Automatic operation and maintenance system for abnormity of photovoltaic panel of photovoltaic power station

Through drone and server-side data analysis technology, combined with QR code stickers and machine learning algorithms, the automated detection of photovoltaic panel faults in photovoltaic power stations is realized, solving the problem of photovoltaic panel fault identification in the existing technology, and improving operation and maintenance efficiency Level of intelligence.

CN118967098APending Publication Date: 2024-11-15JIANG SU ZHONG DUO LV NENG XIN NENG YUAN KE JI YOU XIAN GONG SI
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411016511.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-28
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The existing photovoltaic power station operation and maintenance technology lacks intelligent fault judgment, which makes it difficult to detect small problems of photovoltaic panels, low degree of automation, large manual consumption, and failure to effectively identify internal cracks, dust and poor contact of photovoltaic panels, resulting in waste of costs.

Method used

UAVs carry visible light + infrared cameras, combined with QR code stickers and server-side data analysis, to achieve automated fault detection and classification. The drone takes images at the photovoltaic power station. The server side establishes a database through the comparison picture and patrol position entry module. The fault discovery module analyzes the output data in real time, automatically recognizes the fault type, and judges the photovoltaic panel fault through machine learning algorithms.

Benefits of technology

Automatic and intelligent detection of photovoltaic panel faults in photovoltaic power stations has been realized, which has significantly reduced the workload of operation and maintenance personnel, improved the fault response speed and intelligent operation and maintenance level, and reduced cost waste.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118967098A_ABST
    Figure CN118967098A_ABST
Patent Text Reader

Abstract

The invention is suitable for the technical field of photovoltaic power station operation and maintenance, and provides an automatic operation and maintenance system for abnormity of a photovoltaic panel of a photovoltaic power station, and the system comprises a hardware part which comprises an unmanned aerial vehicle nest, an unmanned aerial vehicle, a visible light and infrared camera and a two-dimensional code sticker; and the algorithm part comprises a comparison picture and patrol point posture input module and a fault discovery module. The unmanned aerial vehicle is dispatched after a fault of a branch is found in an algorithm part, a visible light and infrared camera is utilized to shoot each photovoltaic panel, an image is sent to a server side, which branch has the fault is analyzed by using an algorithm, and then whether the image sent back by the unmanned aerial vehicle has faults such as cracks and poor contact of an interface is analyzed. According to the invention, whether the field end branch has a fault can be identified in real time. Once the fault is detected, the system automatically and intelligently analyzes the main fault type of the photovoltaic panel at the field end, so that the workload of operation and maintenance personnel is remarkably reduced, and the fault response speed and the operation and maintenance intelligent level are greatly improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of photovoltaic power station operation and maintenance, and in particular relates to an automated operation and maintenance system for photovoltaic panels of a photovoltaic power station when the photovoltaic panels are abnormal. Background Art

[0002] Photovoltaic power station is a new energy system that can convert light energy into electrical energy and then provide it to the power grid through inversion. Its structure includes photovoltaic panels, combiner boxes, inverters and other equipment. At present, the operation of photovoltaic power stations requires a large number of professionals to carry out daily operation and maintenance, and the main work content includes fault inspection and maintenance of equipment. Due to the lack of intelligent fault judgment, the current solution mainly relies on the fixed-cycle one-by-one inspection by operation and maintenance personnel. This solution has a low degree of automation, high labor consumption, and it is difficult to use the naked eye to find small problems such as cracks in photovoltaic panels within the visible light band, resulting in ineffective cost waste. Among them, photovoltaic panel failures account for the largest proportion of overall failures, so automated photovoltaic panel fault detection is the most urgent issue. The lowest-level device of the existing data collector is the inverter, so the most detailed physical device granularity that can be used for data analysis is the branch line. According to the current power output data, it is possible to analyze in real time which branch line has a problem, improve the response speed, and reduce the workload of fixed-cycle manual inspection.

[0003] Among the known faults of photovoltaic panels at the field end, there are three main types that need to be addressed: 1) fine internal cracks, which will cause the power output of the photovoltaic panel to drop sharply at a certain ratio over time, or even directly to zero; 2) dust precipitation on the surface of the photovoltaic panel will cause the power output of the photovoltaic panel to drop steadily at a certain ratio over time; 3) poor contact when output to the combiner box, which has a relatively small probability of occurrence but serious consequences, will cause its power output to be directly zero. These problems require automatic and intelligent analysis of the main fault types of photovoltaic panels at the field end. In a more favorable situation, the operation and maintenance team only needs to find the problematic photovoltaic panels and does not need to repair them on site. They only need to dismantle the problematic photovoltaic panels, replace them with new photovoltaic panels, and hand over the problematic photovoltaic panels to the manufacturer for repair. Therefore, in the above problems, cracks do not need to be specifically located on site. In order to alleviate this problem, it is mainly necessary to solve how to use data on the server side to detect and classify the faults at the field end, and how to automatically and intelligently analyze the faults of photovoltaic panels at the field end. The existing technology does not have a comprehensive solution to solve the above problems. Therefore, an automated operation and maintenance system for photovoltaic panel abnormalities in photovoltaic power plants is proposed. Summary of the invention

[0004] The purpose of the present invention is to provide an automated operation and maintenance system for photovoltaic panels in photovoltaic power plants, aiming to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] An automated operation and maintenance system for photovoltaic panels in photovoltaic power plants, comprising:

[0007] A drone nest, a drone, a visible light + infrared camera and a QR code sticker, wherein the drone carries a visible light + infrared camera, and the visible light + infrared camera is used to shoot a target image; the drone nest is used to receive a control instruction issued by a server of a photovoltaic power station, and send the control instruction to the drone; the drone is used to fly to each photovoltaic panel of a designated branch after receiving the instruction from the drone nest, and use the visible light + infrared camera to shoot the target image, and the drone sends the image to the server or uploads it to the drone nest, and the network where the drone nest is located sends it to the server; the QR code sticker is placed at the four corners of each photovoltaic panel and at the male and female heads of the interface;

[0008] The comparison image and patrol point posture input module is used to establish the original comparison image library and the posture database for UAV troubleshooting;

[0009] The fault detection module is used to analyze the output data of the photovoltaic power station collected in real time by the server and monitor branch faults, and send instructions containing the patrol point position to the drone nest when a fault is found.

[0010] Furthermore, the specific process of the comparison picture and patrol point posture input module is as follows:

[0011] Establish an original comparison image library: the drone hovers at the center of each photovoltaic panel, facing the front of all photovoltaic panels in the photovoltaic power station, and uses visible light + infrared cameras to take photos and mark them as original comparison images Ima (ti, bi, si); where ti represents the ti-th image, from 1 to timax, timax is the maximum number of images taken by the current photovoltaic panel (bi, si), when ti = 1, it represents the original normal operation standard image; bi represents the branch number, from 1 to bimax; si represents the number of the photovoltaic panel in the branch, from 1 to simax; the drone uses visible light + infrared cameras to take photos of the interface between each photovoltaic panel and the combiner box, and marks them as original comparison images Imb (ti, bi, si);

[0012] Establish a posture database for UAV troubleshooting: Save the posture Pz(1,bi,si) of the UAV when taking each photo together with Ima and Imb into the database on the server side as the posture Pz(1,bi,si) for UAV troubleshooting.

[0013] Furthermore, the working steps of the fault detection module are as follows:

[0014] Step 1: Normalize the output data by branch: divide all the real output data from each inverter by the real startup capacity of the corresponding inverter to obtain a normalized output data value Ptn1(bi,ti), where bi represents the bi-th inverter branch and ti represents the ti-th time point;

[0015] Step 2: Remove the data during the periods other than sunrise and sunset and maintenance periods to obtain the output data value Ptn2(bi,ti);

[0016] Step 3: Conduct preliminary fault analysis of the photovoltaic panels and send instructions to the drone;

[0017] Step 4. The drone performs the inspection task at the appropriate time according to the task instructions: after arriving at the designated location, the drone hovers over the photovoltaic panel for a period of time, uses the downdraft to blow away the dust deposited on the photovoltaic panel, and then takes images to obtain Ima2(ti,bi,si,ci) and Imb2(ti,bi,si), where Ima2 is the front image of the photovoltaic panel, and Imb2 is the image of the interface between each photovoltaic panel and the junction box. Ima2 and Imb2 are both four-channel ci, which are red, green, blue and infrared of visible light, and ci are 1, 2, 3, and 4 respectively; ti is the time, bi is the inverter branch, and si is the photovoltaic panel number on the bi inverter branch; the drone sends the image to the server;

[0018] Step 5: Analyze the fault of the photovoltaic panel, obtain the judgment result and analyze the judgment result;

[0019] Step 6, connector fault judgment: find the QR code on the male and female connectors and calculate the center position, and calculate the Displug by the distance between the two centers in pixels;

[0020] If Displug>TDisplug, the connector may be faulty, and TDisplug is 5mm.

[0021] Furthermore, the step 3 comprises the following steps:

[0022] Step 31: For each branch bi, the gradient of the force data value at time ti is calculated separately, that is:

[0023] DrPtn(bi,ti)=Ptn2(bi,ti)-Ptn2(bi,(ti-1));

[0024] Then calculate the average value avgDrPtn(ti) and variance varDrPtn(ti) of each valid branch of DrPtn(bi,ti) at time ti;

[0025] Step 32: Using the 3S standard, analyze the output change of each branch at time ti;

[0026] If present:

[0027] DrPtn(bi,ti)>avgDrPtn(ti)+3*varDrPtn(ti);

[0028] or DrPtn(bi,ti) <avgDrPtn(ti)+3*varDrPtn(ti);

[0029] Then further analysis is performed on the branch;

[0030] Step 33: Based on step 32, at a time point where a failure may occur, calculate the average value avgPtn2(ti) and the variance varPtn2(ti) for all Ptn2(bi,ti);

[0031] If present:

[0032] Ptn2(bi,ti)>avgPtn2(ti)+3*varPtn2(ti);

[0033] or Ptn2(bi,ti) <avgPtn2(ti)+3*varPtn2(ti);

[0034] Then further analysis is performed on the branch;

[0035] Step 34: If the output of an inverter changes at three consecutive time points in accordance with the judgment of step 33, proceed to the next step; if the output of an inverter changes at only one or two time points out of the three time points in accordance with the judgment of step 33, return to step 31;

[0036] Step 35: If the output data in step 34 is maintained for 3 consecutive times:

[0037] Ptn2(bi,ti)<5%*avgPtn(ti);

[0038] If not, mark fault 1 and proceed to the next step;

[0039] Step 36, mark the branch number bi of the fault point, call the original posture Pz(1,bi,si) of all photovoltaic panels si on the branch and send it to the drone nest, and the drone nest sends a drone to perform the inspection task.

[0040] Further, the step 5 comprises the following steps:

[0041] Step 51, preprocessing, the specific steps are as follows:

[0042] Step 511, corner alignment: identify the centers of the four QR codes on the four corners of the photovoltaic panel, and then affine the quadrilateral formed by the four corners into a front-view rectangle. During installation, each photovoltaic panel is in normal working condition, and the affine rectangle Rec1_Ima(ti,bi,si) is extracted from Imb(ti,bi,si); then the photovoltaic panel on the branch with possible faults in step 4 is photographed, and the affine rectangle Rec2_Ima(ti,bi,si) is extracted from Ima2(ti,bi,si);

[0043] Step 512, histogram equalization: using limited contrast adaptive histogram equalization CLAHE, respectively obtaining Rhe1_Ima(ti,bi,si) and Rhe2_Ima(ti,bi,si) from Rec1_Ima(ti,bi,si) and Rec2_Ima(ti,bi,si);

[0044] Step 52: Machine training. The specific steps are as follows:

[0045] Step 521, sample set generation: Cracks are carved in the copper wires between solar cells in a standard photovoltaic panel, and 500 different fault samples and 500 daily normal samples are collected to form a data set;

[0046] Step 522, machine learning algorithm training: frame difference between two images:

[0047] Rhe3_Ima(ti,bi,si)=Rhe1_Ima(ti,bi,si)-Rhe2_Ima(ti,bi,si);

[0048] Then, VGG19 is used to conduct supervised learning on Rhe3_Ima(ti,bi,si): 500 normal samples in the data set are randomly frame-differed to extract 200 comparison samples S1 without front and back faults; 500 daily fault samples and 500 normal samples in the data set are randomly frame-differed to extract 200 comparison samples S2 with front and back faults; then, a VGG19 network M1 after supervised learning is trained for comparison samples S1 and comparison samples S2;

[0049] Step 523, actual judgment: On the server side, for all photovoltaic panels si on the branch bi that may have a fault, M1 is used to make a judgment on the actual Rhe3_Ima(ti,bi,si) using VGG19, and the judgment result is analyzed.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] The present invention can identify in real time whether a branch circuit at the field end is faulty by performing data detection and classification at the server end. Once a fault is detected, the system will automatically and intelligently analyze the main fault types of the photovoltaic panel at the field end. This intelligent operation and maintenance method not only significantly reduces the workload of the operation and maintenance personnel, but also greatly improves the speed of fault response and the intelligent level of operation and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 Schematic diagram of the connection relationship of various hardware in the present invention.

[0053] Figure 2 It is a working flow chart of the fault discovery module in the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention 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 invention and are not used to limit the present invention.

[0055] The specific implementation of the present invention is described in detail below in conjunction with specific embodiments.

[0056] like Figure 1 As shown, an automatic operation and maintenance system for abnormal photovoltaic panels in a photovoltaic power station provided by an embodiment of the present invention includes a hardware part and an algorithm part;

[0057] The hardware part adds a drone nest, drones, visible light + infrared cameras and QR code stickers to the hardware structure of the classic photovoltaic power station. When the algorithm finds a branch fault, the drone is dispatched. The visible light + infrared camera on the drone takes pictures of each photovoltaic panel and sends the images to the server.

[0058] The functions of the added hardware are as follows:

[0059] Drone nest: It has the functions of storing drones, charging drones, receiving control instructions from the server, waking up drones, issuing control instructions to drones, launching drones, and recovering drones. In addition to common functions, in the present invention, the drone nest mainly receives control instructions sent by the server of the photovoltaic power station, requiring the drone to fly to the coordinate point of the branch where a fault may exist, and launching and receiving drones. In the embodiment of the present invention, the drone nest is customized.

[0060] Drone: Receives instructions from the drone nest, flies to each photovoltaic panel of the designated branch, aims at the target, and uses its own visible light + infrared camera to capture the image of the target. The image can be sent back to the server by the drone, or it can be uploaded to the drone nest after returning to the drone nest, and then sent to the server by the network where the drone nest is located. In the embodiment of the present invention, the drone is a DJI M300RTK drone.

[0061] Visible light + infrared camera: When the drone flies to each photovoltaic panel of the designated branch and aims at the target, it takes an image of the target. In this embodiment, a Zenmuse H20N visible light + infrared camera is used.

[0062] QR code stickers: placed at the four corners of each photovoltaic panel and at the male and female heads of the interface. They are placed manually at the beginning. The content data of the QR code sticker is its number in the database, so that it is easy to align when the algorithm takes the image later.

[0063] The algorithm part includes the comparison picture and patrol point posture input module and the fault detection module;

[0064] The specific process of the comparison picture and patrol point posture input module is as follows:

[0065] When installing the photovoltaic power station, the operation and maintenance personnel clean each photovoltaic panel. The operator manually operates the drone to hover at the center of each photovoltaic panel, facing the front of all photovoltaic panels in the photovoltaic power station, and uses the visible light + infrared camera to take high-resolution visible light + infrared photos, and mark them as the original comparison image Ima (ti, bi, si). At this moment, ti = 1. Among them, ti represents the ti-th image, from 1 to timax, timax is the maximum number of images taken by the current photovoltaic panel (bi, si), when ti = 1, it represents the standard image of the original normal operation; bi represents the branch number, from 1 to bimax; si represents the number of the photovoltaic panel in the branch, from 1 to simax; the two numbers bi and si also indicate which photovoltaic panel si belongs to which branch bi (photovoltaic panel, junction box, inverter). In addition, the drone uses the visible light + infrared camera to take photos of the interface between each photovoltaic panel and the junction box, and marks them as the original comparison image Imb (ti, bi, si);

[0066] The drone's position Pz(1,bi,si) when taking each photo, including GPS, altitude, drone angle, camera angle and other information, is saved together with Ima and Imb in the server-side database as the drone's position Pz(1,bi,si) when troubleshooting. In this way, when troubleshooting, when taking new photos of each photovoltaic panel, the drone can maintain the patrol point position to obtain a picture that is basically consistent with the original comparison image Ima(1,bi,si) and Imb(1,bi,si), so that it can be quickly positioned and reduce the calculation and operation costs of drone repositioning.

[0067] Since photovoltaic panels do not produce power at night and their output becomes weaker when the weather is cloudy, the optimal working time is when the photovoltaic panels produce the maximum power. In this way, the contrast between cracks and normal parts is most obvious, and the analysis results are more reliable. In this embodiment, the time used is around 12:20 noon on a sunny day.

[0068] The fault detection module runs in the platform software on the server platform. According to the time interval set by the user, it continuously analyzes the output data of the photovoltaic power station collected in real time by the platform software to detect branch faults. If a fault is found, a command containing the patrol point position is sent to the drone nest. The drone nest dispatches the drone. After the drone arrives at the target location, it calls and adjusts to the patrol point position in the command, and then uses its visible light + infrared camera to take pictures of each photovoltaic panel and send the image to the server platform for further analysis.

[0069] In an embodiment of the present invention, the input data of the fault detection module is the output data of the photovoltaic power station collected in real time by the platform software on the server platform. Generally, this data is collected every 5 minutes and reported to the server platform from the inverter on the site of the photovoltaic power station through remote communication. The fault detection module analyzes it once every morning.

[0070] like Figure 2 As shown, the working steps of the fault detection module are as follows:

[0071] Step 1. Normalize the output data by branch: In terms of physical structure, the DC power of each branch is finally aggregated to the inverter, which converts it into AC power and connects it to the grid. The inverter also sends the corresponding power data to the server platform. This power comes from photovoltaic panels. Since the photovoltaic panels on each branch are not necessarily installed in exactly the same way, and the tasks scheduled every day are also different, the actual startup capacity is different. The output data of each inverter needs to be normalized to see the changes in its actual performance. Since the output data is collected with the inverter branch as the minimum resolution of the physical object, all the actual output data from each inverter is divided by the actual startup capacity of the corresponding inverter to obtain a normalized output data value Ptn1(bi,ti), where bi represents the bi-th inverter branch and ti represents the ti-th time point;

[0072] Step 2, remove data except for sunrise and sunset and outage time periods: Although the inverter data is collected 24 hours a day, the data between sunrise and sunset is meaningful in terms of daily meteorology, so it is necessary to remove the data from yesterday's sunset to today's sunrise. This method is well known in the art. The outage time period is when the operation and maintenance personnel arrange the operation and maintenance, they input the relevant object number and time period into the system, turn off the input of the inverter, and the inverter has no output, so the data during this period also needs to be removed. In this way, the output data value Ptn2(bi,ti) is obtained;

[0073] Step 3: Conduct a preliminary fault analysis of the photovoltaic panel, including the following steps:

[0074] Step 31: For each branch bi, the gradient of the force data value at time ti is calculated separately, that is:

[0075] DrPtn(bi,ti)=Ptn2(bi,ti)-Ptn2(bi,(ti-1));

[0076] Then calculate the average value avgDrPtn(ti) and variance varDrPtn(ti) of each valid (i.e., excluding the operation and maintenance work in step 2) branch of DrPtn(bi,ti) at time ti;

[0077] Step 32: Using the 3S standard, analyze the output change of each branch at time ti;

[0078] If present:

[0079] DrPtn(bi,ti)>avgDrPtn(ti)+3*varDrPtn(ti);

[0080] or DrPtn(bi,ti) <avgDrPtn(ti)+3*varDrPtn(ti);

[0081] This means that the change of this branch exceeds the reasonable change range of all sites in the overall site, and there may be a fault, which requires further analysis.

[0082] Step 33: Based on step 32, at a time point where a failure may occur, calculate the average value avgPtn2(ti) and the variance varPtn2(ti) for all Ptn2(bi,ti);

[0083] If present:

[0084] Ptn2(bi,ti)>avgPtn2(ti)+3*varPtn2(ti);

[0085] or Ptn2(bi,ti) <avgPtn2(ti)+3*varPtn2(ti);

[0086] This means that the normalized processing of this branch exceeds the reasonable variation range of all sites in the overall site, and there may be a fault, which requires further analysis.

[0087] Step 34: If the output of an inverter changes at three consecutive time points in accordance with the judgment of step 33, proceed to the next step; if the output of an inverter changes at only one or two time points out of the three time points in accordance with the judgment of step 33, return to step 31;

[0088] Step 35: If the output data in step 34 is maintained for 3 consecutive times:

[0089] Ptn2(bi,ti)<5%*avgPtn(ti);

[0090] Then it means that it is likely to be fault 3: the probability of occurrence is relatively small, but the consequence is serious, the output to the combiner box is poor contact, which will cause its power output to be directly zero. Mark fault 3 and go to the next step; if not, then it is probably fault 1, mark fault 1 and go to the next step;

[0091] Step 36, mark the branch number bi of the fault point, call the original posture Pz(1,bi,si) of all photovoltaic panels si on the branch and send it to the drone nest, and the drone nest sends a drone to perform the inspection task.

[0092] Step 4. The drone will perform inspection tasks according to the task instructions at the appropriate time, mainly during the noon time period when the sunshine is better. After arriving at the designated location, the drone will hover over the photovoltaic panel for a period of time (1 minute), use the downdraft to blow away the dust deposited on the photovoltaic panel, and then take images to obtain Ima2 (ti, bi, si, ci) and Imb2 (ti, bi, si), where Ima2 is the front picture of the photovoltaic panel, and Imb2 is the picture of the interface between each photovoltaic panel and the junction box. Ima2 and Imb2 are both four-channel ci, which are visible light red (ci = 1), green (ci = 2), blue (ci = 3), and infrared (ci = 4); ti is the time, bi is the inverter branch, and si is the photovoltaic panel number on the bi inverter branch; then the drone sends the picture to the server;

[0093] Step 5: Analyze the failure of the photovoltaic panel; Principle of analysis: On the photovoltaic panel, as the use time increases, some circuits will have tiny cracks, which will cause the photovoltaic panel circuit to be disconnected, so that the corresponding part of the current cannot be provided, resulting in a partial decrease in the current output. Step 5 includes the following steps:

[0094] Step 51, preprocessing, the specific steps are as follows:

[0095] Step 511, corner alignment: Identify the centers of the four QR codes on the four corners of the photovoltaic panel, and then affine the quadrilateral formed by the four corners into a front-facing rectangle. In this step, the recognition of the QR code, in the embodiment of the present invention, since the type of the QR code is selected by the user, you can refer to the "QR Two-Dimensional Code Image Processing and Detection Algorithm" to select the type of the QR code, and identify the digital semantics of the QR code, and then use the table lookup method in the preset system to find the corresponding photovoltaic panel number and QR code position number according to the digital semantics. Secondly, how to convert an irregular quadrilateral into a front-facing rectangle is well known in the art, and the specific content can refer to the book "Basics of Robotics Machine Vision and Control MATLAB Algorithm".

[0096] During installation, each photovoltaic panel is in normal working condition, and an affine rectangle Rec1_Ima(ti,bi,si) is extracted from Imb(ti,bi,si); then the photovoltaic panel on the branch with possible fault in step 4 is photographed, and an affine rectangle Rec2_Ima(ti,bi,si) is extracted from Ima2(ti,bi,si);

[0097] Step 512, histogram equalization: Since the illumination of the images taken under different weather conditions is different, both images need to be histogram equalized. Since the traditional OTSU histogram may be random, the present invention adopts limited contrast adaptive histogram equalization CLAHE to minimize the inconsistency of the histogram equalization of the two images due to illumination. In this way, Rhe1_Ima(ti,bi,si) is obtained from Rec1_Ima(ti,bi,si), and Rhe2_Ima(ti,bi,si) is obtained from Rec2_Ima(ti,bi,si);

[0098] Step 52: Machine training. The specific steps are as follows:

[0099] Step 521, sample set generation: By manually setting faults, a fault sample set of a photovoltaic panel can be obtained. Use a knife to micro-carve cracks in the copper wires between solar cells in a standard photovoltaic panel, collect 500 different fault samples and 500 daily normal samples to form a data set. Since the specific cracks are not repaired by the operation and maintenance personnel, they only find faulty photovoltaic panels and submit them to the manufacturer. Therefore, in this task, only the task of finding faulty photovoltaic panels is involved.

[0100] Step 522, machine learning algorithm training: Taking into account nonlinear problems such as dust and various deformations, the present invention does not adopt the classic solution similar to the frame difference method and then the threshold, but combines the speed advantage of the frame difference method, and on this basis adopts a solution based on deep learning.

[0101] The specific method is to do traditional frame difference between the two pictures:

[0102] Rhe3_Ima(ti,bi,si)=Rhe1_Ima(ti,bi,si)-Rhe2_Ima(ti,bi,si);

[0103] Then, VGG19 is used to conduct supervised learning on Rhe3_Ima(ti,bi,si): 500 normal samples in the data set are randomly frame-differed to extract 200 comparison samples S1 without front and back faults; 500 daily fault samples and 500 normal samples in the data set are randomly frame-differed to extract 200 comparison samples S2 with front and back faults; then, a VGG19 network M1 after supervised learning is trained for comparison samples S1 and comparison samples S2;

[0104] Step 523, actual judgment: On the server side, for all photovoltaic panels si on the branch bi that may have a fault, M1 is used to judge the actual Rhe3_Ima(ti,bi,si) using VGG19, and the judgment result is analyzed. If a photovoltaic panel is judged to have a problem by M1, then this result is told to the operation and maintenance personnel, who will replace it and hand over the faulty photovoltaic panel to the manufacturer. If there is no problem with the photovoltaic panels on the entire branch suspected of having a fault, then this result is also submitted to the operation and maintenance personnel.

[0105] Regardless of the result, proceed to step 6.

[0106] Step 6, connector fault judgment: find the QR code on the male and female connectors and calculate the center position, and calculate the Displug by the distance between the two centers in pixels;

[0107] If Displug>TDisplug, then there may be a problem with the connector, and the diagram should be submitted to the maintenance personnel. In this embodiment, TDisplug is about 5 mm.

[0108] The above are only preferred embodiments of the present invention. It should be pointed out that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present invention. These should also be regarded as the protection scope of the present invention. These will not affect the effect of the implementation of the present invention and the practicality of the patent.

Claims

1. An automated operation and maintenance system for photovoltaic panels in photovoltaic power plants. It is characterized in that include: A drone nest, a drone, a visible light + infrared camera and a QR code sticker, wherein the drone carries a visible light + infrared camera, and the visible light + infrared camera is used to shoot a target image; the drone nest is used to receive a control instruction issued by a server of a photovoltaic power station, and send the control instruction to the drone; the drone is used to fly to each photovoltaic panel of a designated branch after receiving the instruction from the drone nest, and use the visible light + infrared camera to shoot the target image, and the drone sends the image to the server or uploads it to the drone nest, and the network where the drone nest is located sends it to the server; the QR code sticker is placed at the four corners of each photovoltaic panel and at the male and female heads of the interface; The comparison image and patrol point posture input module is used to establish the original comparison image library and the posture database for UAV troubleshooting; The fault detection module is used to analyze the output data of the photovoltaic power station collected in real time by the server and monitor branch faults, and send instructions containing the patrol point position to the drone nest when a fault is found.

2. The automated operation and maintenance system for photovoltaic panels abnormality in photovoltaic power plants according to claim 1, It is characterized in that The specific process of the comparison picture and patrol point posture input module is as follows: Establish an original comparison image library: the drone hovers at the center of each photovoltaic panel, facing the front of all photovoltaic panels in the photovoltaic power station, and uses visible light + infrared cameras to take photos and mark them as original comparison images Ima (ti, bi, si); where ti represents the ti-th image, from 1 to timax, timax is the maximum number of images taken by the current photovoltaic panel (bi, si), when ti = 1, it represents the original normal operation standard image; bi represents the branch number, from 1 to bimax; si represents the number of the photovoltaic panel in the branch, from 1 to simax; the drone uses visible light + infrared cameras to take photos of the interface between each photovoltaic panel and the combiner box, and marks them as original comparison images Imb (ti, bi, si); Establish a posture database for UAV troubleshooting: Save the posture Pz(1,bi,si) of the UAV when taking each photo together with Ima and Imb into the database on the server side as the posture Pz(1,bi,si) for UAV troubleshooting.

3. The automated operation and maintenance system for photovoltaic panels abnormality in photovoltaic power plants according to claim 1, It is characterized in that The working steps of the fault detection module are as follows: Step 1: Normalize the output data by branch: divide all the real output data from each inverter by the real startup capacity of the corresponding inverter to obtain a normalized output data value Ptn1(bi,ti), where bi represents the bi-th inverter branch and ti represents the ti-th time point; Step 2: Remove the data during the periods other than sunrise and sunset and maintenance periods to obtain the output data value Ptn2(bi,ti); Step 3: Conduct preliminary fault analysis of the photovoltaic panels and send instructions to the drone; Step 4. The drone performs the inspection task at the appropriate time according to the task instructions: after arriving at the designated location, the drone hovers over the photovoltaic panel for a period of time, uses the downdraft to blow away the dust deposited on the photovoltaic panel, and then takes images to obtain Ima2(ti,bi,si,ci) and Imb2(ti,bi,si), where Ima2 is the front image of the photovoltaic panel, and Imb2 is the image of the interface between each photovoltaic panel and the junction box. Ima2 and Imb2 are both four-channel ci, which are red, green, blue and infrared of visible light, and ci are 1, 2, 3, and 4 respectively; ti is the time, bi is the inverter branch, and si is the photovoltaic panel number on the bi inverter branch; the drone sends the image to the server; Step 5: Analyze the fault of the photovoltaic panel, obtain the judgment result and analyze the judgment result; Step 6, connector fault judgment: find the QR code on the male and female connectors and calculate the center position, and calculate the Displug by the distance between the two centers in pixels; If Displug>TDisplug, the connector may be faulty, and TDisplug is 5mm.

4. The automated operation and maintenance system for photovoltaic panels abnormalities in photovoltaic power plants according to claim 3, It is characterized in that The step 3 comprises the following steps: Step 31: For each branch bi, the gradient of the force data value at time ti is calculated separately, that is: DrPtn(bi,ti)=Ptn2(bi,ti)-Ptn2(bi,(ti-1)); Then calculate the average value avgDrPtn(ti) and variance varDrPtn(ti) of each valid branch of DrPtn(bi,ti) at time ti; Step 32: Using the 3S standard, analyze the output change of each branch at time ti; If present: DrPtn(bi,ti)>avgDrPtn(ti)+3*varDrPtn(ti); or DrPtn(bi,ti) <avgDrPtn(ti)+3*varDrPtn(ti); Then further analysis is performed on the branch; Step 33: Based on step 32, at a time point where a failure may occur, calculate the average value avgPtn2(ti) and the variance varPtn2(ti) for all Ptn2(bi,ti); If present: Ptn2(bi,ti)>avgPtn2(ti)+3*varPtn2(ti); or Ptn2(bi,ti) <avgPtn2(ti)+3*varPtn2(ti); Then further analysis is performed on the branch; Step 34: If the output of an inverter changes at three consecutive time points in accordance with the judgment of step 33, proceed to the next step; if the output of an inverter changes at only one or two time points out of the three time points in accordance with the judgment of step 33, return to step 31; Step 35: If the output data in step 34 is maintained for 3 consecutive times: Ptn2(bi,ti)<5%*avgPtn(ti); If not, mark fault 1 and proceed to the next step; Step 36, mark the branch number bi of the fault point, call the original posture Pz(1,bi,si) of all photovoltaic panels si on the branch and send it to the drone nest, and the drone nest sends a drone to perform the inspection task.

5. The automated operation and maintenance system for photovoltaic panels abnormality in photovoltaic power plants according to claim 3, It is characterized in that The step 5 comprises the following steps: Step 51, preprocessing, the specific steps are as follows: Step 511, corner alignment: identify the centers of the four QR codes on the four corners of the photovoltaic panel, and then affine the quadrilateral formed by the four corners into a front-view rectangle. During installation, each photovoltaic panel is in normal working condition, and the affine rectangle Rec1_Ima(ti,bi,si) is extracted from Imb(ti,bi,si); then the photovoltaic panel on the branch with possible faults in step 4 is photographed, and the affine rectangle Rec2_Ima(ti,bi,si) is extracted from Ima2(ti,bi,si); Step 512, histogram equalization: using limited contrast adaptive histogram equalization CLAHE, respectively obtaining Rhe1_Ima(ti,bi,si) and Rhe2_Ima(ti,bi,si) from Rec1_Ima(ti,bi,si) and Rec2_Ima(ti,bi,si); Step 52: Machine training. The specific steps are as follows: Step 521, sample set generation: Cracks are carved in the copper wires between solar cells in a standard photovoltaic panel, and 500 different fault samples and 500 daily normal samples are collected to form a data set; Step 522, machine learning algorithm training: frame difference between two images: Rhe3_Ima(ti,bi,si)=Rhe1_Ima(ti,bi,si)-Rhe2_Ima(ti,bi,si); Then, VGG19 is used to conduct supervised learning on Rhe3_Ima(ti,bi,si): 500 normal samples in the data set are randomly frame-differed to extract 200 comparison samples S1 without front and back faults; 500 daily fault samples and 500 normal samples in the data set are randomly frame-differed to extract 200 comparison samples S2 with front and back faults; then, a VGG19 network M1 after supervised learning is trained for comparison samples S1 and comparison samples S2; Step 523, actual judgment: On the server side, for all photovoltaic panels si on the branch bi that may have a fault, M1 is used to make a judgment on the actual Rhe3_Ima(ti,bi,si) using VGG19, and the judgment result is analyzed.

Citation Information

Cited By

  • Photovoltaic power station safety monitoring method and system based on safety inspection

    CN119543823A