A high-precision substation three-dimensional automatic modeling method

Abstract

A high-precision substation three-dimensional automatic modeling method, comprising: collecting laser point cloud data of a substation and multi-view pictures of the substation; dense point cloud data registration, camera parameter reconstruction of multi-view pictures, multi-view picture projection and adaptive point cloud density control; approximating the point cloud data with a three-dimensional Gaussian distribution function, summing all three-dimensional Gaussian distribution functions to obtain a Gaussian distribution geometry field of the three-dimensional Gaussian distribution point cloud representation, and obtaining a three-dimensional grid model through three-dimensional grid reconstruction and rendering; setting constraint conditions of devices on the basis of the three-dimensional grid model, generating three-dimensional device models, device coordinates and device classification names by using a volume segmentation algorithm and a three-dimensional marking algorithm; converting the device models into GIM model files, generating PMS files of the device coordinates and the device classification names, and synchronizing to a GIM and PMS 3.0 system. The method can effectively solve the problems of low efficiency and inconsistent quality in the traditional manual modeling process.

Description

Technical Field

[0001] The present invention relates to the fields of digital twin and digital power grid construction, and specifically to a high-precision three-dimensional automatic modeling method for substations. Background Art

[0002] As a critical infrastructure of the power grid, substation planning, design, construction, and operation and maintenance are crucial to the reliability and safety of the power system. Currently, substation modeling based on the Grid Information Model (GIM) plays a vital role in improving the efficiency and safety of substation construction. However, traditional GIM modeling methods rely primarily on manual data collection, drawing processing, and model creation. This is not only inefficient and prone to errors and omissions, but also difficult to update dynamically in real time. With the increasing demand for information and digital management of the power grid, traditional manual modeling methods can no longer meet the accuracy, real-time, and efficiency requirements of modern substations.

[0003] The Power Distribution Management System (PMS3.0), combined with GIM, enables real-time monitoring, diagnosis, and optimization of substation equipment. However, updating and maintaining equipment data in PMS3.0 still requires significant manual effort, resulting in a high workload and difficulty ensuring data accuracy and timeliness.

[0004] Therefore, how to efficiently and automatically generate a three-dimensional model of a substation and update the equipment data in the GIM and PMS3.0 systems in real time is a major challenge in current technology. Summary of the Invention

[0005] The present invention provides an artificial intelligence (AI) high-precision automatic three-dimensional substation modeling method, which aims to generate an accurate three-dimensional substation model through automation technology and seamlessly connect it with the GIM platform and PMS3.0 system, thereby effectively solving the problems of low efficiency and inconsistent quality in the traditional manual modeling process.

[0006] The present invention provides a high-precision three-dimensional automatic modeling method for a substation, comprising the following steps:

[0007] Collect laser point cloud data and multi-view images of the substation;

[0008] Dense point cloud data registration is performed on the substation's laser point cloud data. Camera parameter reconstruction is performed on the substation's multi-view images to obtain the camera's internal and external parameters. The multi-view stereo (MVS) method is used to project the multi-view images onto the corresponding laser point cloud plane based on the camera's internal and external parameters. The density of the point cloud data is adjusted based on the resolution and clarity of the projected images.

[0009] The output point cloud data is approximated by a 3D Gaussian distribution function using a 3D Gaussian distribution point cloud representation algorithm. The Gaussian distribution geometric field represented by the 3D Gaussian distribution point cloud is obtained by summing all the 3D Gaussian distribution functions. The obtained Gaussian distribution geometric field is then fused with the NeRF method to achieve 3D mesh reconstruction and rendering, resulting in a 3D mesh model.

[0010] Set the constraints of the equipment based on the generated 3D mesh model, and use the volume segmentation algorithm and 3D labeling algorithm to generate the 3D equipment model, equipment coordinates and equipment classification name;

[0011] Convert the generated equipment model into a GIM model file, generate a PMS file with the equipment coordinates and equipment classification name, and synchronize them to the GIM and PMS3.0 systems.

[0012] Furthermore, the laser point cloud data is collected by a laser scanner to record the three-dimensional spatial coordinate information of the substation; and the multi-view picture is two-dimensional image information taken by a camera at different positions.

[0013] Furthermore, when collecting substation laser point cloud data, distance constraints are adopted to control the accuracy of the collected data. The distance constraints are as follows:

[0014] d≤d max (1)

[0015] Where d is the distance between the target and the laser scanner, d max The maximum distance between the target and the laser scanner;

[0016] When collecting multi-view images of the substation, image parameter constraints are adopted to ensure the quality of the collected images. The image parameter constraints are as follows:

[0017]

[0018] Among them, C is contrast, L is brightness, C max is the maximum contrast threshold, C min is the minimum contrast threshold, L max is the maximum brightness threshold, L min is the minimum brightness threshold.

[0019] Furthermore, when performing dense point cloud data registration on the laser point cloud data of the substation, the iterative closest point ICP method, the normal distribution transformation method NDT, the three-dimensional shape context 3DSC method or the fast point feature histogram FPFH method are used for processing.

[0020] Furthermore, when reconstructing camera parameters from multi-view images of the substation, the self-calibrated neural radiance field (NeRF) method is used to set the camera's intrinsic and extrinsic parameters as learnable parameters, and the intrinsic parameters are obtained by training with a multi-layer perceptron (MLP). Alternatively, simulation calculations are performed through the motion structure and multi-view stereo method COLMAP software tool pipeline, and the motion structure (SfM) method is used to obtain the camera's position and orientation in three-dimensional space, and then the intrinsic and extrinsic parameters are estimated.

[0021] Furthermore, the three-dimensional Gaussian distribution point cloud representation algorithm is used to approximate the point cloud data output in the second step with a three-dimensional Gaussian distribution function, and all three-dimensional Gaussian distribution functions are summed to obtain a Gaussian distribution geometric field represented by the three-dimensional Gaussian distribution point cloud. The obtained Gaussian distribution geometric field is fused with the NeRF method to realize three-dimensional mesh reconstruction and rendering to obtain a three-dimensional mesh model, which specifically includes:

[0022] After reading the registered point cloud data, perform 3D Gaussian calculation on each point cloud data and randomly generate some disturbance points around each point to increase the density and diversity of the point cloud data;

[0023] Use the three-dimensional Gaussian distribution function to fit each point cloud data and its disturbance point, and obtain the mean vector, covariance matrix and weight coefficient of each point to form a three-dimensional Gaussian distribution point cloud representation to describe the position, direction and shape of the point cloud data. For a given i-th point cloud data, its three-dimensional Gaussian distribution function is:

[0024]

[0025] Among them, x is a three-dimensional vector, representing a point in space; w i is the weight coefficient, indicating the importance of the point cloud data; μ i is the mean vector, indicating the center position of the point cloud data; Σ i is the covariance matrix, which represents the direction and shape of the point cloud data;

[0026] The three-dimensional Gaussian distribution functions of all point cloud data are added together to obtain a mixed Gaussian distribution function. The Gaussian distribution geometric field SUGAR of the entire point cloud data set is expressed as formula (4):

[0027]

[0028] Read the camera's internal and external parameters output during camera parameter reconstruction as input; Neural Radiance Field (NeRF) algorithm

[0029] The Neural Radiation Field (NeRF) algorithm is integrated with the Gaussian distribution geometric field SUGAR represented by formula (4) to finally render a three-dimensional mesh model Mesh.

[0030] Furthermore, the Neural Radiation Field (NeRF) algorithm is integrated with the Gaussian distribution geometric field (SUGAR) represented by formula (4), and finally renders a three-dimensional mesh model Mesh, which specifically includes:

[0031] In the Neural Radiation Field (NeRF) algorithm, x represents a point in space, P represents the intrinsic and extrinsic parameters of the camera, and each inputs a deep learning network to output the density value of the point ρ(x) = MLP. density (x, P) and the color value of the point c(x) = MLP color (x, P), the volume density function σ(x) is obtained by optimization, and the density field of the entire substation is described by formula (5):

[0032] σ(x)=[ρ(x),c(x)] (5)

[0033] Fusion of Gaussian distribution geometry field SUGAR and density field:

[0034] By weighted fusion of the geometric field f(x) of formula (4) and the density field σ(x) of formula (5), the fused geometric density field σ'(x) is obtained:

[0035] σ'(x)=α·f(x)+(1-α)·σ(x) (6)

[0036] Texture generation and mapping, using the NeRF method to generate the color field c(x,d), as shown below:

[0037] c(x,d)=MLP c (x,d) (7)

[0038] The obtained color field c(x,d) is mapped to the Gaussian distribution geometric field SUGAR. On the surface of the initial three-dimensional mesh InitialMesh generated by SUGAR, the color field generated by NeRF is mapped to the vertices or patches of the initial three-dimensional mesh InitialMesh to achieve high-quality texture mapping. A color value is assigned to each vertex of the initial three-dimensional mesh InitialMesh through the vertex coloring method, and a continuous texture is obtained through interpolation. In addition, the color field c(x,d) generated by NeRF is projected onto the surface of the initial three-dimensional mesh InitialMesh through the texture projection method to obtain a high-resolution texture map model.

[0039] The Marching Cubes algorithm is used to extract the mesh surface from the geometric density field σ'(x), and the uncertainty information of SUGAR is used to optimize the smoothness and details of the initial three-dimensional mesh InitialMesh surface. The optimization function E is expressed as (8): total :

[0040] Etotal =E geometry +λ·E texture (8)

[0041] Among them, E geometry represents the geometric error, i.e., minimizing the error between the point cloud and the mesh; E texture represents the texture error, minimizing the error between the color generated by NeRF and the grid, and N is the number of points in the point cloud data;

[0042] Finally, the three-dimensional mesh model of the substation is obtained after optimization.

[0043] Furthermore, the constraints of the device include geometric constraint parameters and texture constraint parameters. The geometric constraint parameters include the area, size, spacing, and position of the device, and the texture constraint parameters include texture coordinates, filtering, and sampling.

[0044] Furthermore, the volume segmentation algorithm divides the obtained three-dimensional mesh model Mesh into different areas by utilizing geometric constraint parameters, and each area represents a device or component; the three-dimensional labeling algorithm assigns a texture map to each area based on texture constraint parameters, and gives each area a classification name according to predefined rules.

[0045] The present invention has the following advantages and effects:

[0046] 1. Automated process: The entire process is automated, requiring no manual intervention, greatly improving modeling efficiency.

[0047] 2. High precision: Combining point cloud data and multi-view images ensures high precision of 3D modeling.

[0048] 3. Real-time update: Through seamless integration with GIM and PMS3.0 systems, substation models and equipment data can be updated in real time.

[0049] 4. Reduce human errors: Automated modeling and data processing processes effectively avoid errors and omissions in manual modeling. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a flow chart of the high-precision three-dimensional automatic modeling method of a substation according to the present invention. DETAILED DESCRIPTION

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0052] Figure 1 FIG. 1 is a flow chart of a high-precision 3D automatic substation modeling method according to the present invention, wherein the method comprises the following steps:

[0053] The first step is to collect laser point cloud data and multi-view images of the substation;

[0054] The laser point cloud data is collected by a laser scanner (lidar) to record the three-dimensional spatial coordinate information of the substation; the multi-view picture is two-dimensional image information taken by a camera at different positions (including aerial photography).

[0055] When collecting substation laser point cloud data, distance constraints are adopted to control the accuracy of the collected data. The distance constraints are as follows:

[0056] d≤d max (1)

[0057] Where d is the distance between the target and the laser scanner, d max The maximum distance between the target and the laser scanner;

[0058] When collecting multi-view images of the substation, image parameter constraints are adopted to ensure the quality of the collected images. The image parameter constraints are as follows:

[0059]

[0060] Among them, C is contrast, L is brightness, C max is the maximum contrast threshold, C min is the minimum contrast threshold, L max is the maximum brightness threshold, L min is the minimum brightness threshold.

[0061] In the second step, dense point cloud data registration is performed on the laser point cloud data of the substation. The laser point cloud data at different positions and angles are aligned to the same coordinate system through rigid body transformation and least squares method to improve the accuracy and integrity of the point cloud data; the camera parameters of the multi-view images of the substation are reconstructed to obtain the internal and external parameters of the camera, and the multi-view stereo MVS method is used to project the multi-view images onto the corresponding laser point cloud plane according to the internal and external parameters of the camera. At the same time, the density of the point cloud data is adjusted according to the resolution and clarity of the projected images to ensure the quality and efficiency of the point cloud data.

[0062] When performing dense point cloud registration on substation laser point cloud data, considering the large-scale and high-precision requirements of substation scenarios, ICP (Iterative Closest Point) methods can be used. For other scenarios, such as small scenes requiring rapid registration, NDT (Normal Distribution Transform) can be used; for non-rigid objects (moving targets or people), 3DSC (Three-Dimensional Shape Context) can be used; and for scenes with minimal environmental interference, FPFH (Fast Point Feature Histogram) can be used. All of the above methods can achieve point cloud data registration.

[0063] Among them, when reconstructing camera parameters for multi-view images of the substation, the NeRF-(self-calibrated neural radiance field) method can be used. By setting the camera's intrinsic parameters (focal length, focus, height) and extrinsic parameters (position, rotation) as learnable parameters, the intrinsic parameters can be obtained by training with an MLP (multi-layer perceptron). Alternatively, simulation calculations can be performed through the COLMAP (Structure from Motion and Multi-view Stereo Method) software tool pipeline. Using the SfM (Structure from Motion) method, the position and orientation of the camera in three-dimensional space can be obtained, and the intrinsic parameters (focal length, focus, and height) and extrinsic parameters (position, rotation) can be estimated.

[0064] For adaptive point cloud density control, Open3D (a 3D data algorithm library that can process data including point clouds, meshes, and depth maps) or PCL (Point Cloud Library) can be used for general scenarios to downsample the number of point clouds. For scenarios requiring higher precision, specific algorithms can also be used, such as sampling methods based on CCVT (Constrained Capacity Voronoi Tile Method) and optimal transmission theory. All of these methods can achieve adaptive control of point cloud data density.

[0065] The third step is to use the 3D Gaussian distribution point cloud representation algorithm to approximate the point cloud data output in the second step with a 3D Gaussian distribution function, which can describe the position, direction and shape of the point cloud data. Then, all 3D Gaussian distribution functions are summed to obtain the Gaussian distribution geometric field (SUGAR) of the 3D Gaussian distribution point cloud representation. The obtained Gaussian distribution geometric field (SUGAR) is fused with the NeRF method to achieve 3D mesh reconstruction and rendering, and finally obtain a 3D mesh model (Mesh). The specific implementation steps of the third step are:

[0066] 1. Read the registered point cloud data output in the second step, perform 3D Gaussian calculation on each point cloud data, and randomly generate some disturbance points around each point to increase the density and diversity of the point cloud data;

[0067] 2. Use the three-dimensional Gaussian distribution function to fit each point cloud data and its perturbation point to obtain the mean vector, covariance matrix and weight coefficient of each point, forming a three-dimensional Gaussian distribution point cloud representation to describe the position, direction and shape of the point cloud data. For a given point cloud data, its three-dimensional Gaussian distribution function is:

[0068]

[0069] Among them, x is a three-dimensional vector, representing a point in space; w i is the weight coefficient, indicating the importance of the point cloud data; μ i is the mean vector, indicating the center position of the point cloud data; Σ i Is the covariance matrix, which represents the direction and shape of the point cloud data.

[0070] 3. Add the three-dimensional Gaussian distribution functions of all point cloud data to obtain a mixed Gaussian distribution function, and express the Gaussian distribution geometric field (SUGAR) of the entire point cloud data set as formula (4):

[0071]

[0072] 4. Read the camera’s intrinsic parameters (focal length, focus, and height) and extrinsic parameters (position, rotation) output during the second step of camera parameter reconstruction as one of the inputs of NeRF.

[0073] 5. The NeRF (Neural Radiance Field) algorithm is integrated with the Gaussian distribution geometric field (SUGAR) represented by formula (4) to finally render a three-dimensional mesh model (Mesh).

[0074] The NeRF (Neural Radiance Field) algorithm is specifically implemented. x represents a point in space, P represents the camera's intrinsic parameters (focal length, focus, and height) and extrinsic parameters (position, rotation), and is input into a deep learning network (such as a multi-layer perceptron MLP) to output the density value of the point ρ(x) = MLP.density (x, P) and the color value of the point c(x) = MLP color (x,P), and the volume density function σ(x) is obtained by optimization. In this way, the density field of the entire substation can be described by formula (5):

[0075] σ(x)=[ρ(x),c(x)] (5)

[0076] Fusion of Gaussian distribution geometry field (SUGAR) and density field:

[0077] By weighted fusion of the geometric field f(x) of formula (4) and the density field σ(x) of formula (5), the fused geometric density field σ'(x) is obtained:

[0078] σ'(x)=α·f(x)+(1-α)·σ(x) (6)

[0079] Texture generation and mapping, using the NeRF method to generate the color field c(x,d), as shown below:

[0080] c(x,d)=MLP c (x,d) (7)

[0081] The resulting color field c(x,d) is mapped to a Gaussian distribution geometric field (SUGAR). On the surface of the initial 3D mesh (Initial Mesh) generated by SUGAR, the color field generated by NeRF is mapped to the vertices or patches of the initial 3D mesh (Initial Mesh) to achieve high-quality texture mapping. Vertex shading is used to assign a color value to each vertex of the initial 3D mesh (Initial Mesh), and a continuous texture is obtained through interpolation. Furthermore, the color field c(x,d) generated by NeRF is projected onto the surface of the initial 3D mesh (Initial Mesh) using texture projection, resulting in a high-resolution texture map model.

[0082] The Marching Cubes algorithm is used to extract the mesh surface from the geometric density field σ'(x), and the uncertainty information of SUGAR is used to optimize the smoothness and details of the initial three-dimensional mesh (InitialMesh) surface. Here, the optimization function E is expressed as (8): total :

[0083] E total =E geometry +λ·E texture (8)

[0084] Among them, E geometry represents the geometric error, i.e., minimizing the error between the point cloud and the mesh; E texturerepresents the texture error, minimizing the error between the color generated by NeRF and the mesh. N is the number of points in the point cloud data.

[0085] Finally, the three-dimensional mesh model (Mesh) of the substation is obtained after optimization.

[0086] Step 4. Based on the generated three-dimensional mesh model, set the device constraints and use the volume segmentation algorithm and three-dimensional labeling algorithm to generate a three-dimensional device model, device coordinates and device classification name. The device constraints include geometric constraint parameters and texture constraint parameters. The geometric constraint parameters include the area, size, spacing, and position of the device. The texture constraint parameters include texture coordinates, filtering, and sampling.

[0087] The geometric constraint parameters of the equipment are used to constrain the layout of the building or equipment, defining the equipment area A device , total area A workspace Length L device , width W device , total length L workspace , total width W workspace ; Distance D between devices device , safety distance D safety , device position P device , other equipment positions P others , respectively expressed by formula (9), formula (10), formula (11), and formula (12):

[0088] Area constraint: A device ≤A workspace (9)

[0089] Required equipment area A device Cannot exceed the total area A of the workspace workspace ;

[0090] Size constraints:

[0091] Required equipment length L device and width W device Plus safety distance D safety , cannot exceed the length L of the working space workspace and width W workspace ;

[0092] Spacing constraint: D device ≥D safety (11)

[0093] The distance between devices must be greater than or equal to the safety distance;

[0094] Position constraint: P device1 ≠Pothers (12)

[0095] Required device position P device1 Cannot be used with other devices' location P others overlapping.

[0096] The texture constraint parameters constrain the texture coordinate range, filtering and sampling of the building or equipment, define the texture coordinate T, texture magnification T enlarge , texture reduction T shrink , texture sampling T sample , texture interpolation T interpolation , the constraints are expressed using equations (13), (14), and (15) respectively:

[0097] Texture coordinate range constraint: 0≤T≤1(13)

[0098] Requirements: Texture coordinates are between 0 and 1;

[0099] Texture filtering constraints:

[0100] The requirement is that the texture of one pixel is magnified to cover multiple pixels, which is greater than 1; the texture of multiple pixels is mapped to one pixel, which is less than 1;

[0101] Texture sampling constraint: T sample =T interpolation (T enlarge ,T shrink )(15)

[0102] Requirements, texture sampling constraints, that is, interpolation is performed within the range of texture magnification and texture reduction constraints.

[0103] The volume segmentation algorithm uses geometric features such as device area, size, and spacing to divide the obtained 3D mesh model into different regions, each representing a device or component. The specific implementation steps are as follows:

[0104] 1. Read the device area, size, spacing and other constraint parameters.

[0105] 2. Use the volume segmentation algorithm to divide the three-dimensional mesh model into different areas according to the constraint parameters, and record the boundaries and center points of each area.

[0106] The 3D labeling algorithm assigns a texture map to each region based on image features such as texture coordinates, filtering, and sampling, and gives each region a classification name based on predefined rules, such as transformer, switch, wire, etc. The specific implementation steps are as follows:

[0107] 1. Read texture coordinates, filtering, sampling and other constraint parameters to initialize texture mapping.

[0108] 2. Use a 3D labeling algorithm to assign a texture map to each region and give each region a classification name based on predefined rules.

[0109] 3. Output the three-dimensional device model, coordinates and classification name results.

[0110] Step 5: Convert the generated equipment model into a GIM model file, generate a PMS file with the equipment coordinates and equipment classification name, and synchronize it to the GIM and PMS3.0 systems.

[0111] The GIM model conversion method is implemented as follows:

[0112] 1. Input the generated 3D device model;

[0113] 2. Automatically check model integrity;

[0114] 3. Repair normals;

[0115] 4. Reduce the number of polygons;

[0116] 5. Remove redundant vertices and faces;

[0117] 6. Adjust UV coordinates;

[0118] 7. Adjust the model scale and position;

[0119] 8. Convert to GIM format and write to GIM file.

[0120] The PMS file generation method and specific algorithm are implemented as follows:

[0121] 1. Obtain device data and device location data from the PMS system, including the classification name, relative coordinates, and asset code of each device;

[0122] 2. Calculate the distance, match the device to the nearest location, traverse each device, calculate its distance to each location, and find the nearest location, and add the matching result between the device and the nearest location to the intermediate result list;

[0123] 3. Write the intermediate results generated in the previous step, including the device's asset code, category name, device coordinates, matching location ID, matching coordinates, and distance, into the PMS file.

[0124] Take the 3D reconstruction function in the digital twin system of a 110kV substation as an example:

[0125] The first step is to collect laser point cloud data and multi-view images of the substation.

[0126] The second step is dense point cloud data registration, camera parameter reconstruction of multi-view images, multi-view image projection and adaptive point cloud density control.

[0127] Among them, dense point cloud data registration is performed on the laser point cloud data of the substation, and the laser point cloud data at different positions and angles are aligned to the same coordinate system through rigid body transformation and least squares method to improve the accuracy and integrity of the point cloud data.

[0128] To achieve automated 3D reconstruction of substations, preprocessing of the input laser point cloud data and multi-view images is required. This includes operations such as denoising, filtering, and downsampling to improve data quality and reduce computational complexity. Point cloud data preprocessing includes motion distortion compensation, denoising, filtering, elevation normalization, and point cloud rasterization. Multi-view image preprocessing includes camera parameter estimation, conversion, and optimization.

[0129] Camera parameter reconstruction for multi-view images involves matching feature points from the images to calculate the camera's intrinsic and extrinsic parameters for each image, including the camera's position, orientation, and focal length. Using the NeRF method, the camera's intrinsic parameters (focal length, focus, height) and extrinsic parameters (position and rotation) are set as learnable parameters and trained using an MLP (Multi-Layer Perceptron).

[0130] Multi-view image projection refers to projecting each image onto the corresponding laser point cloud plane according to the camera parameters to increase the color information of the point cloud data.

[0131] Adaptive point cloud density control refers to adjusting the density of point cloud data according to the resolution and clarity of the projected image to ensure the quality and efficiency of point cloud data.

[0132] In the third step, the point cloud data output in the second step is approximated by a three-dimensional Gaussian distribution function using a three-dimensional Gaussian distribution point cloud representation algorithm, which can describe the position, direction, and shape of the point cloud data. Subsequently, all three-dimensional Gaussian distribution functions are summed to obtain a Gaussian distribution geometric field (SUGAR) representing a three-dimensional Gaussian distribution point cloud. The obtained Gaussian distribution geometric field (SUGAR) is fused with the NeRF method to realize three-dimensional mesh reconstruction and rendering, and finally a three-dimensional mesh model (Mesh) is obtained.

[0133] The fourth step is to set the equipment area, size, spacing constraint parameters and texture coordinate, filtering, and sampling constraint parameters according to the characteristics of the 110kV substation, and use the volume segmentation algorithm and the three-dimensional labeling algorithm to generate the three-dimensional equipment model, coordinates, and classification name results. The volume segmentation algorithm refers to dividing the three-dimensional mesh model (Mesh) into different areas according to the geometric characteristics such as equipment area, size, spacing, and constraint parameters. Each area represents a device or component. The three-dimensional labeling algorithm refers to assigning a texture map to each area based on image features such as texture coordinates, filtering, and sampling, and giving each area a classification name according to predefined rules, such as transformer, switch, wire, etc. The specific implementation steps are as follows:

[0134] 1. Read the device area, size, spacing and other constraint parameters.

[0135] 2. Use the volume segmentation algorithm to divide the image into different regions and record the boundary and center point of each region.

[0136] 3. Read texture coordinates, filtering, sampling and other constraint parameters, and initialize texture mapping.

[0137] 4. Use a 3D labeling algorithm to assign a texture map to each region and give each region a classification name based on predefined rules.

[0138] 5. Output the three-dimensional device model, coordinates and classification name results.

[0139] In the fifth step, the 3D device model generated in the fourth step is converted into a GIM model file using the GIM model conversion method. The coordinates and category names generated in the fourth step are then converted into a PMS file using the PMS file generation method. This concludes the method. The GIM model conversion method converts the 3D device model into a universal 3D modeling format based on the format requirements of the GIM model file, allowing for easy use across different platforms and software. The PMS file generation method converts the coordinates and category names into a universal attribute management format based on the format requirements of the PMS file, allowing for easy use across different platforms and software. The specific implementation steps are as follows:

[0140] 1. Read the 3D device model file generated in the fourth step, analyze its structure and properties, and obtain the device's geometric shape, position, orientation, material and other information.

[0141] 2. According to the format requirements of the GIM model file, create an empty GIM model file and define its metadata, coordinate system, units and other information.

[0142] 3. Traverse all devices in the 3D device model, generate corresponding GIM device objects based on their types and properties, and add them to the GIM model file. GIM device objects include the device's unique identifier, name, type, location, orientation, shape, material, and other properties.

[0143] 4. Save the GIM model file to complete the conversion from the 3D device model to the GIM model file.

[0144] 5. Read the coordinate and category name file generated in step 4, parse its content, and obtain the coordinate and category name information of the device.

[0145] 6. According to the format requirements of the PMS file, create an empty PMS file and define its metadata, version, encoding and other information.

[0146] 7. Traverse all records in the coordinate and classification name files, generate corresponding PMS device objects based on their contents, and add them to the PMS file. The PMS device object includes the device's unique identifier, coordinates, classification name, and other attributes.

[0147] 8. Save the PMS file and complete the conversion of coordinates and classification names to PMS files.

[0148] The output files and output data examples of each step in the embodiment of the present invention are shown in Table 1:

[0149] Table 1

[0150]

[0151]

[0152] Table 2 shows an example of a table output of the device location matching results between the device and the PMS system:

[0153] Table 2

[0154]

[0155] Table field description:

[0156] Asset Code: The asset of the equipment, from the PMS system.

[0157] CategoryName: The category name of the device, which comes from the PMS system.

[0158] Device Coordinates: The coordinates of the device, in the format of (X, Y, Z), from the result of this method.

[0159] Matched Location ID: The location ID in the PMS system that is matched, which comes from the result of this method.

[0160] Matched Coordinates: The matched position coordinates, in the format of (X, Y, Z), are the results of this method.

[0161] Distance: The Euclidean distance between the device coordinates and the matching location coordinates, from the result of this method.

[0162] In this way, the location of the equipment and the PMS system can be clearly associated, facilitating subsequent management and maintenance.

[0163] The present invention has the following characteristics:

[0164] (1) Different from manual modeling and conventional point cloud-based or depth image-based methods, the present invention uses both point cloud and image as input, combining the advantages of traditional single point cloud-based or visible light (such as oblique photography) modeling, and using artificial intelligence methods (NeRF) and Gaussian distribution geometric field (SUGAR), while improving the accuracy and speed of modeling.

[0165] (2) The present invention can achieve high-precision modeling of accurate three-dimensional scenes and objects in specific scenarios (such as substation scenarios) by proposing equipment geometric constraint parameters (area, size, spacing) and texture constraints (coordinates, filtering, sampling).

[0166] (3) The present invention proposes to use a volume segmentation algorithm and a three-dimensional labeling algorithm to respectively realize entity segmentation of the three-dimensional scene model representation and output a general GIM file (.GIM), as well as to detect the three-dimensional scene model representation using a three-dimensional labeling algorithm, and generate a compatible file of the power grid PMS3.0 system containing device coordinates and type names, so as to achieve seamless connection between the three-dimensional automated modeling and the technical middle platform and business application system of power production.

[0167] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A high-precision three-dimensional automatic modeling method for substations, characterized in that: The steps include: Collect laser point cloud data and multi-view images of the substation; Dense point cloud data registration is performed on the substation's laser point cloud data. Camera parameter reconstruction is performed on the substation's multi-view images to obtain the camera's internal and external parameters. The multi-view stereo (MVS) method is used to project the multi-view images onto the corresponding laser point cloud plane based on the camera's internal and external parameters. The density of the point cloud data is adjusted based on the resolution and clarity of the projected images. The output point cloud data is approximated by a 3D Gaussian distribution function using a 3D Gaussian distribution point cloud representation algorithm. The Gaussian distribution geometric field represented by the 3D Gaussian distribution point cloud is obtained by summing all the 3D Gaussian distribution functions. The obtained Gaussian distribution geometric field is then fused with the NeRF method to achieve 3D mesh reconstruction and rendering, resulting in a 3D mesh model. Set the constraints of the equipment based on the generated 3D mesh model, and use the volume segmentation algorithm and 3D labeling algorithm to generate the 3D equipment model, equipment coordinates and equipment classification name; Convert the generated equipment model into a GIM model file, generate a PMS file with the equipment coordinates and equipment classification name, and synchronize them to the GIM and PMS3.0 systems; The three-dimensional Gaussian distribution point cloud representation algorithm is used to approximate the point cloud data output in the second step with a three-dimensional Gaussian distribution function, and all three-dimensional Gaussian distribution functions are summed to obtain a Gaussian distribution geometric field for the three-dimensional Gaussian distribution point cloud representation. The obtained Gaussian distribution geometric field is fused with the NeRF method to realize three-dimensional mesh reconstruction and rendering to obtain a three-dimensional mesh model. Specifically, the method includes: After reading the registered point cloud data, perform 3D Gaussian calculation on each point cloud data and randomly generate some disturbance points around each point to increase the density and diversity of the point cloud data; Use the three-dimensional Gaussian distribution function to fit each point cloud data and its disturbance point, and obtain the mean vector, covariance matrix and weight coefficient of each point to form a three-dimensional Gaussian distribution point cloud representation to describe the position, direction and shape of the point cloud data. Point cloud data, its three-dimensional Gaussian distribution function is: （3）； in, is a three-dimensional vector representing a point in space; is the weight coefficient, which indicates the importance of the point cloud data; is the mean vector, indicating the center position of the point cloud data; is the covariance matrix, which represents the direction and shape of the point cloud data; The three-dimensional Gaussian distribution functions of all point cloud data are added together to obtain a mixed Gaussian distribution function. The Gaussian distribution geometric field SUGAR of the entire point cloud data set is expressed as formula (4): （4）； Read the camera's internal and external parameters output during camera parameter reconstruction as input to the Neural Radiance Field (NeRF) algorithm; The Neural Radiation Field (NeRF) algorithm is integrated with the Gaussian distribution geometric field SUGAR represented by formula (4) to finally render a three-dimensional mesh model Mesh.

2. The high-precision three-dimensional automatic modeling method for substations according to claim 1, characterized in that: The laser point cloud data is collected by a laser scanner to record the three-dimensional spatial coordinate information of the substation; the multi-view pictures are two-dimensional image information taken by a camera at different positions.

3. The high-precision three-dimensional automatic modeling method for substations according to claim 1, characterized in that: When collecting substation laser point cloud data, distance constraints are used to control the accuracy of the collected data. The distance constraints are as follows: （1）； in is the distance between the target and the laser scanner, Laser scanning for target distance Maximum distance of the instrument; When collecting multi-view images of the substation, image parameter constraints are adopted to ensure the quality of the collected images. The image parameter constraints are as follows: （2）； in, is the contrast, brightness, is the maximum contrast threshold, is the minimum contrast threshold, is the maximum brightness threshold, is the minimum brightness threshold.

4. The high-precision three-dimensional automatic modeling method for substations according to claim 1, characterized in that: When performing dense point cloud data registration on the substation laser point cloud data, the iterative closest point ICP method and the normal distribution transformation method NDT are used. The three-dimensional shape context 3DSC method or the fast point feature histogram FPFH method is used for processing.

5. The high-precision three-dimensional automatic modeling method for substations according to claim 1, characterized in that: When reconstructing camera parameters from multi-view images of the substation, the self-calibrated neural radiance field (NeRF) method is used to set the camera's intrinsic and extrinsic parameters as learnable parameters, and the intrinsic parameters are obtained by training a multi-layer perceptron (MLP). Alternatively, simulation calculations can be performed through the COLMAP software tool pipeline using the structure from motion and multi-view stereo method, and the position and orientation of the camera in three-dimensional space can be obtained using the structure from motion (SfM) method, thereby estimating the intrinsic and extrinsic parameters.

6. The high-precision three-dimensional automatic modeling method for substations according to claim 1, characterized in that: The Neural Radiation Field (NeRF) algorithm integrates the Gaussian distribution geometric field SUGAR represented by formula (4) and finally renders a three-dimensional mesh model Mesh, specifically including: In the Neural Radiance Field (NeRF) algorithm, represents a point in space, Represents the intrinsic and extrinsic parameters of the camera, inputs a deep learning network respectively, and outputs the density value of the point respectively and the color value of the point , the volume density function is obtained by optimization , use formula (5) to describe the density field of the entire substation: （5）； Fusion of Gaussian distribution geometry field SUGAR and density field: By transforming the geometric field of Equation (4) The density field of formula (5) Weighted fusion to obtain the fused geometric density field : （6）； Texture generation and mapping, using NeRF method to generate color field , as shown below: （7）； The resulting color field Mapped to the Gaussian distribution geometric field SUGAR, on the surface of the initial three-dimensional mesh Initial Mesh generated by SUGAR, the color field generated by NeRF is mapped to the vertices or faces of the initial three-dimensional mesh Initial Mesh to achieve high-quality texture mapping, wherein a color value is assigned to each vertex of the initial three-dimensional mesh Initial Mesh by the vertex coloring method, and a continuous texture is obtained by interpolation; in addition, the color field generated by NeRF is mapped to the vertex or face of the initial three-dimensional mesh Initial Mesh by the texture projection method. Project it onto the surface of the initial 3D mesh to obtain a high-resolution texture map model; Using the Marching Cubes algorithm, from the geometric density field The mesh surface is extracted from the mesh, and the uncertainty information of SUGAR is used to optimize the smoothness and details of the initial three-dimensional mesh surface. The optimization function is expressed as (8): : （8）； in, Represents the geometric error, that is, minimizing the error between the point cloud and the mesh; represents the texture error, minimizing the error between the color generated by NeRF and the grid, is the number of points in the point cloud data; Finally, the three-dimensional mesh model of the substation is obtained after optimization.

7. The high-precision three-dimensional automatic modeling method for substations according to claim 1, characterized in that: The constraints of the device include geometric constraint parameters and texture constraint parameters. The geometric constraint parameters include the area, size, spacing, and position of the device. The texture constraint parameters include texture coordinates, filtering, and sampling.

8. The high-precision three-dimensional automatic modeling method for substations according to claim 7, characterized in that: The volume segmentation algorithm divides the obtained three-dimensional mesh model Mesh into different areas by using geometric constraint parameters, and each area represents a device or component; the three-dimensional labeling algorithm assigns a texture map to each area according to texture constraint parameters, and gives each area a classification name according to predefined rules.

Images