On-line inspection method and inspection system for photovoltaic module

Abstract

The invention provides an online inspection method and system for a photovoltaic module, and the method comprises the steps: generating an optimal inspection path through a path planning algorithm P = f (L, A) based on a photovoltaic array topological structure; the unmanned aerial vehicle carries a multispectral camera, an infrared thermal imager and a current sensor to fly, and a visible light image, an infrared thermal image and electrical parameters are synchronously obtained; an improved YOLOv5 algorithm is adopted to analyze the collected data in real time to identify a fault component; performing correlation analysis F (T (x, y), I) on the temperature field distribution characteristic T (x, y) and the current characteristic I to distinguish fault types; and generating a fault map containing GPS coordinates and a maintenance suggestion according to an analysis result, and relates to the technical field of photovoltaic power station operation and maintenance, the unmanned aerial vehicle automatically performs routing inspection according to a planned path, the routing inspection time is greatly shortened, the routing inspection efficiency is improved, and the application of multi-source data acquisition and a deep learning algorithm can more accurately detect and classify faults, so that the fault detection efficiency is improved. The false detection rate is reduced, and the fault detection accuracy is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power station operation and maintenance, and particularly to an online inspection method and inspection system for photovoltaic modules. Background Art

[0002] Disadvantages of manual inspection: At present, the manual inspection method of photovoltaic power stations is inefficient, consuming a large amount of human and time costs. Staff need to check each photovoltaic module one by one. Facing a large area of photovoltaic power stations, this process is extremely time-consuming and laborious. Moreover, there are safety hazards in manual inspection, and staff may face risks such as falling from heights and electric shock.

[0003] Deficiencies of traditional drone inspection: Although traditional drone inspection improves the inspection efficiency to a certain extent, it lacks intelligent analysis capabilities. The data collected by drones need to be processed manually later, and problems cannot be discovered and processed in real time during the inspection process, resulting in delays in fault discovery and repair.

[0004] Defects of infrared detection: The existing infrared detection technology is easily affected by the ambient temperature. Under different ambient temperatures, the infrared characteristics of photovoltaic modules are unstable, resulting in a high false detection rate. At the same time, the existing technology lacks real-time fault diagnosis and location functions, making it difficult to quickly and accurately determine the location and type of faulty components, which is not conducive to timely maintenance and affects the power generation efficiency and economic benefits of photovoltaic power stations. Summary of the Invention

[0005] In order to overcome the existing problems, the embodiments of the present application provide an online inspection method and inspection system for photovoltaic modules. The drone automatically conducts inspections according to the planned path, greatly shortening the inspection time and improving the inspection efficiency. Moreover, the application of multi-source data collection and deep learning algorithms can more accurately detect and classify faults, reduce the false detection rate, and enhance the accuracy of fault detection.

[0006] The technical solution adopted by the embodiments of the present application to solve its technical problems is:

[0007] An online inspection method and inspection system for photovoltaic modules, the online inspection method for photovoltaic modules includes:

[0008] Based on the photovoltaic array topology structure, use the path planning algorithm P = f(L, A) to generate the optimal inspection path, where P is the inspection path, L is the photovoltaic array layout parameter, and A is the component arrangement parameter. In the path planning algorithm P = f(L, A), through the analysis of L and A, a variant of Dijkstra's algorithm is used to achieve the shortest path planning, ensuring that the drone completes the inspection in the shortest time and the most reasonable trajectory;

[0009] Among them, the steps of UAV path planning: Based on the topological structure of the photovoltaic array, use professional path planning algorithms to generate the optimal inspection path. By analyzing the layout of the photovoltaic array, the arrangement of components, etc., ensure that the UAV can complete the inspection of all photovoltaic components with the shortest time and the most reasonable flight trajectory, avoiding repeated inspections and omissions;

[0010] The UAV is equipped with a multi-spectral camera, an infrared thermal imager, and a current sensor to fly, and synchronously obtain visible light images, infrared thermal images, and electrical parameters. The image information S(γ) of different spectral bands obtained by the multi-spectral camera, combined with the temperature distribution T(x, y) detected by the infrared thermal imager and the electrical parameter I measured by the current sensor, jointly serve as data support for fault analysis, where γ is the spectral wavelength;

[0011] Among them, the multi-source data acquisition step: The UAV is equipped with devices such as a multi-spectral camera, an infrared thermal imager, and a current sensor, and synchronously obtains visible light images, infrared thermal images, and electrical parameters during flight. The multi-spectral camera can obtain image information of different spectral bands. The infrared thermal imager is used to detect the temperature distribution of components, and the current sensor is used to measure the electrical parameters of components, providing comprehensive data support for subsequent fault analysis;

[0012] An improved YOLOv5 algorithm is used to analyze and identify faulty components in the collected data in real time;

[0013] Among them, the real-time analysis step: An improved YOLOv5 algorithm is used to perform real-time analysis on the collected data to achieve rapid identification of faulty components. Through learning and training on a large number of fault samples, the improved algorithm can accurately identify various fault characteristics from images and data, such as hot spots, hidden cracks, power attenuation, etc.;

[0014] The fault type is distinguished by performing correlation analysis F(T(x, y), I) on the temperature field distribution feature T(x, y) and the current characteristic I, where F is the fault type judgment function;

[0015] Among them, the fault classification step: The fault type is distinguished by performing correlation analysis on the temperature field distribution feature and the current characteristic. Different faults show different characteristics in the temperature field and current characteristic. For example, hot spot faults are usually accompanied by local abnormal temperature increases, while power attenuation faults may be manifested as a decrease in the current value. By establishing a fault feature library and an analysis model, the fault type can be accurately judged;

[0016] Generate a fault map containing GPS coordinates and maintenance suggestions according to the analysis results;

[0017] Among them, the result output step: Generate a fault map containing GPS coordinates and maintenance suggestions according to the analysis results. The fault map intuitively shows the location of the faulty components, and the maintenance suggestions provide targeted maintenance guidance for the staff, facilitating quick repairs.

[0018] Preferably, the improved YOLOv5 algorithm constructs a fault feature recognition model M, fault feature F through learning and training on a large number of fault samples t The recognition formula is:

[0019] F t = M(S(γ), T(x, yy), I)

[0020] To accurately identify faults such as hot spots, hidden cracks, and power attenuation.

[0021] Preferably, in the fault classification step, a fault feature library is established. For hot spot faults, the condition of abnormal temperature rise T(x, y) > T th , where T th is the hot spot temperature threshold; for power attenuation faults, the condition of current value drop I < I th , where I th is the normal current threshold.

[0022] Preferably, the online inspection system for photovoltaic modules includes:

[0023] The UAV subsystem module, equipped with a multi-spectral camera, an infrared thermal imager, and a current sensor, flies along the planned path to collect data. The flight stability of the UAV in the UAV subsystem module is ensured by an attitude control algorithm , where θ, δ are the pitch angle, roll angle, and yaw angle of the UAV respectively;

[0024] Among them, the UAV subsystem module: Equipped with devices such as a multi-spectral camera, an infrared thermal imager, and a current sensor, it is responsible for flying over the photovoltaic power station along the planned path and collecting data. The UAV has stable flight performance and high-precision positioning ability to ensure the accuracy and comprehensiveness of data collection;

[0025] The ground station subsystem module includes a path planning module that generates the optimal inspection path according to the photovoltaic array topology using the formula P = f(L, A) and transmits it to the UAV, and a data fusion module that receives and preprocesses and fuses the data collected by the UAV;

[0026] Among them, the ground station subsystem module: includes a path planning module and a data fusion module. The path planning module generates an optimal inspection path according to the topological structure of the photovoltaic array and transmits the path information to the drone. The data fusion module receives various data collected by the drone, preprocesses and fuses them, and provides a unified data format for subsequent analysis;

[0027] The cloud analysis platform module deploys a fault diagnosis model to analyze and judge the fault type and location based on the fusion data by using a deep learning algorithm, and the knowledge base system stores fault cases and maintenance knowledge;

[0028] Among them, the cloud analysis platform module: deploys a fault diagnosis model and a knowledge base system. The fault diagnosis model analyzes the fused data based on a deep learning algorithm to judge the fault type and location. The knowledge base system stores a large number of photovoltaic component fault cases and maintenance knowledge, providing reference and support for fault diagnosis;

[0029] The visualization terminal module displays a three-dimensional fault distribution map to show the position and type information of the fault components. The three-dimensional fault distribution map of the visualization terminal module converts the world coordinates [X s , Y s , Z s of the fault components into screen coordinates [X w , Y w , Z w through the three-dimensional coordinate transformation formula [X w , Y w , Z w = T[X s , Y s , Z s for display, where T is the coordinate transformation matrix;

[0030] Among them, the visualization terminal module: displays a three-dimensional fault distribution map, and shows the position, type and other information of the fault components to the user in an intuitive three-dimensional graphic way, facilitating the user to quickly understand the fault situation of the photovoltaic power station.

[0031] Preferably, the data fusion module of the ground station subsystem module fuses multi-source data by using a weighted fusion algorithm, and the formula is as follows:

[0032] D = w1S(γ) + w2T(x, y) + w3I

[0033] where D is the fused data, and w1, w2, and w3 are the weights of different data respectively.

[0034] Preferably, the fault diagnosis model of the cloud analysis platform module is based on the convolutional neural network CNN structure and is trained and optimized through a loss function. The loss function formula is as follows:

[0035]

[0036] where y i is the true value, and y ii is the predicted value, and n is the number of samples.

[0037] The advantages of the embodiments of the present application are as follows:

[0038] 1. The drone automatically conducts inspections according to the planned path, greatly shortening the inspection time, improving the inspection efficiency. Moreover, the application of multi-source data collection and deep learning algorithms can more accurately detect and classify faults, reduce the misdetection rate, and enhance the accuracy of fault detection.

[0039] 2. The real-time analysis and generation of the fault map enable the staff to quickly locate the faulty components, perform repairs in a timely manner, reduce the power generation loss of the photovoltaic power station, and at the same time reduce the workload and cost of manual inspections, improving the intelligent level of the operation and maintenance of the photovoltaic power station. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is a schematic flowchart of the steps of the online inspection method for the photovoltaic components of the present invention;

[0041] Figure 2 is a schematic diagram of the framework of the online inspection system for the photovoltaic components of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts belong to the scope of protection of the present invention. In addition, for the convenience of description below, the "upper", "lower", "left", "right", etc. cited are consistent with the upper, lower, left, right, etc. of the accompanying drawings themselves. The "first", "second", etc. in the following text are for distinction in description and have no other special meanings.

[0043] The embodiments of the present application provide an online inspection method and inspection system for photovoltaic components to solve the problems in the prior art. The drone automatically conducts inspections according to the planned path, greatly shortening the inspection time, improving the inspection efficiency. Moreover, the application of multi-source data collection and deep learning algorithms can more accurately detect and classify faults, reduce the misdetection rate, and enhance the accuracy of fault detection. The real-time analysis and generation of the fault map enable the staff to quickly locate the faulty components, perform repairs in a timely manner, reduce the power generation loss of the photovoltaic power station, and at the same time reduce the workload and cost of manual inspections, improving the intelligent level of the operation and maintenance of the photovoltaic power station.

[0044] The technical solutions in the embodiments of this application are to solve the above problems, and the general idea is as follows:

[0045] Embodiment 1

[0046] This embodiment provides an on-line inspection method and inspection system for photovoltaic modules, as Figure 1 shown. The on-line inspection method for photovoltaic modules includes:

[0047] Based on the photovoltaic array topology structure, use the path planning algorithm P = f(L, A) to generate the optimal inspection path, where P is the inspection path, L is the photovoltaic array layout parameter, and A is the component arrangement parameter. In the path planning algorithm P = f(L, A), through the analysis of L and A, a variant of Dijkstra's algorithm is used to achieve the shortest path planning, ensuring that the drone completes the inspection with the shortest time and the most reasonable trajectory;

[0048] Among them, the drone path planning step: Based on the photovoltaic array topology structure, use a professional path planning algorithm to generate the optimal inspection path. By analyzing the layout of the photovoltaic array, the component arrangement method, etc., ensure that the drone can complete the inspection of all photovoltaic modules with the shortest time and the most reasonable flight trajectory, avoiding repeated inspections and omissions;

[0049] The drone is equipped with a multi-spectral camera, an infrared thermal imager, and a current sensor to fly, and synchronously obtains visible light images, infrared thermal images, and electrical parameters. The image information S(γ) of different spectral bands obtained by the multi-spectral camera, combined with the temperature distribution T(x, y) detected by the infrared thermal imager and the electrical parameter I measured by the current sensor, jointly serves as the data support for fault analysis, where γ is the spectral wavelength;

[0050] Among them, the multi-source data acquisition step: The drone is equipped with devices such as a multi-spectral camera, an infrared thermal imager, and a current sensor, and synchronously obtains visible light images, infrared thermal images, and electrical parameters during flight. The multi-spectral camera can obtain image information of different spectral bands. The infrared thermal imager is used to detect the temperature distribution of the components, and the current sensor is used to measure the electrical parameters of the components, providing comprehensive data support for subsequent fault analysis;

[0051] Use the improved YOLOv5 algorithm to analyze and identify faulty components in the collected data in real time;

[0052] Among them, the real-time analysis step: Use the improved YOLOv5 algorithm to perform real-time analysis on the collected data to achieve rapid identification of faulty components. Through the learning and training of a large number of fault samples, the improved algorithm can accurately identify various fault features from images and data, such as hot spots, hidden cracks, power attenuation, etc.;

[0053] The fault types are distinguished by correlating the temperature field distribution characteristics T(x, y) with the current characteristics I through the correlation analysis F(T(x, y), I), where F is the fault type judgment function;

[0054] Among them, the fault classification steps: By correlating the temperature field distribution characteristics with the current characteristics, the fault types are distinguished. Different faults show different characteristics in the temperature field and current characteristics. For example, the hot spot fault is usually accompanied by a local abnormal temperature rise, while the power attenuation fault may be manifested as a decrease in the current value. By establishing a fault feature library and an analysis model, the fault types can be accurately judged;

[0055] Generate a fault map including GPS coordinates and maintenance suggestions according to the analysis results;

[0056] Among them, the result output steps: Generate a fault map including GPS coordinates and maintenance suggestions according to the analysis results. The fault map intuitively shows the location of the faulty components, and the maintenance suggestions provide targeted maintenance guidance for the staff, facilitating quick maintenance.

[0057] The improved YOLOv5 algorithm constructs a fault feature recognition model M through the learning and training of a large number of fault samples, and the recognition formula for the fault feature F t is:

[0058] F t = M(S(γ), T(x, yy), I)

[0059] to accurately identify faults such as hot spots, hidden cracks, and power attenuation.

[0060] In the fault classification steps, a fault feature library is established. For the hot spot fault, the condition of abnormal temperature rise T(x, y) > T th is satisfied, where T th is the hot spot temperature threshold; for the power attenuation fault, the condition of current value decrease I < I th is satisfied, where I th is the normal current threshold.

[0061] The online inspection system for photovoltaic modules includes:

[0062] The unmanned aerial vehicle (UAV) subsystem module, which is equipped with a multi-spectral camera, an infrared thermal imager, and a current sensor, flies along the planned path to collect data. For the UAV in the UAV subsystem module, its flight stability is ensured by the attitude control algorithm where θ, δ are the pitch angle, roll angle, and yaw angle of the UAV respectively;

[0063] Among them, the UAV subsystem module: Equipped with devices such as multi-spectral cameras, infrared thermal imagers, and current sensors, it is responsible for flying over the photovoltaic power station along the planned path and collecting data. The UAV has stable flight performance and high-precision positioning capabilities to ensure the accuracy and comprehensiveness of data collection;

[0064] The ground station subsystem module includes a path planning module that generates the optimal inspection path according to the photovoltaic array topology using the formula P = f(L, A) and transmits it to the UAV, and a data fusion module that receives and preprocesses and fuses the data collected by the UAV;

[0065] Among them, the ground station subsystem module: Includes a path planning module and a data fusion module. The path planning module generates the optimal inspection path according to the photovoltaic array topology and transmits the path information to the UAV. The data fusion module receives various data collected by the UAV, performs preprocessing and fusion, and provides a unified data format for subsequent analysis;

[0066] The cloud analysis platform module deploys a fault diagnosis model to analyze and judge the fault type and location based on the fused data using deep learning algorithms, and a knowledge base system stores fault cases and maintenance knowledge;

[0067] Among them, the cloud analysis platform module: Deploys a fault diagnosis model and a knowledge base system. The fault diagnosis model analyzes the fused data based on deep learning algorithms to judge the fault type and location. The knowledge base system stores a large number of photovoltaic component fault cases and maintenance knowledge, providing reference and support for fault diagnosis;

[0068] The visualization terminal module displays a three-dimensional fault distribution map to show the location and type information of the faulty components. The three-dimensional fault distribution map of the visualization terminal module converts the world coordinates [X s , Y s , Z s of the faulty components into screen coordinates [X w , Y w , Z w through the three-dimensional coordinate transformation formula [X w , Y w , Z w = T[X s , Y s , Z s for display, where T is the coordinate transformation matrix;

[0069] Among them, the visualization terminal module: Displays a three-dimensional fault distribution map, showing the location, type, etc. information of the faulty components to the user in an intuitive three-dimensional graphic way, facilitating the user to quickly understand the fault situation of the photovoltaic power station.

[0070] The data fusion module of the ground station subsystem module uses a weighted fusion algorithm to fuse multi-source data. The formula is as follows:

[0071] D = w1S(γ) + w2T(x, y) + w3I

[0072] Where D is the fused data, and w1, w2, and w3 are the weights of different data respectively.

[0073] The fault diagnosis model of the cloud analysis platform module is based on the convolutional neural network CNN structure and is trained and optimized through a loss function. The loss function formula is as follows:

[0074]

[0075] Where y i is the true value, y ii is the predicted value, and n is the number of samples.

[0076] By adopting the above technical solutions:

[0077] Scenario setting: The photovoltaic power station is located on the roof of an industrial park, covering an area of 5,000 square meters. There are 5,000 photovoltaic modules in total, arranged in a row.

[0078] UAV path planning: The ground station subsystem module obtains the photovoltaic array layout data and uses a path planning program improved based on the A* algorithm to divide the photovoltaic power station into 20 rectangular sub-areas. Considering the safety distance at the roof edge and the turning radius of the UAV, an S-shaped flight path is planned, enabling the UAV to traverse each sub-area in turn. The total flight distance is shortened by 15% compared to the traditional full-coverage path.

[0079] Multi-source data acquisition: The UAV is equipped with a multi-spectral camera with a resolution of 48 million pixels, an infrared thermal imager with an accuracy of ±0.5°C, and a current sensor with a measurement accuracy of 0.1 A. At a flight altitude of 50 meters, visible light images are taken at a speed of 1 image per second, and infrared thermal images and current data are recorded every 5 seconds.

[0080] Real-time analysis: The collected data is transmitted to the ground station in real time for preliminary noise reduction and format conversion, and then uploaded to the cloud analysis platform. The platform uses the improved YOLOv5 algorithm to analyze the data at a speed of processing 10 frames of images per second and identify potential fault points.

[0081] Fault classification: By analyzing the temperature field distribution and current data, the cloud platform compares the suspected fault points with the fault feature library. If it is found that the temperature of a certain component is 5°C higher than the surrounding area and the current is 20% lower than the normal range, it is determined as a composite fault of hot spot and power attenuation.

[0082] Result output: The cloud platform generates a fault map, accurately marks the positions of faulty components, and gives maintenance suggestions such as disassembling and replacing damaged components and checking circuit connections, which are displayed to the operation and maintenance personnel through a visualization terminal.

[0083] Embodiment 2

[0084] This embodiment provides an online inspection method and inspection system for photovoltaic modules, as Figure 1 shown, the online inspection method for photovoltaic modules includes:

[0085] Based on the photovoltaic array topology, use the path planning algorithm P = f(L, A) to generate the optimal inspection path, where P is the inspection path, L is the photovoltaic array layout parameter, and A is the component arrangement parameter. In the path planning algorithm P = f(L, A), through the analysis of L and A, a variant of Dijkstra's algorithm is used to achieve the shortest path planning, ensuring that the drone completes the inspection with the shortest time and the most reasonable trajectory;

[0086] Among them, the drone path planning steps: Based on the photovoltaic array topology, use a professional path planning algorithm to generate the optimal inspection path. By analyzing the layout of the photovoltaic array, the component arrangement method, etc., ensure that the drone can complete the inspection of all photovoltaic modules with the shortest time and the most reasonable flight trajectory, avoiding repeated inspections and omissions;

[0087] The drone is equipped with a multispectral camera, an infrared thermal imager, and a current sensor to fly, and synchronously obtain visible light images, infrared thermal images, and electrical parameters. The image information S(γ) of different spectral bands obtained by the multispectral camera, combined with the temperature distribution T(x, y) detected by the infrared thermal imager and the electrical parameter I measured by the current sensor, jointly serve as the data support for fault analysis, where γ is the spectral wavelength;

[0088] Among them, the multi-source data acquisition steps: The drone is equipped with devices such as a multispectral camera, an infrared thermal imager, and a current sensor, and synchronously obtains visible light images, infrared thermal images, and electrical parameters during flight. The multispectral camera can obtain image information of different spectral bands. The infrared thermal imager is used to detect the temperature distribution of the components, and the current sensor is used to measure the electrical parameters of the components, providing comprehensive data support for subsequent fault analysis;

[0089] Use the improved YOLOv5 algorithm to analyze and identify faulty components in the collected data in real time;

[0090] Among them, the real-time analysis steps: Use the improved YOLOv5 algorithm to perform real-time analysis on the collected data to achieve rapid identification of faulty components. Through the learning and training of a large number of fault samples, the improved algorithm can accurately identify various fault features from images and data, such as hot spots, hidden cracks, power attenuation, etc.;

[0091] By performing a correlation analysis F(T(x, y), I) on the temperature field distribution characteristics T(x, y) and the current characteristics I to distinguish the fault types, where F is the fault type judgment function;

[0092] Among them, the fault classification steps: By performing a correlation analysis on the temperature field distribution characteristics and the current characteristics to distinguish the fault types. Different faults exhibit different characteristics in the temperature field and current characteristics. For example, a hot spot fault is usually accompanied by a local abnormal temperature rise, while a power attenuation fault may be manifested as a decrease in the current value. By establishing a fault feature library and an analysis model, the fault types can be accurately judged;

[0093] Generate a fault map containing GPS coordinates and maintenance suggestions according to the analysis results;

[0094] Among them, the result output steps: Generate a fault map containing GPS coordinates and maintenance suggestions according to the analysis results. The fault map intuitively shows the location of the faulty component, and the maintenance suggestions provide targeted maintenance guidance for the staff, facilitating quick maintenance.

[0095] The improved YOLOv5 algorithm constructs a fault feature recognition model M through learning and training on a large number of fault samples, and the recognition formula for the fault feature F t is:

[0096] F t = M(S(γ), T(x, yy), I)

[0097] to accurately identify faults such as hot spots, hidden cracks, and power attenuation.

[0098] In the fault classification steps, establish a fault feature library. For a hot spot fault, it satisfies the condition of abnormal temperature rise T(x, y) > T th , where T th is the hot spot temperature threshold; for a power attenuation fault, it satisfies the condition of current value decrease I < I th , where I th is the normal current threshold.

[0099] The on-line inspection system for photovoltaic modules includes:

[0100] The unmanned aerial vehicle (UAV) subsystem module, equipped with a multi-spectral camera, an infrared thermal imager, and a current sensor, flies along the planned path to collect data. For the UAV in the UAV subsystem module, its flight stability is ensured by an attitude control algorithm where θ, δ are the pitch angle, roll angle, and yaw angle of the UAV respectively;

[0101] Among them, the UAV subsystem module: Equipped with devices such as multispectral cameras, infrared thermal imagers, and current sensors, it is responsible for flying over the photovoltaic power station along the planned path and collecting data. The UAV has stable flight performance and high-precision positioning capabilities to ensure the accuracy and comprehensiveness of data collection;

[0102] The ground station subsystem module includes a path planning module that generates the optimal inspection path according to the photovoltaic array topology using the formula P = f(L, A) and transmits it to the UAV, and a data fusion module that receives and preprocesses and fuses the data collected by the UAV;

[0103] Among them, the ground station subsystem module: Includes a path planning module and a data fusion module. The path planning module generates the optimal inspection path according to the photovoltaic array topology and transmits the path information to the UAV. The data fusion module receives various data collected by the UAV and performs preprocessing and fusion to provide a unified data format for subsequent analysis;

[0104] The cloud analysis platform module deploys a fault diagnosis model to analyze and judge the fault type and location based on the fused data using deep learning algorithms, and a knowledge base system stores fault cases and maintenance knowledge;

[0105] Among them, the cloud analysis platform module: Deploys a fault diagnosis model and a knowledge base system. The fault diagnosis model analyzes the fused data based on deep learning algorithms to judge the fault type and location. The knowledge base system stores a large number of photovoltaic component fault cases and maintenance knowledge to provide reference and support for fault diagnosis;

[0106] The visualization terminal module displays a 3D fault distribution map to show the location and type information of the faulty components. The 3D fault distribution map of the visualization terminal module uses the 3D coordinate transformation formula [X s , Y s , Z s = T[X w , Y w , Z w to convert the world coordinates [X w , Y w , Z w of the faulty components into screen coordinates [X s , Y s , Z s for display, where T is the coordinate transformation matrix;

[0107] Among them, the visualization terminal module: Displays a 3D fault distribution map, presenting the location, type and other information of the faulty components to users in an intuitive 3D graphical way, facilitating users to quickly understand the fault situation of the photovoltaic power station.

[0108] The data fusion module of the ground station subsystem module fuses multi-source data using a weighted fusion algorithm. The formula is as follows:

[0109] D = w1S(γ) + w2T(x, y) + w3I

[0110] Where D is the fused data, and w1, w2, and w3 are the weights of different data respectively.

[0111] The fault diagnosis model of the cloud analysis platform module is based on the convolutional neural network CNN structure and is trained and optimized through a loss function. The loss function formula is as follows:

[0112]

[0113] Where y i is the true value, y ii is the predicted value, and n is the number of samples.

[0114] By adopting the above technical solutions:

[0115] Scenario setting: A large-scale photovoltaic power station is located in a desert area, covering an area of 5 square kilometers, with 1 million photovoltaic modules, arranged in a large-scale square array.

[0116] UAV path planning: Based on the photovoltaic array topology structure, the ground station adopts a partitioned spiral path planning strategy, divides the power station into 100 areas, plans an efficient inspection route, reduces the number of UAV flight turnbacks, and improves the inspection efficiency by 30%.

[0117] Multi-source data acquisition: Use a long-endurance UAV equipped with a high-resolution multi-spectral camera, a high-sensitivity infrared thermal imager, and a high-precision current sensor. At a flight altitude of 80 meters, collect visible light images at a frequency of 2 seconds per image, and record infrared and current data every 10 seconds to ensure comprehensive coverage and accurate data.

[0118] Real-time analysis: The collected data is simply processed by the ground station and then transmitted to the cloud platform. The improved YOLOv5 algorithm is run using an optimized GPU cluster to achieve 30 frames of image processing per second and quickly identify potential faults.

[0119] Fault classification: By establishing a complex fault feature library and a deep neural network classification model, the cloud platform analyzes temperature and electrical parameters. For example, if the components in a certain area show abnormal low temperature and large current fluctuations, it is judged as a mixed fault of occlusion and poor line contact in combination with the feature library.

[0120] Result output: Generate a 3D fault distribution map, visually display the fault location and severity, and provide targeted maintenance solutions, such as clearing occlusions and tightening line connection points, to facilitate rapid response by maintenance personnel.

[0121] Specific implementation method: UAV path planning:

[0123] First, obtain the detailed topological structure data of the photovoltaic array, including the number of rows and columns, spacing, arrangement method, etc. of the photovoltaic modules. Divide the photovoltaic array into multiple sub-regions. According to the position and shape of each sub-region, use classical path planning algorithms such as the A* algorithm or Dijkstra algorithm to generate the optimal path from the UAV take-off point to each sub-region.

[0124] Considering the flight performance and battery endurance of the UAV, optimize the generated path to ensure the continuity and feasibility of the path.

[0125] Multi-source data acquisition:

[0126] During the flight of the UAV, the multi-spectral camera takes visible light images at set time intervals or flight distances, covering the entire photovoltaic array.

[0127] The infrared thermal imager collects the infrared thermal images of the photovoltaic modules in real time and records the temperature distribution of the modules.

[0128] The current sensor is connected to the circuit of the photovoltaic module to measure the current parameters of the module in real time.

[0129] Real-time analysis:

[0130] Transmit the collected visible light images, infrared thermal images and electrical parameters to the ground station subsystem module for preprocessing, including image enhancement, denoising, data normalization, etc.

[0131] The preprocessed data is transmitted to the cloud analysis platform module for real-time analysis using the improved YOLOv5 algorithm. The improved algorithm adds convolutional layers and attention mechanisms for the fault characteristics of photovoltaic modules on the basis of the original YOLOv5 network structure, improving the fault recognition ability.

[0132] Fault classification:

[0133] Establish a fault feature library, including the temperature field distribution characteristics and current characteristics of different fault types.

[0134] According to the temperature field and current data obtained from real-time analysis, compare with the fault feature library, and use algorithms such as decision tree algorithm or support vector machine algorithm for fault classification.

[0135] Result output:

[0136] According to the fault classification result, combined with the GPS positioning information of the UAV, generate a fault map containing the GPS coordinates of the faulty module.

[0137] Obtain maintenance suggestions for different fault types from the knowledge base system, and...Figure 1 Output to the visualization terminal module for staff to view and use.

[0138] Finally, it should be noted that: Obviously, the above embodiments are merely examples given to clearly illustrate the present invention, rather than limitations on the implementation manners. For those of ordinary skill in the art, other different forms of changes or variations can be made based on the above description. It is not necessary and impossible to enumerate all implementation manners here. And the obvious changes or variations derived therefrom are still within the protection scope of the present invention.

Claims

1. An on-line inspection method for a photovoltaic module, characterized in that, The online inspection method for the photovoltaic module includes: Based on the photovoltaic array topology structure, use the path planning algorithm P = f(L, A) to generate the optimal inspection path, where P is the inspection path, L is the photovoltaic array layout parameter, and A is the component arrangement parameter; The unmanned aerial vehicle (UAV) is equipped with a multispectral camera, an infrared thermal imager, and a current sensor to fly and synchronously obtain visible light images, infrared thermal images, and electrical parameters; Adopt the improved YOLOv5 algorithm to analyze and identify the faulty components in the collected data in real time; Through the correlation analysis F(T(x, y), I) of the temperature field distribution feature T(x, y) and the current characteristic I to distinguish the fault types, where F is the fault type judgment function; Generate a fault map including GPS coordinates and maintenance suggestions according to the analysis results.

2. The on-line inspection method of a photovoltaic module according to claim 1, characterized in that, In the path planning algorithm P = f(L, A), through the analysis of L and A, use the variant of Dijkstra's algorithm to achieve the shortest path planning, ensuring that the UAV completes the inspection in the shortest time and the most reasonable trajectory.

3. The on-line inspection method of a photovoltaic module according to claim 1, characterized in that, The image information S(γ) of different spectral bands obtained by the multispectral camera, combined with the temperature distribution T(x, y) detected by the infrared thermal imager and the electrical parameter I measured by the current sensor, jointly serve as the data support for fault analysis, where γ is the spectral wavelength.

4. The on-line inspection method for a photovoltaic module according to claim 1, characterized in that, The improved YOLOv5 algorithm constructs a fault feature recognition model M and fault feature F through learning and training on a large number of fault samples t The recognition formula is as follows: F t = M(S(γ), T(x, yy), I) To accurately identify faults such as hot spots, hidden cracks, and power attenuation.

5. The on-line inspection method for a photovoltaic module according to claim 1, characterized in that In the fault classification step, a fault feature library is established. For the hot spot fault, the condition of abnormal temperature rise T(x, y)>T th is satisfied, where T th is the hot spot temperature threshold; for the power attenuation fault, the condition of current value drop I<I th is satisfied, where I th is the normal current threshold.

6. An on-line inspection system for a photovoltaic module, characterized in that, The online inspection system for the photovoltaic module includes: The UAV subsystem module, equipped with a multispectral camera, an infrared thermal imager, and a current sensor, flies according to the planned path to collect data; The ground station subsystem module includes a path planning module that generates the optimal inspection path according to the photovoltaic array topology structure using the formula P = f(L, A) and transmits it to the UAV, and a data fusion module that receives and preprocesses and fuses the data collected by the UAV; The cloud analysis platform module deploys a fault diagnosis model to analyze and judge the fault types and locations based on the deep learning algorithm for the fused data, and a knowledge base system stores fault cases and maintenance knowledge; The visualization terminal module displays a three-dimensional fault distribution map to show the location and type information of the faulty components.

7. The on-line inspection system for a photovoltaic module according to claim 6, characterized in that, The drone in the drone subsystem module ensures its flight stability through an attitude control algorithm where θ, δ are the pitch angle, roll angle, and yaw angle of the drone respectively.

8. The on-line inspection system for a photovoltaic module according to claim 6, characterized in that, The data fusion module of the ground station subsystem module uses a weighted fusion algorithm to fuse the multi-source data, and the formula is as follows: D = w1S(γ)+w2T(x, y)+w3I Where D is the fused data, and w1, w2, and w3 are the weights of different data respectively.

9. The on-line inspection system for a photovoltaic module according to claim 6, wherein, The fault diagnosis model of the cloud analysis platform module is based on the convolutional neural network (CNN) structure and is trained and optimized through a loss function. The loss function formula is as follows: where y i is the true value, and y ii is the predicted value, and n is the number of samples.

10. The on-line inspection system for a photovoltaic module according to claim 6, characterized in that, The three-dimensional fault distribution map of the visualization terminal module converts the world coordinates [X s , Y s , Z s of the faulty component into screen coordinates [X w , Y w , Z w for display through the three-dimensional coordinate transformation formula [X w , Y w , Z w = T[X s , Y s , Z s , where T is the coordinate transformation matrix.

Images