Photovoltaic module fault navigation correction method and system

Abstract

The present invention discloses a photovoltaic module fault navigation and correction method and system, which relates to the field of photovoltaic power generation and operation and maintenance technology, including: using drones to collect photovoltaic field images, generating and splicing visible light tile base maps, and forming a photovoltaic field map with a traversable path; using drones for infrared inspections, and identifying and locating photovoltaic module faults based on infrared defect recognition algorithms to identify and analyze infrared inspection images; sending photovoltaic module fault images and positioning information to the operation and maintenance navigation mobile terminal of the operation and inspection personnel, combining the real-time position of the operation and inspection personnel with dynamic correction, and accurately navigating to the fault location. The method of the present invention solves the problem that the terrain of the photovoltaic field is rugged and changeable, the arrangement of strings is complex, and the map software cannot provide path navigation. At the same time, a dynamic position correction method is used to solve the problem that the field is located in a remote area, the satellite signal is weak, and the positioning jump deviation is large from time to time, so that the operation and inspection personnel can more accurately navigate to the target location during the photovoltaic fault inspection.

Description

Technical Field

[0001] The present invention relates to the field of photovoltaic power generation and operation and maintenance technology, and in particular to a photovoltaic component fault navigation and correction method and system. Background Art

[0002] In the field of photovoltaic field fault inspection, with the development of technology in recent years, the trend of intelligence, automation and digitization has become increasingly obvious. In particular, the introduction of unmanned aerial vehicle (UAV) technology has brought revolutionary changes to the inspection of wind power and photovoltaic fields. UAV inspection technology has become an important tool for inspection of new energy fields with its high efficiency, low cost and flexibility, and has greatly promoted the intelligence and automation of new energy technology operation and maintenance. By planning the inspection route of drones, taking visible light and infrared images of photovoltaic panels in the photovoltaic field, and combining image recognition technology, abnormal conditions in the field can be quickly discovered. However, drones cannot replace manual inspection and maintenance of faults, so field operation and maintenance personnel are still required to participate in operation and maintenance. However, since photovoltaic fields usually cover a large area, have rugged and changeable terrain, and have complex string arrangement, various map software cannot perform path navigation within the photovoltaic field. In addition, the photovoltaic field is located in a remote area, the satellite signal is weak, and the positioning position often jumps, resulting in large deviations. This brings great difficulties for operation and maintenance personnel to navigate to the location of the faulty panel. Summary of the invention

[0003] In view of the above-mentioned problems, the present invention is proposed.

[0004] Therefore, the technical problem solved by the present invention is that the photovoltaic field is located in a remote area, the satellite signal is weak, and the positioning position often jumps, resulting in large deviations.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for navigation and correction of photovoltaic module faults, comprising: using a drone to collect photovoltaic field images, generating and splicing visible light tile base maps, and forming a photovoltaic field map with a traversable path; using a drone to perform infrared inspections, and analyzing the infrared inspection images based on an infrared defect recognition algorithm to identify and locate photovoltaic module faults; sending the photovoltaic module fault image and positioning information to the operation and maintenance navigation mobile terminal of the operation and maintenance personnel, combining the real-time position of the operation and maintenance personnel with dynamic correction, and accurately navigating to the fault location.

[0006] As a preferred solution of the photovoltaic component fault navigation and correction method described in the present invention, wherein: the photovoltaic field area image includes reference points evenly distributed in the entire photovoltaic field area to ensure that the reference points cover the entire shooting area, the reference point density is A, additional reference points are arranged in key areas, the reference point density in key areas is 1.5A, a low-resolution UAV pre-flight is performed to obtain a preliminary image, and the area with drastic terrain changes is judged through preliminary image analysis. The reference point density in the area with drastic terrain changes is 2A, and each reference point is measured using a high-precision GPS device to obtain and record precise geographic coordinates to form a reference point layout document; according to a predetermined flight plan, the UAV takes orthophoto images over the photovoltaic field area to obtain a photovoltaic field area image.

[0007] As a preferred solution of the photovoltaic module fault navigation and correction method described in the present invention, the visible light tile base map includes stitching the orthophoto image into a complete base map through image stitching, calculating the SIFT feature to obtain scale invariance, and then extracting the SURF feature to enhance the stability of the feature point, which is expressed as:

[0008] C(x,y,σ)=α·S(x,y,σ)+β·H(x,y)

[0009] Among them, C(x, y, σ) represents the response value of the composite feature point; S(x, y, σ) represents the SIFT feature point detection result; H(x, y) represents the SURF feature point detection result; α represents the weight factor; β represents the weight factor; the FLANN algorithm is used to construct an approximate nearest neighbor index, the extracted feature points are preliminarily matched, the preliminary error threshold is set, the RANSAC algorithm is used to filter the wrong matching points, and the dynamic threshold is set to dynamically adjust the error tolerance according to the current matching quality, and the matching points are further screened, which is expressed as:

[0010]

[0011] Among them, T d (k) represents the dynamic threshold of the kth iteration; D(f i1 ,f i2 ) represents the distance of the matching points, T i represents the preliminary error threshold; f i1 represents the feature point descriptor; f i2 Represents the feature point descriptor; the affine transformation matrix of the image is calculated using the screened matching points to accurately align the images, and the aligned images are fused at multiple scales to build a pyramid model, and the images are fused layer by layer to obtain a complete base map, which is expressed as:

[0012]

[0013] Among them, If (x, y) represents the stitched image; L represents the total number of layers; I l (x, y) represents the image of layer l; G l (x, y) represents the weight of the Gaussian pyramid at the first layer; the spliced ​​base map is cut into multiple small tiles, each tile represents a part of the base map, and the tiles are organized according to the zoom level to form a multi-level visible light tile base map, which is then imported into the photovoltaic inspection system platform.

[0014] As a preferred solution of the photovoltaic component fault navigation and correction method described in the present invention, wherein: the photovoltaic field map includes, using the U-Net model to process the visible light tile base map, outputting a segmentation mask, each pixel point is assigned a category label in the segmentation mask, and the category labels are divided into photovoltaic panels, channels and obstacles; by analyzing the segmentation mask, the channels between the photovoltaic panels are identified, the channels between the photovoltaic panels are extracted as independent pixel sets, the actual width of each channel is calculated, and the preset minimum passable width standard M is used for judgment. If the actual width of the channel is greater than or equal to M, it is marked as a passable channel, and if the actual width of the channel is less than M, it is marked as an obstacle; the area with drastic terrain changes judged by the preliminary image is marked as an obstacle, if there is a road built in the area with drastic terrain changes Then mark the road as a passable channel; create a two-dimensional grid map with the same size as the segmentation mask, for each pixel, mark the passable channel as 1, and mark the obstacle and the photovoltaic field boundary as 0, traverse each pixel of the grid map, and when a pixel marked as 1 is found and has not been visited, add this pixel as the initial unit to the queue and mark it as visited, take a unit from the queue, check the adjacent units in four directions, if the adjacent unit is passable and has not been visited, add the adjacent unit to the queue and mark it as visited; repeat the process of traversing each pixel of the grid map and checking the adjacent units until all passable areas are traversed, summarize all path segments, obtain all traversable paths, overlay the traversable path data on the visible light tile base map, and form a photovoltaic field map with traversable paths.

[0015] As a preferred solution of the photovoltaic component fault navigation and correction method described in the present invention, the use of the drone for infrared inspection includes: the drone mobile inspection terminal receives the inspection task including the flight path and the target area, automatically takes off and flies according to the predetermined path, uses the infrared camera to perform real-time image acquisition, and obtains the infrared inspection image; the infrared inspection image is transmitted back to the photovoltaic inspection system platform in real time, the fault area identified by the infrared defect recognition algorithm is accurately located, the EXIF ​​information of the infrared inspection image is used to determine the specific location of the fault in the photovoltaic field area, and a fault report including the fault type and location is generated.

[0016] As a preferred solution of the photovoltaic module fault navigation and correction method of the present invention, the infrared defect recognition algorithm includes identifying hot spots and cold spots on the panel by analyzing the temperature distribution of the infrared inspection image, which is expressed as:

[0017] T hot >T avg +k hot ·σ T

[0018] T cold <T avg -k cold ·σ T

[0019] Among them, T hot Indicates hot spot temperature; k hot Represents the coefficient for adjusting the sensitivity of hotspot detection; T cold Indicates the cold spot temperature; k cold Indicates the coefficient for adjusting the cold spot detection sensitivity; T avg Represents the average temperature of the entire photovoltaic panel; σ T Represents the standard deviation of the temperature of the entire photovoltaic panel; the temperature is greater than or equal to T hot The photovoltaic panel area is the hot spot area, and the area of ​​the hot spot area is A hot , the maximum temperature in the hot spot area is Temperature is less than or equal to T cold The photovoltaic panel area is the cold spot area, and the area of ​​the cold spot area is A cold , the lowest temperature in the cold spot area is

[0020] Determine the fault type and short circuit based on the identified hot and cold spots:

[0021]

[0022] Open circuit judgment:

[0023]

[0024] Hot spot judgment:

[0025] Among them, P short Indicates the probability score of short circuit fault; P open Indicates the probability score of open circuit fault; P hotspot represents the probability score of hot spot failure; w 1 ,……,w 7 Represents the weight coefficient; A total Represents the total area of ​​photovoltaic panels; Represents temperature gradient; A thresholdIndicates the preset hot spot area threshold; when P short >max(P open ,P hotspot ) and P short >threshold short When P open >max(P short ,P hotspot ) and P open >threshold open When P hotspot >max(P short ,P open ) and P hotspot >threshold hotspot , judged as a hot spot fault; where threshold short Indicates the threshold for short-circuit fault judgment; threshold open Indicates the threshold for open circuit fault judgment; threshold hotspot Indicates the threshold for hot spot fault judgment.

[0026] As a preferred solution of the photovoltaic module fault navigation correction method described in the present invention, the dynamic correction includes that when the maintenance personnel start navigation through the operation and maintenance navigation mobile terminal, they initialize and obtain positioning data and wait for the positioning data to stabilize. At this time, the correction position s 1 To display the position, the correction position s 1 Equal to the positioning position s 0 ; After the first move, use the positioning position s 0 and correction position s 1 The subtraction result s is compared with the empirical jump threshold a. When s>a, it indicates that the positioning is unstable and reinitializes to obtain stable positioning data. When s≤a, it is considered that the positioning is normal and continues; the correction position s is assigned 1 Equal to the positioning position s 0 , record the speed v at the previous moment 1 and time t 1 ; After the next move, the dynamic jump threshold is v 1 ×t 1 If the target position has been reached, the process ends. If the target position has not been reached, the process continues, using the positioning position s 0 and correction position s 1 The subtraction result s and the dynamic jump threshold v 1 ×t 1 Compare, when s>v 1 ×t 1 It is judged that the moving distance exceeds the threshold and recorded. When s≤v 1 ×t1 Then the positioning is judged to be normal, and the correction position s is reassigned 1 Equal to the positioning position s 0 And record the speed and time of the last moment; when the moving distance recorded for n consecutive times exceeds the threshold, reinitialize to obtain stable positioning data; when the moving distance exceeds the threshold but does not meet the requirement of n consecutive records, assign the correction position s based on the moving direction 1 =s 1 +v 1 ×t 1 , record the speed v at the previous moment 1 and time t 1 , calculate the dynamic jump threshold again and determine whether the positioning is normal.

[0027] In the second aspect, the present invention also provides a photovoltaic component fault navigation and correction system, including: a path generation module, which stitches the photovoltaic field image into a complete base map through image stitching, cuts and organizes the stitched base map into a multi-level visible light tile base map, and uses a U-Net model to mark the navigable channels to form a photovoltaic field map with a traversable path; a fault identification module, which performs temperature distribution analysis through infrared defect recognition based on the infrared inspection images obtained by drone inspections, identifies photovoltaic panel faults, and uses the EXIF ​​information of the infrared inspection images to locate the fault location and generate a fault report; a navigation module, which dynamically corrects the fault based on the photovoltaic panel fault and positioning information after each movement in combination with the real-time position of the inspection personnel, and accurately navigates to the fault location.

[0028] In a third aspect, the present invention further provides a computing device, comprising: a memory and a processor;

[0029] The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the photovoltaic component fault navigation and correction method are implemented.

[0030] In a fourth aspect, the present invention further provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the photovoltaic component fault navigation and correction method.

[0031] Beneficial effects of the present invention: The fault inspection path navigation method provided by the present invention solves the problem that the terrain of the photovoltaic field is rugged and changeable, the string arrangement is complex, and the map software cannot provide path navigation. At the same time, the dynamic position correction method is used to solve the problem that the field is located in a remote area, the satellite signal is weak, and the positioning jump deviation is often large, so that the operation and maintenance personnel can navigate to the target location more accurately during the photovoltaic fault inspection. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0033] Figure 1 An overall flow chart of a photovoltaic module fault navigation and correction method provided by an embodiment of the present invention;

[0034] Figure 2 A structural diagram of a correction system for a photovoltaic module fault navigation correction method provided by a second embodiment of the present invention;

[0035] Figure 3 A flowchart of a correction method for a photovoltaic module fault navigation correction method provided by the second embodiment of the present invention. DETAILED DESCRIPTION

[0036] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0037] Example 1, reference Figure 1-3 , as an embodiment of the present invention, provides a photovoltaic module fault navigation and correction method, comprising:

[0038] S1: Use drones to collect images of the photovoltaic field, generate and stitch visible light tile base maps, and form a photovoltaic field map with a navigable path.

[0039] Furthermore, drones are used to take orthophotos of the photovoltaic area and collect images of the photovoltaic area. The orthophoto shooting corrects the image distortion caused by terrain undulations and camera tilt during the photography process to obtain images with real geographic coordinates of each point on the ground, which is suitable for the complex terrain environment of the photovoltaic area.

[0040] To correct the distortion, it is necessary to set up geographic reference points in the photovoltaic field in advance. First, evenly distribute the geographic reference points in the entire photovoltaic field to ensure that the reference points cover the entire shooting area. The reference point density is A.

[0041] Additional reference points are set up in key locations such as areas where photovoltaic panels are concentrated, site boundaries, and around important facilities. The density of reference points in key areas is 1.5A. Key areas are the core parts of photovoltaic sites. The accuracy of images directly affects the inspection and maintenance of photovoltaic modules. Adding reference points in these areas can improve the accuracy and reliability of images.

[0042] Conduct a low-resolution UAV pre-flight to obtain preliminary images. Through topographic map analysis, on-site surveys and UAV pre-flights, determine the areas with large terrain undulations, such as hillsides, gullies, etc. The reference point density in areas with drastic terrain changes is 2A. Areas with drastic terrain changes are prone to geometric distortion of images. Increasing the reference point density can provide more correction data, significantly improve the geometric accuracy of images, and ensure the accuracy of images in these areas.

[0043] Place geographic reference points at selected locations. Use highly reflective materials or brightly colored markers to ensure they are clearly visible in the image. Use high-precision GPS equipment to measure each reference point, obtain precise geographic coordinates (latitude, longitude, and elevation) and record them. Record each reference point's coordinate information, location description, marker features, etc. to form a detailed reference point placement document.

[0044] Import the images taken by the drone into the image processing software for preliminary data processing, and use the automatic alignment function of the image processing software to perform preliminary alignment of the images to ensure that the approximate position and direction of the images are consistent.

[0045] Import the coordinate information of the laid out geographic reference points (GCPs) into the image processing software, manually calibrate each geographic reference point in the image, ensure that the position of the reference point in the image matches the actual coordinates, use the geometric correction function of the image processing software, and perform geometric correction on the image according to the calibrated GCPs to eliminate geometric distortion in the image and ensure the geographic accuracy of the image.

[0046] Furthermore, the geometrically corrected images are stitched together to form a complete base map of the photovoltaic field. Combining the advantages of SIFT and SURF, the feature points of the photovoltaic panels are detected and enhanced. First, the SIFT features are calculated to obtain scale invariance, and then the SURF features are extracted to enhance the stability of the feature points. By adjusting the weight factor, the detection result is more accurate, which is expressed as:

[0047] C(x,y,σ)=α·S(x,y,σ)+β·H(x,y)

[0048] Among them, C(x, y, σ) represents the composite feature point response value; S(x, y, σ) represents the SIFT feature point detection result; H(x, y) represents the SURF feature point detection result; α represents the weight factor; β represents the weight factor.

[0049] Use the FLANN algorithm to build an approximate nearest neighbor index, perform preliminary matching on the extracted feature points, set a preliminary error threshold, use the RANSAC algorithm to filter out incorrect matching points, set a dynamic threshold to dynamically adjust the error tolerance according to the current matching quality, and further screen the matching points, which is expressed as:

[0050]

[0051] Among them, T d (k) represents the dynamic threshold of the kth iteration; D(f i1 ,f i2 ) represents the distance of the matching points, T i represents the preliminary error threshold; f i1 represents the feature point descriptor; f i2 Represents the feature point descriptor.

[0052] The affine transformation matrix of the image is calculated using the screened matching points to accurately align the images. The aligned images are fused at multiple scales to build a pyramid model. The images are fused layer by layer to obtain a complete base map, which is expressed as:

[0053]

[0054] Among them, I f (x, y) represents the stitched image; L represents the total number of layers; I l (x, y) represents the image of layer l; G l (x,y) represents the weight of the Gaussian pyramid at level l.

[0055] The spliced ​​base map is cut into multiple small tiles, each tile represents a part of the base map, and the tiles are organized according to the zoom level to form a multi-level visible light tile base map, which is then imported into the photovoltaic inspection system platform.

[0056] Furthermore, in the process of path generation in the photovoltaic field, the image region segmentation is first required. The image segmentation task is completed using the U-Net model. U-Net is a convolutional neural network (CNN) architecture consisting of an encoder and a decoder. It can effectively capture the contextual information of the image and is suitable for segmentation tasks in the fields of medical imaging and satellite imaging. The advantage of U-Net is that its encoder-decoder structure can capture the global and local features of the image and is suitable for processing complex scenes.

[0057] Collect annotated photovoltaic field image datasets, including annotations of tiles, channels, and obstacles, and use the U-Net model for training. Input the original image containing annotations and output the segmentation mask to identify tiles, channels, and obstacles. The loss function uses cross entropy loss to optimize the segmentation accuracy of the model.

[0058] The trained U-Net model is used to process the visible light tile base map and output a segmentation mask. Each pixel in the segmentation mask is assigned a category label, which is divided into photovoltaic panels, channels, and obstacles.

[0059] By analyzing the segmentation mask, the channels between photovoltaic panels are identified, and the channels between photovoltaic panels are extracted as independent pixel sets. The actual width of each channel is calculated, and the preset minimum passable width standard M is used for judgment. If the actual width of the channel is greater than or equal to M, it is marked as a passable channel. If the actual width of the channel is less than M, it is marked as an obstacle.

[0060] The areas with drastic terrain changes judged by the preliminary image are marked as obstacles. If there are roads built in the areas with drastic terrain changes, the roads are marked as passable passages.

[0061] Create a two-dimensional grid map with the same size as the segmentation mask. For each pixel, mark the traversable channel as 1 and the obstacle and photovoltaic field boundary as 0. Traverse each pixel of the grid map. When a pixel marked as 1 is found and has not been visited, add this pixel as the initial unit to the queue and mark it as visited. Take a unit from the queue and check the adjacent units in four directions. If the adjacent unit is traversable and has not been visited, add the adjacent unit to the queue and mark it as visited.

[0062] The process of traversing each pixel of the raster map and checking adjacent cells is repeated until all traversable areas are traversed, all path segments are summarized, all traversable paths are obtained, and the traversable path data are overlaid on the visible light tile base map to form a photovoltaic field map with traversable paths.

[0063] S2: Use drones for infrared inspections, analyze infrared inspection images based on infrared defect recognition algorithms, identify and locate photovoltaic module faults.

[0064] Furthermore, the mobile inspection terminal receives an inspection task including a flight path and a target area, automatically takes off and flies along a predetermined path, and uses an infrared camera to collect real-time images to obtain infrared inspection images.

[0065] The infrared inspection image is transmitted back to the photovoltaic inspection system platform in real time, and the fault area is accurately located based on the infrared defect recognition algorithm. The EXIF ​​information of the infrared inspection image is used to determine the specific location of the fault in the photovoltaic field area, and a fault report containing the fault type and location is generated.

[0066] Furthermore, photovoltaic panels include multiple fault types, among which short circuit, open circuit and hot spot are relatively common and can be easily identified through infrared inspection images. Therefore, the fault type identification mentioned in the present invention is carried out for these three faults.

[0067] By analyzing the temperature distribution of infrared inspection images, hot spots and cold spots on the board can be identified, which can be expressed as:

[0068] T hot >T avg +k hot ·σ T

[0069] T cold <T avg -k cold ·σ T

[0070] Among them, T hot Indicates hot spot temperature; k hot Represents the coefficient for adjusting the sensitivity of hotspot detection; T cold Indicates the cold spot temperature; k cold Indicates the coefficient for adjusting the cold spot detection sensitivity; T avg Represents the average temperature of the entire photovoltaic panel; σ T Represents the standard deviation of the temperature of the entire photovoltaic panel.

[0071] Temperature greater than or equal to T hot The photovoltaic panel area is the hot spot area, and the area of ​​the hot spot area is A hot , the maximum temperature in the hot spot area is Temperature is less than or equal to T cold The photovoltaic panel area is the cold spot area, and the area of ​​the cold spot area is A cold , the lowest temperature in the cold spot area is

[0072] The temperature gradient of photovoltaic panels is calculated using the central difference method combined with boundary processing. For the interior of photovoltaic panels, the central difference method has higher accuracy (second-order accuracy) than the forward or backward difference method, and can capture temperature changes more accurately. For the edge of the photovoltaic panel, forward or backward difference can be used to ensure that the gradient can be calculated for the entire panel.

[0073] For a temperature matrix, we need to calculate for each point and Use central difference for interior points:

[0074]

[0075] Use forward or backward differencing on boundary points:

[0076] Left border:

[0077] Right border:

[0078] Upper Boundary:

[0079] Lower Boundary:

[0080] Wherein, T represents temperature; Δx represents the distance between adjacent data points in the x direction; Δy represents the distance between adjacent data points in the y direction; T(i+1,j) represents the temperature value at position (i,j); n represents the maximum index value in the x direction; m represents the maximum index value in the y direction.

[0081] According to the calculation results, two new matrices with the same size as the original temperature matrix are obtained. For each point (i, j), the corresponding position is used and Value to calculate the gradient magnitude

[0082]

[0083] The fault type is determined based on the identified hot spots and cold spots. Short circuits usually cause local hot spots and significant increases in local temperature. Therefore, the larger the proportion of the hot spot area to the total area, the higher the possibility of a short circuit. At the same time, short circuits can cause drastic changes in temperature. A large temperature gradient is also an indicator of a short circuit. Therefore, short circuit judgment is expressed as:

[0084]

[0085] Open circuit usually leads to a decrease in local temperature, so the larger the proportion of cold spot area to the total area, the more significant the local temperature drop, and the higher the possibility of open circuit. The open circuit judgment is expressed as:

[0086]

[0087] When the hot spot area approaches or exceeds the threshold, the comparison between the hot spot area and the threshold area becomes smaller, indicating that it is more likely to be a hot spot. Secondly, hot spots usually show that the local temperature is significantly higher than the average temperature. Therefore, the hot spot judgment is expressed as:

[0088]

[0089] Among them, Pshort Indicates the probability score of short circuit fault; P open Indicates the probability score of open circuit fault; P hotspot represents the probability score of hot spot failure; w 1 ,……,w 7 Represents the weight coefficient; A total represents the total area of ​​photovoltaic panels; Represents temperature gradient; A threshold Indicates the preset hot spot area threshold.

[0090] When P short >max(P open ,P hotspot ) and P short >threshold short It is judged as a short circuit fault.

[0091] When P open >max(P short ,P hotspot ) and P open >threshold open When , it is judged as an open circuit fault.

[0092] When P hotspot >max(P short ,P open ) and P hotspot >threshold hotspot , judged as a hot spot failure.

[0093] Among them, threshold short Indicates the threshold for short-circuit fault judgment; threshold open Indicates the threshold for open circuit fault judgment; threshold hotspot Indicates the threshold for hot spot fault judgment.

[0094] It should be noted that in actual situations, more changes may occur. For example, when two power values ​​are equal and greater than the third, manual inspection is required according to the specific equipment, or auxiliary sensor data is used to make a more precise judgment. short and P hotspot If both thresholds are exceeded at the same time, there may be a combined short circuit and hot spot fault, and the fault with greater impact should be dealt with first. Here, the situation is simplified to illustrate the judgment effect of the present invention on the fault type. When abnormal hot spots or cold spots are found in actual detection, an alarm needs to be issued for further investigation.

[0095] S3: The fault image and positioning information of the photovoltaic module are sent to the operation and maintenance navigation mobile terminal of the operation and maintenance personnel, and the fault location is accurately navigated to by combining the real-time position of the operation and maintenance personnel with dynamic correction.

[0096] Furthermore, the operation and maintenance personnel use the operation and maintenance navigation mobile terminal to obtain the fault pictures and fault locations of the photovoltaic panels from the photovoltaic inspection system platform, and combine their own real-time dynamic correction position and movable path to select the best route to navigate to the target location to complete the fault operation and inspection work.

[0097] like Figure 3 As shown in the figure, the process of dynamic correction includes: when the operation and maintenance personnel start navigation through the operation and maintenance navigation mobile terminal, they initialize to obtain positioning data and wait for the positioning data to stabilize. At this time, the correction position s 1 To display the position, the correction position s 1 Equal to the positioning position s 0 .

[0098] After the first move, use the position s 0 and correction position s 1 The subtraction result s is compared with the empirical jump threshold a. When s>a, it indicates that the positioning is unstable and reinitialization is required to obtain stable positioning data. When s≤a, the positioning is considered normal and continues.

[0099] Assign correction position s 1 Equal to the positioning position s 0 , record the speed v at the previous moment 1 and time t 1 .

[0100] After the next move, the dynamic jump threshold is v 1 ×t 1 If the target position has been reached, the process ends. If the target position has not been reached, the process continues, using the positioning position s 0 and correction position s 1 The subtraction result s and the dynamic jump threshold v 1 ×t 1 Compare, when s>v 1 ×t 1 It is judged that the moving distance exceeds the threshold and recorded. When s≤v 1 ×t 1 Then the positioning is judged to be normal, and the correction position s is reassigned 1 Equal to the positioning position s 0 And record the speed and time of the previous moment.

[0101] When the moving distance recorded for n consecutive times exceeds the threshold, reinitialization is performed to obtain stable positioning data.

[0102] When the moving distance exceeds the threshold but does not meet the requirement of n consecutive records, the correction position s is assigned based on the moving direction. 1 =s 1 +v 1 ×t1 , record the speed v at the previous moment 1 and time t 1 , calculate the dynamic jump threshold again and determine whether the positioning is normal.

[0103] The present embodiment also provides a photovoltaic component fault navigation and correction system, including: a path generation module, which stitches photovoltaic field images into a complete base map through image stitching, cuts and organizes the stitched base map into a multi-level visible light tile base map, and uses a U-Net model to mark navigable channels to form a photovoltaic field map with a traversable path; a fault identification module, which performs temperature distribution analysis through infrared defect recognition based on infrared inspection images obtained by drone inspections, identifies photovoltaic panel faults, and uses EXIF ​​information of infrared inspection images to locate the fault location and generate a fault report; a navigation module, which dynamically corrects the fault based on the photovoltaic panel fault and positioning information after each movement in combination with the real-time position of the inspection personnel, and accurately navigates to the fault location.

[0104] This embodiment also provides a computing device, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the photovoltaic component fault navigation and correction method proposed in the above embodiment.

[0105] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the photovoltaic component fault navigation and correction method proposed in the above embodiment is implemented.

[0106] The storage medium proposed in this embodiment and the photovoltaic component fault navigation and correction method proposed in the above embodiment belong to the same inventive concept. The technical details not described in detail in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0107] Through the above description of the implementation method, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ReadOnly, Memory, ROM), random access memory (RandomAccessMemory, RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform the methods of various embodiments of the present invention.

[0108] Example 2, below is an embodiment of the present invention, which provides a photovoltaic module fault navigation and correction method. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0109] In this example, a photovoltaic power station is selected as the test site. First, reference points are evenly distributed throughout the photovoltaic field to ensure that the reference points cover the entire shooting area. The density of reference points in general areas is set to 1 per 100 square meters, and additional reference points are distributed in key areas (such as power station boundaries and major road intersections), with the density increased to 1 per 50 square meters.

[0110] A low-resolution drone pre-flight was conducted to obtain preliminary images. By analyzing the preliminary images, it was determined that there was an area with drastic topographic changes in the northwest corner of the site, accounting for about 5% of the total area. In this area, the density of reference points was further increased to 1 per 25 square meters. RTK-GPS equipment was used to measure each reference point, obtain precise geographic coordinates and record them to form a reference point layout document.

[0111] Subsequently, a DJI Matrice 300RTK drone equipped with a 60MP high-resolution camera was used to capture orthophoto images over the photovoltaic field according to a predetermined flight plan. The flight altitude was set to 120 meters, the route overlap was 80%, and the lateral overlap was 60% to ensure that high-quality orthophoto images were obtained. The image stitching was then performed using the method of the present invention. The experimental data are shown in Table 1.

[0112] Table 1 Image stitching parameter comparison table

[0113]

[0114] By comparing the accuracy of the affine transformation matrix of each experiment, it can be seen that the accuracy of the affine transformation matrix generally improves with the increase in the number of images and feature points, reaching up to 96%. This shows that increasing the number of images and feature points can effectively improve the accuracy of image stitching, thereby improving the quality of the base map and the accuracy of subsequent applications.

[0115] Pathfinding and dynamic deviation correction experiments were conducted based on the above-mentioned spliced ​​photovoltaic field map. First, the U-Net model was used to process the visible light tile base map and output a segmentation mask. The category labels in the segmentation mask include photovoltaic panels, channels, and obstacles. By analyzing the segmentation mask, the channels between photovoltaic panels were identified and extracted as independent pixel sets.

[0116] The actual width of each channel is calculated and judged using the preset minimum passable width standard of 1.2 meters. If the actual width of the channel is greater than or equal to 1.2 meters, it is marked as a passable channel; if it is less than 1.2 meters, it is marked as an obstacle. At the same time, the previously identified area with drastic terrain changes (approximately 5% of the area in the northwest corner of the site) is marked as an obstacle, but the main road constructed therein is still marked as a passable channel.

[0117] Next, a DJI Matrice 300RTK drone equipped with a FLIRVue Pro R 640 infrared camera was used for infrared inspection. The drone flew along a predetermined path at an altitude of 50 meters and a speed of 5 meters per second to ensure high-quality infrared images. The infrared images were transmitted back to the photovoltaic inspection system platform in real time and analyzed using the developed infrared defect recognition algorithm.

[0118] The infrared defect recognition algorithm first analyzes the temperature distribution of the infrared inspection image and identifies the hot spots and cold spots on the panel. The hot spot detection sensitivity coefficient k hot Set to 2.5, the cold spot detection sensitivity coefficient k cold Set to 2.0. Based on the identified hot and cold spots, calculate the probability scores for short, open, and hot spot faults.

[0119] Finally, a dynamic deviation correction test was conducted. Five operation and maintenance personnel were selected, each equipped with an operation and maintenance navigation mobile terminal, to perform simulated inspection tasks in the photovoltaic field. During initialization, the empirical jump threshold was set to 5 meters, the coefficient c in the dynamic jump threshold calculation was set to 1.5, and the number of continuous recordings N was set to 3 times. The experimental data obtained are shown in Table 2.

[0120] Table 2 Fault identification and dynamic correction data table

[0121]

[0122] Through many experiments, the success rate of fault identification of the present invention is about 94%, that is, through the method of the present invention, short circuit, open circuit and hot spot faults can be effectively identified according to hot spots and cold spots. However, due to the complex and changeable fault types of photovoltaic panels, the accuracy of specific judgment of fault types is low, and it is also necessary to combine manual actual detection.

[0123] In the table data, the dynamic correction efficiency indicates the effectiveness of the dynamic correction process, specifically, the efficiency with which the system can adjust the navigation path according to the positioning data during the navigation process. The higher the value, the more effective the system can correct the deviation. For example, if the dynamic correction efficiency is 85%, it means that in all navigation scenarios that require correction, the system effectively corrects 85%.

[0124] From the table data, we can see that the positioning error is between 2.0 and 2.4 meters, which shows that the system can locate within a reasonable error range, which is very important for navigation accuracy in actual operation. Smaller positioning error values ​​mean that the system can determine the fault location more accurately.

[0125] The navigation correction time is between 18 and 22 seconds, which means that the system can complete navigation correction in a shorter time. Shorter correction time helps improve operation and maintenance efficiency and reduce fault handling time.

[0126] The dynamic correction efficiency is between 80% and 87%, indicating that the system can effectively adjust the navigation path when processing navigation correction, ensuring that the operation and maintenance personnel can accurately reach the fault location. The higher dynamic correction efficiency means that the system can make better adjustments and reduce errors when facing the instability of positioning data.

[0127] In summary, the photovoltaic module fault navigation and correction method of the present invention performs well in fault identification, positioning accuracy, navigation correction efficiency and dynamic correction effect. Compared with the prior art, it has significant innovation and advantages in improving identification accuracy, shortening correction time and optimizing dynamic correction efficiency. These improvements not only improve the operation and maintenance efficiency of photovoltaic modules, but also enhance the reliability of the system in practical applications.

[0128] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A photovoltaic module fault navigation and correction method, characterized in that: include: Use drones to collect images of photovoltaic fields, generate and stitch visible light tile base maps, and form photovoltaic field maps with traversable paths; The photovoltaic field image includes: evenly distributing reference points throughout the photovoltaic field to ensure that the reference points cover the entire shooting area, with a reference point density of A; arranging additional reference points in key areas, with a reference point density of 1.5A in key areas; conducting a low-resolution drone pre-flight to obtain preliminary images; judging areas with drastic terrain changes through preliminary image analysis, with a reference point density of 2A in areas with drastic terrain changes; using a high-precision GPS device to measure each reference point, obtain accurate geographic coordinates and record them to form a reference point layout document; According to the predetermined flight plan, the UAV takes orthophoto images over the photovoltaic field to obtain the photovoltaic field image; The visible light tile base map includes stitching orthophoto images into a complete base map through image stitching, calculating SIFT features to obtain scale invariance, and then extracting SURF features to enhance the stability of feature points, which is expressed as: C(x,y,σ)=α·S(x,y,σ)+β·H(x,y) Where C(x,y,σ) represents the composite feature point response value; S(x,y,σ) represents the SIFT feature point detection result; H(x,y0) represents the SURF feature point detection result; α represents the weight factor; β represents the weight factor; Use the FLANN algorithm to build an approximate nearest neighbor index, perform preliminary matching on the extracted feature points, set a preliminary error threshold, use the RANSAC algorithm to filter out incorrect matching points, set a dynamic threshold to dynamically adjust the error tolerance according to the current matching quality, and further screen the matching points, which is expressed as: Among them, T d (k) represents the dynamic threshold of the kth iteration; D(f i1 ,f i2 ) represents the distance of the matching points, T i represents the preliminary error threshold; f i1 represents the feature point descriptor; f i2 Represents the feature point descriptor; The affine transformation matrix of the image is calculated using the screened matching points to accurately align the images. The aligned images are fused at multiple scales to build a pyramid model. The images are fused layer by layer to obtain a complete base map, which is expressed as: Among them, I f (x, y) represents the stitched image; L represents the total number of layers; I l (x, y) represents the image of layer l; G l (x,y) represents the weight of the Gaussian pyramid at level l; Cut the spliced ​​base map into multiple small tiles, each tile represents a part of the base map, organize the tiles according to the zoom level, form a multi-level structure of visible light tile base map, and import the visible light tile base map into the photovoltaic inspection system platform; The photovoltaic field map includes processing a visible light tile base map using a U-Net model, outputting a segmentation mask, each pixel point in the segmentation mask is assigned a category label, and the category labels are divided into photovoltaic panels, channels and obstacles; By analyzing the segmentation mask, the channels between the photovoltaic panels are identified, and the channels between the photovoltaic panels are extracted as independent pixel sets. The actual width of each channel is calculated, and the preset minimum passable width standard M is used for judgment. If the actual width of the channel is greater than or equal to M, it is marked as a passable channel. If the actual width of the channel is less than M, it is marked as an obstacle. Mark the area with drastic terrain changes judged by the preliminary image as an obstacle. If there is a road built in the area with drastic terrain changes, mark the road as a passable passage; Create a two-dimensional grid map with the same size as the segmentation mask. For each pixel, mark the passable channel as 1, and mark the obstacle and the photovoltaic field boundary as 0. Traverse each pixel of the grid map. When a pixel marked as 1 is found and has not been visited, add this pixel as the initial unit to the queue and mark it as visited. Take a unit from the queue and check the adjacent units in four directions. If the adjacent unit is passable and has not been visited, add the adjacent unit to the queue and mark it as visited. Repeat the process of traversing each pixel of the grid map and checking the adjacent cells until all traversable areas are traversed, summarize all path segments, obtain all traversable paths, overlay the traversable path data on the visible light tile base map, and form a photovoltaic field map with traversable paths; Use drones for infrared inspections, analyze infrared inspection images based on infrared defect recognition algorithms, identify and locate photovoltaic module faults; The fault image and positioning information of the photovoltaic module are sent to the operation and maintenance navigation mobile terminal of the operation and maintenance personnel, and the dynamic deviation correction is combined with the real-time position of the operation and maintenance personnel to accurately navigate to the fault location.

2. The photovoltaic module fault navigation and correction method according to claim 1, characterized in that: Using the drone for infrared inspection includes: the drone mobile inspection terminal receives an inspection task including a flight path and a target area, automatically takes off and flies along a predetermined path, and uses an infrared camera to collect real-time images to obtain infrared inspection images; The infrared inspection image is transmitted back to the photovoltaic inspection system platform in real time, and the fault area is accurately located based on the infrared defect recognition algorithm. The EXIF ​​information of the infrared inspection image is used to determine the specific location of the fault in the photovoltaic field area, and a fault report containing the fault type and location is generated.

3. The photovoltaic module fault navigation and correction method according to claim 2, characterized in that: The infrared defect recognition algorithm includes identifying hot spots and cold spots on the panel by analyzing the temperature distribution of the infrared inspection image, which is expressed as: T hot >T avg +k hot ·σ T T cold <T avg -k cold ·s T Among them, T hot Indicates hot spot temperature; k hot Represents the coefficient for adjusting the sensitivity of hotspot detection; T cold Indicates the cold spot temperature; k cold Indicates the coefficient for adjusting the cold spot detection sensitivity; T avg Represents the average temperature of the entire photovoltaic panel; σ T Represents the standard deviation of the temperature of the entire photovoltaic panel; Temperature greater than or equal to T hot The photovoltaic panel area is the hot spot area, and the area of ​​the hot spot area is A hot , the maximum temperature in the hot spot area is Temperature is less than or equal to T cold The photovoltaic panel area is the cold spot area, and the area of ​​the cold spot area is A cold , the lowest temperature in the cold spot area is Determine the fault type and short circuit based on the identified hot and cold spots: Open circuit judgment: Hot spot judgment: Among them, P short Indicates the probability score of short circuit fault; P open Indicates the probability score of open circuit fault; P hotspot represents the probability score of hot spot failure; w1, ..., w7 represent weight coefficients; A total represents the total area of ​​photovoltaic panels; Represents temperature gradient; A threshold Indicates the preset hot spot area threshold; When P short >max(P open ,P hotspot ) and P short >threshold short When , it is judged as a short circuit fault; When P open >max(P short ,P hotspot ) and P open >threshold open When , it is judged as an open circuit fault; When P hotspot >max(P short ,P open ) and P hotspot >threshold hotspot , judged as a hot spot fault; Among them, threshold short Indicates the threshold for short circuit fault judgment; threshold open Indicates the threshold for open circuit fault judgment; threshold hotspot Indicates the threshold for hot spot fault judgment.

4. The photovoltaic module fault navigation and correction method according to claim 3, characterized in that: The dynamic deviation correction includes that when the operation and maintenance personnel start navigation through the operation and maintenance navigation mobile terminal, they initialize and obtain positioning data and wait for the positioning data to stabilize. At this time, the deviation correction position s1 is the display position, and the deviation correction position s1 is equal to the positioning position s0; After the first move, the result s of subtracting the positioning position s0 from the correction position s1 is compared with the empirical jump threshold a. If s>a, it indicates that the positioning is unstable and reinitializes to obtain stable positioning data. If s≤a, it is considered that the positioning is normal and continues; Assign the correction position s1 equal to the positioning position s0, and record the speed v1 and time t1 at the previous moment; After the next move, the dynamic jump threshold is v1×t1. If the target position has been reached, the process ends. If the target position has not been reached, the process continues. The result s of subtracting the positioning position s0 from the correction position s1 is compared with the dynamic jump threshold v1×t1. When s>v1×t1, it is judged that the moving distance exceeds the threshold and recorded. When s≤v1×t1, it is judged that the positioning is normal, and the correction position s1 is reassigned to be equal to the positioning position s0, and the speed and time of the previous moment are recorded. When the moving distance recorded for n consecutive times exceeds the threshold, reinitialize to obtain stable positioning data; When the moving distance exceeds the threshold but does not meet the requirement of n consecutive records, the correction position s1=s1+v1×t1 is assigned based on the moving direction, the speed v1 and time t1 at the previous moment are recorded, the dynamic jump threshold is calculated again, and it is determined whether the positioning is normal.

5. A photovoltaic module fault navigation and correction system using the method according to any one of claims 1 to 4, characterized in that: include, The path generation module stitches the photovoltaic field images into a complete base map through image stitching, cuts and organizes the stitched base map into a multi-level visible light tile base map, and uses the U-Net model to mark the traversable channels to form a photovoltaic field map with traversable paths; The fault identification module uses infrared inspection images obtained by drone inspections to analyze temperature distribution and identify photovoltaic panel faults through infrared defect recognition. It also uses the EXIF ​​information of infrared inspection images to locate the fault location and generate a fault report. The navigation module dynamically corrects the fault according to the fault and positioning information of the photovoltaic panels and the real-time position of the maintenance personnel after each movement, and accurately navigates to the fault location.

6. A computer device comprising: Memory and processor; The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the method described in any one of claims 1 to 4 are implemented.

7. A computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions, when executed by a processor, implement the steps of the method according to any one of claims 1 to 4.

Images