A point cloud data-driven 3D modeling adjustment method

Abstract

This invention relates to the field of 3D modeling technology, and more particularly to a point cloud data-driven 3D modeling adjustment method. The method includes the following steps: acquiring 3D point cloud data of a power facility area and performing point cloud downsampling and repair processing and semantic segmentation of power components to obtain semantic labels for the power facility components; performing geometric analysis of the power facility components based on the semantic labels to obtain their geometric features; performing topological connection and 3D point cloud modeling of the power facility components based on the semantic labels and geometric features to generate a 3D point cloud model of the power facility area; acquiring a BIM design model of the power facility area and performing point cloud-BIM alignment evaluation to generate a 3D alignment deviation matrix for the power facility; constructing a corresponding 3D deformation field for the power facility and adjusting the 3D deformation constraints to generate a 3D deformation reconstruction model of the power facility area. This invention enables automated optimization and adjustment of the 3D modeling process.

Description

Technical Field

[0001] This invention relates to the field of 3D modeling technology, and in particular to a point cloud data-driven 3D modeling adjustment method. Background Technology

[0002] With the continuous expansion of power infrastructure, especially in complex geographical environments, the construction and maintenance of power facilities face increasingly severe challenges. Point cloud data has become a major data source for 3D modeling. Point cloud data is typically obtained through laser scanning, photogrammetry, or other 3D acquisition technologies, and can accurately record the spatial information of objects or scenes. Existing point cloud data processing technologies usually rely on static or semi-static modeling methods, processing and reconstructing point cloud data through preset algorithms. However, due to noise, missing data, and sparsity issues during the point cloud data acquisition process, existing technologies suffer from positional deviations and low overlap when processing large-scale point cloud data, resulting in inaccurate 3D modeling results. Summary of the Invention

[0003] Therefore, it is necessary for the present invention to provide a point cloud data-driven 3D modeling adjustment method to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a point cloud data-driven 3D modeling and adjustment method includes the following steps:

[0005] Step S1: Obtain 3D point cloud data of the power facility area, and perform point cloud downsampling repair processing on the 3D point cloud data of the power facility area to obtain optimized point cloud downsampling data of the power facility area.

[0006] Step S2: Perform semantic segmentation of the power facility component on the downsampled optimization data of the power facility area point cloud to obtain semantic labels for the power facility component; perform geometric analysis of the power component on the downsampled optimization data of the power facility area point cloud based on the semantic labels of the power facility component to obtain the geometric features of the power facility component; perform topological connection of the power facility component based on the semantic labels and geometric features of the power facility component to generate a topological network diagram of the power facility component.

[0007] Step S3: Based on the topology network diagram of power facility components, perform point cloud 3D modeling on the point cloud downsampling optimization data of the power facility area to generate a 3D point cloud model of the power facility area; obtain the BIM design model of the power facility area, and perform point cloud-BIM alignment evaluation on the 3D point cloud model of the power facility area based on the BIM design model of the power facility area to generate a 3D alignment deviation matrix of the power facility.

[0008] Step S4: Construct the corresponding three-dimensional deformation field of the power facility based on the three-dimensional alignment deviation matrix of the power facility, and adjust the three-dimensional deformation constraint of the three-dimensional point cloud model of the power facility area based on the three-dimensional deformation field of the power facility to generate a three-dimensional deformation reconstruction model of the power facility area.

[0009] Furthermore, step S1 includes the following steps:

[0010] Step S11: Acquire 3D point cloud data of the power facility area;

[0011] Step S12: Obtain the corresponding point cloud density distribution of the power facility area through the three-dimensional point cloud data of the power facility area, and perform point cloud downsampling processing on the three-dimensional point cloud data of the power facility area based on the point cloud density distribution of the power facility area to obtain the point cloud downsampling data of the power facility area.

[0012] Step S13: Obtain the corresponding regional spatial distribution through the power facility area, and perform spatial hierarchical differentiation analysis on the point cloud downsampling data of the power facility area based on the regional spatial distribution to obtain the hierarchical differentiation data of the point cloud of the power facility area.

[0013] Step S14: Based on the hierarchical differential data of the power facility area point cloud, perform boundary difference correction on the power facility area point cloud downsampled data to obtain the power facility area point cloud downsampled corrected data.

[0014] Step S15: Perform missing point cloud repair processing based on adaptive interpolation on the downsampled correction data of the power facility area point cloud to obtain the optimized downsampled data of the power facility area point cloud.

[0015] Furthermore, step S2 includes the following steps:

[0016] Step S21: Extract the outlines of regional power components from the downsampled and optimized point cloud data of the power facility area to obtain the outlines of different power facility components within the power facility area, including power facility components corresponding to transmission towers, insulator strings, cables, and transformers.

[0017] Step S22: Based on the different outlines of power facility components within the power facility area, the point cloud downsampling optimization data of the power facility area is divided into component spatial regions to obtain spatial region point cloud data corresponding to different components within the power facility area;

[0018] Step S23: Use a deep learning-based semantic segmentation network to perform semantic segmentation of the spatial point cloud data corresponding to different components within the power facility area to obtain semantic labels for the power facility components.

[0019] Step S24: Perform geometric analysis of power facility components on the downsampled optimization data of the power facility area point cloud based on the semantic tags of power facility components to obtain the geometric features of power facility components;

[0020] Step S25: Based on the semantic tags and geometric features of power facility components, perform topological connections of power facility components to generate a topological network diagram of power facility components.

[0021] Furthermore, step S24 includes the following steps:

[0022] Based on the semantic tags of power facility components, the point cloud downsampling optimization data of power facility area is divided into component semantic point clouds to obtain the point cloud subsets of power facility components with semantic tags.

[0023] Statistical analysis of point cloud curvature is performed on the point cloud subsets of each power facility component with semantic labels to obtain the point cloud geometric curvature corresponding to each power facility component.

[0024] Based on the point cloud geometric curvature of each power facility component, the flatness of the corresponding power facility component point cloud subset is estimated to obtain the point cloud geometric flatness of each power facility component.

[0025] The normal vector distribution of each power facility component is obtained by using a subset of the point cloud of each power facility component with a semantic label.

[0026] The point cloud geometric curvature, point cloud geometric flatness, and normal vector distribution corresponding to each power facility component are combined as geometric features to obtain the geometric features of the power facility component.

[0027] Furthermore, step S25 includes the following steps:

[0028] Step S251: Perform multi-dimensional comprehensive mapping on the semantic labels and geometric features of power facility components to associate and map the semantic labels and geometric features corresponding to each power facility component, thereby obtaining power facility component feature fusion data;

[0029] Step S252: Based on the feature fusion data of power facility components, perform spatial coupling quantitative analysis on the spatial relationship between various components within the power facility area to calculate the relative position, contact surface and functional interaction relationship between the components, and obtain spatial coupling data between power facility components;

[0030] Step S253: Based on the spatial coupling data between power facility components, perform topological connection analysis on the spatial relationship between each component within the power facility area to obtain the corresponding topological connection relationship between each component within the power facility area, including the connection relationship between each component, the support relationship between components, and the distance constraint relationship between components;

[0031] Step S254: Based on the corresponding power component topology connection relationship between each component in the power facility area, perform power component topology connection to generate a power facility component topology network diagram.

[0032] Furthermore, step S3 includes the following steps:

[0033] Step S31: Based on the power facility component topology network diagram, perform point cloud topology constraint analysis on the power facility area point cloud downsampling optimization data to obtain the power facility component point cloud topology connection constraints and power facility component point cloud geometric constraints.

[0034] Step S32: Based on the topological connection constraints and geometric constraints of the point cloud of power facility components, perform 3D constraint modeling of the point cloud downsampling optimization data of the power facility area to generate a 3D constraint reconstruction model of the power facility area.

[0035] Step S33: Perform global point cloud smoothing and homogenization on the three-dimensional constrained reconstruction model of the power facility area to generate a three-dimensional point cloud model of the power facility area.

[0036] Step S34: Obtain the BIM design model of the power facility area;

[0037] Step S35: Based on the BIM design model of the power facility area, perform point cloud-BIM alignment evaluation on the 3D point cloud model of the power facility area to generate a 3D alignment deviation matrix of the power facility.

[0038] Furthermore, step S35 includes the following steps:

[0039] Step S351: Spatial alignment and registration of the BIM design model and the 3D point cloud model of the power facility area under the same spatial coordinate system are performed to generate the corresponding BIM model and 3D point cloud model of the power facility area under the same coordinate system.

[0040] Step S352: Calculate the point cloud-BIM geometric structure deviation between the BIM model and the 3D point cloud model of the power facility area under the same coordinate system to obtain the power structure alignment deviation between the point cloud model and the BIM model.

[0041] Step S353: Based on the alignment deviation of the power structure, perform a point cloud-BIM surface overlap assessment on the corresponding power facility area BIM model and the power facility area 3D point cloud model in the same coordinate system to obtain the surface alignment overlap degree between the point cloud model and the BIM model.

[0042] Step S354: Combine the alignment deviation of the power structure between the point cloud model and the BIM model with the surface alignment overlap to construct the corresponding alignment deviation matrix, so as to generate the three-dimensional alignment deviation matrix of the power facility.

[0043] Furthermore, step S352 includes the following steps:

[0044] Perform point cloud-BIM geometric alignment matching on the BIM model of the power facility area and the 3D point cloud model of the power facility area under the same coordinate system to obtain geometric alignment matching data of the point cloud model and the power facilities in the BIM model in terms of angle, edge and surface.

[0045] Alignment deviation is calculated by matching the geometric alignment data of power facilities in the point cloud model and the BIM model at angles, edges and surfaces, and the alignment deviation of the power structure between the point cloud model and the BIM model is obtained.

[0046] Furthermore, step S353 includes the following steps:

[0047] The surface contact area is divided into the BIM model and the 3D point cloud model of the power facility area under the same coordinate system to obtain the relative position of the contact area between the surface of the point cloud model and the BIM model.

[0048] The contact space distance is calculated based on the relative position of the contact area between the point cloud model and the BIM model surface to obtain the alignment space distance between each pair of contact points within the contact area between the point cloud model and the BIM model surface.

[0049] Based on the alignment deviation of the power structure and the alignment space distance between each pair of contact points in the contact area between the point cloud model and the BIM model, the point cloud-BIM surface overlap assessment is performed on the corresponding power facility area BIM model and the power facility area 3D point cloud model in the same coordinate system to obtain the surface alignment overlap degree between the point cloud model and the BIM model.

[0050] Furthermore, step S4 includes the following steps:

[0051] Step S41: Perform radial deviation mapping on each point within the three-dimensional alignment deviation matrix of the power facility to generate a three-dimensional radial deviation scheduling field for the power facility.

[0052] Step S42: Based on the three-dimensional radial deviation scheduling field of the power facility, perform three-dimensional deformation correction on the three-dimensional alignment deviation matrix of the power facility to generate the three-dimensional deformation field of the power facility;

[0053] Step S43: Perform three-dimensional deformation inversion analysis on the three-dimensional point cloud model of the power facility area based on the three-dimensional deformation field of the power facility to obtain the elastic modulus, connection stiffness and boundary condition deformation constraints corresponding to the three-dimensional point cloud model.

[0054] Step S44: Based on the elastic modulus, connection stiffness and boundary condition deformation constraints corresponding to the three-dimensional point cloud model, adjust the three-dimensional deformation constraints of the three-dimensional point cloud model of the power facility area to generate a three-dimensional deformation reconstruction model of the power facility area.

[0055] The beneficial effects of this invention are:

[0056] The point cloud data-driven 3D modeling adjustment method proposed in this invention has the following advantages over existing technologies: by acquiring 3D point cloud data of power facility areas, it can comprehensively and accurately capture the spatial morphology and geographical information of power facilities. The point cloud data is acquired through technologies such as laser scanning and lidar, providing the original spatial data foundation for the digital modeling of power facilities. In the subsequent point cloud downsampling and repair process, by simplifying the point cloud data (e.g., denoising, reducing point cloud density), the noise, missing data, and sparsity problems in the point cloud data acquisition process can be effectively reduced, thereby improving computational efficiency and the accuracy of subsequent processing. The downsampled point cloud data not only retains the basic shape and features of the power facility area but also reduces unnecessary computational overhead, making data processing more efficient. This processing lays a clearer and simpler foundation for subsequent analysis and modeling. Secondly, semantic segmentation of point cloud data of power facilities enables automatic identification and labeling of different components. Semantic segmentation utilizes machine learning and deep learning technologies to accurately separate different components (such as transformers, conductors, and poles) based on the spatial distribution and morphological features of point cloud data, assigning semantic labels to each component. This not only lays the foundation for automated modeling of power facilities but also allows for personalized processing of different components in subsequent analysis. Furthermore, geometric analysis based on semantic labels can extract the geometric features of power facility components, such as size, shape, and positional relationships. Acquiring these geometric features not only helps optimize facility design and layout but also provides accurate information for maintenance and fault diagnosis. By constructing a topological network diagram of the power facility based on semantic labels and geometric features, the connection relationships and structure between various components can be clearly displayed. Then, using the topological network diagram of power facility components and point cloud data, high-precision 3D modeling of the power facility can be performed. 3D modeling of point cloud data can generate a digital 3D model of the power facility, accurately reflecting its spatial layout and structural form. Compared to traditional 2D design drawings, 3D point cloud models have a higher spatial representation capability, better reflecting the actual situation of facilities. This helps engineers fully understand the spatial relationships of power facilities. After obtaining the BIM design model of the power facility area, the point cloud model can be aligned and evaluated based on the BIM model. This allows for comparison between the design model and the actual point cloud data, identifying differences between the model and reality. This process not only helps reduce positional deviations during 3D point cloud modeling but also improves the overlap between the model and reality. By generating a 3D alignment deviation matrix, the error between the design and reality of power facilities can be quantified, helping engineers accurately locate 3D modeling problems and develop effective solutions to ensure that excessive errors do not occur during facility construction and maintenance.Finally, by using the 3D alignment deviation matrix of the power facilities, a 3D deformation field of the power facility area can be constructed. The deformation field reflects the deformation of the power facilities due to factors such as time and environment. This process can clearly show the deformation and displacement of the facilities during long-term use. Based on this deformation information, the 3D deformation constraint adjustment of the 3D point cloud model of the power facility area can accurately reconstruct and correct the structure of the facilities. Through this deformation reconstruction, a more accurate 3D model that conforms to the actual situation can be generated. It can correct the facilities in a timely manner when deformation or damage occurs, and predict possible deformation trends, effectively avoiding major failures or safety hazards of the facilities, thereby improving the accuracy of 3D modeling of power facilities. Attached Figure Description

[0057] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0058] Figure 1 This is a flowchart illustrating the steps of the point cloud data-driven 3D modeling and adjustment method of the present invention.

[0059] Figure 2 for Figure 1 A detailed flowchart of step S1;

[0060] Figure 3 for Figure 1 A detailed flowchart of step S1. Detailed Implementation

[0061] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0062] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0063] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0064] To achieve the above objectives, please refer to Figures 1 to 3 This invention provides a point cloud data-driven 3D modeling adjustment method, the method comprising the following steps:

[0065] Step S1: Obtain 3D point cloud data of the power facility area, and perform point cloud downsampling repair processing on the 3D point cloud data of the power facility area to obtain optimized point cloud downsampling data of the power facility area.

[0066] Step S2: Perform semantic segmentation of the power facility component on the downsampled optimization data of the power facility area point cloud to obtain semantic labels for the power facility component; perform geometric analysis of the power component on the downsampled optimization data of the power facility area point cloud based on the semantic labels of the power facility component to obtain the geometric features of the power facility component; perform topological connection of the power facility component based on the semantic labels and geometric features of the power facility component to generate a topological network diagram of the power facility component.

[0067] Step S3: Based on the topology network diagram of power facility components, perform point cloud 3D modeling on the point cloud downsampling optimization data of the power facility area to generate a 3D point cloud model of the power facility area; obtain the BIM design model of the power facility area, and perform point cloud-BIM alignment evaluation on the 3D point cloud model of the power facility area based on the BIM design model of the power facility area to generate a 3D alignment deviation matrix of the power facility.

[0068] Step S4: Construct the corresponding three-dimensional deformation field of the power facility based on the three-dimensional alignment deviation matrix of the power facility, and adjust the three-dimensional deformation constraint of the three-dimensional point cloud model of the power facility area based on the three-dimensional deformation field of the power facility to generate a three-dimensional deformation reconstruction model of the power facility area.

[0069] In the embodiments of this invention, please refer to Figure 1 The diagram shown illustrates the steps of the point cloud data-driven 3D modeling adjustment method of the present invention. In this example, the point cloud data-driven 3D modeling adjustment method includes the following steps:

[0070] Step S1: Obtain 3D point cloud data of the power facility area, and perform point cloud downsampling repair processing on the 3D point cloud data of the power facility area to obtain optimized point cloud downsampling data of the power facility area.

[0071] In this embodiment of the invention, a multi-device collaborative acquisition method is adopted when acquiring 3D point cloud data of power facility areas. A drone equipped with a lidar is used to scan the area along a grid route at a height of 150 meters and a speed of 8 meters per second. The lidar emits laser beams, and the point cloud coordinates are obtained by measuring the light reflection time. The horizontal angular resolution is set to 0.01°, and the vertical angular resolution is set to 0.1° to ensure the capture of details such as power towers and lines. Simultaneously, 3D laser scanner stations are set up every 10 meters on the ground. The scanners perform close-range, high-precision scanning of the facilities at a transmission frequency of 500,000 points per second. For areas difficult for the drone and ground scanner to reach, handheld laser scanning devices are used for supplementary scanning. After data acquisition, point cloud downsampling and repair processing is performed. First, the 3D point cloud data is divided into cubic grid units with a side length of 0.5 meters. The number of points in each unit is counted. For high-density units, a voxel grid filtering method is used to construct a 0.3-meter side length voxel grid. The centroid coordinates of the voxels are used to replace all points inside, reducing the number of points. Low-density units are not processed at this stage. When repairing missing point clouds, a local neighborhood-based method is adopted. A spherical neighborhood with a radius of 0.2 meters is constructed with the missing point as the center. If the number of point clouds in the neighborhood is large and the distribution is uniform, the inverse distance weighting method is used to calculate the coordinates of the missing point by assigning weights based on the distance from the neighborhood point to the missing point. If the number of point clouds in the neighborhood is small or the distribution is uneven, the Kriging interpolation method is used to calculate the coordinates of the missing point by considering the spatial autocorrelation of the point cloud. Finally, the downsampling optimization data of the power facility area point cloud is obtained.

[0072] Step S2: Perform semantic segmentation of the power facility component on the downsampled optimization data of the power facility area point cloud to obtain semantic labels for the power facility component; perform geometric analysis of the power component on the downsampled optimization data of the power facility area point cloud based on the semantic labels of the power facility component to obtain the geometric features of the power facility component; perform topological connection of the power facility component based on the semantic labels and geometric features of the power facility component to generate a topological network diagram of the power facility component.

[0073] In this embodiment of the invention, semantic segmentation of power components is performed on the downsampled and optimized point cloud data of power facility areas using a PointNet++ deep learning-based network model. The point cloud data is divided into groups of 1024 points, each containing three-dimensional coordinates and normal vector information as network input. The network is trained using a dataset of 1000 labeled point cloud samples of power facilities, covering components such as transmission towers, insulator strings, cables, and transformers, with each point having a corresponding semantic label. The training batch size is set to 32, the learning rate is 0.001, and the training is performed for 200 iterations. During inference, the point cloud data is input into the trained network, and the network outputs the probability of each point belonging to different components. The category with the highest probability is taken as the semantic label of that point, thus obtaining the semantic labels of the power facility components. Based on the semantic labels, geometric analysis of the power components is performed. When calculating the point cloud density, the point cloud is divided into a 0.1m × 0.1m × 0.1m voxel grid, and the number of points in each voxel is counted. When calculating the point cloud curvature, a 0.2m radius neighborhood is taken as the center of each point, and the covariance matrix of the neighborhood point cloud is calculated through principal component analysis. The ratio of the maximum to the minimum eigenvalue is taken as the curvature value of that point. When calculating the flatness, a 0.15m radius neighborhood is taken, and the plane is fitted using the least squares method. The root mean square value of the distance from the point to the plane is calculated as the flatness index. When calculating the normal vector, a 0.1m radius neighborhood is taken, and the plane is fitted using the least squares method to obtain the normal vector of that point. Then, the spatial distribution of the normal vectors of all points is statistically analyzed to obtain the geometric features of the power components. Finally, topological connections of power components are established based on semantic labels and geometric features. A graph structure is constructed, with each power component as a node. Nodes contain semantic labels and geometric feature attributes to determine node connections. For physically connected components, such as transmission towers and insulator strings, a direct connection is determined by detecting the minimum distance between their point clouds; if it is less than 0.05 meters, an edge is added between the nodes, with the edge weight set to the contact area. For functionally related components, such as cables and transformers, a functional connection is determined when the cable endpoint is less than 0.1 meters from the transformer input port; an edge is added, with the edge weight set to the matching degree between the cable's current transmission capacity and the transformer's capacity. A force-directed algorithm is used to lay out the topology graph. The node repulsion force is proportional to the mass, and the edge attraction force is proportional to the weight. Force balance is achieved through iterative calculations, ultimately generating a topology network graph of power facility components.

[0074] Step S3: Based on the topology network diagram of power facility components, perform point cloud 3D modeling on the point cloud downsampling optimization data of the power facility area to generate a 3D point cloud model of the power facility area; obtain the BIM design model of the power facility area, and perform point cloud-BIM alignment evaluation on the 3D point cloud model of the power facility area based on the BIM design model of the power facility area to generate a 3D alignment deviation matrix of the power facility.

[0075] In this embodiment of the invention, point cloud 3D modeling is performed on the point cloud downsampling optimization data of the power facility area based on the power facility component topology network diagram. First, the nodes of the topology network diagram are mapped one-to-one with the power components in the point cloud data. For example, the nodes representing transmission towers are mapped to the transmission tower point cloud set in the point cloud data. Topology connection constraint analysis is performed, and the component topology relationship is determined based on the node connection edges in the diagram. If two nodes are connected, the connection relationship of the corresponding component point cloud is determined using a distance threshold of 0.05 meters in the point cloud data. The connection point position is recorded, and geometric constraint analysis is performed, utilizing node attribute information. Compared with point cloud data, if the transformer node attribute is a cuboid, the principal axis direction of the transformer point cloud is calculated through principal component analysis. When the deviation from the length, width, and height directions of the cuboid exceeds 15°, the geometric shape of the point cloud is adjusted according to the node attribute. A constraint-based triangulation algorithm is adopted, and virtual connection lines are added to the point cloud boundary to construct a topological skeleton according to the topological connection constraints. The shape, size, and other constraints are set for each component point cloud according to the geometric constraints. Taking the transmission line as an example, if the geometric constraint requires a straight line, the Laplace deformation algorithm is used to adjust the point cloud that deviates from the straight line shape during triangulation. When constructing the triangular mesh, ensure that the interior angle of each triangle is between 30° and 120°. Merge the triangular meshes of the point clouds of each component, and use a weighted average method to merge the point clouds of overlapping parts in the connecting areas to generate a 3D point cloud model of the power facility area. Obtain the BIM design model of the power facility area, and use professional BIM design software to accurately draw the 3D models of each component according to the design drawings and technical specifications. Assign component material type, specifications and other attribute information, and follow the unified WGS84 coordinate system to ensure accurate positioning. Export the model as an IFC format file. Based on the BIM design model, perform point cloud-BIM alignment evaluation on the 3D point cloud model. First, use a coarse registration method based on feature points to extract feature points such as key nodes of power towers from the BIM model. Then, use the Harris corner detection algorithm to extract feature points from the point cloud model. Calculate the Euclidean distance of the 3D coordinates of the feature points to construct the initial correspondence. Use singular value decomposition to solve the initial transformation matrix to achieve preliminary alignment. Then, the iterative nearest point algorithm is used for fine registration. The maximum number of iterations is set to 40, and the convergence threshold is 0.005 meters. In each iteration, the nearest point is found to calculate the translation and rotation parameters and update the transformation matrix. After the alignment is completed, the geometric deviation is calculated using the projection method. The surface overlap is evaluated by setting a distance threshold of 0.04 meters. The geometric deviation and surface overlap data are integrated to construct a 4-row, 4-column three-dimensional alignment deviation matrix for power facilities.

[0076] Step S4: Construct the corresponding three-dimensional deformation field of the power facility based on the three-dimensional alignment deviation matrix of the power facility, and adjust the three-dimensional deformation constraint of the three-dimensional point cloud model of the power facility area based on the three-dimensional deformation field of the power facility to generate a three-dimensional deformation reconstruction model of the power facility area.

[0077] In this embodiment of the invention, a three-dimensional deformation field of the power facility is constructed based on the three-dimensional alignment deviation matrix. The three-dimensional alignment deviation matrix is ​​regarded as a set of discrete data points in three-dimensional space. Each point corresponds to the spatial position of the model and carries geometric deviation and surface overlap information. The radial basis function interpolation method is used, with an influence radius of 0.1 meters centered on the matrix point. The Euclidean distance between any interpolation point in the power facility area and each point in the matrix is ​​calculated. The influence weight of each matrix point on the interpolation point is determined based on the distance. The radial deviation value is assigned to the interpolation point by weighted summation. Radial deviation values ​​are also assigned to the 0.01m × 0.01m × 0.01m grid nodes in the area, forming a three-dimensional radial deviation scheduling field. The three-dimensional model of the power facility is divided into hexahedral grid elements with a side length of 0.02 meters using the finite element mesh generation method. The radial deviation value corresponding to the center of the grid element is obtained based on the position of the grid element in the three-dimensional radial deviation scheduling field. Combined with the knowledge of mechanics of materials, the deformation of the element in the three coordinate axis directions is calculated. For example, a grid cell of a power transmission tower has a calculated deformation of 0.002 meters in the x-direction, 0.0015 meters in the y-direction, and 0.0025 meters in the z-direction. Combining the deformation values ​​of each cell into a three-dimensional vector forms a three-dimensional deformation field for the power facility. Based on this field, the three-dimensional point cloud model of the power facility area is adjusted for three-dimensional deformation constraints. A deformation algorithm based on physical constraints is used, treating each point in the three-dimensional point cloud model as a particle with mass, connected by virtual springs. The spring elasticity coefficient is determined by the connection stiffness. For fixed boundary particles, the position is locked; for elastic boundary particles, elastic forces are applied according to the boundary condition deformation constraints. Taking the power tower point cloud model as an example, considering the bottom fixed boundary and component connection stiffness, the point cloud particle positions are updated iteratively based on Newton's second law (each iteration has a time step of 0.01 seconds, and 100 iterations). For areas where the point cloud density changes by more than 20% due to deformation, a density-based resampling method is used to add or delete points, ultimately generating a three-dimensional deformation reconstruction model of the power facility area, achieving precise adjustment of the three-dimensional model.

[0078] Furthermore, step S1 includes the following steps:

[0079] Step S11: Acquire 3D point cloud data of the power facility area;

[0080] Step S12: Obtain the corresponding point cloud density distribution of the power facility area through the three-dimensional point cloud data of the power facility area, and perform point cloud downsampling processing on the three-dimensional point cloud data of the power facility area based on the point cloud density distribution of the power facility area to obtain the point cloud downsampling data of the power facility area.

[0081] Step S13: Obtain the corresponding regional spatial distribution through the power facility area, and perform spatial hierarchical differentiation analysis on the point cloud downsampling data of the power facility area based on the regional spatial distribution to obtain the hierarchical differentiation data of the point cloud of the power facility area.

[0082] Step S14: Based on the hierarchical differential data of the power facility area point cloud, perform boundary difference correction on the power facility area point cloud downsampled data to obtain the power facility area point cloud downsampled corrected data.

[0083] Step S15: Perform missing point cloud repair processing based on adaptive interpolation on the downsampled correction data of the power facility area point cloud to obtain the optimized downsampled data of the power facility area point cloud.

[0084] As an embodiment of the present invention, reference is made to... Figure 2 As shown, Figure 1 A detailed flowchart of step S1 is shown below. In this embodiment, step S1 includes the following steps:

[0085] Step S11: Acquire 3D point cloud data of the power facility area;

[0086] In this embodiment of the invention, a multi-platform collaborative acquisition method is used to acquire three-dimensional point cloud data of the power facility area. First, a drone equipped with a lidar is used to scan the power facility area at a flight altitude of 150 meters and a flight speed of 8 meters per second, following a grid-like flight path. The lidar emits a laser beam with a wavelength of 1550 nanometers. By measuring the time difference between the laser emission and its reflection back to the sensor, the distance from each sampling point to the sensor is calculated, thereby obtaining spatial coordinate information. The horizontal angular resolution of the scan is set to 0.01°, and the vertical angular resolution is set to 0.1° to ensure that details of the power facilities can be captured. Simultaneously, a three-dimensional laser scanner is deployed on the ground, with a scanning station set every 10 meters centered on the base of the power tower. The scanner emits at a frequency of 500,000 points per second, performing close-range, high-precision scanning of facilities such as power towers and transmission lines. In addition, a handheld laser scanning device is used to supplement the scanning of areas that are difficult for the drone and ground scanner to reach, such as narrow gaps in power equipment and complex structures at high altitudes. Finally, the point cloud data acquired by drones, ground scanners, and handheld devices are fused together. Using a feature-matching-based point cloud registration algorithm, the geometric features of power facilities, such as the columns and crossarms of power towers, are utilized to unify the data collected by different devices into the same coordinate system, forming complete three-dimensional point cloud data of the power facility area.

[0087] Step S12: Obtain the corresponding point cloud density distribution of the power facility area through the three-dimensional point cloud data of the power facility area, and perform point cloud downsampling processing on the three-dimensional point cloud data of the power facility area based on the point cloud density distribution of the power facility area to obtain the point cloud downsampling data of the power facility area.

[0088] In this embodiment of the invention, to obtain the point cloud density distribution of the power facility area, the three-dimensional point cloud data is divided into cubic grid cells with a side length of 0.5 meters. Each grid cell is traversed, and the number of point clouds falling into it is counted. If the number of point clouds in a certain grid cell exceeds 500, the area is determined to be a high-density area; if the number of point clouds is between 100 and 500, it is a medium-density area; if it is less than 100, it is a low-density area. For high-density areas, a voxel grid filtering method is used for downsampling processing to construct a voxel grid with a side length of 0.3 meters. The centroid coordinates of all point clouds in each voxel are calculated, and the centroid coordinates are used as representative points to replace all point clouds in the original cell, thereby reducing the number of point clouds while preserving the geometric features of the power facilities. For medium-density regions, the Random Sample Consensus (RANSAC) algorithm is used for optimization. The sampling ratio is set to 0.6. A certain number of points are randomly selected from the point cloud set of each medium-density region. Through iterative calculation, outliers are removed, and point clouds that better reflect the structure of power facilities are retained. For low-density regions, downsampling is not performed to avoid losing important information. The processed point cloud data of each region are merged to finally obtain the downsampled point cloud data of the power facility region.

[0089] Step S13: Obtain the corresponding regional spatial distribution through the power facility area, and perform spatial hierarchical differentiation analysis on the point cloud downsampling data of the power facility area based on the regional spatial distribution to obtain the hierarchical differentiation data of the point cloud of the power facility area.

[0090] In this embodiment of the invention, based on the spatial distribution characteristics of power facilities, the power facility area is divided into three levels: an overhead transmission line area, a power tower area, and a ground equipment area. For the overhead transmission line area, using the elevation information of point cloud data, a set of point clouds with an elevation of more than 10 meters and a linear distribution is extracted. By calculating the Euclidean distance between point clouds and setting a distance threshold of 0.8 meters, point clouds with a distance less than the threshold are divided into the same transmission line segment. The direction, curvature, and other characteristics of each line segment are analyzed. For the power tower area, a spatial clustering algorithm based on point clouds, using DBSCAN clustering algorithm... The method sets a neighborhood radius of 1.5 meters and a minimum number of points of 30. Point clouds are aggregated into different tower point cloud clusters. The height, structural composition, and other characteristics of each tower are analyzed. For ground equipment areas, point cloud data with elevations between 0 and 2 meters are extracted. Using the normal vector information of the point clouds, the angle between the normal vectors of the point clouds is calculated, and point clouds with a normal vector angle of less than 15° are divided into the same plane region. The shape, size, and other characteristics of the ground equipment are analyzed. Through feature analysis of point cloud data in different layers of regions, the differential data of point cloud layers in the power facility area are finally obtained, highlighting the spatial distribution and structural characteristics of power facilities at different levels.

[0091] Step S14: Based on the hierarchical differential data of the power facility area point cloud, perform boundary difference correction on the power facility area point cloud downsampled data to obtain the power facility area point cloud downsampled corrected data.

[0092] In this embodiment of the invention, the boundary point cloud in the downsampled point cloud data of the power facility area is corrected based on the differential data of the point cloud hierarchy. For the transmission line area, since downsampling may cause the boundary point cloud of the line to be missing or offset, the boundary is corrected by using the direction characteristics of the line through a linear interpolation method. For example, for a straight transmission line, the coordinates of two adjacent boundary points are known, and the coordinates of the missing boundary points are calculated according to the direction of the line and added to the point cloud data. For the power tower area, a model-based boundary correction method is adopted. A three-dimensional model of the power tower is pre-established, and the tower point cloud data is registered with the model. The model boundary and the point cloud boundary are compared. For boundary parts that exist in the model but are missing in the point cloud, the point cloud data is supplemented by surface fitting according to the model shape and size. For the ground equipment area, the boundary point cloud is identified based on the normal vector change and elevation information of the point cloud. When the boundary point cloud is discontinuous, the boundary surface is constructed by using the normal vector and elevation trend of the adjacent point cloud through triangulation algorithm to supplement the missing boundary point cloud, making the boundary smoother and more accurate. After the above processing, the point cloud downsampling correction data of the power facility area is finally obtained.

[0093] Step S15: Perform missing point cloud repair processing based on adaptive interpolation on the downsampled correction data of the power facility area point cloud to obtain the optimized downsampled data of the power facility area point cloud.

[0094] In this embodiment of the invention, a local neighborhood-based adaptive interpolation algorithm is used to repair missing point clouds in the downsampling correction data of power facility area point clouds. First, for each missing point, a spherical neighborhood with a radius of 0.2 meters is constructed with that point as the center. The known point clouds in the neighborhood are searched. The interpolation weights are adaptively adjusted according to the number and distribution of point clouds in the neighborhood. If there are many point clouds in the neighborhood and they are evenly distributed, the inverse distance weighted interpolation method is used. According to the distance from each known point to the missing point, points that are closer are given a larger weight and points that are farther away are given a smaller weight. The coordinates of the missing point are then calculated. If the number of point clouds in the neighborhood is small or the distribution is uneven, Kriging interpolation is used. Considering the spatial autocorrelation of the point clouds, the influence of each known point on the missing point is calculated through the covariance function, and interpolation is performed. For example, for the missing point cloud area on the power tower caused by shading, the missing point cloud is gradually filled by the point cloud data around the tower through an adaptive interpolation algorithm, making the tower structure more complete. After repairing all the missing point clouds, the downsampling and optimization data of the power facility area point cloud is obtained, providing a high-quality point cloud data foundation for subsequent 3D modeling.

[0095] Furthermore, step S2 includes the following steps:

[0096] Step S21: Extract the outlines of regional power components from the downsampled and optimized point cloud data of the power facility area to obtain the outlines of different power facility components within the power facility area, including power facility components corresponding to transmission towers, insulator strings, cables, and transformers.

[0097] Step S22: Based on the different outlines of power facility components within the power facility area, the point cloud downsampling optimization data of the power facility area is divided into component spatial regions to obtain spatial region point cloud data corresponding to different components within the power facility area;

[0098] Step S23: Use a deep learning-based semantic segmentation network to perform semantic segmentation of the spatial point cloud data corresponding to different components within the power facility area to obtain semantic labels for the power facility components.

[0099] Step S24: Perform geometric analysis of power facility components on the downsampled optimization data of the power facility area point cloud based on the semantic tags of power facility components to obtain the geometric features of power facility components;

[0100] Step S25: Based on the semantic tags and geometric features of power facility components, perform topological connections of power facility components to generate a topological network diagram of power facility components.

[0101] As an embodiment of the present invention, reference is made to... Figure 3 As shown, Figure 1 A detailed flowchart of step S2 is shown below. In this embodiment, step S2 includes the following steps:

[0102] Step S21: Extract the outlines of regional power components from the downsampled and optimized point cloud data of the power facility area to obtain the outlines of different power facility components within the power facility area, including power facility components corresponding to transmission towers, insulator strings, cables, and transformers.

[0103] In this embodiment of the invention, when extracting the outline of regional power components from the downsampled and optimized point cloud data of power facility areas, a straight line and surface fitting method based on the RANSAC algorithm is adopted. First, the calculation radius of the point cloud normal vector is set to 0.1 meters. A k-nearest neighbor search is performed on the 100 neighboring points of each point. The local plane is fitted using the least squares method to obtain the normal vector of each point. For transmission tower components, utilizing the straight line characteristics of their columns, the distance threshold for RANSAC straight line fitting is set to 0.05 meters, and the number of iterations is 1000. The straight line parameters of the columns are extracted from the point cloud, and then the outline frame of the tower is generated through straight line extension. For insulator strings, utilizing the surface characteristics of their skirts, a RANSAC algorithm is used. NSAC quadratic surface fitting, with a distance threshold of 0.03 meters and 500 iterations, fits the spherical parameters of the shed. The outline of the insulator string is formed by combining multiple spheres. For cables, utilizing their linear distribution characteristics, RANSAC curve fitting is performed with a distance threshold of 0.02 meters and 800 iterations to fit the spatial curve parameters of the cable and generate its outline. For transformers, utilizing the planar characteristics of their enclosure, RANSAC plane fitting is used with a distance threshold of 0.04 meters and 600 iterations to fit various planes of the enclosure. The outline of the transformer is formed by combining the planes. Finally, the accurate outlines of various power facility components are obtained.

[0104] Step S22: Based on the different outlines of power facility components within the power facility area, the point cloud downsampling optimization data of the power facility area is divided into component spatial regions to obtain spatial region point cloud data corresponding to different components within the power facility area;

[0105] In this embodiment of the invention, based on the extracted outlines of power facility components, the point cloud downsampling optimization data of the power facility area is divided into component spatial regions. For transmission tower components, a bounding box is generated based on the straight line parameters and outline frame of its column. The size of the bounding box extends 0.5 meters beyond the straight line of the column. The point cloud within the bounding box is divided into point cloud data of the transmission tower spatial region. For insulator strings, a spherical bounding body is generated based on the spherical parameters and outline of its skirts. The radius is increased by 0.2 meters based on the spherical parameters. The point cloud within the spherical bounding body is divided into point cloud data of the insulator string spatial region. For cables, based on their spatial curve parameters and contours, a tubular bounding box with a diameter of 0.3 meters is generated. The point cloud within the tubular bounding box is then divided into point cloud data for the cable spatial region. For transformers, based on their enclosure planar parameters and contours, an enclosure box is generated, with the box's dimensions extending 0.4 meters beyond the enclosure plane. The point cloud within the enclosure box is then divided into point cloud data for the transformer spatial region. This contour-based bounding box method accurately divides the point cloud data into point cloud data for different components, providing clean point cloud data for subsequent semantic segmentation.

[0106] Step S23: Use a deep learning-based semantic segmentation network to perform semantic segmentation of the spatial point cloud data corresponding to different components within the power facility area to obtain semantic labels for the power facility components.

[0107] In this embodiment of the invention, a deep learning-based semantic segmentation network is used to perform semantic segmentation of power components from point cloud data of spatial regions. The PointNet++ network architecture is employed, with the input layer containing the 3D coordinates and normal vector of each point. Through multiple feature extraction and aggregation layers, the semantic label probability of each point is output. During training, a dataset containing 1000 point cloud samples of power facilities is used. Each sample includes point clouds of components such as transmission towers, insulator strings, cables, and transformers, and each point has a corresponding semantic label. The training batch size is set to 32, the learning rate to 0.001, and the number of training iterations to 200. During the inference phase, the point cloud data of spatial regions of each component is input into the trained network. The network outputs the probability value of each point belonging to different components, and the category with the highest probability value is taken as the semantic label of that point. For example, for point cloud data of insulator strings, the network outputs the probability that each point belongs to an insulator string. When the probability is greater than 0.8, the point is determined to be an insulator string component. Finally, by statistically analyzing the semantic labels of all points, the semantic labels of the power facility components are obtained, thus achieving semantic segmentation of the point cloud data.

[0108] Step S24: Perform geometric analysis of power facility components on the downsampled optimization data of the power facility area point cloud based on the semantic tags of power facility components to obtain the geometric features of power facility components;

[0109] In this embodiment of the invention, geometric analysis of power components is performed on the point cloud downsampling optimization data of power facility areas based on semantic tags of power facility components. For the point cloud of a component corresponding to each semantic tag, its point cloud density is first calculated, and the point cloud is divided into a voxel grid of 0.1m × 0.1m × 0.1m. The number of points in each voxel is counted to obtain the point cloud density distribution. Then, the curvature feature of the point cloud is calculated. Taking each point as the center, a neighborhood with a radius of 0.2m is taken, and the covariance matrix of the neighborhood point cloud is calculated through principal component analysis. The ratio of the largest eigenvalue to the smallest eigenvalue is used as the curvature value of that point. The curvature distribution of the entire component point cloud is then counted. Next, the flatness of the point cloud is calculated. For each point, a neighborhood with a radius of 0.15 meters is taken, and a plane is fitted using the least squares method. The distance from each point to the plane is calculated, and the root mean square value of the distance is used as the flatness index. Finally, the normal vector distribution of the point cloud is calculated. For each point, a neighborhood with a radius of 0.1 meters is taken, and a plane is fitted using the least squares method to obtain the normal vector of that point. The distribution of the normal vectors of all points in space is statistically analyzed, thereby obtaining the geometric characteristics of the power facility components, including point cloud density, curvature distribution, flatness, and normal vector distribution.

[0110] Step S25: Based on the semantic tags and geometric features of power facility components, perform topological connections of power facility components to generate a topological network diagram of power facility components.

[0111] In this embodiment of the invention, power facility components are topologically connected based on semantic tags and geometric features. First, a graph of nodes is constructed, with each node representing a power facility component. Node attributes include semantic tags and geometric feature parameters. Then, the connection relationships between nodes are determined. For physically connected components, such as transmission towers and insulator strings, the minimum distance between their point clouds is detected. When the distance is less than 0.05 meters, they are considered directly connected, and connecting edges are added between the nodes. The weight of the edge is the contact area between the two components. For functionally related components, such as cables and transformers, their semantic tags and spatial locations are analyzed. When the endpoint of the cable is less than 0.1 meters from the input port of the transformer, they are considered functionally connected, and functional connecting edges are added between the nodes. The weight of the edge is the matching degree between the current transmission capacity of the cable and the capacity of the transformer. Finally, a force-directed algorithm is used to lay out the topology graph. The repulsive force between nodes is proportional to the node mass, and the attractive force of the connecting edge is proportional to the edge weight. Through iterative calculation, a force balance state is achieved, ultimately generating an intuitive power facility component topology network graph that clearly shows the physical connections and functional relationships between the components.

[0112] Furthermore, step S24 includes the following steps:

[0113] Based on the semantic tags of power facility components, the point cloud downsampling optimization data of power facility area is divided into component semantic point clouds to obtain the point cloud subsets of power facility components with semantic tags.

[0114] In this embodiment of the invention, the point cloud downsampling optimization data of the power facility area is divided into component semantic point clouds based on a pre-defined semantic labeling system for power facility components. The semantic labeling system covers more than 20 categories of power facility components, such as transmission lines, power towers, insulators, and transformers. A machine learning-based classification algorithm is used to train the classification model using features such as the spatial distribution and geometric shape of the point cloud. For example, a random forest classifier can be trained using 10 features, such as elevation, density, and distribution patterns of neighboring point clouds, as input. By training on labeled point cloud data of power facility components, the model achieves 92% accuracy on the test set. During segmentation, each point cloud in the downsampled and optimized point cloud data is used as a sample input to the trained classification model. The model outputs the semantic label of the component to which that point cloud belongs. Point clouds with the same semantic label are aggregated to form corresponding subsets of power facility component point clouds. For example, for transmission line components, the model identifies all point clouds belonging to transmission lines and integrates them into a subset of transmission line component point clouds; similarly, for insulator components, related point clouds are grouped into a subset of insulator component point clouds. Finally, each subset of power facility component point clouds with semantic labels is obtained, achieving precise semantic segmentation of point cloud data.

[0115] Preferably, point cloud curvature statistical analysis is performed on the point cloud subsets of each power facility component with semantic tags to obtain the point cloud geometric curvature corresponding to each power facility component;

[0116] In this embodiment of the invention, point cloud curvature statistical analysis is performed on point cloud subsets of power facility components with semantic tags. A curvature calculation method based on local fitting is adopted. For each point cloud subset, a spherical neighborhood with a radius of 0.1 meters is selected centered on that point. Using the point cloud data within the neighborhood, a quadratic surface is fitted using the least squares method. Let the equation of the quadratic surface be z=ax. 2 +bxy+cy 2 By solving the system of equations that minimizes the sum of squared errors, we determine the coefficients a, b, c, d, e, and f, and then calculate the first-order partial derivative of the surface at that point. , and second-order partial derivatives , , , According to the Gaussian curvature calculation formula and the formula for calculating the mean curvature The Gaussian curvature and mean curvature of the point cloud are obtained and used as the geometric curvature of the point cloud. For example, in the point cloud subset of insulator components, for the smooth spherical insulator part, the calculated Gaussian curvature and mean curvature values ​​are small and uniformly distributed; while at the connection part of the insulator, due to the complex structure, the curvature values ​​are relatively large and vary significantly. The above calculation is performed on all point clouds in the point cloud subset of each component to finally obtain the geometric curvature distribution of the point cloud corresponding to each power facility component.

[0117] Preferably, the flatness of the point cloud subset corresponding to each power facility component is estimated based on the point cloud geometric curvature of each power facility component, so as to obtain the point cloud geometric flatness of each power facility component.

[0118] In this embodiment of the invention, the flatness of the corresponding power facility component point cloud subsets is estimated based on the geometric curvature of the point cloud corresponding to each power facility component. Using a principal component analysis (PCA)-based method, the covariance matrix of the point cloud is constructed for each component point cloud subset. The eigenvalues ​​and eigenvectors of the covariance matrix are calculated using singular value decomposition (SVD). The direction of the eigenvector corresponding to the smallest eigenvalue is the normal vector direction of the local plane of the point cloud. The eigenvectors corresponding to the other two larger eigenvalues ​​determine the orientation of the plane. The sum of squared distances from the point cloud to the fitted plane is then calculated. ,in Let i be the distance from the point cloud to the fitted plane, and let the sum of squared distances be... Divide the result by the number of points n in the point cloud subset, and then take the square root to obtain the root mean square error (RMSE). This RMSE is used as a measure of the geometric flatness of the point cloud. For example, for the point cloud subset of the beam component of a power tower, a smaller RMSE value indicates that the surface of the component is relatively flat; while for the complex connection components of the power tower, a larger RMSE value indicates that the flatness is poor. Through this method, the geometric flatness of the point cloud corresponding to each power facility component is finally obtained, quantifying the flatness of the component surface.

[0119] Preferably, the normal vector distribution corresponding to each power facility component is obtained through the point cloud subset of each power facility component with semantic labels;

[0120] In this embodiment of the invention, the normal vector distribution corresponding to each power facility component is obtained through a subset of point clouds containing semantically tagged power facility components. A normal vector calculation method based on local neighborhood fitting is employed. For each point cloud subset, a spherical neighborhood with a radius of 0.05 meters is selected centered on that point. Using the point cloud data within the neighborhood, a plane is fitted using the least squares method. Let the plane equation be Ax + By + Cz + D = 0. By solving the system of equations that minimizes the sum of squared errors, the plane coefficients A, B, C, and D are determined. Then, the normal vector of the plane is... =(A,B,C), and normalize it to a modulus of 1. Traverse all point clouds in each component point cloud subset, and calculate the normal vector of each point cloud. For the point cloud subset of transmission line components, since the line is linear, its normal vector is roughly perpendicular to the line direction; while for the point cloud subset of power tower column components, the normal vector is approximately perpendicular to the column surface. Statistically analyze the direction distribution and angle of the normal vector in each component point cloud subset, such as calculating the distribution frequency of the normal vector in each direction, and finally obtain the normal vector distribution characteristics corresponding to each power facility component, reflecting the directional characteristics of the component surface.

[0121] Preferably, the point cloud geometric curvature, point cloud geometric flatness, and normal vector distribution corresponding to each power facility component are combined as geometric features to obtain the geometric features of the power facility component.

[0122] In this embodiment of the invention, the geometric features of the power facility components are obtained by merging the point cloud geometric curvature, point cloud geometric flatness, and normal vector distribution corresponding to each power facility component as geometric features. The Gaussian curvature and average curvature values ​​in the point cloud geometric curvature, the RMSE value of the point cloud geometric flatness, and the directional frequency and normal vector angle of the normal vector distribution are integrated according to a fixed format. For example, for insulator components, the resulting geometric feature data structure is: {Gaussian curvature: [0.01, 0.02, …, 0.03], mean curvature: [0.005, 0.01, …, 0.015], flatness RMSE: 0.02, normal vector direction frequency: {0°-30°: 20%, 30°-60°: 30%, …}, normal vector angle: [10°, 15°, …]}. This integration operation is performed on all power facility components to obtain a complete set of geometric features for each component. These geometric features comprehensively describe the shape, surface characteristics, and other information of the power facility components, providing crucial data support for subsequent 3D modeling and component analysis.

[0123] Furthermore, step S25 includes the following steps:

[0124] Step S251: Perform multi-dimensional comprehensive mapping on the semantic labels and geometric features of power facility components to associate and map the semantic labels and geometric features corresponding to each power facility component, thereby obtaining power facility component feature fusion data;

[0125] In this embodiment of the invention, a three-dimensional hash index structure is constructed by comprehensively mapping the semantic tags and geometric features of power facility components in multiple dimensions. The first dimension index is the semantic tag of the component, the second dimension index is the geometric curvature range, and the third dimension index is the flatness threshold range. The semantic tag of each power facility component (such as "transmission line" or "insulator") is used as the first dimension key value, the combination range of Gaussian curvature and average curvature (such as Gaussian curvature 0.01-0.03 and average curvature 0.005-0.015) is used as the second dimension key value, and the quantile range of the flatness RMSE value (such as 0.01-0.02) is used as the third dimension key value. The specific component data bucket is quickly located through the three-level index, and the semantic tags are associated with the geometric features. For example, for the "insulator" component, when its Gaussian curvature is detected to be in the range of 0.01-0.03 and its flatness RMSE value is in the range of 0.01-0.02, all geometric feature vectors (including normal vector distribution, etc.) of the component are bound to the "insulator" semantic label. This operation is performed on all power facility components to form a hierarchical data structure with semantic labels as root nodes and geometric features as child nodes. Finally, feature fusion data of power facility components is obtained, realizing a deep association between semantic information and geometric information.

[0126] Step S252: Based on the feature fusion data of power facility components, perform spatial coupling quantitative analysis on the spatial relationship between various components within the power facility area to calculate the relative position, contact surface and functional interaction relationship between the components, and obtain spatial coupling data between power facility components;

[0127] In this embodiment of the invention, spatial coupling quantitative analysis is performed on the spatial relationships between components within a power facility area based on the feature fusion data of power facility components. A spatial octree index is constructed, and the power facility area is divided into cubic grid units with a side length of 0.5 meters. Each grid unit is then recursively subdivided into 8 sub-units until the number of components in each sub-unit does not exceed 10. The neighboring components of each component are quickly retrieved through the octree. The neighborhood radius is set to 2 meters. For each component, all other components in its neighborhood are retrieved. When calculating the relative positional relationship between components, the Euclidean distance, azimuth angle, and pitch angle are calculated based on the coordinates of the center point of the component. For example, the distance between the transmission line and the insulator is accurate to 0.01 meters, and the azimuth angle is calculated to 0.1°. When analyzing the contact surface, the normal vector distribution of the components is used. When the angle between the normal vectors of two adjacent surfaces of the components is less than 15° and the distance is less than 0.05 meters, it is determined that a contact surface exists. The area and shape characteristics of the contact surface are calculated. When quantifying the functional interaction relationship, a functional correlation matrix of power facility components is established. For transmission lines and transformers, parameters such as current transmission efficiency and voltage drop are set. The functional coupling strength between the two is calculated through a physical model. For example, when the current intensity of the transmission line changes, the change in the transformer output voltage is calculated based on the transformer's turns ratio parameters. This is used as a quantitative index of functional interaction. All calculation results are integrated to form spatial coupling data between power facility components that includes a relative position matrix, a contact surface feature table, and a functional coupling tensor.

[0128] Step S253: Based on the spatial coupling data between power facility components, perform topological connection analysis on the spatial relationship between each component within the power facility area to obtain the corresponding topological connection relationship between each component within the power facility area, including the connection relationship between each component, the support relationship between components, and the distance constraint relationship between components;

[0129] In this embodiment of the invention, the spatial relationship between components within the power facility area is analyzed based on the spatial coupling data between power facility components to construct a weighted directed graph data structure. Nodes in the graph represent power facility components, and edges represent the connection relationship between components. For physically connected components, such as insulators and transmission lines, a directed edge is added to the graph, and the weight of the edge is set as the connection strength, calculated using parameters such as contact area and number of bolts. For example, consider a contact surface with four bolt connections and a connection strength of 4. When analyzing the support relationships of components, the support hierarchy is determined based on the relative positions and gravity directions of the components. For power towers and transmission lines, the power tower acts as the supporting component, and the transmission line as the supported component. A directed edge is added to the graph from the power tower to the transmission line, with the edge weight equal to the magnitude of the support force. A distance threshold of 5 meters is set based on the weight and span of the transmission line. When the distance between two components is less than this threshold and there is a functional connection, a directed edge is added to the graph, with the edge weight being the reciprocal of the distance, forming a component distance constraint relationship. A depth-first search algorithm is used to traverse the graph structure to determine the connectivity and hierarchical structure between components. For example, starting from the transformer node, all reachable nodes are traversed to determine all components directly or indirectly connected to the transformer. This ultimately forms a topological connection relationship of power components, including a connection matrix, a support relationship tree, and a distance constraint network, providing a complete description of the topological structure between components within the power facility area.

[0130] Step S254: Based on the corresponding power component topology connection relationship between each component in the power facility area, perform power component topology connection to generate a power facility component topology network diagram.

[0131] In this embodiment of the invention, power component topology connections are established based on the corresponding power component topology connections between components within a power facility area. A force-directed layout algorithm is employed, treating power facility components as nodes in a graph and the topology connections between components as edges. Each node is assigned a mass attribute, determined based on the component's physical size and importance; for example, the mass of a transformer is set to 10, and the mass of an insulator is set to 1. Each edge is assigned an elasticity coefficient and a natural length attribute. The elasticity coefficient is determined based on the connection strength, and the natural length is determined based on the actual distance between components. For example, the elasticity coefficient of the edge between a transmission line and an insulator is set to 0.8, and the natural length is set to the actual distance between them. In three-dimensional space, attractive and repulsive forces are applied to the nodes. Attraction brings connected nodes closer, while repulsion moves unconnected nodes further apart. Through iterative calculation, the nodes reach an equilibrium state under the influence of these forces, forming an intuitive topology network layout. Attribute labels are added to the nodes and edges in the topology network graph. Node labels contain information such as component semantic labels and geometric feature statistics, while edge labels contain information such as connection type and weight value. For example, the transformer node label displays "Transformer - Gaussian curvature 0.02 - Flatness RMSE 0.015", and the edge label between the transmission line and the insulator displays "Physical connection - Strength 4". Finally, a topology network diagram of power facility components containing information such as node location coordinates, connection relationships, and attribute labels is generated, providing a structured data foundation for the three-dimensional modeling and analysis of power facilities.

[0132] Furthermore, step S3 includes the following steps:

[0133] Step S31: Based on the power facility component topology network diagram, perform point cloud topology constraint analysis on the power facility area point cloud downsampling optimization data to obtain the power facility component point cloud topology connection constraints and power facility component point cloud geometric constraints.

[0134] In this embodiment of the invention, point cloud topology constraint analysis is performed on the point cloud downsampling optimization data of the power facility area based on the power facility component topology network diagram. First, the nodes in the topology network diagram are mapped one-to-one with the power facility components in the point cloud data. For example, the nodes representing transmission towers in the diagram are mapped to the point cloud set of transmission towers in the point cloud data. For topology connection constraint analysis, the topological relationship between components is determined based on the connection edges between nodes in the network diagram. If there is a connection edge between two nodes, such as the node of the transmission tower being connected to the node of the insulator string, the connection relationship of the corresponding component point cloud is determined in the point cloud data by the spatial distance threshold method. A distance threshold of 0.05 meters is set. When the distance between the nearest point in the point cloud of a transmission tower and the point cloud of an insulator string is less than this threshold, it is determined that there is a topological connection constraint between the two. The location information of the connection point is recorded. In geometric constraint analysis, the attribute information of the nodes in the topological network diagram (such as the geometric shape description of the component) is compared with the point cloud data. For the transformer component, it is known that its attribute in the topological diagram is a cuboid shape. The principal axis direction of the transformer point cloud is calculated in the point cloud data through principal component analysis (PCA). If the deviation of the principal axis direction from the length, width, and height directions of the cuboid exceeds 15°, the geometric shape is constrained and adjusted according to the topological diagram attribute. The deviation value between the point cloud and the ideal geometric shape is calculated, and finally the topological connection constraint and geometric constraint data of the power facility component point cloud are obtained.

[0135] Step S32: Based on the topological connection constraints and geometric constraints of the point cloud of power facility components, perform 3D constraint modeling of the point cloud downsampling optimization data of the power facility area to generate a 3D constraint reconstruction model of the power facility area.

[0136] In this embodiment of the invention, three-dimensional constraint modeling is performed on the downsampled and optimized point cloud data of a power facility area based on point cloud topological connection constraints and geometric constraints. A constraint-based triangulation algorithm is employed. First, the connection relationships between point clouds are determined according to the topological connection constraints. Virtual connection lines are added at the boundaries of point clouds with connection constraints to construct an initial topological skeleton. For geometric constraints, constraints are set for each power facility component point cloud based on its geometric constraint information (such as shape and size). Taking a transmission line as an example, if the geometric constraints indicate that it should be a straight line, the Laplace deformation algorithm is used during triangulation to adjust point clouds that deviate from a straight line shape to conform to the geometric constraints. When constructing the triangular mesh, it is ensured that the interior angle of each triangle is not less than 30° and not greater than 120° to guarantee mesh quality. The triangular meshes of all component point clouds are merged. For the connection areas between components, a weighted average method is used to merge the overlapping point clouds to eliminate gaps at the connections. Finally, a three-dimensional constraint reconstruction model of the power facility area containing accurate topological relationships and geometric shapes is generated. This model fully presents the spatial structure and connection relationships of each component of the power facility.

[0137] Step S33: Perform global point cloud smoothing and homogenization on the three-dimensional constrained reconstruction model of the power facility area to generate a three-dimensional point cloud model of the power facility area.

[0138] In this embodiment of the invention, global point cloud smoothing and homogenization are performed on the 3D constrained reconstruction model of the power facility area. For smoothing, the moving least squares (MLS) method is used. A neighborhood with a radius of 0.1 meters is selected centered on each point cloud, and a local surface is fitted using least squares. The point cloud is then projected onto the fitted surface to achieve a smooth transition. For areas with high curvature, such as the edges of insulator string skirts, the order of the fitted surface is appropriately reduced to enhance the smoothing effect and avoid over-smoothing that could lead to loss of detail. For homogenization, a voxel resampling method is used. The model space is divided into cubic voxels with a side length of 0.03 meters. The number of point clouds within each voxel is counted. If the number of point clouds in a voxel is too high (more than 50), the number of point clouds is reduced through random sampling. If the number of point clouds is too low (less than 5), the point clouds in the voxel's neighborhood are used to supplement the point clouds using inverse distance weighted interpolation. After multiple iterations, the point clouds are evenly distributed throughout the entire model space, ultimately generating a smooth 3D point cloud model of the power facility area with consistent point cloud density.

[0139] Step S34: Obtain the BIM design model of the power facility area;

[0140] In this embodiment of the invention, a BIM design model of the power facility area is acquired and created using professional BIM design software (such as Autodesk Revit). During the design process, based on the design drawings and technical specifications of the power facility, the 3D models of each component are accurately drawn. For transmission towers, their height, crossarm dimensions, and column cross-sectional shape are determined according to design parameters. For transformers, structures such as enclosures, windings, and terminals are constructed according to model specifications. During model construction, a unified coordinate system standard (such as the WGS84 coordinate system) is strictly followed to ensure the accurate spatial position of each component. Simultaneously, detailed attribute information is assigned to each component, including material type, specifications, and technical parameters. After the design is completed, the BIM model is exported as a common IFC (Industry Foundation Classes) format file. This file contains complete geometric information and attribute data, facilitating subsequent comparison and analysis with the 3D point cloud model.

[0141] Step S35: Based on the BIM design model of the power facility area, perform point cloud-BIM alignment evaluation on the 3D point cloud model of the power facility area to generate a 3D alignment deviation matrix of the power facility.

[0142] In this embodiment of the invention, a point cloud-BIM alignment evaluation is performed on a 3D point cloud model of a power facility area based on a BIM design model of the power facility area. First, a coarse registration method based on feature points is used to extract key nodes of power towers, cable endpoints, and other feature points from the BIM model. Then, the Harris corner detection algorithm is used to extract feature points in the point cloud model. An initial correspondence is established by calculating the Euclidean distance of the feature points' 3D coordinates. Singular value decomposition (SVD) is used to solve the initial transformation matrix to achieve preliminary alignment. Finally, the Iterative Closest Point (ICP) algorithm is used for fine registration, with a maximum of 40 iterations and a convergence threshold of 0.0. In each iteration, the nearest point pair between the point cloud model and the BIM model is found. Translation and rotation parameters are calculated, and the transformation matrix is ​​updated. After alignment, the geometric deviation is calculated using a projection method. The point cloud is projected onto the triangular mesh of the BIM model, and the vertical distances corresponding to the points are calculated, including angles, edges, and surfaces. The average distance, standard deviation, maximum distance, and minimum distance are statistically analyzed. For surface overlap assessment, a distance threshold of 0.04 meters is set. Points in the point cloud that are less than this threshold from the surface of the BIM model are considered overlapping points, and the consistency of the normal vectors of the overlapping points is calculated. When the angle between the normal vectors is less than 12°, it is considered a valid overlapping point, and the overlap degree is calculated. Finally, the geometric deviation and surface overlap data are integrated to construct a 4x4 3D alignment deviation matrix for power facilities, which visually displays the alignment status of the two models and provides a quantitative basis for model optimization.

[0143] Furthermore, step S35 includes the following steps:

[0144] Step S351: Spatial alignment and registration of the BIM design model and the 3D point cloud model of the power facility area under the same spatial coordinate system are performed to generate the corresponding BIM model and 3D point cloud model of the power facility area under the same coordinate system.

[0145] In this embodiment of the invention, spatial alignment and registration are performed on the BIM design model and the 3D point cloud model of the power facility area under the same spatial coordinate system. A combination of coarse registration based on feature points and fine registration using the Iterative Closest Point (ICP) algorithm is employed. First, feature points are extracted from both the BIM design model and the 3D point cloud model. For the BIM model, obvious geometric feature points such as the vertices of power towers and the endpoints of cables are selected. For the 3D point cloud model, the Harris corner detection algorithm, with a search radius of 0.1 meters, detects points with significant curvature changes as feature points. These feature points are then used for coarse registration, employing a method based on Euclidean geometry. The nearest neighbor matching algorithm based on the Stanford distance is used to calculate the correspondence between feature points of the BIM model and the point cloud model, construct an initial transformation matrix, and initially align the two models to keep the initial error between them within 0.5 meters. Then, the ICP algorithm is used for fine registration, with a maximum number of iterations of 50 and a convergence threshold of 0.01 meters. By continuously finding the nearest point pair between the point cloud model and the BIM model, calculating translation and rotation parameters, and iteratively updating the transformation matrix, the convergence condition is met. Finally, a BIM model of the power facility area and a 3D point cloud model of the power facility area are generated in the same coordinate system, ensuring that the two models are accurately aligned in space.

[0146] Step S352: Calculate the point cloud-BIM geometric structure deviation between the corresponding BIM model and the 3D point cloud model of the power facility area under the same coordinate system to obtain the power structure alignment deviation between the point cloud model and the BIM model.

[0147] In this embodiment of the invention, the geometric deviation between the point cloud and BIM models of the power facility area and the 3D point cloud model under the same coordinate system is calculated. The geometric structure of the BIM model is discretized into triangular meshes, with each triangular mesh having a side length not exceeding 0.2 meters. For the 3D point cloud model, voxelization is performed at intervals of 0.05 meters. A point-by-point comparison method is used. For each point in the point cloud model, the nearest point is found in the triangular mesh of the BIM model using the projection method. The 3D Euclidean distance between the point and the nearest point is calculated using the formula: ,in The coordinates of a point in a point cloud model. The coordinates of the nearest point in the BIM model are calculated, and the distances between all point cloud points and corresponding points in the BIM model (including angles, edges, and surfaces) are statistically analyzed. The average, standard deviation, maximum, and minimum values ​​of these distances are calculated as the basic indicators of the alignment deviation of the power structure. These values ​​comprehensively reflect the overall deviation and local deviation fluctuations between the point cloud model and the BIM model in terms of geometric structure, and finally, complete data on the alignment deviation of the power structure are obtained.

[0148] Step S353: Based on the alignment deviation of the power structure, perform a point cloud-BIM surface overlap assessment on the corresponding power facility area BIM model and the power facility area 3D point cloud model in the same coordinate system to obtain the surface alignment overlap degree between the point cloud model and the BIM model.

[0149] In this embodiment of the invention, a point cloud-BIM surface overlap assessment is performed on the BIM model and the 3D point cloud model of the power facility area in the same coordinate system based on the alignment deviation of the power structure. A distance threshold of 0.05 meters is set, and points in the point cloud model whose distance to the corresponding point in the BIM model is less than this threshold are considered overlap points. The proportion of overlap points to the total number of points in the point cloud model is calculated as a preliminary indicator of surface alignment overlap. To more accurately assess the overlap, a normal vector-based overlap optimization method is adopted. For each overlap point, its normal vector in the point cloud model and the BIM model is calculated. The consistency of the surface orientation is assessed by calculating the cosine of the angle between the normal vectors, as shown in the formula: ,in and The normal vectors of the overlapping points in the point cloud model and the BIM model are respectively set. The included angle threshold is set to 15°. Only overlapping points with an included angle less than this threshold are included in the final set of overlapping points. The proportion of overlapping points is recalculated to obtain a more accurate surface alignment overlap. For example, if the final calculated overlap ratio is 85%, it indicates that the point cloud model and the BIM model have reached a high degree of surface overlap.

[0150] Step S354: Combine the alignment deviation of the power structure between the point cloud model and the BIM model with the surface alignment overlap to construct the corresponding alignment deviation matrix, so as to generate the three-dimensional alignment deviation matrix of the power facility.

[0151] In this embodiment of the invention, an alignment deviation matrix is ​​constructed by combining the alignment deviation of the power structure between the point cloud model and the BIM model with the surface alignment overlap. The alignment deviation matrix has a 4-row, 4-column structure. The first to third columns correspond to the alignment deviation data of the power structure (including the alignment deviation of angles, edges, and surfaces), and the fourth column corresponds to the surface alignment overlap. This matrix is ​​used as the three-dimensional alignment deviation matrix of the power facility. This matrix comprehensively and intuitively shows the quantitative relationship between the point cloud model and the BIM model in terms of geometric structure deviation and surface overlap, providing important data basis for subsequent adjustment, optimization, design verification, and other work of the three-dimensional model of the power facility.

[0152] Furthermore, step S352 includes the following steps:

[0153] Perform point cloud-BIM geometric alignment matching on the BIM model of the power facility area and the 3D point cloud model of the power facility area under the same coordinate system to obtain geometric alignment matching data of the point cloud model and the power facilities in the BIM model in terms of angle, edge and surface.

[0154] In this embodiment of the invention, point cloud-BIM geometric alignment matching is performed on the BIM model and the 3D point cloud model of the power facility area under the same coordinate system. Regarding angle alignment matching, for components with obvious directional characteristics, such as the columns of power towers and the routing of cables, the axial direction vector of the component in the BIM model is first extracted. For example, for the columns of power towers, the axial direction vector is calculated using the coordinates of its upper and lower endpoints. In the point cloud model, Principal Component Analysis (PCA) is used to calculate the principal component direction vectors based on the column point cloud data. This vector represents the main distribution direction of the point cloud data, and is used as the direction vector of the pillars in the point cloud model. The angle between the two direction vectors is calculated. The angle alignment matching data is obtained. For edge alignment matching, the edge contour of the BIM model is discretized into a series of ordered edge points. For the edge of the transmission line, an edge point is taken every 0.1 meters. For the point cloud model, the Canny edge detection algorithm is used with a gradient threshold of 0.2 and a non-maximum suppression threshold of 0.15 to extract the edge point set of the point cloud. A bidirectional nearest neighbor matching algorithm is used to find the nearest point in the edge point set of the point cloud model for each edge point of the BIM model, and at the same time, to find the nearest point in the edge point set of the BIM model for each edge point of the point cloud model. The coordinate information of these matching point pairs is recorded to form edge alignment matching data. During surface alignment matching, the BIM model edge point is further refined. The surface of the IM model is triangulated, with each triangular mesh having a side length not exceeding 0.2 meters. The point cloud model is voxelized, with each voxel being 0.05 meters in size. For each triangular mesh in the BIM model, the point cloud covering the mesh is found in the point cloud model using a projection method. The percentage of overlap between the point cloud and the triangular mesh is calculated and used as the surface alignment matching data. For example, if the overlap percentage of a triangular mesh in the point cloud model is 0.7, it means that the matching degree of that part of the surface in the two models is 70%. Through the above operations, the complete geometric alignment matching data of the power facilities in the point cloud model and the BIM model in terms of angles, edges, and surfaces is finally obtained.

[0155] Preferably, the alignment deviation is calculated by performing geometric alignment matching data of the power facilities in the point cloud model and the BIM model at the corresponding angles, edges and surfaces, to obtain the power structure alignment deviation between the point cloud model and the BIM model.

[0156] In this embodiment of the invention, alignment deviation between the point cloud model and the BIM model is calculated based on geometric alignment matching data. For angle deviation calculation, the included angle of each component is determined according to previously obtained angle alignment matching data. Set the ideal angle to 0° (i.e., the angle when perfectly aligned), and use the formula... Calculate angular deviations. For example, if the included angle of a power tower column is 8°, then its angular deviation is 8°. Summarize the angular deviations of all components and calculate the average, standard deviation, maximum, and minimum values ​​as indicators of angular alignment deviation in the power structure. For edge deviation calculation, based on the coordinate information of the matching point pairs, calculate the Euclidean distance between each pair of matching points. ,in and The coordinates of the matched edge points in the BIM model and point cloud model are given, the distances of all edge point pairs are counted, and the average, standard deviation, maximum, and minimum distances are calculated to obtain the alignment deviation in terms of the edges. In the surface deviation calculation, based on the overlap area ratio data of surface alignment matching, the overlap area ratio is set to 1 when there is complete overlap, and the formula is used... Calculate surface deviation, The actual overlapping area ratio is used. For example, if the overlapping area ratio of a certain surface is 0.7, then its surface deviation is 0.3. Similarly, the average, standard deviation, maximum, and minimum values ​​of the deviations for all surface areas are calculated. Finally, the deviation calculation results of the angle, edge, and surface are integrated to obtain the complete power structure alignment deviation between the point cloud model and the BIM model, comprehensively reflecting the differences in the geometric structure between the two models.

[0157] Furthermore, step S353 includes the following steps:

[0158] The surface contact area is divided into the BIM model and the 3D point cloud model of the power facility area under the same coordinate system to obtain the relative position of the contact area between the surface of the point cloud model and the BIM model.

[0159] In this embodiment of the invention, the surface contact area is divided into a BIM model and a 3D point cloud model of a power facility area under the same coordinate system. The surface of the BIM model is triangularly meshed, ensuring that the maximum side length of each triangular mesh does not exceed 0.15 meters. Simultaneously, the point cloud model is voxelized at intervals of 0.03 meters. Using a spatial projection method, each triangular mesh of the BIM model is projected onto the space of the point cloud model. If the projected area contains point cloud voxels, the area is determined to be a potential contact area. To accurately determine the contact area, a distance threshold of 0.02 meters is set. For each vertex of the triangular mesh in the BIM model, the five nearest points in the point cloud model are found using a k-nearest neighbor search (k set to 5). The average distance from the vertex to these five points is calculated. If the average distance is less than 0.02 meters, the triangular mesh is marked as a contact area. For example, for a certain part of the surface of a power tower, after the above operations, multiple triangular meshes that meet the conditions are determined. These meshes together constitute the contact area of ​​the BIM model for that part. Similarly, using the point cloud model as a reference, the BIM model is reverse-engineered to obtain the relative position of the contact area between the point cloud model and the BIM model surface, thus clarifying the specific area range of the contact between the two model surfaces.

[0160] Preferably, the contact space distance is calculated based on the relative position of the contact area between the point cloud model and the BIM model surface to obtain the alignment space distance between each pair of contact points within the contact area between the point cloud model and the BIM model surface.

[0161] In this embodiment of the invention, the contact space distance is calculated based on the relative positions of the contact areas between the point cloud model and the BIM model surface. For each pair of contact points within the contact area, the three-dimensional Euclidean distance formula is used. Perform calculations, where These are the coordinates of the contact points in the point cloud model. To determine the coordinates of the corresponding contact points in the BIM model, a bidirectional nearest neighbor matching strategy is adopted when identifying contact point pairs. For each point in the contact area of ​​the BIM model, the nearest point in the contact area of ​​the point cloud model is found as the matching point. At the same time, for each point in the contact area of ​​the point cloud model, the nearest point in the contact area of ​​the BIM model is found. To avoid duplicate matching, a set of already matched points is established to ensure that each pair of contact points is calculated only once. For example, in the contact area between the insulator string and the transmission line, this method calculates the distance between hundreds of pairs of contact points, accurately quantifies the spatial distance relationship between the points of the two models in the contact area, and finally obtains the alignment spatial distance data between each pair of contact points.

[0162] Preferably, based on the alignment deviation of the power structure and combined with the alignment spatial distance between each pair of contact points in the contact area between the point cloud model and the BIM model surface, the point cloud-BIM surface overlap assessment of the corresponding power facility area BIM model and the power facility area 3D point cloud model in the same coordinate system is performed to obtain the surface alignment overlap degree between the point cloud model and the BIM model.

[0163] In this embodiment of the invention, the point cloud-BIM surface overlap assessment of the power facility area BIM model and the 3D point cloud model is performed based on the alignment deviation of the power structure and the alignment spatial distance between contact points within the contact area. First, a distance weighting coefficient is set. =0.6, deviation weighting coefficient =0.4, for each pair of contact points within the contact area, calculate the distance score based on their alignment spatial distance d and the set ideal overlap distance (set to 0 meters). ,in The comprehensive score is calculated by taking the maximum distance among all contact point pairs and combining it with the previously obtained power structure alignment deviation. ,in This is the weighted average of the electrical structure alignment deviation (such as angle deviation, edge deviation, and surface deviation) corresponding to the contact area. The maximum deviation across all contact areas is calculated by summing the comprehensive scores of all contact point pairs and dividing by the total number of contact point pairs to obtain the average comprehensive score. This average score is used as the surface alignment overlap. For example, if the calculated average comprehensive score is 0.85, it means that the surface alignment overlap between the point cloud model and the BIM model is 85%. This clearly quantifies the degree of surface overlap between the two models, providing a crucial quantitative basis for the subsequent adjustment and optimization of the 3D model of power facilities.

[0164] Furthermore, step S4 includes the following steps:

[0165] Step S41: Perform radial deviation mapping on each point within the three-dimensional alignment deviation matrix of the power facility to generate a three-dimensional radial deviation scheduling field for the power facility.

[0166] In this embodiment of the invention, radial deviation mapping is performed on each point within the three-dimensional alignment deviation matrix of the power facility. The three-dimensional alignment deviation matrix is ​​regarded as a set of discrete data points in three-dimensional space. Each point corresponds to a spatial location in the power facility model and carries information on geometric deviation and surface overlap. Radial basis function interpolation is used for mapping. With each matrix point as the center, the influence radius is set to 0.1 meters. For any point to be interpolated within the power facility area, the Euclidean distance between it and each point in the matrix is ​​calculated. And according to the distance weight formula (in To prevent extremely small constants with a denominator of zero, a value of 10 is used. −6 Determine the influence weight of each matrix point on the interpolation point using multiple quadratic functions. ( Taking 0.05 as the radial basis function, the radial deviation value of the interpolation point is calculated by weighted summation: For example, in the area corresponding to the transformer component, the above calculation assigns a radial deviation value to each 0.01m × 0.01m × 0.01m grid node in the area, ultimately forming a three-dimensional radial deviation scheduling field covering the entire power facility area, which intuitively shows the deviation distribution at each location.

[0167] Step S42: Based on the three-dimensional radial deviation scheduling field of the power facility, perform three-dimensional deformation correction on the three-dimensional alignment deviation matrix of the power facility to generate the three-dimensional deformation field of the power facility;

[0168] In this embodiment of the invention, the three-dimensional alignment deviation matrix of the power facility is constructed by performing three-dimensional deformation correction based on the three-dimensional radial deviation scheduling field of the power facility. A finite element mesh generation method is used to divide the three-dimensional model of the power facility into hexahedral mesh elements with a side length of 0.02 meters, ensuring that each component can be reasonably discretized. For each mesh element, the radial deviation value corresponding to the center of the element is obtained based on its position in the three-dimensional radial deviation scheduling field. Combining Hooke's law from mechanics of materials, the deformation of the element in the three coordinate axes (x, y, z) is calculated. Assuming the radial deviation value at the center of a certain element is... The elastic modulus of the unit material is E (here, it is uniformly set to the elastic modulus of steel, 2.06 × 10⁻⁶). 11 Pa), with Poisson's ratio ν taken as 0.3, using the formula ( Unit in Calculation of the original dimensions of the direction Similarly, the deformation in the y and z directions is calculated. Taking a certain grid cell of a transmission tower as an example, the deformation in the x direction is 0.002 meters, the deformation in the y direction is 0.0015 meters, and the deformation in the z direction is 0.0025 meters. The above calculation is performed on all grid cells, and the deformation of each cell is combined into a three-dimensional vector to form a three-dimensional deformation field of the power facility, which describes the deformation trend and degree of each part of the model.

[0169] Step S43: Perform three-dimensional deformation inversion analysis on the three-dimensional point cloud model of the power facility area based on the three-dimensional deformation field of the power facility to obtain the elastic modulus, connection stiffness and boundary condition deformation constraints corresponding to the three-dimensional point cloud model.

[0170] In this embodiment of the invention, a three-dimensional deformation inversion analysis is performed on a three-dimensional point cloud model of a power facility area based on the three-dimensional deformation field of the power facility. An inversion analysis algorithm is employed, and an objective function is constructed based on the least squares method. ,in The deformation observed in the three-dimensional deformation field. To calculate the deformation using a theoretical model, for elastic modulus inversion, with fixed connection stiffness and boundary conditions, the elastic modulus value is iteratively adjusted (initial value set at 2.06 × 10⁻⁶). 11 Pa, with a step size of 10 for each iteration. 9 To minimize the objective function F, for example, after 12 iterations, when the elastic modulus is 2.02 × 10⁻⁶ Pa, the objective function F is minimized. 11 When Pa, the objective function reaches its minimum value. In the connection stiffness inversion, the determined elastic modulus and boundary conditions are fixed. Taking the connection between the insulator string and the transmission tower as an example, the stiffness coefficient of the connection spring is adjusted (initial value set to 10). 8 N / m, iteration step size is 10 6 The calculation was performed using N / m, and after 8 iterations, the connection stiffness coefficient was obtained as 1.2 × 10⁻⁶. 8 For the boundary condition deformation constraints, by analyzing the actual installation of power facilities, fixed boundaries (such as the bottom of the tower being completely fixed) and elastic boundaries (such as the elastic constraints of the cable suspension points) are determined. The results are then corrected by combining the deformation field data, and finally, the elastic modulus, connection stiffness and boundary condition deformation constraint parameters corresponding to the accurate three-dimensional point cloud model are obtained.

[0171] Step S44: Based on the elastic modulus, connection stiffness and boundary condition deformation constraints corresponding to the three-dimensional point cloud model, adjust the three-dimensional deformation constraints of the three-dimensional point cloud model of the power facility area to generate a three-dimensional deformation reconstruction model of the power facility area.

[0172] In this embodiment of the invention, the three-dimensional deformation constraint adjustment of the three-dimensional point cloud model of the power facility area is performed based on the elastic modulus, connection stiffness, and boundary condition deformation constraints corresponding to the three-dimensional point cloud model. A deformation algorithm based on physical constraints is adopted, treating each point in the three-dimensional point cloud model as a particle with mass, with particles connected by virtual springs. The elastic coefficient of the springs is determined by the connection stiffness. For boundary condition constraints, the positions of particles at fixed boundaries are locked, preventing displacement. For particles at elastic boundaries, corresponding elastic forces are applied according to the boundary condition deformation constraints. Based on Newton's second law F=ma, the force exerted on each particle under the constraint is calculated. Taking a point cloud model of a power tower as an example, the acceleration, velocity, and displacement under force conditions are calculated iteratively (with a time step of 0.01s for each iteration, and 100 iterations) to update the position of each point cloud particle. For areas where the point cloud density changes due to deformation, a density-based resampling method is used. When the point cloud density change in a certain area exceeds 20%, adjustments are made by adding or deleting points to ensure the integrity and accuracy of the model. Finally, a three-dimensional deformation reconstruction model of the power facility area that conforms to the actual physical characteristics is generated, achieving accurate optimization of the three-dimensional model of the power facility.

[0173] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0174] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A point cloud data driven 3D modeling adjustment method, characterized in that, The method comprises the following steps: Step S1: Obtain power facility area three-dimensional point cloud data, and perform point cloud down-sampling repair processing on the power facility area three-dimensional point cloud data to obtain power facility area point cloud down-sampling optimization data; Step S2: Perform power component semantic segmentation on the power facility area point cloud down-sampling optimization data to obtain power facility component semantic labels; and perform power component geometric analysis on the power facility area point cloud down-sampling optimization data based on the power facility component semantic labels to obtain power facility component geometric features; Perform power component topology connection based on the power facility component semantic labels and the power facility component geometric features to generate a power facility component topology network graph; Step S3: Perform point cloud three-dimensional modeling on the power facility area point cloud down-sampling optimization data based on the power facility component topology network graph to generate a power facility area three-dimensional point cloud model; obtain a power facility area BIM design model, and perform point cloud-BIM alignment evaluation on the power facility area three-dimensional point cloud model based on the power facility area BIM design model to generate a power facility three-dimensional alignment deviation matrix; wherein the point cloud-BIM alignment evaluation comprises the following steps: Step S351: Perform spatial alignment registration on the power facility area BIM design model and the power facility area three-dimensional point cloud model in the same spatial coordinate system to generate corresponding power facility area BIM models and power facility area three-dimensional point cloud models in the same coordinate system; Step S352: Perform point cloud-BIM geometric structure deviation calculation on the corresponding power facility area BIM models and power facility area three-dimensional point cloud models in the same coordinate system to obtain power structure alignment deviation amounts between the point cloud model and the BIM model; Step S353: Perform point cloud-BIM surface coincidence evaluation on the corresponding power facility area BIM models and power facility area three-dimensional point cloud models in the same coordinate system based on the power structure alignment deviation amounts to obtain surface alignment coincidence degrees between the point cloud model and the BIM model; Step S354: Combine the power structure alignment deviation amounts and the surface alignment coincidence degrees between the point cloud model and the BIM model to construct a corresponding alignment deviation matrix to generate the power facility three-dimensional alignment deviation matrix; Step S4: Construct a corresponding power facility three-dimensional deformation field based on the power facility three-dimensional alignment deviation matrix, and perform three-dimensional deformation constraint adjustment on the power facility area three-dimensional point cloud model based on the power facility three-dimensional deformation field to generate a power facility area three-dimensional deformation reconstruction model; wherein step S4 comprises the following steps: Step S41: Perform radial deviation mapping on each point corresponding to the power facility three-dimensional alignment deviation matrix to generate a power facility three-dimensional radial deviation scheduling field; Step S42: Perform three-dimensional deformation correction construction on the power facility area three-dimensional point cloud model based on the power facility three-dimensional radial deviation scheduling field to generate a power facility three-dimensional deformation field; Step S43: Perform three-dimensional deformation inversion analysis on the power facility area three-dimensional point cloud model based on the power facility three-dimensional deformation field to obtain an elastic modulus, a connection stiffness, and a boundary condition deformation constraint corresponding to the three-dimensional point cloud model; Step S44: based on the elastic modulus, connection stiffness and boundary condition deformation constraints corresponding to the three-dimensional point cloud model, the three-dimensional deformation constraint adjustment is performed on the three-dimensional point cloud model of the power facility area to generate a three-dimensional deformation reconstruction model of the power facility area.

2. The point cloud data driven three-dimensional modeling adjustment method of claim 1, wherein, Step S1 includes the following steps: Step S11: obtaining three-dimensional point cloud data of the power facility area; Step S12: obtaining the corresponding power facility area point cloud density distribution through the power facility area three-dimensional point cloud data, and performing point cloud down-sampling processing on the power facility area three-dimensional point cloud data based on the power facility area point cloud density distribution to obtain power facility area point cloud down-sampling data; Step S13: obtaining the corresponding regional spatial distribution through the power facility area, and performing spatial hierarchical differentiation analysis on the power facility area point cloud down-sampling data based on the regional spatial distribution to obtain power facility area point cloud hierarchical differentiation data; Step S14: based on the power facility area point cloud hierarchical differentiation data, the boundary difference of the power facility area point cloud boundary in the power facility area point cloud down-sampling data is corrected to obtain power facility area point cloud down-sampling correction data; Step S15: performing missing point cloud repair processing based on adaptive interpolation on the power facility area point cloud down-sampling correction data to obtain power facility area point cloud down-sampling optimization data.

3. The point cloud data driven three-dimensional modeling adjustment method of claim 1, wherein, Step S2 includes the following steps: Step S21: extracting the power facility area point cloud down-sampling optimization data to obtain different power facility component contours in the power facility area, including power transmission tower, insulator string, cable and transformer corresponding power facility components; Step S22: based on the different power facility component contours in the power facility area, the component spatial region is divided to obtain the spatial region point cloud data corresponding to the different components in the power facility area; Step S23: using a deep learning-based semantic segmentation network to perform power component semantic segmentation on the spatial region point cloud data corresponding to the different components in the power facility area to obtain power facility component semantic labels; Step S24: based on the power facility component semantic labels, the power facility component geometry analysis is performed on the power facility area point cloud down-sampling optimization data to obtain power facility component geometric features; Step S25: based on the power facility component semantic labels and the power facility component geometric features, the power component topology connection is performed to generate a power facility component topology network graph.

4. The point cloud data driven three-dimensional modeling adjustment method of claim 3, wherein, Step S24 includes the following steps: based on the power facility component semantic labels, the component semantic point cloud is divided to obtain the power facility component point cloud subset corresponding to each semantic label; performing point cloud curvature statistical analysis on each power facility component point cloud subset corresponding to each semantic label to obtain the point cloud geometric curvature corresponding to each power facility component; based on the point cloud geometric curvature corresponding to each power facility component, the component flatness estimation is performed on the corresponding power facility component point cloud subset to obtain the point cloud geometric flatness corresponding to each power facility component; The normal vector distribution corresponding to each power facility component is obtained by the point cloud subset of each power facility component corresponding to the semantic label of each belt; The point cloud geometry curvature, point cloud geometry flatness and normal vector distribution corresponding to each power facility component are combined as geometric features to obtain power facility component geometric features.

5. The point cloud data driven three-dimensional modeling adjustment method of claim 3, wherein, Step S25 includes the following steps: Step S251: multi-dimensional comprehensive mapping is performed on the power facility component semantic label and the power facility component geometric features to associate and map the semantic label and the geometric features corresponding to each power facility component to obtain power facility component feature fusion data; Step S252: spatial coupling quantitative analysis is performed on the spatial relationship between each component in the power facility area based on the power facility component feature fusion data to calculate the relative position, contact surface and functional interaction relationship between each component to obtain spatial coupling data between power facility components; Step S253: based on the spatial coupling data between power facility components, the spatial relationship between each component in the power facility area is analyzed to obtain the corresponding power component topological connection relationship between each component in the power facility area, including the connection relationship between each component, the component support relationship and the component distance constraint relationship; Step S254: power component topology connection is performed based on the corresponding power component topological connection relationship between each component in the power facility area to generate a power facility component topological network graph.

6. The point cloud data driven three-dimensional modeling adjustment method of claim 1, wherein, Step S3 includes the following steps: Step S31: based on the power facility component topological network graph, point cloud topological constraint analysis is performed on the power facility area point cloud downsampling optimization data to obtain power facility component point cloud topological connection constraints and power facility component point cloud geometric constraints; Step S32: based on the power facility component point cloud topological connection constraints and the power facility component point cloud geometric constraints, point cloud three-dimensional constraint modeling is performed on the power facility area point cloud downsampling optimization data to generate a power facility area three-dimensional constraint reconstruction model; Step S33: global point cloud smoothing and uniformization is performed on the power facility area three-dimensional constraint reconstruction model to generate a power facility area three-dimensional point cloud model; Step S34: obtaining a power facility area BIM design model; Step S35: based on the power facility area BIM design model, point cloud-BIM alignment evaluation is performed on the power facility area three-dimensional point cloud model to generate a power facility three-dimensional alignment deviation matrix.

7. The point cloud data driven three-dimensional modeling adjustment method of claim 1, wherein, Step S352 includes the following steps: Point cloud-BIM geometric alignment matching is performed on the corresponding power facility area BIM model and the power facility area three-dimensional point cloud model in the same coordinate system to obtain geometric alignment matching data of the power facilities in the angle, edge and surface corresponding to the point cloud model and the BIM model; The geometric alignment matching data of the power facilities in the angle, edge and surface corresponding to the point cloud model and the BIM model are calculated to obtain the power structure alignment deviation amount between the point cloud model and the BIM model.

8. The point cloud data driven three-dimensional modeling adjustment method of claim 1, wherein, Step S353 includes the following steps: The corresponding power facility area BIM model and the power facility area three-dimensional point cloud model under the same coordinate system are subjected to surface contact area division to obtain the relative position of the contact area between the point cloud model and the BIM model surface; The contact space distance is calculated according to the relative position of the contact area between the point cloud model and the BIM model surface to obtain the alignment space distance between each pair of contact points in the contact area between the point cloud model and the BIM model surface; Based on the power structure alignment deviation amount and in combination with the alignment space distance between each pair of contact points in the contact area between the point cloud model and the BIM model surface, the point cloud-BIM surface coincidence evaluation is performed on the corresponding power facility area BIM model and the power facility area three-dimensional point cloud model under the same coordinate system to obtain the surface alignment coincidence degree between the point cloud model and the BIM model.

Images