processed by a spatial relationship modeling algorithm to obtain the spatial distribution data of the components. Based on three-dimensional coordinates and geometric features, the spatial relationship modeling algorithm can accurately locate the position of each component in three-dimensional space and model the spatial layout of each photovoltaic module according to its tilt angle, orientation, and other parameters. This process is achieved through three-dimensional coordinate transformation algorithms and matrix calculations. Subsequently, the component spatial distribution data is clustered using a region partitioning algorithm to obtain regional grouping data for the components. The region partitioning algorithm divides all photovoltaic modules into multiple sub-regions according to certain rules. These regions can be divided based on the similarity of photovoltaic modules, or based on the functional characteristics and failure probability of the regions. Cluster analysis can identify high-density areas or areas that may have potential faults, thereby improving the accuracy of inspections.

[0046] After obtaining the spatial distribution data and regional grouping data of the components, power data and electrical parameter data are acquired and processed in a timely manner via the POWERBUS bus. POWERBUS is a high-efficiency bus technology that enables synchronous data acquisition from multiple devices and provides high-bandwidth communication capabilities. Through the POWERBUS bus, the operating status data of the components (including power output, voltage, current, etc.) can be acquired synchronously, ensuring data time consistency. Next, the STM32F334 digital power synchronization processing chip is used to process the synchronously acquired data. The STM32F334 chip has powerful processing capabilities, ensuring the time consistency of acquisition from multiple data sources (such as power, temperature, voltage, etc.), avoiding delays or asynchrony issues between data, thus providing accurate data support for subsequent analysis. Finally, the component regional grouping data and synchronously acquired data are mapped using a data association algorithm to obtain a digital twin model data package for the photovoltaic modules. The digital twin model is a virtual representation of the component, reflecting its spatial location, structural features, electrical parameters, etc. The data association algorithm, by fusing information from different data sources, can accurately match the actual data of each photovoltaic module with its digital twin model. Digital twin models not only reflect the physical location and structure of components, but can also be dynamically updated to reflect the working status and health status of components in real time.

[0047] For example, suppose that in a photovoltaic power station, a group of modules exhibits significant fluctuations in power, voltage, and current data. By processing location and structural data, the system identifies the spatial distribution characteristics of the area. Combined with synchronously acquired data, the system discovers that the modules in this area show a clear heating trend, with large voltage fluctuations and low power output. Using a region segmentation algorithm, the system marks the modules in this area as high-risk areas. Then, through a data association algorithm, it maps the monitoring data of this area to a digital twin model, generating a detailed diagnostic data package. This data package contains various electrical, temperature, and power data for the area, which is compared with the digital twin model to ultimately generate a comprehensive health assessment report for subsequent inspection operations.

[0048] In one specific embodiment, the process of performing step S102 may specifically include the following steps:

[0049] (1) The component layout features are divided into regions using a grid segmentation algorithm to obtain basic inspection area data, and the spacing features are analyzed using minimum spacing calculation to obtain flightable area data.

[0050] (2) Collision risk assessment is performed on obstacle features through three-dimensional grid modeling to obtain safe route constraint data, and key areas are marked on performance features through weight allocation algorithm to obtain area priority data;

[0051] (3) The flightable area data is processed by waypoint generation algorithm to plan path nodes to obtain candidate waypoint data, and the candidate waypoint data is processed by route optimization based on safe route constraint data to obtain basic route data.

[0052] (4) The basic route data is adjusted according to the regional priority data to obtain the target waypoint sequence data. The route parameter configuration is performed on the target waypoint sequence data through flight parameter calculation to obtain the inspection route data package.

[0053] Specifically, in the process of drone inspection of photovoltaic power plants, the first step is to process the component layout characteristics. A grid-based segmentation algorithm is used to divide the photovoltaic power plant area into several small grids, thus obtaining basic inspection area data. Grid segmentation involves setting a reasonable grid size and organizing and dividing the photovoltaic components within the area according to this grid, making subsequent inspection path planning clearer and more efficient. Next, the spacing characteristics are processed by calculating the minimum spacing to analyze the minimum distance between each photovoltaic component and further determine the passage space between components. This process ensures the safety and flyability of the drone inspection path, avoiding collisions or obstructions between the drone and photovoltaic components during inspection. Finally, the resulting flyable area data provides clear boundaries for inspection path planning. Then, obstacle characteristics are assessed using 3D grid modeling to evaluate potential collision risks. 3D grid modeling involves modeling the entire photovoltaic power plant area as a 3D grid, further identifying potential obstacles such as large equipment, supports, or other objects. By calculating the obstacle density within the grid and combining it with machine learning algorithms, collision risk assessment is performed, ultimately yielding safe flight path constraint data. At this point, the known obstacle information provides a safety guarantee for the planned flight path, ensuring that the drone does not enter dangerous areas. Simultaneously, performance characteristics are processed by weighting different areas using a weighted allocation algorithm to determine which areas require priority inspection. By analyzing the importance of different areas, such as the power generation efficiency, age, and wear and tear of photovoltaic modules, priorities can be reasonably assigned, ensuring that important areas receive more inspection attention. The obtained area priority data is crucial for subsequent inspection path planning.

[0054] Path node planning is performed using a waypoint generation algorithm. In this step, the algorithm generates multiple candidate waypoints based on the determined flyable area and safe flight path constraints. These waypoints are critical points for the UAV's flight and need to be planned within a specified safe range to avoid collisions during flight while ensuring coverage of all important areas. After the candidate waypoints are generated, they are further optimized using safe flight path constraints. The optimization goal is to ensure that the flight path avoids obstacles while effectively covering all areas that must be inspected. The optimization algorithm calculates the shortest path for each route and selects the optimal combination of waypoints to obtain the basic flight path data. After obtaining the basic flight path data, the next step is to adjust the inspection density based on regional priority data. This process ensures that the inspection frequency and path density are higher in the most important areas to guarantee more comprehensive and accurate detection of key areas. By adjusting the inspection density, inspection efficiency and coverage can be improved without increasing flight time. Finally, the target waypoint sequence is configured with flight parameter calculations to complete the setting of the entire inspection path and obtain the final inspection route data package.

[0055] For example, in a real photovoltaic power plant, a grid-based segmentation algorithm divides the entire power plant area into 50×50 meter grids, with multiple photovoltaic modules placed within each grid according to the actual layout. After spacing calculations, the minimum distance between modules in certain areas is determined to be 2 meters; these areas are marked as flyable zones. Subsequently, 3D grid modeling identifies several areas with high obstacle risks, such as supports or other equipment. Therefore, the safety flight path constraints for these areas are specifically marked to ensure that drones avoid these areas during flight. Finally, based on the importance and age of the modules, certain areas are assigned higher weights and prioritized in path planning, thus ensuring the coverage and inspection priority of critical areas.

[0056] In one specific embodiment, the process of executing step S103 may specifically include the following steps:

[0057] (1) The surface of the photovoltaic module is subjected to spectral imaging processing by a multispectral camera to obtain the spectral reflectance data of the module, and the spectral reflectance data of the module is subjected to feature separation processing by a band analysis algorithm to obtain multi-band feature data;

[0058] (2) The temperature distribution of the photovoltaic module surface is scanned by an infrared thermal imager to obtain module temperature field data, and the module temperature field data is analyzed by temperature gradient calculation to obtain temperature characteristic data.

[0059] (3) The electrical parameter data is time-synchronized through the IEEE1588 clock node to obtain the reference timing data, and the multi-band characteristic data and temperature characteristic data are time-stamped to obtain the synchronization detection data packet.

[0060] Specifically, a multispectral camera is used to perform spectral imaging on the surface of the photovoltaic module, obtaining spectral reflectance data. The multispectral camera can capture light reflection information at different wavelengths, reflecting the reflective characteristics of the photovoltaic module's surface material. Based on different wavelengths, this reflectance data can provide crucial information about the surface condition of the photovoltaic module, such as contamination, aging, and microcracks. The obtained spectral reflectance data is then processed by a band analysis algorithm to extract reflectance characteristics of different bands. The band analysis algorithm uses data from multiple spectral bands to identify the reflectance characteristics of different areas on the photovoltaic module surface and separates these features from the data of different bands. The resulting multi-band feature data provides a necessary foundation for subsequent fault diagnosis. Temperature distribution of the photovoltaic module is also a very important parameter during inspection. An infrared thermal imager is used to scan the temperature distribution of the photovoltaic module, obtaining temperature field data of the module surface. By detecting the temperature at different points on the module surface, the infrared thermal imager generates a temperature distribution map, reflecting whether the module has overheating, temperature differences, or other abnormalities. By calculating the temperature gradient from the temperature field data, the changes in the module surface temperature can be further analyzed, identifying potential areas of thermal anomaly. This analysis is highly effective for detecting poor contact, wiring problems, and localized damage on solar panels. Temperature gradient calculation, by modeling temperature changes in the temperature field data, can identify areas with excessive temperature differences, especially in parts susceptible to temperature stress during long-term operation. The resulting temperature characteristic data helps diagnose component failures or identify maintenance needs.

[0061] After acquiring the spectral reflectance and temperature characteristic data of the photovoltaic modules, the next step is to synchronize the electrical parameters. In a photovoltaic power plant, multiple sensors (such as voltage, current, temperature, and spectral reflectance sensors) simultaneously acquire different types of data, and the time synchronization of these data is crucial for comprehensive analysis. To ensure that all data can be accurately matched, time synchronization is required using the IEEE 1588 clock node. IEEE 1588 is a precise clock synchronization protocol that ensures clock consistency between distributed devices, thereby guaranteeing the synchronization of multiple data types at the same point in time. Through this protocol, reference time-series data can be obtained, laying the time foundation for subsequent data processing and analysis. After synchronization, the next step is to align the multi-band characteristic data and temperature characteristic data. Data alignment refers to matching multiple data sources (such as spectral reflectance and temperature data) from different sensors using timestamps, ensuring that various types of data collected at the same time can be analyzed within the same time frame. Timestamp marking is a key step, ensuring that measurement results from different data sources (such as spectral cameras and infrared thermal imagers) at the same moment can be correlated. This processing step effectively fuses detection results from different sensors to obtain a comprehensive synchronous detection data packet, which facilitates subsequent fault diagnosis and performance analysis.

[0062] For example, suppose a photovoltaic (PV) power plant uses a multispectral camera and an infrared thermal imager to simultaneously inspect the PV modules during a routine inspection. The multispectral camera captures surface reflectance data showing a significant drop in reflectivity in a certain wavelength band, suggesting potential contamination or other surface issues in that area. Simultaneously, the infrared thermal imager detects a high temperature in the same area, indicating possible localized overheating. After time-synchronizing this data, the spectral and temperature data are precisely matched using an IEEE 1588 clock. Timestamps help align these two types of data, ultimately generating a synchronized inspection data package that provides comprehensive information about the overall condition of the area. These data packages will be used to analyze whether cleaning, repair, or replacement of the modules is necessary, thereby improving the operational efficiency and reliability of the PV power plant.

[0063] In one specific embodiment, the process of executing step S104 may specifically include the following steps:

[0064] (1) The multi-band feature data in the synchronous detection data packet is processed by the spectral analysis algorithm to extract the defect features and obtain the spectral defect feature data. The spectral defect feature data is then processed by the thermal imaging analysis algorithm to identify the hot spot features and obtain the temperature anomaly feature data.

[0065] (2) Multidimensional data integration processing is performed on spectral defect feature data and temperature anomaly feature data through feature fusion algorithm to obtain component status feature data, and parameter correlation processing is performed on electrical parameters through electrical performance analysis algorithm to obtain performance impact factor data;

[0066] (3) The component status characteristic data and performance impact factor data are processed by the mapping association algorithm to model the feature relationship and obtain the component health assessment model.

[0067] Specifically, the obtained synchronous detection data package contains multi-band feature data and temperature anomaly feature data from multispectral cameras and infrared thermal imagers. For this data, the first step is to extract defect features using spectral analysis algorithms. Based on the reflectance changes in each band of the spectral reflectance data, spectral analysis algorithms can identify different types of defects on the module surface. For example, by analyzing the reflectance of a certain band, spectral anomalies caused by contamination or other physical damage on the module surface can be identified, thus obtaining spectral defect feature data. This data reflects potential problems on the photovoltaic module surface and provides a foundation for subsequent analysis. Based on the extraction of spectral defect feature data, the next task is to identify temperature anomaly features using thermal imaging analysis algorithms. After processing, the temperature data obtained by the infrared thermal imager can reveal potential hotspots on the photovoltaic module, which are usually caused by localized circuit problems or module aging. The thermal imaging analysis algorithm, based on the temperature distribution in the temperature field, calculates the temperature gradient to identify temperature anomaly areas. These temperature anomalies usually represent potential faults or performance degradation. Through this process, temperature anomaly feature data is obtained, which, together with the spectral defect feature data, helps to further assess the health status of the module.

[0068] Next, spectral defect feature data and temperature anomaly feature data are combined using a feature fusion algorithm for multi-dimensional data integration. The feature fusion algorithm integrates data from different sensors (such as spectral and temperature data) through data fusion technology to obtain a unified and comprehensive module status feature data. By fusing spectral and temperature information, the health status of photovoltaic modules can be identified more accurately. For example, if spectral data indicates a significant decrease in reflectivity on the module surface, and thermal imaging analysis also indicates a temperature anomaly in that area, then these pieces of information together indicate a possible serious electrical problem or damage in that area. Therefore, feature fusion can provide higher accuracy and reliability than relying on a single data source. Simultaneously, electrical parameters (such as voltage and current) must also be included in the analysis. By processing the correlations between electrical parameters through electrical performance analysis algorithms, a deeper understanding of the influencing factors of various electrical performance parameters can be obtained. For example, changes in current and voltage are closely related to module surface damage or temperature increases. By analyzing the interrelationships of different electrical parameters, electrical performance analysis algorithms can reveal which electrical factors are key to the performance degradation of photovoltaic modules. The correlation processing of electrical performance yields performance influencing factor data, which indicates the relationship between module health status and its electrical performance, providing a basis for comprehensive diagnosis.

[0069] Finally, a comprehensive component health assessment model is constructed by modeling the feature relationships between component status characteristic data and performance influencing factor data using a mapping-association algorithm. The mapping-association algorithm is a data modeling method that can model the relationships between multiple features (such as spectral defects, temperature anomalies, electrical parameters, etc.) to obtain a predictive health assessment model. This model can automatically predict the health status of the component based on the input data, thereby determining whether maintenance or replacement is necessary. This model enables a comprehensive assessment of the photovoltaic module's status and provides data support for maintenance decisions. For example, during an inspection, spectral data obtained from a multispectral camera showed a significant decrease in the reflectivity of a certain photovoltaic module in the mid-wave band (e.g., 800-1000nm), while infrared thermal imager scans also indicated a high temperature in this area. This suggests a possible localized electrical fault or overheating problem in this area. In the data integration stage, spectral and temperature data were integrated using a feature fusion algorithm to form the component's status characteristic data. After electrical parameter analysis, correlation analysis of current and voltage showed that the module's voltage was abnormally lower than normal. Finally, by combining these data through a mapping and association algorithm, a health assessment model for the component was obtained. This model predicts that the component may fail in the short term and recommends further maintenance or replacement.

[0070] In one specific embodiment, the process of executing step S105 may specifically include the following steps:

[0071] (1) The state characteristic data in the component health assessment model is compared with historical trends by time series analysis algorithm to obtain performance change trend data, and the performance change trend data is evaluated for abnormality by threshold judgment algorithm to obtain fault level data.

[0072] (2) The performance impact factor data is processed by the association rule mining algorithm to extract fault features, and the fault type feature data is processed by the fault mode recognition algorithm to obtain the component diagnostic data package.

[0073] Specifically, in drone inspections of photovoltaic power plants, accurately assessing the health status and fault types of photovoltaic modules is crucial for ensuring the long-term operation of the power plant. In-depth analysis of the state characteristic data in the module health assessment model can identify the health change trends and potential fault risks of the modules. In this process, the application of time-series analysis algorithms plays a vital role. Time-series analysis algorithms can compare the current performance of modules with historical data to identify performance change trends. This analysis not only reveals the health changes of photovoltaic modules within a specific time period but also provides a foundation for subsequent fault prediction. For example, suppose that through comparative analysis of historical and current data, it is found that the current output of a certain module has been continuously declining over the past few months. The time-series analysis algorithm calculates the performance change trend of the module by backtracking and comparing historical data, thus obtaining clear performance change trend data. Based on this data, a threshold judgment algorithm can be further applied to assess the degree of anomaly in the performance change trend. The threshold judgment algorithm assesses the degree of anomaly in performance changes based on preset performance index thresholds. If the performance of the module exceeds the preset normal fluctuation range, it can be determined that the module has a fault risk, and the calculated fault level data provides a quantitative result of the current health status of the module. This data provides a reliable basis for determining whether components require further maintenance or replacement. In further analysis of fault detection, the processing of performance impact factor data is also particularly important. Performance impact factor data can be processed for fault feature extraction using association rule mining algorithms. Association rule mining is a common data mining technique that can discover potential relationships between different performance indicators, helping to identify component fault characteristics. For example, if multiple parameters of a component, such as voltage, temperature, and current, change simultaneously, association rule mining can reveal the connections between these changes and extract possible fault modes. For instance, by mining association rules for voltage drops and abnormal temperatures, it can be found that these changes may be related to poor internal contact or material aging within the component. In this way, the obtained fault type feature data can indicate the possible fault types of the component, providing a basis for subsequent fault mode identification.

[0074] Next, the fault mode recognition (FMR) algorithm performs further diagnostic analysis on these fault type characteristic data. The FMR algorithm can identify specific fault modes by comparing historical fault data with current monitoring data and then analyze them. For example, if a component exhibits abnormal spectral reflectance across multiple wavelengths accompanied by a significant temperature increase, these symptoms can be compared with known fault modes using the FMR algorithm to determine if the component may have an internal short circuit or overheating. Through this diagnostic process, the resulting component diagnostic data package records the component's current health status and possible fault types in detail. These data packages not only contain the fault type but also suggested handling methods, such as whether certain components need to be replaced or repaired. Through these analytical steps, an accurate component health assessment system can be constructed. In practical applications, suppose a routine inspection shows that the spectral reflectance of a photovoltaic module has significantly decreased in the near-infrared band (1200nm to 1300nm), while temperature data also shows a continuous increase in temperature in this area. A time-series analysis algorithm compares this data with historical trends, showing that the module's performance has been continuously declining since the last maintenance. Next, the threshold judgment algorithm determined that the performance degradation exceeded the preset tolerance range, classifying it as a moderate fault. Further analysis using an association rule mining algorithm revealed a strong correlation between the temperature rise and abnormal spectral reflectance, indicating that the component may have material aging or poor electrical contact issues. Finally, the fault mode recognition algorithm, combined with a historical fault mode database, confirmed the component's fault type as "localized overheating caused by poor contact," and generated a data package containing this fault diagnosis information, suggesting further repair or replacement.

[0075] In one specific embodiment, the process of executing step S106 may specifically include the following steps:

[0076] (1) The fault level data in the component diagnostic data package is processed by the hierarchical weight allocation algorithm to evaluate the regional importance, and the regional weight coefficient data is processed by the inspection frequency calculation to configure the time density, and the inspection cycle data is obtained.

[0077] (2) The sampling requirements of the fault type feature data are calculated and processed by the detection requirement analysis algorithm to obtain the sampling parameter data. The sampling parameter data and the inspection cycle data are processed by the strategy generation algorithm to formulate the inspection plan and obtain the inspection strategy data package.

[0078] Specifically, fault level data is processed using a hierarchical weighted allocation algorithm to assess regional importance. The core of this algorithm is assigning different weights to photovoltaic modules in different regions based on their fault levels, reflecting the importance of each region in the overall system operation. Specifically, individual modules in a photovoltaic power plant may have different levels of criticality due to factors such as installation location and sunlight intensity. Some modules may have a significant impact on the power plant's output efficiency, therefore their fault levels need to be assigned higher weights so that these areas can be prioritized during inspections. For example, if modules in a certain region frequently experience overheating or abnormal temperature issues, and these modules account for a high proportion of the power plant's power generation, then that region's weight coefficient will be assessed as high. After processing by this algorithm, the resulting regional weight coefficient data not only reflects the fault risk of each region but also provides a basis for prioritizing inspections. In actual inspection processes, the calculation of inspection frequency is particularly important. The calculation of inspection frequency is usually optimized based on the relationship between fault level and regional importance, ensuring that high-risk areas receive more frequent inspections, while low-risk areas can have their inspection cycles appropriately extended. The time density configuration process generates inspection cycle data by calculating the relationship between regional weighting coefficients and inspection frequency. For example, if a module in a certain area is assessed as a high-risk area due to abnormal temperature, the inspection frequency for that area will be appropriately increased, possibly set to once a month, while other low-risk areas may be set to once a quarter. This ensures the rational allocation of power plant resources and maximizes the operational safety of the photovoltaic power plant.

[0079] Next, sampling requirements are calculated using a detection demand analysis algorithm based on the fault type characteristic data. The goal of this algorithm is to calculate the specific parameters that need to be sampled based on different fault types and their impact on the overall performance of the power plant. For example, if a component exhibits an abnormal phenomenon of voltage drop accompanied by temperature increase, real-time monitoring of the current, voltage, and temperature in the surrounding environment of that component needs to be increased. By analyzing historical fault data, the algorithm can identify which sampling parameters are crucial for fault diagnosis, thereby determining whether to increase or decrease the sampling of monitoring data. For example, when the fault type is an electrical short circuit, the algorithm will prioritize collecting voltage and current data to more quickly locate the fault source. The resulting sampling parameter data provides accurate monitoring points and parameters for subsequent inspections. Finally, the formulation of the inspection plan relies on the combination of sampling parameter data and inspection cycle data. A strategy generation algorithm combines these two data sets to formulate a specific inspection strategy data package. This algorithm automatically generates the optimal inspection plan, taking into account the fault characteristics, sampling requirements, and inspection frequency of different areas. These inspection plans not only ensure high-frequency inspections of critical areas but also allocate time and resources rationally to ensure timely monitoring in all areas. For example, for a photovoltaic module exhibiting a prolonged voltage drop, the inspection plan would include increasing the frequency of voltage and temperature monitoring and meticulously recording the module's status. Meanwhile, for other areas without obvious anomalies, the inspection plan might only require periodic visual inspections and basic data collection.

[0080] Taking real-world data as an example, suppose a certain area of ​​a photovoltaic power station has had its module failure level assessed as moderate over the past three months, with a regional weighting coefficient of 0.75. Based on the calculated inspection frequency, the inspection cycle is once a month. The main failure type in this area is performance degradation caused by overheating; therefore, the monitoring frequency of temperature and current is increased through a detection demand analysis algorithm. The inspection strategy data package explicitly specifies that this area should be inspected monthly, and requires special attention to module temperature changes and current data during each inspection. In contrast, other areas without obvious failures may have an inspection cycle of once a quarter, with inspection content limited to basic visual inspection and routine electrical performance testing.

[0081] The above describes the drone inspection method for photovoltaic power plants in the embodiments of this application. The following describes the drone inspection device for photovoltaic power plants in the embodiments of this application. Please refer to [link / reference]. Figure 2 One embodiment of the drone inspection device for photovoltaic power plants in this application includes:

[0082] The acquisition module 201 is used to synchronously acquire and process the location data, structural data, power data, voltage data, and current data of each photovoltaic module in the photovoltaic power station via the POWERBUS bus to obtain a digital twin model data package of the photovoltaic module.

[0083] The calculation module 202 is used to perform regional optimization calculations on the component layout features, spacing features, obstacle features, and performance features in the digital twin model data packet to obtain the inspection route data packet;

[0084] The association module 203 is used to perform multimodal imaging scanning on photovoltaic modules and to perform time correlation processing on the scanning data and electrical parameter data through the IEEE1588 clock synchronization protocol to obtain a synchronous detection data packet;

[0085] The mapping module 204 is used to perform feature recognition processing on the synchronous detection data packet through a multi-dimensional feature extraction algorithm, and to perform mapping and association processing with electrical parameters to obtain a component health assessment model;

[0086] The identification module 205 is used to perform abnormal feature identification processing on the component health assessment model through a historical data comparison algorithm to obtain a component diagnostic data package;

[0087] The processing module 206 is used to perform inspection parameter adjustment processing on the component diagnostic data packet through regional weight calculation to obtain the inspection strategy data packet.

[0088] Through the collaborative efforts of the aforementioned components, and by conducting time-series analysis of the state characteristic data in the component health assessment model, the changing trends of component performance can be effectively identified, and potential failures can be predicted in advance, thus providing a scientific basis for subsequent inspections and maintenance. By allocating hierarchical weights to fault level data, not only can the fault risk in different areas be accurately assessed, but inspection resources can also be rationally allocated, prioritizing inspections of high-risk areas and effectively reducing maintenance time and power plant downtime after a failure. Compared to traditional periodic inspection methods, dynamic inspection strategies based on fault level and regional importance assessment can more effectively improve inspection coverage and efficiency, ensuring that critical areas receive more frequent and timely inspections. This data-driven intelligent inspection method significantly improves the accuracy and flexibility of inspections, avoiding the inefficient practice of applying the same inspection frequency to all areas. Furthermore, by combining performance impact factor data with inspection cycle data and configuring the inspection cycle based on regional weight coefficients, a more scientific and reasonable inspection plan can be achieved. Adjustments to the inspection cycle are not only based on historical component failure data but also consider the specific needs and importance of each region, thus avoiding unnecessary resource waste. For example, for areas showing signs of fault or exhibiting long-term anomalies, the system automatically adjusts the inspection frequency to ensure these areas receive more inspections and data collection, effectively reducing the risk of power plant failures. Furthermore, based on sampling requirements calculated from fault type characteristic data, the system can automatically adjust sampling parameters according to the characteristics of different fault types, ensuring sufficiently accurate and comprehensive data support during fault diagnosis, avoiding the omission of key data, and thus improving the accuracy and timeliness of fault diagnosis. Through the application of strategy generation algorithms, personalized inspection plans can be formulated based on sampling parameters and inspection cycle data, avoiding the limitations of fixed plans in traditional inspection methods and further enhancing the flexibility and adaptability of inspections. Depending on the fault type, the inspection plan can automatically adjust the inspection content and frequency, making the inspection plan not only more efficient but also capable of accurately identifying and repairing different types of faults.

[0089] The above-described 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 of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for unmanned aerial vehicle (UAV) inspection of a photovoltaic power station, characterized in that, The drone inspection method for the photovoltaic power station includes: The location, structure, power, voltage, and current data of each photovoltaic module in the photovoltaic power station are synchronously acquired and processed using the POWERBUS bus to obtain a digital twin model data package of the photovoltaic module. The component layout features, spacing features, obstacle features, and performance features in the digital twin model data package are processed by regional optimization calculation to obtain the inspection route data package; Multimodal imaging scanning of photovoltaic modules is performed, and the scanning data and electrical parameter data are time-correlated processed using the IEEE 1588 clock synchronization protocol to obtain a synchronous detection data packet; The synchronous detection data packets are processed for feature recognition using a multi-dimensional feature extraction algorithm, and then mapped and associated with electrical parameters to obtain a component health assessment model. The component health assessment model is processed using a historical data comparison algorithm to identify abnormal features, resulting in a component diagnostic data package; The inspection strategy data packet is obtained by adjusting the inspection parameters of the component diagnostic data packet through regional weight calculation.

2. The method for unmanned aerial vehicle (UAV) inspection of a photovoltaic power station according to claim 1, characterized in that, The process involves synchronously acquiring and processing the location, structural, power, voltage, and current data of each photovoltaic module in the photovoltaic power station via the POWERBUS bus to obtain a digital twin model data package for the photovoltaic modules, including: The location data is processed by UTM projection coordinate transformation to obtain component standardized coordinate data, and the tilt angle parameter and orientation parameter in the structural data are processed by geometric parameter standardization to obtain structural feature data; The power data is acquired and processed by a high-precision ADC sampling circuit to obtain power sampling data, and the power sampling data is encapsulated into data frames using the MODBUS protocol to obtain power parameter data. The voltage and current data are sampled and processed by the INA210 shunt monitor to obtain electrical sampling data. The electrical sampling data is then encapsulated into data frames using the MODBUS protocol to obtain voltage and current parameter data. The standardized coordinate data and structural feature data of the components are processed by a three-dimensional spatial mapping algorithm to model spatial relationships, thereby obtaining component spatial distribution data. The component spatial distribution data is then processed by a region partitioning algorithm to obtain component region grouping data. The power parameter data and voltage and current parameter data are acquired and processed in a timing manner through the POWERBUS bus to obtain component operating status data. The component operating status data is then processed synchronously through the STM32F334 digital power supply to obtain synchronous acquisition data. The component area grouping data and synchronously collected data are mapped using a data association algorithm to obtain the digital twin model data package of the photovoltaic module.

3. The method for unmanned aerial vehicle (UAV) inspection of a photovoltaic power station according to claim 1, characterized in that, The process of performing regional optimization calculations on the component layout features, spacing features, obstacle features, and performance features in the digital twin model data package yields the inspection route data package, including: The component layout features are divided into regions using a grid segmentation algorithm to obtain basic inspection area data, and the spacing features are analyzed for passage space using minimum spacing calculation to obtain flightable area data. The obstacle features are modeled using a 3D grid to perform collision risk assessment, resulting in safe flight path constraint data. The performance features are then marked with key areas using a weight allocation algorithm to obtain area priority data. The flyable area data is processed by a waypoint generation algorithm to plan path nodes and obtain candidate waypoint data. The candidate waypoint data is then optimized according to the safe route constraint data to obtain basic route data. The basic route data is adjusted for inspection density based on the regional priority data to obtain target waypoint sequence data. The target waypoint sequence data is then processed for route parameter configuration through flight parameter calculation to obtain the inspection route data package.

4. The method for unmanned aerial vehicle (UAV) inspection of a photovoltaic power station according to claim 1, characterized in that, The process involves performing multimodal imaging scanning on the photovoltaic module and correlating the scan data with electrical parameter data using the IEEE 1588 clock synchronization protocol to obtain a synchronization detection data packet, including: The surface of the photovoltaic module is subjected to spectral imaging processing using a multispectral camera to obtain the module's spectral reflectance data. The spectral reflectance data is then processed by a band analysis algorithm to separate features, resulting in multi-band feature data. The surface of the photovoltaic module is scanned for temperature distribution using an infrared thermal imager to obtain module temperature field data. The module temperature field data is then analyzed for thermal anomalies by calculating temperature gradients to obtain temperature characteristic data. The electrical parameter data is time-synchronized using the IEEE 1588 clock node to obtain reference timing data. The multi-band characteristic data and temperature characteristic data are then time-aligned using timestamps to obtain a synchronization detection data packet.

5. The unmanned aerial vehicle (UAV) inspection method for photovoltaic power plants according to claim 1, characterized in that, The synchronous detection data packet is processed for feature recognition using a multi-dimensional feature extraction algorithm, and then mapped and correlated with electrical parameters to obtain a component health assessment model, including: The multi-band feature data in the synchronous detection data packet is processed by a spectral analysis algorithm to extract defect features, thereby obtaining spectral defect feature data. The spectral defect feature data is then processed by a thermal imaging analysis algorithm to identify hotspot features, thereby obtaining temperature anomaly feature data. The spectral defect feature data and temperature anomaly feature data are processed by a feature fusion algorithm to integrate multidimensional data and obtain component state feature data. The electrical parameters are processed by an electrical performance analysis algorithm to obtain performance influence factor data. The component status characteristic data and performance impact factor data are processed by a mapping association algorithm to model the feature relationship, thereby obtaining the component health assessment model.

6. The method for unmanned aerial vehicle (UAV) inspection of a photovoltaic power station according to claim 5, characterized in that, The component health assessment model is processed using a historical data comparison algorithm to identify abnormal features, resulting in a component diagnostic data package, including: The state characteristic data in the component health assessment model is processed by historical trend comparison using a time series analysis algorithm to obtain performance change trend data. The performance change trend data is then processed by an anomaly assessment algorithm using a threshold judgment algorithm to obtain fault level data. The performance impact factor data is processed by an association rule mining algorithm to extract fault features, resulting in fault type feature data. The fault type feature data is then processed by a fault mode recognition algorithm to obtain a component diagnostic data package.

7. The method for unmanned aerial vehicle (UAV) inspection of a photovoltaic power station according to claim 6, characterized in that, The inspection strategy data packet is obtained by adjusting the inspection parameters of the component diagnostic data packet through regional weight calculation, including: The fault level data in the component diagnostic data package is processed by a hierarchical weight allocation algorithm to evaluate the regional importance, and the regional weight coefficient data is then processed by the inspection frequency calculation to configure the time density, and the inspection cycle data is obtained. The fault type feature data is processed by a detection requirement analysis algorithm to calculate sampling requirements and obtain sampling parameter data. The sampling parameter data and inspection cycle data are then processed by a strategy generation algorithm to formulate an inspection plan and obtain the inspection strategy data package.

8. A drone inspection device for a photovoltaic power station, used to implement the drone inspection method for a photovoltaic power station as described in any one of claims 1-7, characterized in that, The unmanned aerial vehicle (UAV) inspection device for the photovoltaic power station includes: The data acquisition module is used to synchronously acquire and process the location data, structural data, power data, voltage data, and current data of each photovoltaic module in the photovoltaic power station via the POWERBUS bus, and obtain a digital twin model data package of the photovoltaic module. The calculation module is used to perform regional optimization calculations on the component layout features, spacing features, obstacle features, and performance features in the digital twin model data package to obtain the inspection route data package; The correlation module is used to perform multimodal imaging scanning of photovoltaic modules and to perform time correlation processing on the scanning data and electrical parameter data through the IEEE1588 clock synchronization protocol to obtain a synchronous detection data packet; The mapping module is used to perform feature recognition processing on the synchronous detection data packet using a multi-dimensional feature extraction algorithm, and to perform mapping and association processing with electrical parameters to obtain a component health assessment model; The identification module is used to perform abnormal feature identification processing on the component health assessment model through historical data comparison algorithm to obtain component diagnostic data package; The processing module is used to adjust the inspection parameters of the component diagnostic data packet by calculating the regional weight, so as to obtain the inspection strategy data packet.

Images