Intelligent management method of photovoltaic base based on Beidou +5G

Abstract

The invention discloses an intelligent management method of a photovoltaic base based on Beidou + 5G. The method comprises the following steps: step 1, measuring position information of a photovoltaic module through Beidou + 5G + GPS fusion positioning, storing the position information in a database, numbering photovoltaic panels according to the position information of the photovoltaic module by using N areas, N rows, N columns and edition of the photovoltaic module and taking the photovoltaic panels as basic units, and realizing approximate position positioning through Beidou + 5G + GPS fusion positioning and image data transmission, in the invention, the unmanned aerial vehicle is equipped with an infrared camera and a high-definition camera, a fault photovoltaic panel is judged through hot spots according to thermal imaging influence, and hot spot thermal imaging identification and high-definition image linkage calling of the photovoltaic panel are carried out. The positioning of the photovoltaic panel is realized through comparison, hot spot frame selection, pixel coordinate positioning and digital identification, the influence of signals and the environment is avoided, and the positioning is accurate and reliable.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of communication and photovoltaic facilities, and particularly relates to a smart management method for a photovoltaic base based on Beidou and 5G. BACKGROUND

[0002] The core of the photovoltaic base station refers to a communication / data acquisition / scheduling core node powered by photovoltaic power generation or used for operation and maintenance of a photovoltaic power station, which includes a photovoltaic power supply communication base station to solve the communication coverage of a remote photovoltaic power station and a photovoltaic power station dedicated operation and maintenance base station.

[0003] In the unmanned aerial vehicle photovoltaic inspection scene, Beidou+5G+GPS fusion positioning is a core scheme for solving high-precision positioning, long-distance data transmission and anti-interference redundancy, which can meet the requirements of accurate planning of an inspection path and accurate positioning of a defect position of a photovoltaic power station (especially a large distributed / mountainous power station), and can realize real-time return and remote control of inspection data (images / laser point clouds) through 5G, so as to form a closed loop of positioning-transmission-application.

[0004] The biggest problem of the unmanned aerial vehicle photovoltaic inspection is signal reception and environmental interference, and signal shielding, satellite signal reflection by photovoltaic panels and metal supports, electromagnetic interference, system fusion and cooperation, hardware and algorithm limitations can all cause GPS positioning deviation of the unmanned aerial vehicle, and the accuracy of the unmanned aerial vehicle GPS positioning of a fault photovoltaic panel is easily affected by signal reception and environmental interference. In order to improve the positioning accuracy of the unmanned aerial vehicle thermal imaging fault photovoltaic panel and eliminate the signal and environmental interference phenomenon of the fault photovoltaic panel positioning, the application is provided. SUMMARY

[0005] The application aims to provide a smart management method for a photovoltaic base based on Beidou and 5G to solve the problems in the background.

[0006] To achieve the above-mentioned purpose, the application provides the following technical scheme.

[0007] The smart management method for the photovoltaic base based on Beidou and 5G comprises the following steps.

[0008] Step 1: Measure the position information of the photovoltaic component through Beidou+5G+GPS fusion positioning and store it in a database, according to the position information of the photovoltaic component, number the photovoltaic panel based on the photovoltaic panel as a basic unit by using the N area, N row, N column and editing of the photovoltaic component, and prepare a number plate, arrange the photovoltaic panel according to the corresponding position, and fix the number plate on the frame of the photovoltaic panel, and the angle of the number plate is ensured to be able to be shot by the unmanned aerial vehicle.

[0009] Step 2, the unmanned aerial vehicle is equipped with an infrared camera and a high-definition camera, the infrared camera and the high-definition camera have the same angle of view, the shooting time is synchronized, and the camera on the unmanned aerial vehicle shoots the photovoltaic module downward to obtain a thermal imaging video and a high-definition video;

[0010] Step 3, real-time analysis and processing are performed on each image in the thermal imaging video, a hot spot of the photovoltaic panel in the thermal imaging is identified, a hot spot thermal imaging time with the largest (closest) shape of the photovoltaic panel hot spot is captured, and a corresponding high-definition camera image is called according to the captured time;

[0011] Step 4, the position of the hot spot in the thermal imaging image is obtained through pixel column, and the position of the hot spot photovoltaic panel in the high-definition image is located;

[0012] Step 5, the shape and area of the photovoltaic panel are identified and framed through AI, a number (photovoltaic panel number) in the imaging in the area of the hot spot photovoltaic panel is extracted, and the GPS position of the photovoltaic panel is called in the called database through the photovoltaic panel number.

[0013] As a further scheme of the application: the photovoltaic panel number is set according to the scale of the photovoltaic power station, and large photovoltaic power stations are arranged according to zones / rows / columns as follows:

[0014] A / 6 / 5;

[0015] The number and the background color form a contrast, the shooting process of the high-definition camera is clearer, and the AI number recognition is more accurate.

[0016] As a further scheme of the application: the photovoltaic panels are synchronously regulated in a centralized manner, 2-3 rows of photovoltaic panels are usually arranged in the centralized synchronous regulation, A / 6 / 5 can only represent A zone, 6th row and 5th column, and cannot accurately represent the row of photovoltaic panels on the centralized synchronous regulation, in this case, the number is represented as A / 6 / 5-1, which represents A zone, 6th row, 5th column and 1st row.

[0017] As a further scheme of the application: the hot spot in the thermal imaging is marked as a hot spot area through visual color difference and by comparing normal / abnormal temperature distribution rules,

[0018] Key parameters for marking the hot spot;

[0019] Position: hot spot center coordinates (Xc, Yc) pixel coordinates or physical coordinates;

[0020] Size: hot spot area S, pixel number or actual area cm 2 , equivalent diameter d = 2 .

[0021] As a further scheme of the application: the photovoltaic panel hot spot thermal imaging identification and high-definition image linkage calling technology;

[0022] Screen the target hot spot according to the maximum area (the maximum shape is equivalent to the maximum pixel area of the hot spot region), extract the corresponding thermal imaging timestamp, synchronously call the matching image of the high-definition camera based on the timestamp, the time error is less than or equal to 100 ms, and the visualization tracing of the hot spot scene is realized.

[0023] As a further scheme of the present application: through the phase velocity comparison of the thermal imaging picture and the high-definition image, when the similarity is more than 95%, the thermal imaging picture and the high-definition image can be replaced.

[0024] As a further scheme of the present application: the image recognition of the single photovoltaic panel adopts an image segmentation extraction method.

[0025] The original image is subjected to Otsu threshold segmentation, Gaussian filtering, Sobel edge detection, LSD straight line detection, rectangular contour screening and photovoltaic panel region extraction.

[0026] As a further scheme of the present application: the pixel frame selection coordinate information of the hot spot region of the thermal imaging image is extracted and copied into the high-definition image, the numbers in the frame selection region of the high-definition image are recognized and extracted, the high-definition image selection can be the picture with a time error of ±100 ms and a global threshold S of more than 95%, the numbers recognized by multiple images are compared, and the numbers recognized by multiple images are consistent and without any objection.

[0027] As a further scheme of the present application: the image number recognition preprocessing step is number positioning and segmentation, feature extraction, post-processing optimization and core deep learning model analysis.

[0028] As a further scheme of the present application: if the numbers recognized by multiple images are inconsistent, the number recognition result with a large proportion is referred to, if the multiple image recognition is unclear, the image information of the adjacent photovoltaic panel of the hot plate is called, the number plate of the adjacent photovoltaic panel is recognized, and the number plate of the photovoltaic panel of the hot plate is determined according to the database arrangement.

[0029] Compared with the prior art, the present application has the following advantages:

[0030] The intelligent management method of the photovoltaic base based on Beidou+5G realizes rough position positioning through the fusion positioning and transmission of image data of Beidou+5G+GPS, can only accurately determine the range of the area, and the deviation of the row and column positioning caused by the signal and the environment. Therefore, the unmanned aerial vehicle is equipped with an infrared camera and a high-definition camera, the fault photovoltaic panel is judged through thermal imaging, the photovoltaic panel thermal imaging recognition and high-definition image linkage are performed, and the photovoltaic panel is positioned through comparison, thermal spot frame selection, pixel coordinate positioning and number recognition. The positioning is accurate and reliable and is not affected by the signal and the environment. DETAILED DESCRIPTION

[0031] The embodiment of the application discloses a photovoltaic base intelligent management method based on Beidou and 5G.

[0032] Step 1, the position information of the photovoltaic module is measured through Beidou + 5G + GPS fusion positioning and stored in a database, according to the position information of the photovoltaic module, the photovoltaic module is numbered according to the photovoltaic panel as a basic unit, and a number plate is prepared, the photovoltaic panel is arranged according to the corresponding position, and the number plate is fixed on the frame of the photovoltaic panel, and the angle of the number plate is ensured to be capable of being shot by the unmanned aerial vehicle;

[0033] In the unmanned aerial vehicle photovoltaic inspection scene, the Beidou + 5G + GPS fusion positioning is a core scheme for solving "high-precision positioning + long-distance data transmission + anti-interference redundancy", which not only meets the accurate planning of the inspection path of the photovoltaic power station (especially large distributed / mountain power station) and the accurate positioning requirement of the defect position, but also realizes the real-time return and remote control of the inspection data (image / laser point cloud) through 5G, and the three constitute a closed loop of "positioning-transmission-application".

[0034] Step 2, the unmanned aerial vehicle is provided with an infrared camera and a high-definition camera, the infrared camera and the high-definition camera are taken at the same angle, the shooting time is synchronized, and the camera on the unmanned aerial vehicle shoots the photovoltaic module downward to obtain a thermal imaging video and a high-definition video;

[0035] Step 3, the thermal imaging video is first analyzed and processed in real time, the hot spot of the thermal imaging photovoltaic panel is identified, and the hot spot thermal imaging time of the hot spot with the largest (closest) shape of the photovoltaic panel is captured, and the corresponding high-definition camera image is retrieved according to the captured time;

[0036] Step 4, the hot spot position in the thermal imaging image is obtained through pixel column, and the hot spot photovoltaic panel position in the high-definition image is positioned.

[0037] Step 5, the photovoltaic panel shape and area are identified and framed by AI, the number (photovoltaic panel number) in the imaging in the area of the hot spot photovoltaic panel is extracted, and the GPS position of the photovoltaic panel is retrieved in the retrieved database through the photovoltaic panel number.

[0038] In a preferred embodiment, the photovoltaic panel number is set according to the scale of the photovoltaic power station, and the large photovoltaic power station is arranged according to the area / row / column, as follows:

[0039] A / 6 / 5;

[0040] The number and the background color form a contrast, the shooting process of the high-definition camera is clearer, and the AI digital identification is more accurate.

[0041] In a preferred embodiment, the photovoltaic panels are regulated by centralized synchronization, which usually sets 2-3 rows of photovoltaic panels, A / 6 / 5 can only represent the A area, 6 rows, and 5 columns, and cannot accurately represent the row of photovoltaic panels on the centralized synchronization, in which case the numbering is represented as A / 6 / 5-1, which is the A area, 6 rows, 5 columns-1 row.

[0042] In a preferred embodiment, the hot spot in thermal imaging is marked as a hot spot area by locating the fault area exceeding the threshold value through visual color difference and comparing the normal / abnormal temperature distribution rule,

[0043] Key parameters for marking hot spots;

[0044] Location: Hot spot center coordinates (Xc, Yc) in pixel coordinates or physical coordinates;

[0045] Size: Hot spot area S, pixel number or actual area cm 2 , equivalent diameter d = 2 .

[0046] In a preferred embodiment, the photovoltaic panel hot spot thermal imaging identification is linked to the high-definition image retrieval technology;

[0047] Filter the target hot spot by area maximum (shape maximum equivalent to maximum hot spot area pixel), extract its corresponding thermal imaging timestamp, and synchronously retrieve the matching image of the high-definition camera based on the timestamp, with a time error of ≤100ms, to realize the visual traceability of the hot spot scene,

[0048] If there are multiple hot spots with the largest area (area difference ≤5%), take the latest hot spot (recently photographed) with the latest timestamp; output the timestamp T of the target hot spot, find the high-definition image closest to T, with a time error of ≤100ms, retrieve the timestamp range (T±100ms) from the high-definition image index library, if there are multiple images, take the frame with the smallest timestamp difference from T; if no result is found, automatically expand the search range to ±500ms, and output a warning log.

[0049] In a preferred embodiment, by comparing the speed of thermal imaging pictures and high-definition images, the similarity reaches more than 95%, which can be used for thermal imaging picture and high-definition image replacement;

[0050] Thermal imaging pictures and high-definition images are compared for similarity by pixel alignment;

[0051] Preprocessing: uniform dimension, color stripping;

[0052] Size alignment: if the sizes of the two images are different, first scale them to the same WxH scene through interpolation, such as bilinear interpolation, on the premise that there is no scaling / cropping;

[0053] Color stripping: Convert both images to grayscale (formula: Gray=0.299R+0.587G+0.114B) to completely remove color channel interference;

[0054] Denoising preprocessing: Gaussian filtering (σ=1.0, kernel size 3×3) is used to remove slight noise, avoiding noise affecting edge detection and statistical features; features are extracted from each column j (j∈[0,W-1]) of the thermal imaging image and the high-resolution image respectively.

[0055] In-column grayscale histogram (16 levels)

[0056] Perform 16-level quantization on the H gray values ​​(0-255) of column j (each level spans 16: 0-15, 16-31, ..., 240-255);

[0057] The pixel percentage of each order is counted to obtain the feature vector Hist_j (dimension 1×16).

[0058] column edge point coordinates

[0059] Perform Canny edge detection on the entire grayscale image (threshold: low threshold 50, high threshold 150, apertureSize=3) to obtain a binary edge map (edge ​​points=255, non-edge points=0).

[0060] For column j, extract the row coordinates of all edge points. For example, col_j=[10,50,120] means that the 10th, 50th and 120th rows of column j are edges. If there are no edge points in the column, it is recorded as an empty vector.

[0061] Feature similarity matching (column-by-column verification)

[0062] Calculate the feature similarity for corresponding columns j in the two graphs (column j of A vs. column j of B), and set a threshold to determine whether they are consistent:

[0063] Histogram similarity: The similarity between Hist_Aj and Hist_Bj is calculated using Bhattacharyya Distance. The smaller the Bhattacharyya Distance, the more consistent the distribution (the threshold is recommended to be ≤0.1, corresponding to a similarity of ≥95%).

[0064] Edge point coordinate similarity: Calculate the "overlap rate" of edge points in two columns = number of common row coordinates / maximum of the total number of edge points in two columns. The threshold is recommended to be ≥98%, and a small amount of noise-induced edge point offset is allowed.

[0065] Column-level consistency determination: If the histogram similarity and edge point overlap rate of a column both meet the threshold, then the column is determined to be "structurally consistent".

[0066] Global scenario consistency determination

[0067] Calculate the proportion of "structure consistent columns" in all columns (denoted as S);

[0068] Set a global threshold (recommended S≥95%, because only a few columns are allowed to be affected by noise when the scene is completely consistent. If S≥threshold, it is determined that the two images are "scene same, only color different".

[0069] In a preferred embodiment, monolithic photovoltaic panel image recognition uses image segmentation extraction method;

[0070] Original image→Otsu threshold segmentation→Gaussian filtering→Sobel edge detection→LSD line detection→rectangular contour screening→photovoltaic panel region extraction;

[0071] Otsu algorithm automatically finds the optimal threshold T to divide the image grayscale into foreground (photovoltaic panel) and background two categories, so that the inter-class variance of the two categories is maximized. The larger the inter-class variance, the better the target and background separation effect;

[0072] The original RGB image needs to be converted to a grayscale image first, and the grayscale level L=256;

[0073] Key parameters: no manual threshold (automatic calculation), only need to specify the segmentation type (binary: (pixel>T set to 255, otherwise 0;

[0074] Applicable conditions: photovoltaic panel, background, sky, support and license plate have significant grayscale contrast. When AI recognizes and frames the shape and area of the photovoltaic panel, it can frame the photovoltaic panel and license plate together in the area;

[0075] Output: binary image; photovoltaic panel area is white and background is black;

[0076] Gaussian filtering; by Gaussian kernel and image convolution, the pixel neighborhood is weighted and averaged;

[0077] Sobel operator detects edges by calculating the horizontal Gx and vertical Gy components of the image grayscale gradient. The larger the gradient amplitude, the more obvious the edge;

[0078] Photovoltaic panel rectangular contour screening logic; photovoltaic panel is a standard rectangle, whose edge is composed of 2 horizontal lines + 2 vertical lines. By line angle, line length, rectangular fitting and contour verification, the contour of photovoltaic panel and license plate is recognized and the area is framed.

[0079] In a preferred embodiment, the hot spot region pixel frame coordinate information of the extracted thermal imaging image is copied to the high-definition image, and the numbers in the framed region of the high-definition image are identified and extracted. The high-definition image selection can be multiple image identification of numbers and comparison of pictures with a time error of ±100 ms and a global threshold S≥95%.

[0080] In a preferred embodiment, the image number recognition preprocessing step is:

[0081] Grayscale (remove color interference, retain light and dark differences) -> binarization (convert to black and white image, enhance the contrast between numbers and background) -> noise reduction (remove noise and interference lines) -> tilt correction (correct the deviation of scanning or shooting angle);

[0082] Number positioning and segmentation:

[0083] Target detection (use YOLO algorithm to locate the number area in the image) -> character segmentation (separate continuous number string into single character, projection method, connected region analysis) ->

[0084] Feature extraction;

[0085] HOG direction gradient histogram, SIFT scale invariant feature transform -> deep learning (CNN automatically extracts high-dimensional features, captures number edges, contours and structures) -> number recognition (single number classification: CNN directly outputs 0-9 classification results) -> number sequence recognition (CRNN (CNN+RNN+CTC) processes continuous number string, without the need for segmentation before recognition);

[0086] Post-processing optimization (language model corrects recognition errors) -> combines front and back license plates to improve accuracy -> performs structured output on specific format numbers;

[0087] Core deep learning model analysis;

[0088] CRNN (the king of sequence number recognition) -> architecture composition (CNN layer: extracts image spatial features and generates feature sequence) -> RNN layer (LSTM / BLSTM) captures sequence context dependency) -> CTC layer (solves the problem of inconsistent label and predicted sequence length, realizes end-to-end training) -> YOLO+OCR: accurate positioning + efficient recognition -> YOLO: quickly detects the number area in the image -> OCR engine (Tesseract / PaddleOCR): high-precision recognition of the detected area.

[0089] In a preferred embodiment, the numbers identified by the multi-image are inconsistent, the number identification result with a higher reference proportion is referred to, the multi-image identification is unclear, the image information of the adjacent photovoltaic panel of the hot panel is called, the number plate of the adjacent photovoltaic panel is identified, and the number plate number of the hot panel photovoltaic panel is determined according to the database arrangement.

[0090] It should be noted that the above embodiments all belong to the same inventive concept, and the description of each embodiment has its own emphasis. If the description in an individual embodiment is not exhaustive, the description in other embodiments can be referred to.

[0091] The above-described embodiments only express the implementation of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be noted that for ordinary skilled persons in the art, without departing from the inventive concept, a number of modifications and improvements can be made, which are within the scope of the present application. Therefore, the protection scope of the present application patent should be subject to the appended claims.

Claims

1.A method for intelligent management of a photovoltaic base based on Beidou and 5G, characterized in that, It comprises the following steps: Step 1, measure the position information of the photovoltaic module by Beidou + 5G + GPS fusion positioning, and store it in the database. According to the position information of the photovoltaic module, use the N area, N row, N column and edit according to the photovoltaic panel as the basic unit to number the photovoltaic panel, and prepare a number plate. Arrange the photovoltaic panel according to the corresponding position, and fix the number plate on the frame of the photovoltaic panel. The angle of the number plate ensures that the unmanned aerial vehicle can shoot; Step 2, the unmanned aerial vehicle is equipped with an infrared camera and a high-definition camera. The infrared camera and the high-definition camera have the same angle of view, and the shooting time is synchronized. The camera on the unmanned aerial vehicle shoots downward to obtain thermal imaging video and high-definition video of the photovoltaic module; Step 3, first analyze and process each image in the thermal imaging video in real time, identify the hot spot of the thermal imaging photovoltaic panel, and capture the hot spot thermal imaging time of the photovoltaic panel with the largest (closest) shape. According to the captured time, the corresponding high-definition camera image is retrieved; Step 4, obtain the hot spot position in the thermal imaging image by pixel column, and locate the hot spot photovoltaic panel position in the high-definition image; Step 5, identify and frame the photovoltaic panel shape and area by AI, extract the number (photovoltaic panel number) in the image in the hot spot photovoltaic panel area, and retrieve the GPS position of the photovoltaic panel in the database through the photovoltaic panel number. 2.The Beidou+5G-based photovoltaic base intelligent management method according to claim 1, characterized in that, The photovoltaic panel number is set according to the scale of the photovoltaic power station. Large photovoltaic power stations are arranged according to area / row / column, as follows: A / 6 / 5； The number and background color form a contrast, which is clearer during high-definition camera shooting and more accurate for AI digital recognition. 3.The Beidou+5G-based photovoltaic base intelligent management method of claim 2, characterized in that, The photovoltaic panel is adjusted synchronously in a centralized manner. Centralized synchronous adjustment usually sets 2-3 rows of photovoltaic panels. In A / 6 / 5, it can only represent A area, 6 rows, and 5 columns, and cannot accurately represent the row of photovoltaic panel on the centralized synchronous adjustment. In this case, the number is represented as A / 6 / 5-1, which is A area, 6 rows, 5 columns-1 row. 4.The Beidou+5G-based photovoltaic base intelligent management method of claim 3, characterized in that, The hot spot in the thermal imaging is marked as a hot spot area by visualizing the color difference and comparing the normal / abnormal temperature distribution rule to locate the fault area exceeding the threshold value, Key parameters for marking hot spots; Location: Hot spot center coordinates (Xc, Yc) pixel coordinates or physical coordinates; Size: hot spot area S, number of pixels or actual area cm 2 Equivalent diameter d = 2 . 5.The Beidou+5G-based photovoltaic base intelligent management method according to claim 4, characterized in that, Photovoltaic panel hot spot thermal imaging identification and high-definition image linkage retrieval technology; According to the area maximum (shape maximum equivalent to hot spot area pixel area maximum), the target hot spot is screened, the corresponding thermal imaging timestamp is extracted, the matching image of the high-definition camera is retrieved based on the timestamp, the time error is ≤100ms, and the visualization of the hot spot scene is realized. 6.The Beidou+5G-based photovoltaic base intelligent management method according to claim 5, characterized in that, By comparing the speed of the thermal imaging picture and the high-definition picture, the similarity reaches more than 95%, which can be used for thermal imaging picture and high-definition picture replacement. 7.The Beidou+5G-based photovoltaic base intelligent management method according to claim 6, characterized in that, Single photovoltaic panel image recognition uses image segmentation extraction method; Original image→Otsu threshold segmentation→Gaussian filtering→Sobel edge detection→LSD straight line detection→rectangular contour screening→photovoltaic panel area extraction. 8.The Beidou+5G-based photovoltaic base intelligent management method according to claim 7, characterized in that, Extract the hot spot area pixel frame coordinate information of the thermal imaging image, copy it to the high-definition image, identify the numbers in the framed area of the high-definition image and extract them, the high-definition image selection can be multiple image identification of numbers and comparison of pictures with a time error of ±100 ms and a global threshold S≥95%. 9.The Beidou+5G-based photovoltaic base intelligent management method of claim 8, wherein, Image digital recognition preprocessing step → digital positioning and segmentation → feature extraction → post-processing optimization → core deep learning model analysis. 10.The Beidou+5G-based photovoltaic base intelligent management method according to claim 9, characterized in that, If the numbers identified by multiple images are inconsistent, refer to the number identification result with a higher proportion, if the multiple image identification is unclear, retrieve the image information of the adjacent photovoltaic panel, identify the adjacent photovoltaic panel number plate, and arrange it according to the database to determine the hot plate photovoltaic panel number plate number through the adjacent photovoltaic panel number plate.