Photovoltaic string reverse modeling method based on laser point cloud and geometric feature extraction

Abstract

The present application relates to the technical field of photovoltaic power station digital modeling, and more particularly to a photovoltaic string reverse modeling method based on laser point cloud and geometric feature extraction, comprising: collecting high-precision three-dimensional laser point cloud data of a photovoltaic power station, using a deep learning model to perform instance segmentation on the point cloud data, accurately extracting photovoltaic strings and establishing a unique index, identifying the geometric features and contour point set of the photovoltaic strings through vertex extraction, distance-based cluster search and angle-based contour detection methods, calculating the inclination and direction angle of the photovoltaic strings using an interpolation fitting method, calculating the center position of the photovoltaic strings using the contour point set, detecting and calculating the area of the photovoltaic panel shielding or damage using a sliding window method, and finally completing the reverse modeling of the photovoltaic power station. This method significantly improves the anti-interference capability and modeling accuracy of data acquisition, reduces the consumption of computing resources, and is suitable for intelligent management and operation of large-scale photovoltaic power stations under various environmental conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of digital modeling of photovoltaic stations, and in particular to a photovoltaic string reverse modeling method based on laser point cloud and geometric feature extraction. Background Art

[0002] Photovoltaic power generation has developed rapidly as a clean and renewable form of energy. With the large-scale construction of photovoltaic stations, the completion acceptance and operation and maintenance of photovoltaic stations have ushered in huge challenges. The efficiency and reliability of photovoltaic stations are affected by many factors, among which the actual position and angle parameters of photovoltaic strings are one of the keys to the efficiency and reliability of photovoltaic power stations.

[0003] During the final acceptance phase, determining whether the actual spatial position and angular parameters of the PV strings deviate from the designed values ​​is a crucial aspect of project acceptance. Furthermore, in the digital age, digital O&M based on three-dimensional digital twin models is an inevitable trend. The position and angular parameters of PV strings are the core support for digital O&M, enabling visualized, refined, standardized, and intelligent management of PV power plants, thus enhancing their O&M capabilities.

[0004] The existing technical means of using oblique photography to achieve reverse modeling of photovoltaic panels have the following problems:

[0005] 1. Easily affected by environmental factors, data quality is unstable.

[0006] Oblique photography relies on high-quality image data, which is susceptible to significant impacts from geographical conditions (such as undulating terrain and building obstructions), lighting (such as shadows and reflections), and weather conditions (such as fog, haze, rain, and snow), leading to image degradation and distortion. These external factors can lead to unstable quality and accuracy in the generated 3D point cloud data, especially when shooting from a distance or in complex terrain, resulting in low accuracy in the modeling results.

[0007] 2. The processing flow is complex and consumes large amounts of computing resources.

[0008] The 3D reconstruction process using oblique photography involves multiple complex steps, including feature point matching, spatial triangulation and dense matching (aerotriangulation), sparse point cloud extraction, and dense point cloud construction. Each step requires extensive data processing and computation, which is time-consuming and resource-intensive. Furthermore, additional correction operations are required to correct image distortion and occlusion, further increasing the computational burden and modeling time.

[0009] 3. The processing burden in non-target areas is heavy and the efficiency is low.

[0010] When generating 3D models of photovoltaic panels, oblique photography captures a large amount of background information (such as the ground, vegetation, and buildings) in non-PV areas. This non-target area information needs to be removed and corrected during subsequent processing, consuming a large amount of computing resources and reducing overall modeling efficiency.

[0011] The problems existing in the above existing technologies limit the application of oblique photography methods in reverse modeling of photovoltaic panels. Compared with laser point cloud technology, its efficiency and reliability are lower.

[0012] Therefore, studying the reverse modeling technology of photovoltaic stations can provide support for the completion acceptance, digital operation and maintenance, and post-evaluation of photovoltaic stations by determining the spatial position, angle, index, and contour of photovoltaic panels, which has great application value for the development of the photovoltaic power generation industry. Summary of the Invention

[0013] To solve the above technical problems, the present invention provides a photovoltaic string reverse modeling method based on laser point cloud and geometric feature extraction, comprising the following steps:

[0014] S1. Obtaining a 3D laser point cloud and photovoltaic string design parameters for a photovoltaic station, and preprocessing the laser point cloud to obtain point cloud data. The photovoltaic string design parameters include the width, length, and diagonal length of the photovoltaic string;

[0015] S2. Based on the obtained point cloud data, perform instance segmentation on the point cloud data based on deep learning methods, extract the photovoltaic string point cloud and set a unique index for each photovoltaic string, and establish the photovoltaic string index set PI T ;

[0016] S3. Based on the obtained photovoltaic string index set PI T , through the alternative vertex search method and the distance-based vertex cluster search method, the candidate vertex set Point of each photovoltaic string is extracted respectively 0 and feasible vertex set Point 1 And obtain the contour line of the photovoltaic string through the angle judgment method, detect the integrity of the photovoltaic string, and establish the index set PI of the intact photovoltaic string respectively. E The index set PI of the blocked or damaged photovoltaic strings D ;

[0017] S4. Extract PI by sliding pane method E PV strings and PI that are blocked or damaged D Merge and calculate the blocked or damaged area of ​​the blocked or damaged PV strings;

[0018] S5. Contour extraction method based on feasible contour line processing to extract PI EThe contour point set C of all photovoltaic strings E , and based on the principle that there is no laser point on one side of the adjacent edge point line, the PI is extracted D The contour point set C of all non-straight edge parts D , the PI E is associated with C E uniquely through the index, the PI D is associated with C D uniquely through the index;

[0019] S6. According to the contour point set C E and C D , the center position of the photovoltaic string is calculated, and the inclination and direction angle of the photovoltaic string are calculated by the interpolation fitting method;

[0020] S7. According to the center position, angle and contour of the photovoltaic string obtained, the reverse modeling of the photovoltaic station is completed.

[0021] Further, the step S3 includes the following steps:

[0022] S301. The point cloud data in the PI T is processed by an alternative vertex search method to obtain an alternative vertex set Point 0 , the Point 0 is uniquely associated with the PI T through the index;

[0023] S302. The point cloud data in the PI T is processed by a distance-based vertex cluster search method to obtain a feasible vertex set Point 1 , the Point 1 is uniquely associated with the PI T through the index;

[0024] S303. By the contour straight line acquisition method based on angle judgment, it is judged whether the photovoltaic string is complete, and the index set PI D of damaged or blocked photovoltaic strings and the index set PI E of intact photovoltaic strings are established to obtain a right angle vertex set Point 2 and an associated vertex set Point', the PI D and the PI E sum of the number of photovoltaic strings in the PI T , the Point 2 and the Point' are uniquely associated.

[0025] Further, the step S301 includes the following steps:

[0026] S301-1. Project the point cloud data onto the two-dimensional XY plane to obtain a two-dimensional point set:

[0027] P=(A x ,A y );

[0028] A x ={x1,x2…,x n};

[0029] A y ={y1,y2…,y n};

[0030] Where P is a set of plane points; A x A is the set of X-axis coordinate values ​​corresponding to the point; x is the set of Y-axis coordinate values ​​corresponding to the point; 1, 2, ... n is the index of the point;

[0031] S301-2. Get the maximum and minimum values ​​of the X and Y axes respectively, that is: max , x min ,y max ,y min , its calculation can be expressed as:

[0032] x max =max(A x )

[0033] x min =min(A x )

[0034] y max =max(A y )

[0035] y min =min(A y )

[0036] S301-3. Get x max , x min ,y max ,y min The index of the corresponding two-dimensional point set, and then obtain the corresponding three-dimensional point set P0;

[0037] S301-4. Pair the points in P0 with each other and calculate the distance d between the two points. The calculation can be expressed as:

[0038]

[0039] Where i and j represent the subscripts of the two paired points in P0, i≠j;

[0040] S301-5: Setting the distance threshold And take d greater than the distance threshold Point, as the candidate vertex set Point 0 , the distance threshold It is the diagonal length in the PV string design parameters.

[0041] Furthermore, step S302 includes the following steps:

[0042] S302-1, based on the Point obtained in step 301 0 Establish a set to be tested And set the number of iterations n = 0, and set the distance threshold according to the laser point cloud scanning accuracy

[0043] Among them, the distance threshold Based on the laser point cloud resolution setting, the For PI T The set of candidate vertices corresponding to the photovoltaic string with index q;

[0044] S302-2. From Randomly select K data points as the initial cluster centers, calculate the mutual distance d′ between the K data points, and set d′ less than the distance threshold The data points are merged into one cluster to obtain M initial clusters Said M≤K;

[0045] S302-3. Create a new set to be detected Calculate the centroids of M clusters and calculate the distance d″ from other points in P1 to the centroids, and set d″ to be less than the distance threshold The data points are assigned to the corresponding clusters, and vice versa.

[0046] S302-4. Judgment Is it empty? If not, set n=n+1 and repeat S302-2 to S302-3; if it is empty, proceed to the next step;

[0047] S302-5. Obtain candidate vertex clusters Randomly select a point in each cluster as a candidate vertex as the vertex set Point of the photovoltaic string q =(p1,p2,…,p l ), where l and the candidate vertex cluster The number of

[0048] S302-6. Setting the Threshold Distance And remove the vertex set Point qFrom the remaining vertices, select the collinear vertices on the same straight line, and the distance between adjacent vertices along the width direction of the photovoltaic string and the distance along the length direction of the photovoltaic string are equal to the distance threshold. The points of the PV string width and length are obtained to obtain the vertex set The threshold distance Including the width and length of the photovoltaic string in the photovoltaic string design parameters;

[0049] Among them, the vertex set Point is eliminated q Collinear vertices on the same straight line, that is, judging the vertex set Point q Is the cosine value of the straight line formed by any three vertices equal to 1 or -1, that is, to determine whether the absolute value of the cosine value is

[0050] Furthermore, step S303 includes the following steps:

[0051] S303-1. Based on the feasible vertices obtained in step S302 Create a set of right-angle vertices Associated vertex set Point′ q , and Point′ q unique association;

[0052] S303-2. Take any vertex p b , take any two vertices p a 、p c Cosine values ​​between the constituent lines Its calculation can be expressed as:

[0053]

[0054] when When p is equal to 0, b Add to And record p b 、p a 、p c The index of the traversal is continued until the traversal ends; the p a 、p c is divided by p b Other vertices outside

[0055] S303-3. Verify the vertex set Is the number of midpoints equal to 4 and equal to If the number is consistent, the string index is added to PI E Otherwise, it means that the photovoltaic string is blocked or damaged, and the photovoltaic string index is added to the set PI D .

[0056] Furthermore, the sliding pane method in step S4 includes the following steps:

[0057] S401. Set the window step size γ and density coefficient threshold a based on the 3D laser point cloud resolution of the photovoltaic station, and take PI E A photovoltaic string is selected and its point cloud is obtained, wherein the window step length γ is greater than the 3D laser point cloud resolution of the photovoltaic station, and the width and length of the photovoltaic string are integer multiples of the window step length γ;

[0058] S402. Select any vertex in the window pane and slide along the width and length of the PV string at a step size γ to calculate the window pane density coefficient ε. Compare it with the density coefficient threshold to identify all missing or blocked window panes in the PV string and record the coefficient of each window pane until the entire PV string is detected. The number of slides is determined by the length and width of the string. The specific density coefficient comparison model is:

[0059]

[0060] Where v is the number of points in the window pane; u is the ratio of the building window pane area to the laser point cloud resolution; a is the density coefficient threshold, which is determined based on the 3D laser point cloud resolution of the photovoltaic station and the photovoltaic string design parameters. The density coefficient is used to evaluate the density of the point cloud data in each window pane and determine whether the window pane is missing or blocked.

[0061] S403. According to the list of window coefficients of the string, determine the number e of ε = 1. If it is 0, it means that the string is intact. Otherwise, it is considered damaged or blocked, and the damaged or blocked area s is calculated. The calculation can be expressed as:

[0062] s=e×γ 2

[0063] Where s is the damaged or blocked area, e is the number of panes with ε=1, and γ is the pane step size.

[0064] Furthermore, the contour extraction method based on feasible contour straight line processing in step S5 includes the following steps:

[0065] S501. Set the search domain threshold β, for PI E The set of right-angle vertices of the photovoltaic string with index q Associated vertex set Point′ q , choose any k-th right-angle vertex p k , for Point′ q The corresponding incident vertex p in a 、p c Process them separately;

[0066] S502. PKP a The direction vector is the primitive vector, and the points in the cube area with the threshold β as the side length are traversed to determine the angle between the primitive vector and other vectors. The number of points in the cube is determined to be 0. If so, it is considered that the photovoltaic string is blocked or damaged, and the photovoltaic string index is added to the set PI D , and repeat S501; otherwise, execute S503;

[0067] S503. Set an angle threshold δ and calculate the angle between adjacent vectors. If the angle between adjacent vectors is greater than the angle threshold δ, the point is considered a contour point.

[0068] S504. Repeat S502 and S503 until all points in the small cube are traversed;

[0069] S505. Repeat S502-S504 until all small cubes are traversed to obtain the outline extraction result of the photovoltaic string.

[0070] Furthermore, the photovoltaic panel angle parameter estimation method based on interpolation fitting in step S6 includes the following steps:

[0071] S601. Construct a point set based on the point cloud data and construct a photovoltaic string plane equation, which can be expressed as:

[0072] Ax+By+Cz+D=0

[0073] Where A, B, C are the values ​​of the plane normal vector a in the plane equation, and D is the position offset of the plane in the three-dimensional coordinate system;

[0074] S602. Take the z-axis unit vector b as the direction vector and calculate the inclination value of the photovoltaic string. The calculation can be expressed as:

[0075] θ i =arccos(a·b / ||a||×||b||)

[0076] S603. Take the north direction unit vector c of the photovoltaic string as the direction vector and calculate the direction angle value of the photovoltaic string. The calculation can be expressed as:

[0077] θ d =arccos(a·c / ||a||×||c||).

[0078] The photovoltaic string reverse modeling method based on laser point cloud and geometric feature extraction provided by the present invention has the following beneficial effects:

[0079] 1. High accuracy and stability: The application of laser point cloud technology ensures high accuracy and high reliability of data acquisition, reducing the impact of external environmental factors on modeling quality.

[0080] 2. High efficiency and low resource consumption: Deep learning and intelligent point cloud processing methods simplify the data processing process, reduce the computing burden, and improve modeling efficiency.

[0081] 3. Fast and efficient detection: Automated photovoltaic module shading and damage detection and inclination and azimuth angle calculation methods support intelligent management of photovoltaic stations and improve overall power generation efficiency.

[0082] 4. Wide adaptability: This solution is suitable for photovoltaic sites of different sizes and complex terrains. It can operate stably under various environmental conditions and has better scalability and application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 , a flow chart of the photovoltaic string reverse modeling method based on laser point cloud and geometric feature extraction of the present invention.

[0084] Figure 2 , a schematic diagram of the occlusion or damage forms of photovoltaic strings according to the photovoltaic string inverse modeling method based on laser point cloud and geometric feature extraction of the present invention.

[0085] Figure 3 , a schematic diagram of an alternative vertex search method for the photovoltaic string inverse modeling method based on laser point cloud and geometric feature extraction of the present invention.

[0086] Figure 4 , a schematic diagram of the results of the vertex cluster search method of the photovoltaic string inverse modeling method based on laser point cloud and geometric feature extraction of the present invention.

[0087] Figure 5 , schematic diagram of feasible vertex sets of the photovoltaic string reverse modeling method based on laser point cloud and geometric feature extraction of the present invention.

[0088] Figure 6 , a schematic diagram of the sliding pane method of the photovoltaic string reverse modeling method based on laser point cloud and geometric feature extraction of the present invention.

[0089] Figure 7 , a schematic diagram of the contour extraction method of the photovoltaic string inverse modeling method based on laser point cloud and geometric feature extraction of the present invention.

[0090] Figure 8 , Schematic diagram of photovoltaic string scanning in an embodiment of the present invention.

[0091] Figure 9 , a schematic diagram of the photovoltaic string instance segmentation and unique indexing results in an embodiment of the present invention.

[0092] Figure 10 , schematic diagram of a two-dimensional point set of photovoltaic strings in an embodiment of the present invention.

[0093] Figure 11 , schematic diagram of a photovoltaic string candidate point set in an embodiment of the present invention.

[0094] Figure 12 , schematic diagram of feasible vertices of photovoltaic strings in an embodiment of the present invention.

[0095] Figure 13 , schematic diagram of the photovoltaic string outline point cloud in an embodiment of the present invention.

[0096] Figure 14 , schematic diagram of the photovoltaic string outline point cloud with index 9 in an embodiment of the present invention.

[0097] Figure 15 , a schematic diagram of the center position point of a photovoltaic string with index 1 in an embodiment of the present invention. DETAILED DESCRIPTION

[0098] In order to enable those skilled in the art to better understand the present invention and make the purpose, technical solution and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with the embodiments and the accompanying drawings. The illustrative embodiments of the present invention and their description are only used to explain the present invention and are not intended to further limit the present invention.

[0099] Example 1:

[0100] like Figure 1 As shown, an embodiment of the present invention provides a photovoltaic string reverse modeling method based on laser point cloud and geometric feature extraction, comprising the following steps:

[0101] S1. Use airborne LiDAR technology to acquire high-precision three-dimensional laser point clouds of photovoltaic sites, perform point cloud noise reduction, downsampling, and registration to obtain high-quality point cloud data; and obtain the design parameters of the photovoltaic strings in the completed design plan of the photovoltaic site, including length, width, and diagonal length.

[0102] Collect some photovoltaic string point clouds such as Figure 8 As shown in the figure, the design parameters of the photovoltaic strings in the completed design plan of the photovoltaic field are: a string has 3×26 photovoltaic modules, the module size is 2.382×1.134 (m), the photovoltaic string size is: 7.146×29.484 (m), and the diagonal length is 48.73m.

[0103] S2. Based on the photovoltaic field point cloud data, the PointNet++ model is used to extract the photovoltaic string point cloud, set a unique index for each photovoltaic string, and establish the photovoltaic string index set P0 T .

[0104] First, use the PointNet++ model to perform instance segmentation to obtain each photovoltaic string; then set a unique index for the photovoltaic string. Figure 9 As shown in the figure, after instance segmentation, 12 photovoltaic strings are obtained.

[0105] PointNet++ is a deep learning model for 3D point cloud processing. It was developed by the same team behind the PointNet model and aims to address PointNet's shortcomings in handling local structure. PointNet++ introduces a module called "Set Abstraction." This module downsamples the input point cloud and generates a new point cloud with multi-scale features. Specifically, the Set Abstraction module includes the following steps:

[0106] Sampling: A set of key points is selected from the input point cloud using the farthest point sampling strategy. This ensures that the selected key points are representative at different scales.

[0107] Grouping: With each key point as the center, the points near it are divided into small local areas. The size and shape of these local areas can be adjusted as needed to capture features of different scales.

[0108] Feature extraction: Extract features from points within each local area, using a network structure such as PointNet. The extracted features contain both the geometric and contextual information of the local area.

[0109] Aggregation: Aggregate the features of each local area to obtain a multi-scale feature representation of each key point.

[0110] By repeating the above steps, PointNet++ can gradually expand the receptive field and realize multi-scale feature processing of point clouds.

[0111] PointNet++'s network structure consists of multiple Set Abstraction layers, each of which includes a sampling layer, a grouping layer, and a PointNet layer. These layers work together to map point cloud data from its original high-dimensional space to a low-dimensional feature space while preserving the hierarchical structure of the point cloud. Compared to PointNet, PointNet++ is more robust and more accurate when handling complex scenes. In addition, because PointNet++ adopts a hierarchical structure, the network is better able to process large-scale point cloud data and reduce computational costs. By introducing hierarchical feature learning and multi-scale grouping, PointNet++ significantly improves its ability to process point cloud data, especially in capturing local and global information.

[0112] S3. Using alternative vertex search method to PI T Process all photovoltaic string point clouds in the grid and obtain the corresponding candidate vertex set Point 0 , the Point 0 By index and PI T Unique association.

[0113] The specific steps are:

[0114] S301. Obtain the three-dimensional laser point cloud of the photovoltaic string and project it onto the two-dimensional XY plane to obtain a two-dimensional point set; obtain the maximum and minimum values ​​of the maximum XY axis, i.e., x max , x min ,y max ,y min ; Get x max , x min ,y max ,y min The index of the corresponding two-dimensional point set, and then obtain the corresponding three-dimensional point set P0. Figure 10 As shown, a two-dimensional plane point set corresponding to the maximum and minimum values ​​of the photovoltaic string is obtained.

[0115] S302: Pair the points in P0 with each other and calculate the distance d. The calculation can be expressed as:

[0116]

[0117] Where i and j represent the subscripts of the two paired points in P0, i≠j.

[0118] S303. Set the distance threshold according to the collected photovoltaic string design parameters is 48.73m, and d is greater than the distance threshold Point, get the candidate vertex set Point 0 . Get as Figure 11 As shown in FIG, a set of candidate vertices is obtained after processing a photovoltaic string.

[0119] S4. Use distance-based vertex cluster search method to find PI T All photovoltaic string point clouds are processed to obtain the feasible vertex set Point 1 , the Point 1 By index and PI T Unique association.

[0120] The specific steps are:

[0121] S401. Based on the candidate vertex set Point 0 Establish a set to be tested And set the number of iterations n = 0, and set the distance threshold according to the laser point cloud scanning accuracy

[0122] Among them, the distance threshold Based on the laser point cloud resolution setting, the For PI T The set of candidate vertices corresponding to the photovoltaic string with index q;

[0123] S402. From Randomly select K data points as the initial cluster centers, calculate the mutual distance d′ between the K data points, and set d′ less than the distance threshold The data points are merged into one cluster to obtain M initial clusters Said M≤K;

[0124] S403. Create a new set to be detected Calculate the centroids of M clusters and calculate the distance d″ from other points in P1 to the centroids, and set d″ to be less than the distance threshold The data points are assigned to the corresponding clusters, and vice versa.

[0125] S404. Judgment Is it empty? If it is not empty, it means that there are still some items in the current set to be detected that do not meet the distance threshold. If there is a data point, set n=n+1 and repeat steps S402 to S403 until When it is empty, go to the next step;

[0126] S405. Obtain candidate vertex cluster Randomly select a point in each cluster as a candidate vertex as the vertex set Point of the photovoltaic string q =(p1,p2,…,p l ), where l and the candidate vertex cluster The number of

[0127] S406. Set threshold distance Include the width of the photovoltaic string 7.146m and the length of the photovoltaic string 29.484m, and remove the vertex set Point q Collinear vertices on the same straight line, that is, judging the vertex set Point q Is the cosine value of the straight line formed by any three vertices equal to 1 or -1, that is, to determine whether the absolute value of the cosine value is Select from the remaining vertices the distance between adjacent vertices along the width direction of the photovoltaic string and the distance along the length direction of the photovoltaic string are equal to the distance threshold respectively. The width and length of the photovoltaic string are obtained as follows Figure 12 As shown in the figure, the feasible vertex set obtained after processing a photovoltaic string is

[0128] S5. Use the contour line acquisition method based on angle judgment to determine whether the photovoltaic string is complete and establish the index set PI of the damaged or blocked photovoltaic string D and other PV string index sets PI E , get the right-angle vertex set Point 2 Associated with the vertex set Point'. The PI D with PI E The sum of the number of elements in is equal to PI T The sum of the number of elements in the Point 2 Uniquely associated with Point'.

[0129] The specific steps are:

[0130] S501. According to the feasible vertex set Create a set of right-angle vertices Associated vertex set Point′ q , and Point′ p unique association;

[0131] S303-2. Take any vertex p b , take any two vertices p a 、p c Cosine values ​​between the constituent lines Its calculation can be expressed as:

[0132]

[0133] when When p is equal to 0, b Add to And record p b 、p a 、p c The index of the traversal is continued until the traversal ends; the p a 、p c is divided by p b Other vertices outside

[0134] S303-3. Verify the vertex set Is the number of midpoints equal to 4 and equal to If the number is consistent, the string index is added to PI E Otherwise, it means that the photovoltaic string is blocked or damaged, and the photovoltaic string index is added to the set PI D .

[0135] After detection, in this embodiment, there is one blocked and damaged photovoltaic string, with an index of 9.

[0136] S6. For the photovoltaic string with index 9 that is blocked or damaged, extract PI by sliding window method. E PV strings and PI that are blocked or damaged D Merge and calculate the blocked or damaged area of ​​the blocked or damaged PV strings.

[0137] The specific steps are:

[0138] S601. Set the window step size γ and density coefficient threshold a based on the 3D laser point cloud resolution of the photovoltaic station, and take PI E A photovoltaic string is selected and its point cloud is obtained, wherein the window step length γ is greater than the 3D laser point cloud resolution of the photovoltaic station, and the width and length of the photovoltaic string are integer multiples of the window step length γ;

[0139] S602. Select any vertex in the window pane and slide along the width and length of the PV string in step size γ to calculate the window pane density coefficient ε. Compare it with the density coefficient threshold to identify all missing or blocked window panes in the PV string and record the coefficient of each window pane until the entire PV string is detected. The number of slides is determined by the length and width of the PV string. The specific density coefficient comparison model is:

[0140]

[0141] Where ε is the density coefficient, v is the number of points in the window pane, u is the ratio of the building window pane area to the laser point cloud resolution, and a is the density coefficient threshold, which is determined based on the 3D laser point cloud resolution of the photovoltaic station and the photovoltaic string design parameters. The density coefficient is used to evaluate the density of the point cloud data in each window pane and determine whether the window pane is missing or blocked.

[0142] S603. According to the list of window coefficients of the string, determine the number e of ε = 1. If it is 0, it means that the string is intact. Otherwise, it is considered damaged or blocked, and the damaged or blocked area s is calculated. The calculation can be expressed as:

[0143] s=e×γ 2

[0144] S6. Use the contour extraction method based on feasible contour line processing to extract PI E Each photovoltaic string contour point set C E , the PI E with C E Uniquely associated by index.

[0145] The specific steps are:

[0146] S601. Set the search domain threshold β for PI E The set of right-angle vertices of the photovoltaic string with index q Associated vertex set Point′ q , choose any k-th right-angle vertex p k , for Point′ q The corresponding incident vertex p in a 、p c Treat them separately;

[0147] S602. With edge p k p a The direction vector is the primitive vector, and the points in the cube area with the threshold β as the side length are traversed to determine the angle between the primitive vector and other vectors. The number of points in the cube is determined to be 0. If so, it is considered that the photovoltaic string is blocked or damaged, and the photovoltaic string index is added to the set PI D , and repeat S601; otherwise, execute S603;

[0148] S603. Set an angle threshold δ and calculate the angle between adjacent vectors. If the angle between adjacent vectors is greater than the angle threshold δ, the point is considered a contour point.

[0149] S604. Repeat S602 and S603 until all points in the small cube are traversed;

[0150] S605. Repeat S602-S604 until all small cubes are traversed and the following is obtained: Figure 13 As shown in the figure, the feasible region to be searched of a photovoltaic string and the contour point cloud after the search.

[0151] S7. According to the principle that there is no laser point on one side of the line connecting adjacent edge points, D Extract the photovoltaic string contour point cloud set C from each photovoltaic string D , the PI D with C D Uniquely associated by index. Figure 14 As shown in FIG, the extraction results of the two-dimensional contour points and three-dimensional contour points of the photovoltaic string with index 9 that is blocked and damaged.

[0152] S8. Based on the PV string contour points C obtained in S6 and S7 E with C D By calculating the mean of the three-dimensional coordinates, the center position of the photovoltaic string is obtained (455803, 3073913, 70). Figure 15 The center point of the PV string with index 1 is shown;

[0153] S9. Use a photovoltaic panel angle parameter estimation method based on interpolation fitting to calculate the inclination and direction angles of the photovoltaic strings;

[0154] After calculation, the inclination angle and direction angle of the photovoltaic string in this example are shown in Table 1:

[0155] Table 1 PV string inclination and direction angles

[0156] String Index Direction inclination String Index Direction inclination 1 90.07 20.66 7 91.01 21.62 2 90.04 20.10 8 90.05 19.89 3 90.09 20.62 9 91.01 21.02 4 90.04 20.50 10 90.89 20.90 5 89.04 19.89 11 89.94 20.89 6 90.19 20.83 12 90.23 20.54

[0157] S10. Based on the spatial position, inclination, direction angle, and contour of the photovoltaic strings, the reverse modeling of the photovoltaic station can be completed.

[0158] The present invention is explained from the perspectives of purpose of use, effectiveness, progress and novelty. The practical progress it has is in compliance with the functional enhancement and use requirements emphasized by the Patent Law. The above description and drawings of this application are only preferred embodiments of this application and are not intended to limit this application. Therefore, all structures, devices, features, etc. that are similar or identical to those of this application, that is, all equivalent replacements or modifications made in accordance with the scope of this patent application, should fall within the scope of protection of this patent application.

[0159] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A photovoltaic string reverse modeling method based on laser point cloud and geometric feature extraction, characterized by: The following steps are involved: S1. Obtaining a 3D laser point cloud and photovoltaic string design parameters for a photovoltaic station, and preprocessing the laser point cloud to obtain point cloud data. The photovoltaic string design parameters include the width, length, and diagonal length of the photovoltaic string; S2. Based on the obtained point cloud data, perform instance segmentation on the point cloud data based on deep learning methods, extract the photovoltaic string point cloud and set a unique index for each photovoltaic string, and establish the photovoltaic string index set PI T ; S3. Based on the obtained photovoltaic string index set PI T , through the alternative vertex search method and the distance-based vertex cluster search method, the candidate vertex set Point of each photovoltaic string is extracted respectively 0 and feasible vertex set Point 1 And obtain the contour line of the photovoltaic string through the angle judgment method, detect the integrity of the photovoltaic string, and establish the index set PI of the intact photovoltaic string respectively. E The index set PI of the blocked or damaged photovoltaic strings D ; S4. Extract PI by sliding pane method E PV strings and PI that are blocked or damaged D Merge and calculate the blocked or damaged area of ​​the blocked or damaged PV strings; S5. Contour extraction method based on feasible contour line processing to extract PI E The contour point set C of all photovoltaic strings in E , and based on the principle that there is no laser point on one side of the line connecting adjacent edge points, PI is extracted D The set C of all contour points without straight edges in D , the PI E with C E Uniquely associated by index, the PI D with C D Uniquely associated by index; S6. According to the contour point set C E with C D , calculate the center position of the photovoltaic string, and calculate the inclination and direction angle of the photovoltaic string through the interpolation fitting method; S7. Complete the modeling of the photovoltaic station based on the obtained center position, inclination angle, direction angle, and profile of the photovoltaic string; Step S3 includes the following steps: S301. PI by alternative vertex search method T Process the point cloud data in to obtain the candidate vertex set Point 0 , the Point 0 By index and PI T unique association; S302. PI is searched by vertex cluster search method based on distance. T Process the point cloud data in to obtain the feasible vertex set Point 1 , the Point 1 By index and PI T unique association; S303. Determine whether the photovoltaic string is complete by obtaining the contour line based on the angle judgment method, and establish the damaged or blocked photovoltaic string index set PI D and the intact photovoltaic string index set PI E , get the right-angle vertex set Point 2 Associated with the vertex set Point', the PI D with PI E The sum of the number of photovoltaic strings is equal to PI T The sum of the number of photovoltaic strings in the Point 2 Uniquely associated with Point'; Step S301 includes the following steps: S301-1. Project the point cloud data onto the two-dimensional XY plane to obtain a two-dimensional point set: P=(A x ,A y ); A x ={x1,x2…,x n }; A y ={y1,y2…,y n }; Where P is a set of plane points; A x A is the set of X-axis coordinate values ​​corresponding to the point; y is the set of Y-axis coordinate values ​​corresponding to the point; 1, 2, ... n are the indexes of the points; S301-2. Get the maximum and minimum values ​​of the X and Y axes respectively, that is: max , x min ,y max ,y min , its calculation can be expressed as: x max =max(A x ) x min =min(A x ) and max =max(A y ) y min =min(A y ) S301-3. Get x max , x min ,y max ,y min The index of the corresponding two-dimensional point set, and then obtain the corresponding three-dimensional point set P0; S301-4. Pair the points in P0 with each other and calculate the distance d between the two points. The calculation can be expressed as: Where i and j represent the subscripts of the two paired points in P0, i≠j; S301-5: Take d greater than the distance threshold Point, as the candidate vertex set Point 0 , the distance threshold is the diagonal length of the PV string in the PV string design parameters; Step S302 includes the following steps: S302-1. Based on the Point obtained in step 301 0 Establish a collection to be tested Set the number of iterations n = 0 and set the distance threshold according to the laser point cloud scanning accuracy described For PI T The set of candidate vertices corresponding to the photovoltaic string with index q; S302-2. From Randomly select K data points as the initial cluster centers, calculate the mutual distance d′ between the K data points, and set d′ less than the distance threshold The data points are merged into one cluster to obtain M initial clusters Said M≤K; S302-3. Create a new set to be detected Calculate the centroids of M clusters and calculate the distance d″ from other points in P1 to the centroids, and set d″ to be less than the distance threshold The data points are assigned to the corresponding clusters, and vice versa. S302-4. Judgment Is it empty? If not, set n=n+1 and repeat S302-2 to S302-3; if it is empty, proceed to the next step; S302-5. Obtain candidate vertex clusters Randomly select a point in each cluster as a candidate vertex as the vertex set Point of the photovoltaic string q =(p1,p2,…,p l ), where l and the candidate vertex cluster The number of S302-6. Setting the Threshold Distance And remove the vertex set Point q From the remaining vertices, select the collinear vertices on the same straight line, and the distance between adjacent vertices along the width direction of the photovoltaic string and the distance along the length direction of the photovoltaic string are equal to the distance threshold. The points of the PV string width and length are obtained to obtain the vertex set The threshold distance Including the width and length of the photovoltaic string in the photovoltaic string design parameters; The contour extraction method based on feasible contour straight line processing in step S5 includes the following steps: S501. Set the search domain threshold β, for PI E The set of right-angle vertices of the photovoltaic string with index q Associated vertex set Point′ q , choose any k-th right-angle vertex p k , for Point′ q The corresponding incident vertex p in a 、p c Treat them separately; S502. With edge p k p a The direction vector is the primitive vector, and the points in the cube area with the threshold β as the side length are traversed to determine the angle between the primitive vector and other vectors. The number of points in the cube is determined to be 0. If so, it is considered that the photovoltaic string is blocked or damaged, and the photovoltaic string index is added to the set PI D , and repeat S501; otherwise, execute S503; S503. Set an angle threshold δ and calculate the angle between adjacent vectors. If the angle between adjacent vectors is greater than the angle threshold δ, the point is considered a contour point. S504. Repeat S502 and S503 until all points in the small cube are traversed; S505. Repeat S502-S504 until all small cubes are traversed to obtain the outline extraction result of the photovoltaic string.

2. The photovoltaic string reverse modeling method based on laser point cloud and geometric feature extraction according to claim 1 is characterized in that: Step S303 includes the following steps: S303-1. Based on the feasible vertices obtained in step S302 Create a set of right-angle vertices Associated vertex set Point′ q , and Point′ q unique association; S303-2. Take any vertex p b , take any two vertices p a 、p c Cosine values ​​between the constituent lines Its calculation can be expressed as: when When p is equal to 0, b Add to And record p b 、p a 、p c The index of the traversal is continued until the traversal ends; the p a 、o c is divided by p b Other vertices outside S303-3. Verify the vertex set Is the number of midpoints equal to 4 and equal to If the number is consistent, the string index is added to PI E Otherwise, it means that the photovoltaic string is blocked or damaged, and the photovoltaic string index is added to the set PI D .

3. The photovoltaic string reverse modeling method based on laser point cloud and geometric feature extraction according to claim 1 is characterized in that: The sliding pane method in step S4 includes the following steps: S401. Set the window step size γ and density coefficient threshold a based on the 3D laser point cloud resolution of the photovoltaic station, and take PI E A photovoltaic string is selected and its point cloud is obtained, wherein the window step length γ is greater than the 3D laser point cloud resolution of the photovoltaic station, and the width and length of the photovoltaic string are integer multiples of the window step length γ; S402. Select any vertex in the window pane and slide along the width and length of the PV string with a step size γ to calculate the window pane density coefficient ε. Compare it with the density coefficient threshold to identify all missing or blocked window panes in the PV string and record the coefficient of each window pane until the entire PV string is detected. The number of slides is determined by the length and width of the string. The specific density coefficient comparison model is: Where ε is the density coefficient, v is the number of points in the window pane, u is the ratio of the building window pane area to the laser point cloud resolution, and a is the density coefficient threshold, which is determined based on the 3D laser point cloud resolution of the photovoltaic station and the photovoltaic string design parameters. The density coefficient is used to evaluate the density of the point cloud data in each window pane and determine whether the window pane is missing or blocked. S403. According to the list of window coefficients of the string, determine the number e of ε = 1. If it is 0, it means that the string is intact. Otherwise, it is considered damaged or blocked, and the damaged or blocked area s is calculated. The calculation can be expressed as: s=e×γ 2 Where s is the damaged or blocked area, e is the number of panes with ε=1, and γ is the pane step size.

4. The photovoltaic string reverse modeling method based on laser point cloud and geometric feature extraction according to claim 1 is characterized in that: Step S6 calculates the inclination and direction angles of the photovoltaic strings by using an interpolation fitting method, including the following steps: S601. Construct a point set based on the point cloud data and construct a photovoltaic string plane equation, which can be expressed as: Ax+By+Cz+D=0 Where A, B, C are the values ​​of the plane normal vector a in the plane equation, and D is the position offset of the plane in the three-dimensional coordinate system; S602. Take the z-axis unit vector b as the direction vector and calculate the inclination value of the photovoltaic string. The calculation can be expressed as: i i =arccos(a·b / ‖a‖×‖b‖) S603. Take the north direction unit vector c of the photovoltaic string as the direction vector and calculate the direction angle value of the photovoltaic string. The calculation can be expressed as: i d =arccos(a·c / ‖a‖×‖c‖)。

Images