Photovoltaic station defect identification method, system and device based on target detection algorithm
Through the photovoltaic station defect recognition method based on the object detection algorithm, combined with area division and identification algorithm selection, the problem of insufficient accuracy rate and high cost of photovoltaic station defect recognition is solved, and efficient and accurate defect recognition and positioning is achieved.
Patent Information
- Application Number
- CN202410950729.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-16
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-07-16
AI Technical Summary
The existing photovoltaic station defect identification system fails to effectively consider the station scale and environmental factors, resulting in insufficient accuracy and high cost of defect identification.
Using a method based on the object detection algorithm, the area division and numbering is obtained by acquiring photovoltaic station information, the patrol route is designed, and the appropriate identification algorithm (YOLOv8 or improved YOLOv8) is selected according to the area type for defect identification, and combined with the drone patrol Inspection and meteorological data adjustment tasks in real time to ensure accuracy and efficiency.
The accuracy and efficiency of photovoltaic station defect identification are improved, the cost is reduced, the difficulty of defect identification is improved by distinguishing regional types and using different identification algorithms, and the inspection frequency and accuracy are adjusted through real-time meteorological data.
Smart Images

Figure CN118506220B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of intelligent detection technology, and specifically to a photovoltaic station defect identification method, system and device based on a target detection algorithm. Background Art
[0002] As the global demand for renewable energy grows, photovoltaic energy, as a clean and renewable energy form, is increasingly used. However, due to long-term exposure to the outdoor environment, the operating efficiency and safety of photovoltaic stations are affected by many factors. Therefore, rapid and accurate defect identification of photovoltaic panels in photovoltaic stations is of great significance to ensure the stable operation of photovoltaic stations and improve energy conversion efficiency.
[0003] Most existing defect recognition systems for photovoltaic stations focus on optimizing detection algorithms, so that they can accurately identify target objects in complex scenarios, obtain more accurate detection results, and reduce the efficiency of false detection and missed detection. However, most existing technologies do not take into account the specific environmental conditions of each photovoltaic station. Some small-scale photovoltaic stations can use basic detection models to accurately identify photovoltaic panel defects in photovoltaic stations, without the need to build accurate models to increase costs.
[0004] How to comprehensively consider the scene scale of photovoltaic stations, design a suitable target detection algorithm, and improve the accuracy of defect recognition while reducing costs is an issue that needs further research. Summary of the invention
[0005] In order to improve the accuracy and efficiency of photovoltaic station defect detection while reducing costs, the present application provides a photovoltaic station defect identification method, system and device based on a target detection algorithm.
[0006] In a first aspect, the present application provides a photovoltaic station defect identification method based on a target detection algorithm, comprising:
[0007] Acquire photovoltaic station information, divide and number the photovoltaic stations; design inspection routes for photovoltaic stations according to the regional division and numbering of photovoltaic stations, and determine the area type corresponding to each area according to the acquired photovoltaic station information; the area type includes: photovoltaic panel dense area or photovoltaic panel sparse area; the photovoltaic station information includes: photovoltaic station location information, layout information and quantity information of photovoltaic panels in the photovoltaic station;
[0008] Based on the inspection routes of the PV stations, create inspection tasks and dispatch drones to perform the assigned inspection tasks;
[0009] Obtain image data collected during the inspection mission of the drone, and determine the area to which the collected image data belongs and the area type corresponding to the area according to the inspection route, flight speed and flight time of the drone during the inspection mission;
[0010] Different types of recognition tasks are created according to the area type corresponding to the area determined by the collected image data; if the area type corresponding to the area determined by the collected image data is a photovoltaic panel dense area, a first type of recognition task is created accordingly; if the area type corresponding to the area determined by the collected image data is a photovoltaic panel sparse area, a second type of recognition task is created accordingly;
[0011] Different recognition algorithms are used to perform different types of recognition tasks to identify defective photovoltaic panels. If the current recognition task is the first type of recognition task, the YOLOv8 recognition algorithm is used to identify photovoltaic panel defects. If the current recognition task is the second type of recognition task, the improved YOLOv8 recognition algorithm is used to identify photovoltaic panel defects. A photovoltaic station fault report is generated based on the defective photovoltaic panels identified.
[0012] By adopting the above scheme, considering the areas with higher density of photovoltaic panels, the arrangement between photovoltaic panels may be closer. For example, in a dense photovoltaic panel array, the shadows or reflections of adjacent photovoltaic panels may interfere with defect identification, causing some defects to be obscured or difficult to identify, which increases the difficulty of defect detection. Therefore, from the perspective of regional photovoltaic panel density, regional type division is carried out, and the corresponding basic recognition algorithm or improved algorithm is selected to ensure that more accurate defect identification is achieved while reducing costs.
[0013] Preferably, it also includes:
[0014] After the photovoltaic station is divided and numbered, the photovoltaic panels in each divided area are numbered, each photovoltaic panel has a unique corresponding number, and each photovoltaic panel unique number corresponds to a unique location information;
[0015] The layout of the photovoltaic panels in the area determined based on the image data of the defective photovoltaic panel is matched with the image data of the identified defective photovoltaic panel to determine the photovoltaic panel number corresponding to the defective photovoltaic panel; based on the determined photovoltaic panel number, the location information of the defective photovoltaic panel is determined and displayed in the photovoltaic station fault report.
[0016] By adopting the above scheme, considering that the existing defect recognition results generally stay at the level of image photovoltaic panel defect display, and cannot timely and accurately locate the defective photovoltaic panels, by numbering the photovoltaic panels, using the existing photovoltaic panel layout and matching the acquired image data, the position information is determined in time according to the matching numbers, and the defective photovoltaic panels are located, which is convenient for subsequent manual verification and maintenance work.
[0017] Preferably, the photovoltaic station information further includes: the longitude and latitude of the photovoltaic panel reference object, the relative position and deflection angle between the photovoltaic panel reference object and the photovoltaic panel; the photovoltaic panel reference object includes: roads, trees and iconic buildings;
[0018] Using image recognition technology to identify the photovoltaic panel reference object in the image data of the photovoltaic panel with defects, and obtaining the relative position and angle between the identified photovoltaic panel reference object and the photovoltaic panel with defects in the image data;
[0019] Combined with the relative position and deflection angle of the photovoltaic panel reference object and the photovoltaic panel in the photovoltaic station information, and the relative position and deflection angle of the identified photovoltaic panel reference object and the defective photovoltaic panel in the image data, the layout of the photovoltaic panels in the area determined by the image data of the defective photovoltaic panel is matched with the image data of the identified defective photovoltaic panel to determine the photovoltaic panel number corresponding to the defective photovoltaic panel.
[0020] By adopting the above scheme, considering that there are photovoltaic panels with similar layouts, it may be impossible to accurately identify the number of the defective photovoltaic panel during the matching process with the acquired image data. By using the photovoltaic panel reference as a reference, more accurate positioning of the defective photovoltaic panel can be achieved.
[0021] Preferably, it also includes:
[0022] The passing speed of the drone collected by the speed sensor or camera installed along the inspection route of each divided area is obtained. If the difference between the real-time passing speed collected within a preset time and the speed set for the drone to perform the inspection task is greater than the preset speed difference, it is inferred that the drone has a fault, and the inspection task is stopped and the drone is reallocated to perform the current inspection task.
[0023] By adopting the above solution, considering that a UAV failure may lead to an error in determining the area to which the collected image data belongs, thereby making it impossible to achieve more accurate defect identification, a speed sensor or camera device is used to obtain the flight speed of the UAV, and the set flight speed is compared to promptly find and replace the faulty UAV, and re-execute the corresponding inspection task to ensure the accuracy of defect identification.
[0024] Preferably, it also includes: real-time collection of meteorological data of the photovoltaic station, wherein the meteorological data includes visibility, light intensity and amount of rain and snow;
[0025] Determine whether the visibility of the photovoltaic station currently collected is lower than the preset visibility, whether the light intensity is greater than the preset light intensity, and whether the amount of rain and snow is greater than the first preset rain and snow threshold. If the visibility of the photovoltaic station currently collected is lower than the preset visibility or the light intensity is greater than the preset light intensity or the amount of rain and snow is greater than the first preset rain and snow threshold, stop creating and executing the inspection task.
[0026] By adopting the above solution, taking into account the environmental factors that may affect the recognition of photovoltaic panel defects, such as low visibility affecting the accuracy of defect recognition, reflected light caused by high light intensity affecting the accuracy of defect recognition, and large amounts of rain and snow covering the photovoltaic panels affecting the accuracy of defect recognition, etc., real-time collection of environmental data is set and compared with the corresponding preset thresholds to avoid performing inspection tasks in an environment that is not suitable for inspection and to ensure the accuracy of defect recognition.
[0027] Preferably, it also includes:
[0028] The meteorological data collected in real time at the photovoltaic station also includes: wind speed;
[0029] Determine whether the wind speed currently collected at the photovoltaic station is greater than the first preset wind speed threshold and whether the amount of rain and snow is greater than the second preset rain and snow threshold. If the wind speed currently collected at the photovoltaic station is greater than the first preset wind speed threshold or the amount of rain and snow is greater than the second preset rain and snow threshold, increase the creation and execution frequency of inspection tasks after the current time.
[0030] By adopting the above solution, considering that the occurrence of severe weather conditions will increase the probability of defects in photovoltaic panels, it is timely judged whether a relatively severe weather environment has occurred. Once a relatively severe weather environment has occurred, the inspection frequency is increased accordingly to promptly detect defects.
[0031] Preferably, the improved YOLOv8 recognition algorithm can select YOLOv8-GAM, YOLOv8-SA or YOLOv8-DBB; the YOLOv8-GAM is based on the original YOLOv8 model framework, and a GAM module is added after the SPPF module of the backbone network; the YOLOv8-SA is based on the original YOLOv8 model framework, and the SA mechanism is added after the Conv +Concat+ C2f network of the original YOLOv8 model; the YOLOv8-DBB is based on the original YOLOv8 model framework, and the C2f module is replaced by DBB in the neck network.
[0032] By adopting the above scheme, a GAM module is added after the SPPF module of the backbone network to reduce information loss and amplify the global interaction representation, helping the model to better capture the global contextual information in the image; based on the YOLOv8 model, the permutation attention SA is introduced. The SA mechanism can efficiently integrate the spatial attention and channel attention mechanisms, so as to more accurately locate and identify faults on photovoltaic panels; in the neck network, diverse convolution branches DBB are added, and convolution branches of different sizes and structures are introduced to enrich the feature representation ability of the network, effectively integrate features at different levels, and better complete the defect recognition of photovoltaic panels.
[0033] In a second aspect, the present application provides a photovoltaic station defect recognition system based on a target detection algorithm, comprising:
[0034] The inspection route planning and area type determination module is used to obtain photovoltaic station information, divide and number photovoltaic stations into regions; design the inspection route of photovoltaic stations according to the regional division and numbering of photovoltaic stations, and determine the area type corresponding to each area according to the obtained photovoltaic station information; the area type includes: photovoltaic panel dense area or photovoltaic panel sparse area; the photovoltaic station information includes: photovoltaic station location information, layout information and quantity information of photovoltaic panels in the photovoltaic station;
[0035] Inspection task creation and execution module, which is used to create inspection tasks based on the inspection routes of photovoltaic stations and dispatch drones to execute the assigned inspection tasks;
[0036] The inspection image data acquisition module is used to acquire the image data collected by the drone during the inspection mission, and determine the area to which the collected image data belongs and the area type corresponding to the area according to the inspection route, flight speed and flight time during the inspection mission of the drone;
[0037] The inspection image recognition task creation module is used to create different types of recognition tasks according to the area type corresponding to the area determined by the collected image data; if the area type corresponding to the area determined by the collected image data is a photovoltaic panel dense area, then a first type of recognition task is created accordingly; if the area type corresponding to the area determined by the collected image data is a photovoltaic panel sparse area, then a second type of recognition task is created accordingly;
[0038] The inspection image recognition task execution module is used to use different recognition algorithms to perform different types of recognition tasks and identify defective photovoltaic panels; if the current type of recognition task is the first type of recognition task, the YOLOv8 recognition algorithm is used to identify photovoltaic panel defects; if the current type of recognition task is the second type of recognition task, the improved YOLOv8 recognition algorithm is used to identify photovoltaic panel defects;
[0039] The photovoltaic station fault report generation module is used to generate a photovoltaic station fault report based on photovoltaic panels that are identified to have defects.
[0040] By adopting the above solution, through the inspection area division and the determination of the corresponding area type, different recognition algorithms are matched to the image data collected during the inspection task to identify photovoltaic panel defects, thereby achieving more accurate defect identification while reducing costs.
[0041] In a third aspect, the present application provides a photovoltaic station defect identification device based on a target detection algorithm, comprising a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the photovoltaic station defect identification method based on the target detection algorithm as described above is implemented.
[0042] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the method as described above.
[0043] In summary, this application has the following beneficial effects:
[0044] 1. Consider the density of photovoltaic panels in different areas of the photovoltaic station. The higher the density, the more difficult it is to identify defects. Determine the density of photovoltaic panels in the area where the image data collected during the inspection belongs, create different types of recognition tasks accordingly, and use different recognition algorithms to perform corresponding types of recognition tasks, so as to improve the accuracy of photovoltaic panel defect recognition while reducing costs;
[0045] 2. Design the numbering of photovoltaic panels in different areas of the photovoltaic station, correspond to the location information of the photovoltaic panels, compare the layout information of the photovoltaic panels in different areas of the photovoltaic station with the collected image data, and combine the reference of the photovoltaic panels to quickly match the number of the photovoltaic panels and obtain the location of the defective photovoltaic panels;
[0046] 3. Consider the meteorological environment, collect meteorological environment data in real time, compare it with the preset meteorological environment data, adjust the execution and creation of inspection tasks, and ensure the accuracy of the collected image recognition. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 is a flow chart of the method described in a specific embodiment;
[0048] Figure 2 It is a diagrammatic representation of the layout of some photovoltaic panels in a photovoltaic station in the method described in a specific embodiment;
[0049] Figure 3It is a diagram of some photovoltaic panel reference objects in a photovoltaic station in the method described in the specific embodiment;
[0050] Figure 4 It is a flow chart of using the target detection algorithm to identify defects in the method described in the specific embodiment;
[0051] Figure 5 is a diagram of the matching of image data and actual layout information in the method described in the specific embodiment;
[0052] Figure 6 It is a partially enlarged illustration of the matching between image data and actual layout information in the method described in the specific embodiment;
[0053] Figure 7 It is a schematic diagram of the structure of the system described in the specific embodiment. DETAILED DESCRIPTION
[0054] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0055] like Figure 1 As shown, the embodiment of the present application discloses a photovoltaic station defect identification method based on a target detection algorithm, which specifically includes:
[0056] S1. Obtain photovoltaic station information, and divide and number the photovoltaic stations.
[0057] Specifically, the information of the photovoltaic station includes: photovoltaic station location information (latitude and longitude), photovoltaic panel layout information and quantity information in the photovoltaic station, photovoltaic panel size information in the photovoltaic station, photovoltaic panel deflection angle, photovoltaic panel reference position information (latitude and longitude), photovoltaic panel reference and photovoltaic panel relative position and angle, etc.; wherein, photovoltaic panel layout information in the photovoltaic station includes photovoltaic panel location information, arrangement, etc.; Figure 2 As shown in the figure, the deflection angle of the photovoltaic panel is the angle of deviation from the coordinate axis with due north and due east as the coordinate axis; Figure 3 As shown, the reference objects for photovoltaic panels include: roads, trees and other iconic buildings.
[0058] The photovoltaic station is divided into regions according to the location information of the photovoltaic panels in the photovoltaic station and numbered, and divided into region 1, region 2, ... region N; Figure 2 As shown, after the photovoltaic station is divided and numbered, the photovoltaic panels in each divided area are numbered, each photovoltaic panel has a unique corresponding number, and each photovoltaic panel unique number corresponds to a unique location information; the numbers are: 1, 2, 3...
[0059] S2. Design inspection routes for photovoltaic stations based on the regional division information and numbering information of the photovoltaic stations.
[0060] Specifically, the inspection route of the photovoltaic station is designed according to the regional division information and numbering information of the photovoltaic station, such as: the inspection route is designed to pass through area 1-area 3-area 4-area 2...-area N in sequence, and the inspection route in each area is designed to be S-shaped.
[0061] S3. Determine the area type corresponding to each area according to the acquired photovoltaic site information.
[0062] Specifically, the area types include: photovoltaic panel dense area or photovoltaic panel sparse area; the size of each area is determined according to the position of the area in the photovoltaic station information, and the density of photovoltaic panels in each area is calculated in combination with the number of photovoltaic panels in each area; the density of photovoltaic panels in each area is compared with the density of preset photovoltaic panels. If the density of photovoltaic panels in a certain area is greater than the density of preset photovoltaic panels, the area is identified as a photovoltaic panel dense area, otherwise, the area is identified as a photovoltaic panel sparse area.
[0063] S4. Based on the inspection route of the PV station, create inspection tasks and dispatch drones to perform the assigned inspection tasks.
[0064] Specifically, the creation of the inspection task includes: inspection objects, that is, part or all of the areas of the photovoltaic station that conform to the inspection route sequence, such as: Area 1, Area 3; inspection routes, that is, inspection routes designed for each area of the photovoltaic station, such as: Area 1-Area 3; inspection execution objects: drones with specific numbers, such as drone No. 001; flight parameters of the inspection execution objects, including: flight time, flight speed, etc.
[0065] For the inspection tasks created, drones are dispatched to perform the corresponding assigned inspection tasks and collect image data or video data of various areas of the photovoltaic station.
[0066] S5. Obtain image data collected by the drone during the inspection mission, and determine the area to which the collected image data belongs and the area type corresponding to the area.
[0067] Specifically, the shooting data of the drone during the inspection mission is obtained, and if the shooting data is video data, the key frame image data is extracted. Considering that the drone performs the inspection mission by flying to a specific area along the flight inspection route and shooting images of the specific area, the location of the drone corresponding to the image data collected at the current moment is determined according to the inspection route, flight speed and flight time during the inspection mission of the drone corresponding to the acquisition moment, and the area to which the image data content corresponding to the current moment of collection belongs is obtained, and the area type of the determined corresponding area is obtained based on the area type of each area.
[0068] S6. Create different types of recognition tasks according to the area type corresponding to the area determined by the collected image data, and set the execution time.
[0069] Specifically, if the area type corresponding to the area determined by the collected image data is a photovoltaic panel dense area, a first type of recognition task is created accordingly, and the first type of recognition task includes identifying the image data collected and the area type of which is a photovoltaic panel dense area; if the area type corresponding to the area determined by the collected image data is a photovoltaic panel sparse area, a second type of recognition task is created accordingly, and the second type of recognition task includes identifying the image data collected and the area type of which is a photovoltaic panel sparse area.
[0070] When setting the execution time, you can choose to obtain the next moment of the corresponding generation of the recognition task of all image data of a single area as the execution time, or obtain the next moment of the recognition task of the image data in real time as the execution time.
[0071] S7. Use different recognition algorithms to perform different types of recognition tasks and identify defective photovoltaic panels.
[0072] Specifically, if the current type of recognition task is the first type of recognition task, the YOLOv8 recognition algorithm is used to identify the defective photovoltaic panels in the corresponding image data; if the current type of recognition task is the second type of recognition task, the improved YOLOv8 recognition algorithm is used to identify the defective photovoltaic panels in the corresponding image data.
[0073] like Figure 4 As shown, the process of performing the recognition task and identifying defective photovoltaic panels in the image includes:
[0074] The collected image data corresponding to the first type of recognition task is preprocessed, and the preprocessed image data is input into the YOLOv8 recognition model; if the collected data is video data, the key frame is extracted in advance; the features of the input image are extracted through the backbone network (such as DarkNet), and then the semantic information of different scales is captured through the multi-scale feature fusion strategy. Next, the target is predicted through the output layer, and the output includes the coordinates of the target bounding box, the category probability and the confidence score. During the training process, a comprehensive loss function is used to optimize the network, which includes the positioning error of the target position, the cross entropy loss of the category prediction, and the error of the confidence score. Finally, after post-processing steps, such as filtering the bounding boxes with low confidence and applying non-maximum suppression, the final defect recognition result is obtained.
[0075] The collected image data corresponding to the second type of recognition task is preprocessed, and the preprocessed image data is input into the improved YOLOv8 recognition model, and the improved YOLOv8 recognition model is used to identify defects to obtain the final defect recognition result. Specifically, the improved YOLOv8 recognition model can be based on the original YOLOv8 model framework, and a new and efficient multi-scale attention (EMA) module is added to the neck module of YOLOv8. The improved model can be based on the original YOLOv8 model framework, and the DCNv4 is combined with SPPF, the original C2f is replaced by the CSPstage, and the number of detection heads is increased. Improved model; or select YOLOv8-GAM, YOLOv8-SA or YOLOv8-DBB.
[0076] Among them, the YOLOv8-GAM is based on the original YOLOv8 model framework, and a global attention mechanism (GAM) is added to the backbone network (Backbone) of the model. The GAM module is mainly composed of a channel attention submodule and a spatial attention submodule. Among them, the channel uses a three-dimensional module arrangement to retain information in three dimensions of channel, spatial width and spatial height, and a two-layer MLP (multi-layer perceptron) is used to amplify the cross-dimensional channel-space dependency. Two convolutional layers are used for spatial information fusion. At the same time, the maximum pooling is deleted to further retain the feature map. The spatial attention submodule uses two convolutional layers for spatial information fusion. At the same time, the maximum pooling is deleted to further retain the feature map. On the basis of the original model, the GAM module is added to the back of the SPPF module of the backbone network (backbone). After experiments, compared with the traditional YOLOv8 model, the F1 score of YOLOv8-GAM is improved from 0.802 to 0.811.
[0077] The YOLOv8-SA is based on the original YOLOv8 model framework and introduces Shuffle Attention (SA). The SA mechanism can efficiently integrate spatial attention and channel attention mechanisms, so as to more accurately locate and identify faults on photovoltaic panels. The specific principle is that the input feature map X is first divided into G groups along the channel dimension, each group is a sub-feature Xk. Then, the attention is mixed, and the sub-feature Xk is divided into two branches, one branch uses the mutual relationship between channels to generate a channel attention map, and the other branch captures the spatial dependency between features to generate a spatial attention map. The model simultaneously completes the attention of semantic and position information. Then the channel shuffle operation is used to ensure the interaction between each group of sub-features, and finally obtains an attention map with the same dimension as the input feature map X. Specifically, SA is added to the back of the Conv +Concat+ C2f (aligned feature pyramid P5) network of the original model. After experiments, compared with the traditional YOLOv8 model, the F1 score of YOLOv8-SA is improved from 0.802 to 0.816, an increase of 2.99%.
[0078] The YOLOv8-DBB is based on the original YOLOv8 model framework. In the neck network (Neck), a diverse convolution branch (DBB, Diverse Branch Block) is added. The main principle of DBB is to introduce convolution branches of different sizes and structures to enrich the feature representation ability of the network and effectively fuse features at different levels. The DBB module consists of multiple Transforms. Transform I is to fuse convolution layers with batch normalization (batch norm), Transform II is used to merge the outputs of convolution layers with the same configuration, and Transform III is used to merge sequential convolution layers. Transform IV is to merge convolution layers through deep concatenation, Transform V is to integrate the average pooling (AVG) operation into the convolution operation, and TransformVI is used to combine convolution layers of different scales. Adding the modified DBB module can help the model more effectively fuse features at different levels. Specifically, in the neck network (Neck), the C2f module is replaced by DBB. Through experimental verification and comparison, compared with the traditional YOLOv8 model, the F1 score of YOLOv8-DBB is improved from 0.802 to 0.815.
[0079] S8. Generate a photovoltaic station fault report based on the photovoltaic panels identified to have defects.
[0080] Specifically, statistics are collected for defective photovoltaic panels identified in image data collected from different regions, and the current proportion of defective photovoltaic panels and the specific defect type of each photovoltaic panel can be counted; the defect information of the photovoltaic panels includes: breakage, obstruction, hot spots, etc.
[0081] In addition, in order to timely and accurately maintain the defective photovoltaic station, it is necessary to determine the location of the defective photovoltaic panel. The specific positioning process includes: Figure 5 As shown, the layout information of the photovoltaic panels in the corresponding area determined according to the image data of the defective photovoltaic panel is matched with the image data of the photovoltaic panel with the defect, such as Figure 5 and Figure 6 As shown, the collected image data is first matched with the actual layout information to determine the photovoltaic panel number corresponding to the defective photovoltaic panel, and the position information of the defective photovoltaic panel is determined based on the determined photovoltaic panel number. The matching results in this embodiment are shown in Table 1: Table 1
[0082]
[0083] Among them, the bounding box number is that one image may contain multiple defective photovoltaic panels, so the multiple bounding boxes output by the model are numbered; the layout number is the number of the photovoltaic panels in the divided area;
[0084] A photovoltaic station fault report is generated based on the statistical percentage information of defective photovoltaic panels, defect analysis information and positioning information.
[0085] Through the above scheme, by introducing artificial intelligence and machine learning technologies, the intelligent and automated defect identification of photovoltaic stations is realized. Taking into account the differences in photovoltaic panels in different areas of photovoltaic stations, different detection algorithms are used to achieve accurate identification of photovoltaic panel defects in different areas of photovoltaic stations. Compared with traditional inspection methods, this application improves the efficiency and accuracy of defect identification and reduces costs.
[0086] In a specific embodiment, in addition to considering the density, the photovoltaic panels in different areas of the photovoltaic station can also be differentiated and designed from other angles, so as to better achieve accurate detection; the difference from the above embodiment is that:
[0087] The area types include: an area with complex photovoltaic panel sizes or an area with simple photovoltaic panel sizes;
[0088] The process of determining the area type corresponding to each area according to the acquired photovoltaic station information includes: counting the size information of the photovoltaic panels in each area, if the number of photovoltaic panels of different size types in a certain area counted is greater than the number of preset size types, then the area type is considered to be a photovoltaic panel size complex area, otherwise, the photovoltaic panel area size is considered to be a simple area;
[0089] Creating different types of recognition tasks according to the area type corresponding to the area determined by the collected image data specifically includes: if the area type corresponding to the area determined by the collected image data is a photovoltaic panel complex size area, then a first type of recognition task is created accordingly; if the area type corresponding to the area determined by the collected image data is a photovoltaic panel simple size area, then a second type of recognition task is created accordingly;
[0090] In the process of using different recognition algorithms to perform different types of recognition tasks, the improved YOLOv8 recognition algorithm adopts YOLOv8 fused with BiFPN, that is, based on the framework of the YOLOv8 recognition model, an innovative fusion feature network is used to improve the PANet structure to Bifpn, so as to better handle the defect recognition of photovoltaic panels with large size differences in a region.
[0091] A specific embodiment, considering that there may be multiple similar matching photovoltaic panels in the process of matching the captured image with the actual layout, matching is performed in combination with reference objects to improve the accuracy of matching; the difference from the above embodiment is that it also includes:
[0092] Image recognition technology is used to identify the photovoltaic panel reference object in the image data of the photovoltaic panel with defects, and the relative position and angle between the identified photovoltaic panel reference object and the defective photovoltaic panel in the image data are obtained. For the photovoltaic panel reference object image recognition, a lighting reference object recognition model is pre-trained, and the lighting reference object recognition model is used to realize the lighting reference object recognition.
[0093] The relative position and angle between the photovoltaic panel reference object and the photovoltaic panel in the photovoltaic station information, and the relative position and angle between the identified photovoltaic panel reference object and the defective photovoltaic panel in the image data are compared to determine the number of the photovoltaic panel in the photovoltaic panel layout that matches the image data of the defective photovoltaic panel, such as Figure 5 shown.
[0094] A specific embodiment, taking into account that failures may occur during the inspection of the drone, affecting the accuracy of the judgment of the corresponding area of the image data during the inspection, in order to further ensure the accuracy of subsequent defect identification, is different from the above embodiment in that it also includes:
[0095] The passing speed of the drone collected by the speed sensors installed along the inspection routes of each divided area or the passing speed of the drone collected and calculated by the camera device is obtained in real time; wherein, speed sensors or camera devices are installed at intervals along the inspection route of each area, close to the location of the photovoltaic panels in the area, and the speed of the drone when passing through the specific location, that is, the unmanned passing speed, can be obtained through the speed sensor device or the camera device.
[0096] Determine the difference between the real-time passing speed and the speed set for executing the inspection task. If the speed difference is greater than the preset speed difference, it means that the UAV may have a fault. However, considering that other sudden factors such as environmental factors and bird factors affect the flight speed of the UAV, only when the speed difference obtained for a continuous period of time or for multiple calculations is greater than the preset speed difference, such as: the difference between the real-time passing speed collected for a preset number of consecutive times and the speed set for the UAV to execute the inspection task is greater than the preset speed difference, it is inferred that the UAV flight may have a fault, then stop continuing to execute the inspection task, and reallocate the UAV to execute the current inspection task.
[0097] A specific embodiment, taking into account that environmental factors may affect the accuracy of photovoltaic panel defect identification, analyzes the meteorological data collected in real time, does not set the inspection task when the meteorological data has a greater impact, or adjusts the inspection task execution frequency when the adverse meteorological data affects the generation of photovoltaic panel defects, to ensure timely and accurate photovoltaic panel defect identification in the photovoltaic station; the difference from the above embodiment is that it also includes:
[0098] Real-time collection of meteorological data of photovoltaic stations, including visibility, light intensity, amount of rain and snow, and wind speed;
[0099] Determine whether the currently collected visibility of the photovoltaic station is lower than the preset visibility, whether the light intensity is greater than the preset light intensity, and whether the amount of rain and snow is greater than the first preset rain and snow threshold. If the currently collected visibility of the photovoltaic station is lower than the preset visibility or the light intensity is greater than the preset light intensity or the amount of rain and snow is greater than the first preset rain and snow threshold, stop creating and executing the inspection task; wherein the first preset rain and snow threshold is the threshold that does not cause damage to the photovoltaic panels.
[0100] Determine whether the wind speed currently collected at the photovoltaic station is greater than the first preset wind speed threshold and whether the amount of rain and snow is greater than the second preset rain and snow threshold. If the wind speed currently collected at the photovoltaic station is greater than the emcee's preset wind speed threshold or the amount of rain and snow is greater than the second preset rain and snow threshold, increase the creation and execution frequency of inspection tasks after the current moment; among which, the first preset wind speed threshold and the second preset rain and snow threshold are thresholds that will cause damage to photovoltaic panels, which can be specifically obtained based on corresponding meteorological data statistics under historical conditions of photovoltaic panel damage.
[0101] like Figure 7 As shown, the present application provides a photovoltaic station defect recognition system based on a target detection algorithm, comprising:
[0102] Inspection route planning and area type determination module 101 is used to obtain photovoltaic station information, divide and number photovoltaic stations into regions; design inspection routes of photovoltaic stations according to the regional division and numbering of photovoltaic stations, and determine the area type corresponding to each area as a photovoltaic panel intensive area or a photovoltaic panel sparse area according to the obtained photovoltaic station information; the photovoltaic station information includes: photovoltaic station location information, layout information and quantity information of photovoltaic panels in the photovoltaic station;
[0103] Inspection task creation and execution module 102, used to create inspection tasks based on the inspection routes of the photovoltaic stations and dispatch drones to execute the assigned inspection tasks;
[0104] The inspection image data acquisition module 103 is used to acquire the image data collected during the inspection mission of the drone, and determine the area to which the collected image data belongs and the area type corresponding to the area according to the inspection route, flight speed and flight time during the inspection mission of the drone;
[0105] The inspection image recognition task creation module 104 is used to create different types of recognition tasks according to the area type corresponding to the area determined by the collected image data; if the area type corresponding to the area determined by the collected image data is a photovoltaic panel dense area, then a first type of recognition task is created accordingly; if the area type corresponding to the area determined by the collected image data is a photovoltaic panel sparse area, then a second type of recognition task is created accordingly;
[0106] The inspection image recognition task execution module 105 is used to perform different types of recognition tasks using different recognition algorithms to identify defective photovoltaic panels; if the currently executed type of recognition task is the first type of recognition task, the YOLOv8 recognition algorithm is used to identify photovoltaic panel defects; if the currently executed type of recognition task is the second type of recognition task, the improved YOLOv8 recognition algorithm is used to identify photovoltaic panel defects;
[0107] The photovoltaic station fault report generating module 106 is used to generate a photovoltaic station fault report according to the photovoltaic panels identified to have defects.
[0108] In addition, the system comprises:
[0109] The faulty photovoltaic panel positioning module 107 is used to divide and number the photovoltaic panels in each divided area after dividing the photovoltaic station into areas, each photovoltaic panel has a unique corresponding number, and each photovoltaic panel unique number corresponds to a unique location information; the layout of the photovoltaic panels in the corresponding area determined according to the image data of the defective photovoltaic panel is matched with the image data of the defective photovoltaic panel, the photovoltaic panel number corresponding to the defective photovoltaic panel is determined, and the location information of the defective photovoltaic panel is determined based on the determined photovoltaic panel number and sent to the photovoltaic station fault report generation module 106;
[0110] It is also used to use image recognition technology to identify photovoltaic panel reference objects in image data of photovoltaic panels with defects, and obtain the relative position and angle of the identified photovoltaic panel reference objects and the photovoltaic panel with defects in the image data; combine the relative position and angle of the photovoltaic panel reference objects and the photovoltaic panel in the photovoltaic station information, and the relative position and angle of the identified photovoltaic panel reference objects and the photovoltaic panel with defects in the image data, and match the layout of the photovoltaic panels in the area determined by the image data of the photovoltaic panel with defects with the image data of the photovoltaic panel with defects, so as to determine the photovoltaic panel number corresponding to the photovoltaic panel with defects;
[0111] The inspection task creation and execution module 102 is also used to obtain the passing speed of the drone collected by the speed sensor or camera installed along the inspection route of each divided area. If the difference between the real-time passing speed collected within a preset time and the speed set for the drone to execute the inspection task is greater than the preset speed difference, it is inferred that the drone may have a fault, and the inspection task is stopped, and the drone is reallocated to execute the current inspection task;
[0112] It is also used to collect meteorological data of photovoltaic stations in real time, and the meteorological data includes: visibility, light intensity and rain and snow amount; judge whether the visibility of the photovoltaic station currently collected is lower than the preset visibility, whether the light intensity is greater than the preset light intensity, and whether the rain and snow amount is greater than the first preset rain and snow amount threshold; if the visibility of the photovoltaic station currently collected is lower than the preset visibility or the light intensity is greater than the preset light intensity or the rain and snow amount is greater than the first preset rain and snow amount threshold, then stop creating and executing the inspection task;
[0113] Also used for real-time collection of meteorological data of photovoltaic stations, it also includes: wind speed; judging whether the wind speed of the photovoltaic station currently collected is greater than the first preset wind speed threshold and whether the amount of rain and snow is greater than the second preset rain and snow threshold; if the wind speed of the photovoltaic station currently collected is greater than the first preset wind speed threshold or the amount of rain and snow is greater than the second preset rain and snow threshold, then increasing the creation and execution frequency of inspection tasks after the current moment.
[0114] The embodiment of the present application is a photovoltaic station defect identification device based on a target detection algorithm.
[0115] Specifically, the identification device includes a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the photovoltaic station defect identification method based on the target detection algorithm as described above is implemented.
[0116] The embodiment of the present application also discloses a computer-readable storage medium.
[0117] Specifically, the computer-readable storage medium stores a computer program that can be loaded by a processor and executes the photovoltaic station defect identification method based on the target detection algorithm as described above. The computer-readable storage medium includes, for example: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.
[0118] The above are all preferred embodiments of the present application, and are not intended to limit the protection scope of the present application. Any feature disclosed in this specification (including the abstract and drawings), unless otherwise stated, can be replaced by other equivalent or alternative features with similar purposes. That is, unless otherwise stated, each feature is only an example of a series of equivalent or similar features.
Claims
1. A photovoltaic station defect identification method based on target detection algorithm, It is characterized in that include: Obtain PV station information, divide and number the PV stations; design inspection routes for PV stations based on the regional division and numbering of PV stations, and determine the regional type corresponding to each area based on the acquired PV station information; The area type includes: a photovoltaic panel dense area or a photovoltaic panel sparse area; the photovoltaic station information includes: photovoltaic station location information, layout information and quantity information of photovoltaic panels in the photovoltaic station; Based on the inspection routes of the PV stations, create inspection tasks and dispatch drones to perform the assigned inspection tasks; Obtain image data collected during the inspection mission of the drone, and determine the area to which the collected image data belongs and the area type corresponding to the area according to the inspection route, flight speed and flight time of the drone during the inspection mission; Different types of recognition tasks are created according to the area type corresponding to the area determined by the collected image data; if the area type corresponding to the area determined by the collected image data is a photovoltaic panel dense area, a first type of recognition task is created accordingly; if the area type corresponding to the area determined by the collected image data is a photovoltaic panel sparse area, a second type of recognition task is created accordingly; Different recognition algorithms are used to perform different types of recognition tasks to identify defective photovoltaic panels; if the current recognition task is the first type of recognition task, the YOLOv8 recognition algorithm is used to identify photovoltaic panel defects; if the current recognition task is the second type of recognition task, the improved YOLOv8 recognition algorithm is used to identify photovoltaic panel defects; the improved YOLOv8 recognition algorithm can select YOLOv8-GAM, YOLOv8-SA or YOLOv8-DBB; the YOLOv8-GAM is based on the original YOLOv8 model framework, and a GAM module is added after the SPPF module of the backbone network; the YOLOv8-SA is based on the original YOLOv8 model framework, and the SA mechanism is added after the Conv +Concat+ C2f network of the original YOLOv8 model; the YOLOv8-DBB is based on the original YOLOv8 model framework, and the C2f module is replaced by DBB in the neck network; Generate a PV station fault report based on the PV panels identified as defective.
2. According to the photovoltaic station defect identification method based on the target detection algorithm of claim 1, It is characterized in that Also includes: After the photovoltaic station is divided and numbered, the photovoltaic panels in each divided area are numbered, each photovoltaic panel has a unique corresponding number, and each photovoltaic panel unique number corresponds to a unique location information; The layout of the photovoltaic panels in the area determined based on the image data of the defective photovoltaic panel is matched with the image data of the identified defective photovoltaic panel to determine the photovoltaic panel number corresponding to the defective photovoltaic panel; based on the determined photovoltaic panel number, the location information of the defective photovoltaic panel is determined and displayed in the photovoltaic station fault report.
3. According to the photovoltaic station defect identification method based on target detection algorithm according to claim 2, It is characterized in that The photovoltaic station information also includes: the latitude and longitude of the photovoltaic panel reference object, the relative position and angle between the photovoltaic panel reference object and the photovoltaic panel; the photovoltaic panel reference object includes: roads, trees and iconic buildings; Using image recognition technology to identify the photovoltaic panel reference object in the image data of the photovoltaic panel with defects, and obtaining the relative position and angle between the identified photovoltaic panel reference object and the photovoltaic panel with defects in the image data; Combined with the relative position and angle of the photovoltaic panel reference object and the photovoltaic panel in the photovoltaic station information, and the relative position and angle of the identified photovoltaic panel reference object and the defective photovoltaic panel in the image data, the layout of the photovoltaic panels in the area determined by the image data of the defective photovoltaic panel is matched with the image data of the identified defective photovoltaic panel to determine the photovoltaic panel number corresponding to the defective photovoltaic panel.
4. According to the photovoltaic station defect identification method based on target detection algorithm according to claim 1, It is characterized in that Also includes: The passing speed of the drone collected by the speed sensor or camera installed along the inspection route of each divided area is obtained. If the difference between the real-time passing speed collected within a preset time and the speed set for the drone to perform the inspection task is greater than the preset speed difference, it is inferred that the drone has a fault, and the inspection task is stopped and the drone is reallocated to perform the current inspection task.
5. According to the photovoltaic station defect identification method based on target detection algorithm according to claim 1, It is characterized in that Also includes: Real-time collection of meteorological data of photovoltaic stations, including visibility, light intensity, and amount of rain and snow; Determine whether the visibility of the photovoltaic station currently collected is lower than the preset visibility, whether the light intensity is greater than the preset light intensity, and whether the amount of rain and snow is greater than the first preset rain and snow threshold; if the visibility of the photovoltaic station currently collected is lower than the preset visibility or the light intensity is greater than the preset light intensity or the amount of rain and snow is greater than the first preset rain and snow threshold, stop creating and executing the inspection task.
6. According to the photovoltaic station defect identification method based on target detection algorithm according to claim 1, It is characterized in that Also includes: The meteorological data collected in real time at the photovoltaic station also includes: wind speed; Determine whether the wind speed currently collected at the photovoltaic station is greater than the first preset wind speed threshold and whether the amount of rain and snow is greater than the second preset rain and snow threshold. If the wind speed currently collected at the photovoltaic station is greater than the first preset wind speed threshold or the amount of rain and snow is greater than the second preset rain and snow threshold, increase the creation and execution frequency of inspection tasks after the current time.
7. A photovoltaic station defect recognition system based on target detection algorithm, It is characterized in that include: Inspection route planning and area type determination module, used to obtain photovoltaic station information, divide and number photovoltaic stations into regions; design inspection routes for photovoltaic stations according to the regional division and numbering of photovoltaic stations, and determine the area type corresponding to each area according to the obtained photovoltaic station information; The area type includes: a photovoltaic panel dense area or a photovoltaic panel sparse area; the photovoltaic station information includes: photovoltaic station location information, layout information and quantity information of photovoltaic panels in the photovoltaic station; Inspection task creation and execution module, which is used to create inspection tasks based on the inspection routes of photovoltaic stations and dispatch drones to execute the assigned inspection tasks; The inspection image data acquisition module is used to acquire the image data collected by the drone during the inspection mission, and determine the area to which the collected image data belongs and the area type corresponding to the area according to the inspection route, flight speed and flight time during the inspection mission of the drone; The inspection image recognition task creation module is used to create different types of recognition tasks according to the area type corresponding to the area determined by the collected image data; if the area type corresponding to the area determined by the collected image data is a photovoltaic panel dense area, then a first type of recognition task is created accordingly; if the area type corresponding to the area determined by the collected image data is a photovoltaic panel sparse area, then a second type of recognition task is created accordingly; The inspection image recognition task execution module is used to perform different types of recognition tasks using different recognition algorithms to identify defective photovoltaic panels; if the current recognition task is the first type of recognition task, the YOLOv8 recognition algorithm is used to identify photovoltaic panel defects; if the current recognition task is the second type of recognition task, the improved YOLOv8 recognition algorithm is used to identify photovoltaic panel defects; the improved YOLOv8 recognition algorithm can select YOLOv8-GAM, YOLOv8-SA or YOLOv8-DBB; the YOLOv8-GAM is based on the original YOLOv8 model framework, and the GAM module is added after the SPPF module of the backbone network; the YOLOv8-SA is based on the original YOLOv8 model framework, and the SA mechanism is added after the Conv +Concat+ C2f network of the original YOLOv8 model; the YOLOv8-DBB is based on the original YOLOv8 model framework, and the C2f module is replaced by DBB in the neck network; The photovoltaic station fault report generation module is used to generate a photovoltaic station fault report based on photovoltaic panels that are identified to have defects.
8. A photovoltaic station defect identification device based on target detection algorithm, It is characterized in that It comprises a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, a photovoltaic station defect identification method based on a target detection algorithm as described in any one of claims 1 to 6 is implemented.
9. A computer-readable storage medium, It is characterized in that The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the method according to any one of claims 1 to 6.
