A digital-twin-based substation three-dimensional modeling design method

Abstract

The application discloses a kind of based on digital twinning substation three-dimensional modeling design method, belong to substation three-dimensional modeling and digital twinning technical field.The method includes: to the space topology mapping of multi-source heterogeneous equipment, static structure and dynamic parameter are synchronously converted into initial three-dimensional point cloud model;Extract device relative coordinate and connection relationship, construct hierarchical correlation matrix based on space constraint, locate geometric dependence path;Point cloud model is dynamically sampled and adjusted using the path, extracts geometric feature set and establishes multi-scale geometric constraint network;Based on the network, gradually refine reconstruction, generate structured substation three-dimensional basic model;Obtain environmental parameter embedding space coordinate node, generate digital twinning model containing environmental dynamic information.The application realizes static and dynamic parameter unified mapping and high-fidelity reconstruction, provides support for intelligent operation and maintenance.

Description

Technical Field

[0001] This invention relates to the field of substation 3D modeling and digital twin technology, specifically a substation 3D modeling design method based on digital twin. Background Technology

[0002] 3D modeling of substations is a core component in constructing digital twin substations. Existing modeling methods often rely on laser point cloud or image data collected by a single type of sensor, and obtain the 3D model of the substation through direct surface reconstruction or parametric modeling. These methods struggle to deeply integrate the static structural data and dynamic operating parameters of the equipment under a unified spatiotemporal reference, resulting in models that only reflect the external outline of the substation equipment and fail to express the spatial topological constraints and connectivity dependencies between the equipment.

[0003] During model building, due to the lack of quantitative assessment of the importance and spatial sparsity of equipment in the overall substation layout, point cloud sampling and surface reconstruction often adopt a globally uniform processing strategy, failing to adaptively adjust the model's level of detail based on the actual geometric dependencies between equipment. This results in geometric distortion in critical connection areas of the reconstructed model, making it difficult to guarantee the positional continuity and directional consistency of equipment boundaries and docking areas, thus limiting the model's usability in subsequent engineering design, interference detection, and simulation analysis.

[0004] Furthermore, existing 3D models of substations are generally static geometries, lacking the ability to integrate dynamic environmental information such as temperature, humidity, and electromagnetic field strength. Even if environmental parameters can be linked as independent layers or external data, it is difficult to express them synchronously with the geometric structure at the model's spatial coordinate nodes. This makes it impossible for the model to intuitively reflect the spatiotemporal changes in the operating environment of the equipment, and thus fails to meet the needs of real-time state mapping and dynamic prediction in digital twin applications.

[0005] To address the aforementioned shortcomings, two issues need to be resolved: First, how to establish a hierarchical geometric dependency representation based on the spatial constraints and connections between devices, and use this to drive the adaptive sampling of point clouds and the construction of multi-scale geometric constraint networks, thereby generating a geometrically accurate and topologically closed structured model; Second, how to embed synchronously collected environmental dynamic parameters into the attributes of model nodes according to their spatial locations on a basic model with complete device geometric structures, forming a digital twin that can synchronously carry and intuitively display environmental dynamic information. Summary of the Invention

[0006] This invention provides a digital twin-based 3D modeling and design method for substations, aiming to solve the problems that existing substation modeling methods cannot effectively utilize the geometric dependencies between equipment for differentiated modeling, and that the generated 3D models lack the ability to embed dynamic environmental information. It achieves the construction of a digital twin 3D model of a substation with accurate geometric structure, consistent spatial constraints, and dynamic environmental perception capabilities.

[0007] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides a three-dimensional modeling and design method for substations based on digital twins, the method comprising: Spatial topology mapping is performed on multi-source heterogeneous equipment in the substation scenario, and the collected static structural data and dynamic operating parameters of the equipment are synchronously converted into an initial three-dimensional point cloud model. Based on the initial 3D point cloud model, the relative spatial coordinates of each device and the connection relationship between the devices are extracted, and a hierarchical association matrix based on spatial constraints is constructed. The association matrix is ​​used to locate the geometric dependency path between devices. The geometric dependency path is used to perform dynamic sampling and adjustment on the initial 3D point cloud model. Geometric feature sets for each device are extracted from the sampled and adjusted 3D point cloud model, and a multi-scale geometric constraint network is established based on the geometric feature sets. Based on the multi-scale geometric constraint network, the initial three-dimensional point cloud model is progressively refined and reconstructed to generate a three-dimensional basic model of a structured substation. Obtain the substation operating environment parameters, embed the operating environment parameters into the spatial coordinate nodes of the structured substation three-dimensional basic model, and generate a digital twin three-dimensional model containing dynamic environmental information.

[0008] By incorporating hierarchical spatial constraints and multi-scale geometric constraints into the modeling process, the final digital twin 3D model can accurately reflect the physical connections and spatial dependencies between devices, while also carrying real-time operating environment information, providing a high-precision and highly consistent data foundation for the 3D design of substations.

[0009] As a preferred technical solution of the present invention, the spatial topology mapping of multi-source heterogeneous equipment in a substation scenario includes: acquiring equipment outline point cloud data, equipment identification image data, and real-time equipment operating parameters from laser scanning equipment, image acquisition equipment, and operating status monitoring equipment deployed in the substation scenario, respectively; aligning the equipment outline point cloud data, equipment identification image data, and real-time equipment operating parameters according to a unified timestamp; performing spatial coordinate normalization on the aligned equipment outline point cloud data; and mapping the normalized equipment outline point cloud data to the substation global coordinate system to generate the initial three-dimensional point cloud model. By temporally aligning and fusing multi-source heterogeneous data, the inconsistency between time and spatial references is eliminated, improving the information integrity and coordinate accuracy of the initial three-dimensional point cloud model.

[0010] Preferably, the construction of the hierarchical association matrix based on spatial constraints includes: identifying the point cloud cluster corresponding to each device from the initial 3D point cloud model, and extracting the centroid coordinates of each point cloud cluster as the relative spatial coordinates of the device; determining the cable connection relationships and support fixing relationships between devices according to the physical layout rules of the substation, and converting the cable connection relationships and support fixing relationships into connection relationships between devices; constructing the hierarchical association matrix with the relative spatial coordinates of each device as diagonal elements and the connection relationships between devices as off-diagonal elements, and nesting spatial distance constraints and connection direction constraints layer by layer in the hierarchical association matrix to form a matrix structure for locating geometric dependency paths between devices. This matrix structure deeply couples the spatial topology and physical connection characteristics between devices, providing a quantitative basis for subsequent geometric dependency transmission path design.

[0011] As a preferred technical solution of the present invention, the process of establishing a multi-scale geometric constraint network includes: determining the sampling rate adjustment coefficient for each device based on its importance level and spatial location in the geometrically dependent path; performing density increase / decrease operations on the point clouds of the corresponding devices in the initial 3D point cloud model according to the sampling rate adjustment coefficient to obtain a density-adjusted 3D point cloud model; extracting the surface curvature, edge position, and plane normal of each device from the density-adjusted 3D point cloud model, and combining the surface curvature, edge position, and plane normal into a geometric feature set for each device; grouping the geometric feature set of each device according to its hierarchical position in the hierarchical association matrix, and establishing internal geometric constraints and relative geometric constraints between devices at different scales based on the grouped geometric feature sets to form a multi-scale geometric constraint network. Dynamically adjusting the sampling density according to the topological importance and spatial distribution of the devices can effectively reduce point cloud redundancy in non-critical areas while ensuring the geometric feature resolution capability of key areas, improving processing efficiency and enhancing the structural expression capability of the multi-scale constraint network.

[0012] The preferred method for determining the sampling rate adjustment coefficient for each device is as follows: Extract the depth and connectivity values ​​of each device's level from the hierarchical association matrix. Normalize these values ​​and then sum them using weighted averages to obtain the importance level score for each device. Calculate the average spatial distance between each device and all other devices in the hierarchical association matrix based on its relative spatial coordinates, and use this average spatial distance as the spatial sparsity of the device. Multiply the importance level score of each device by a preset importance level weighting coefficient to obtain a first weighting coefficient. Multiply the spatial sparsity of the device by a preset sparsity weighting coefficient to obtain a second weighting coefficient. Add the first and second weighting coefficients to obtain the sampling rate adjustment coefficient for the device. By comprehensively considering topological centrality and spatial distribution characteristics to quantify sampling requirements, the sampling adjustment better reflects the actual geometric dependency distribution of substation equipment, improving the targeting and precision of subsequent geometric feature extraction.

[0013] Preferably, the method for extracting the geometric feature set from the density-adjusted 3D point cloud model is as follows: A local neighborhood search is performed on the point cloud corresponding to each device in the density-adjusted 3D point cloud model. A least-squares plane is fitted within the local neighborhood of each point. The principal curvature and Gaussian curvature of the point are calculated based on the fitted least-squares plane. The average principal curvature and the average Gaussian curvature of all points within the same device are used as the surface curvature of the device. The average Gaussian curvature may be positive, zero, or negative, corresponding to the geometric features of elliptical, parabolic, or hyperbolic points on the surface, respectively. Retaining the sign information can fully characterize the surface concavity / convexity and saddle-shaped structure of the device. Gradient calculation is performed on the point cloud. Points in the point cloud whose normal change rate exceeds a preset threshold are extracted as candidate edge points. Connecting lines are traced on the candidate edge points to obtain the edge position of the device. Principal component analysis is performed on the point cloud. The eigenvector corresponding to the minimum eigenvalue of the point cloud covariance matrix is ​​extracted as the plane normal of each point. The average plane normal of all points within the same device is used as the plane normal of the device. The obtained geometric feature set fully describes the surface morphology, edge orientation, and orientation information of the device, providing high-quality input for multi-scale geometric constraint networks.

[0014] As a technical solution of this invention, the process of generating a three-dimensional basic model of a structured substation includes: extracting the internal geometric constraints of each device at the current scale from a multi-scale geometric constraint network; performing local surface fitting on the point cloud of the corresponding device in the initial three-dimensional point cloud model based on the internal geometric constraints to generate a local refined surface of the device at the current scale; mapping the relative geometric constraints between devices in the multi-scale geometric constraint network to the docking boundary conditions between the local refined surfaces; stitching the local refined surfaces of all devices into a global refined surface based on the docking boundary conditions; and performing meshing processing on the global refined surface to generate a structured mesh model with topological connections. By using multi-scale constraint-driven hierarchical reconstruction and global stitching, the final generated three-dimensional basic model can meet engineering design requirements in terms of surface continuity and geometric compatibility, clearly expressing the physical form of the devices and their interconnections.

[0015] Furthermore, the process of generating the locally refined surface preferably includes: extracting the surface curvature constraints, edge position constraints, and plane normal constraints of the current device at the current scale from the multi-scale geometric constraint network, and combining the surface curvature constraints, edge position constraints, and plane normal constraints into a set of internal geometric constraints at the current scale; selecting the point cloud corresponding to the current device from the initial 3D point cloud model according to the constraint set, and performing bicubic B-spline surface fitting on the selected point cloud according to the surface curvature constraints to obtain a preliminary fitted surface; applying the edge position constraints and plane normal constraints as boundary constraints to the preliminary fitted surface, and adjusting the coordinates of the control points of the preliminary fitted surface to ensure that the surface satisfies positional continuity and normal continuity at the edge positions. Through the forced edge constraints and normal consistency control, the locally refined surface can accurately restore the sharp geometric edges of the device while maintaining a smooth transition, thus improving the geometric fidelity of the device model.

[0016] Furthermore, the preferred process for stitching together the overall refined surfaces includes: extracting the relative position constraints and relative orientation constraints between every two adjacent devices from the multi-scale geometric constraint network, and using these constraints as initial docking boundary conditions between the local refined surfaces of the two adjacent devices; based on the initial docking boundary conditions, calculating the first boundary point sequence on the docking boundary of the first device's local refined surface and the second boundary point sequence on the docking boundary of the second device's local refined surface, optimizing the initial docking boundary conditions by minimizing the spatial distance between the first and second boundary point sequences, and obtaining optimized docking boundary conditions; based on the optimized docking boundary conditions, aligning and stitching the docking boundaries of adjacent local refined surfaces, and merging and retriangulating the mesh vertices at the stitching points, gradually completing the unified stitching of all station equipment surfaces. Through boundary optimization and mesh retriangulation, stitching gaps and topological inconsistencies are eliminated, giving the overall refined surface continuous spatial wiring and a complete topological structure.

[0017] As a preferred embodiment of the present invention, the process of generating a digital twin 3D model containing dynamic environmental information includes: collecting temperature, humidity, and electromagnetic field strength parameters from environmental sensors deployed in the substation; aligning the temperature, humidity, and electromagnetic field strength parameters with the spatial coordinate nodes of the structured substation 3D basic model according to the collection time; mapping the aligned temperature, humidity, and electromagnetic field strength parameters to the attribute fields of the corresponding spatial coordinate nodes in the structured substation 3D basic model, forming a set of node attributes carrying dynamic environmental information; updating the display parameters of the structured substation 3D basic model according to the set of node attributes of each spatial coordinate node, thereby generating the digital twin 3D model containing dynamic environmental information. This allows the 3D model to not only carry geometric and topological information but also reflect the temperature, humidity, and electromagnetic field distribution at the substation site in real time, achieving dynamic visualization and synchronous mapping of the physical substation's operating status.

[0018] The technical effects and advantages provided by the present invention in the above technical solution are as follows: By extracting the relative spatial coordinates and connection relationships of each device from the initial 3D point cloud model, a hierarchical association matrix based on spatial constraints is constructed, which accurately describes the hierarchical dependencies and geometric connectivity paths of devices in the substation space. This matrix nests spatial distance constraints and connection direction constraints layer by layer, clarifying the geometric dependency path of each device in the overall structure. Based on this geometric dependency path, the importance level score and spatial sparsity of each device are further calculated, and a sampling rate adjustment coefficient is generated accordingly to perform differentiated upsampling or downsampling operations on the point cloud. This sampling method can retain more geometric details for devices located in critical connection areas and with dense spatial distribution, while appropriately simplifying the point cloud for isolated and highly redundant devices. This improves reconstruction efficiency while effectively avoiding the loss of important geometric features, overcoming the shortcomings of the uniform sampling strategy where detail loss and data redundancy coexist.

[0019] Based on the density-adjusted point cloud data, surface curvature, edge positions, and plane normals are extracted from the point cloud of each device, combined to form a geometric feature set, and a multi-scale geometric constraint network is established according to the hierarchical grouping in the hierarchical association matrix. This network defines the curvature, edge, and normal constraints within the devices, as well as the relative position and orientation constraints between devices. During the progressive refinement and reconstruction process, local bicubic B-spline surface fitting is performed on the point cloud using the internal constraints of the devices, and control points are adjusted through edge and normal constraints to ensure that the surface satisfies positional and normal continuity at the edges. Then, based on the relative geometric constraints between devices, the locally refined surfaces of adjacent devices are aligned and stitched at the docking boundary, optimizing the spatial distance of the boundary point sequence and completing mesh vertex merging and re-triangulation to form the overall structured basic model. This process unifies the geometric accuracy of the model and the topological consistency between devices within the same multi-scale constraint framework. The reconstructed structured mesh model maintains accurate geometric relationships at device edges, surface transitions, and device interfaces, avoiding splicing misalignment and gap problems in conventional modeling methods.

[0020] Based on a structured 3D substation base model, parameters such as temperature, humidity, and electromagnetic field strength are collected from environmental sensors deployed throughout the substation. These parameters are then time-aligned with the spatial coordinate nodes of the model according to the time of collection, mapping them to the attribute fields of the corresponding nodes and updating the node display parameters. This generates a digital twin 3D model, enabling each spatial node of the substation to simultaneously carry geometric shape and dynamic environmental attributes, thus presenting real-time changes in the operating environment at the equipment's location. This approach, deeply embedding dynamic environmental information into the model's spatial nodes, gives the digital twin model the ability to directly reflect and visualize the multi-physics distribution of the substation. It can be applied to scenarios such as condition monitoring, fault prediction, and operational simulation, compensating for the lack of dynamic environmental information in traditional static 3D models. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0022] Figure 1 This is a flowchart of a digital twin-based 3D modeling and design method for substations. Figure 2 This is a flowchart of multi-source data alignment and initial 3D point cloud model generation; Figure 3 It is a flowchart for the generation and visualization of dynamic environmental information of a digital twin 3D model of a substation. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] See Figure 1 This invention provides a digital twin-based 3D modeling and design method for substations. Specifically, firstly, spatial topology mapping is performed on the multi-source heterogeneous equipment in the substation scenario, synchronously converting the collected static structural data and dynamic operating parameters of the equipment into an initial 3D point cloud model. Then, based on this initial 3D point cloud model, the relative spatial coordinates of each device and the connection relationships between devices are extracted, constructing a hierarchical association matrix based on spatial constraints. This association matrix is ​​used to locate the geometric dependency paths between devices. Dynamic sampling and adjustment are performed on the initial 3D point cloud model using these geometric dependency paths. Geometric feature sets for each device are extracted from the sampled 3D point cloud model, and a multi-scale geometric constraint network is established based on these feature sets. Based on the multi-scale geometric constraint network, the initial 3D point cloud model is progressively refined and reconstructed to generate a structured 3D basic model of the substation. Next, substation operating environment parameters are acquired and embedded into the spatial coordinate nodes of the structured substation 3D basic model to generate a digital twin 3D model containing dynamic environmental information. Example 1

[0025] In specific implementation, please refer to Figure 2Spatial topology mapping is performed on multi-source heterogeneous equipment in the substation scenario, and the collected static structural data and dynamic operating parameters of the equipment are synchronously converted into an initial three-dimensional point cloud model.

[0026] Laser scanning equipment, image acquisition equipment, and operational status monitoring equipment deployed in the substation environment acquire equipment outline point cloud data, equipment identification image data, and real-time equipment operating parameters, respectively. Laser scanning equipment uses multi-station scanning to obtain a set of three-dimensional coordinate points on the surface of equipment within the substation, forming equipment outline point cloud data. Each point in the equipment outline point cloud data carries local coordinate values ​​and reflection intensity information. Image acquisition equipment captures images of the equipment's exterior and extracts equipment identification text or nameplate patterns from these images to generate equipment identification image data. This equipment identification image data includes text recognition results of the equipment number, model, and rated parameters. Operational status monitoring equipment collects real-time measurements of voltage amplitude, current amplitude, and active power during equipment operation as real-time operating parameters. These real-time operating parameters are marked with the sampling time of the status monitoring equipment itself.

[0027] In practice, the equipment outline point cloud data, equipment identification image data, and equipment real-time operating parameters are aligned using a unified timestamp. Using a pre-determined unified time reference sequence, for each time point in the sequence, the absolute value of the time difference between that time point and the corresponding equipment outline point cloud data frame, equipment identification image data frame, and equipment real-time operating parameter sampling frame is searched for. If the timestamps of all three fall within the same time tolerance threshold, a correspondence is established, achieving data alignment at that time point. When no corresponding data frame exists for a certain type of data within the time tolerance threshold, linear interpolation is performed using the data frame of that type with the smallest time distance from that time point to generate a supplementary data frame, which is then incorporated into the alignment. After alignment, a set of mutually corresponding equipment outline point cloud data, equipment identification image data, and equipment real-time operating parameters is obtained at each unified time point.

[0028] Next, spatial coordinate normalization is performed on the aligned device outline point cloud data. The mean and standard deviation of the three-dimensional coordinate components of all points in the aligned device outline point cloud data are calculated. The mean is subtracted from the three-dimensional coordinates of each point and the standard deviation is divided to obtain the spatially normalized device outline point cloud data, so that the distribution of the normalized point cloud in the three coordinate axes has zero mean and unit standard deviation.

[0029] The normalized equipment outline point cloud data is mapped to the substation global coordinate system to generate an initial 3D point cloud model. The mapping process utilizes a rigid transformation matrix, pre-determined from the local scanning coordinate system to the substation global coordinate system when the laser scanning equipment is installed in the substation scene. The local coordinates of each point in the normalized equipment outline point cloud data are multiplied by the rigid transformation matrix to obtain the 3D spatial coordinates in the substation global coordinate system. The rigid transformation matrix includes rotation and translation components. The rotation component is determined by the attitude angle of the laser scanning equipment in the global coordinate system, and the translation component is determined by the installation position of the laser scanning equipment center in the global coordinate system. After mapping, the equipment outline point cloud data in the substation global coordinate system is obtained. This data, along with timestamps and equipment identification information, is organized into an initial 3D point cloud model. Each point in the initial 3D point cloud model is accompanied by global coordinates, a data acquisition timestamp, and an equipment identification number. Example 2

[0030] In practice, based on the initial 3D point cloud model, the relative spatial coordinates of each device and the connection relationships between devices are extracted. Each device's corresponding point cloud cluster is identified from the initial 3D point cloud model, and the centroid coordinates of each point cloud cluster are extracted as the device's relative spatial coordinates. Each point in the initial 3D point cloud model is associated with a device identifier number; these identifiers are used as cluster labels to group points with the same identifier number into the same point cloud cluster. For each device's corresponding point cloud cluster, the arithmetic mean of the 3D spatial coordinates of all points in the cluster is calculated. The average of the three coordinate components constitutes the centroid coordinates, which are then used as the relative spatial coordinates of the corresponding device. The relative spatial coordinates of all devices constitute the device spatial coordinate set.

[0031] In practical implementation, based on the physical layout rules of the substation, the cable connection relationships and support / fixing relationships between various devices are determined, and these relationships are then converted into device connection relationships. The physical layout rules of the substation are pre-stored in the form of a device topology diagram. In this diagram, nodes represent devices, edges represent physical connections, and the edge type attribute is marked as either cable connection or support / fixing. All edges in the device topology diagram are extracted. For each edge, the identifiers of the two devices connected by the edge and their edge type attributes are recorded as a device connection relationship. This connection relationship includes a first device identifier, a second device identifier, and a connection type, which is either cable connection or support / fixing. If two devices have both cable connection and support / fixing relationships, two independent connection relationships are generated, marked as cable connection and support / fixing respectively.

[0032] In practical implementation, a hierarchical association matrix is ​​constructed using the relative spatial coordinates of each device as diagonal elements and the connection relationships between devices as off-diagonal elements. Spatial distance constraints and connection direction constraints are nested layer by layer within this matrix to form a matrix structure used to locate geometrically dependent paths between devices. Before constructing the hierarchical association matrix, each device is assigned a unique index number from 1 to N, where N is the total number of devices. The size of the hierarchical association matrix is ​​N×N. For two devices with index numbers p and q, the element value in the p-th row and q-th column of the hierarchical association matrix is ​​determined as follows: when p equals q, the diagonal element is the relative spatial coordinate of the device with index number p, represented as a three-dimensional vector. When P is not equal to q, check if there is a connection between the device with index number p and the device with index number q. If there is no connection, the off-diagonal elements are zero vectors. If there is at least one connection, the off-diagonal elements are calculated based on the connection type and the spatial relationship between the devices.

[0033] When calculating off-diagonal elements, spatial distance constraint vectors and connection direction constraint vectors are introduced, and these vectors are nested as matrix element values. Spatial distance constraint vector Defined as the unit direction vector from the centroid of device with index number p to the centroid of device with index number q, the spatial distance constraint vector. The magnitude is always 1. Connecting direction constraint vector. Determined based on connection type: When the connection type is cable connection, the connection direction constraint vector is... Take the standard orientation vector of the cable outlet port of the device with index number p in the global coordinate system. The standard orientation vector is obtained by coordinate transformation from the predefined direction of the cable interface in the device's 3D model; when the connection type is fixed support, the connection direction constraint vector is taken as... Let p be the normal vector of the support surface of the device. The normal vector of the support surface is determined by the outward normal direction of the device base plane. When both cable connection and support fixation exist simultaneously, calculate the connection direction constraint vectors for each type separately, and take the average of the two connection direction constraint vectors as the connection direction constraint vector. Obtain the spatial distance constraint vector. With connection direction constraint vector Then, generate the off-diagonal element values ​​of the hierarchical association matrix according to the following formula. :

[0034] in, The off-diagonal element in the p-th row and q-th column of the hierarchical correlation matrix is ​​a three-dimensional vector. The fusion coefficient is the fusion coefficient between spatial distance constraints and connection direction constraints. The range of values ​​is Fusion coefficient The fusion coefficients are obtained through a fusion coefficient function, whose input is the Euclidean distance between the centroids of the device with index p and the device with index q. The fusion coefficient function is in the form of ,in The preset reference distance is the median of the Euclidean distances between all pairs of centroids of all devices in the substation scenario. K is the scaling parameter, with a value of [value missing]. , making when and When the absolute value of the difference is large, the fusion coefficient It rapidly approaches the boundary value. When the Euclidean distance... Smaller and less than the preset reference distance At that time, the fusion coefficient Values ​​close to 0, off-diagonal elements Mainly depends on the connection direction constraint vector When Euclidean distance Larger and greater than the preset reference distance At that time, the fusion coefficient Values ​​close to 1, off-diagonal elements Primarily relies on spatial distance constraint vectors Off-diagonal elements of a hierarchical association matrix The system stores geometric dependency vectors from device p to device q. The direction of the geometric dependency vectors indicates the spatial orientation of the geometric dependency path, and the magnitude of the geometric dependency vectors reflects the dependency strength. After all elements of the hierarchical association matrix are constructed, the nested matrix structure is used to locate the geometric dependency path between any two devices within the substation. Multi-level dependencies are passed through the multiplication of the hierarchical association matrix, forming spatial constraint associations between multiple devices at a higher level.

[0035] Example 3 In practice, the initial 3D point cloud model is dynamically sampled and adjusted using geometrically dependent paths. Geometric feature sets for each device are extracted from the sampled 3D point cloud model, and a multi-scale geometric constraint network is established based on these feature sets. According to the importance level and spatial location of each device in the geometrically dependent path, a sampling rate adjustment coefficient is determined for each device. Based on these sampling rate adjustment coefficients, density increases or decreases are performed on the corresponding point clouds in the initial 3D point cloud model to obtain a density-adjusted 3D point cloud model.

[0036] The depth and connectivity values ​​of each device's level are extracted from the hierarchical association matrix. These values ​​are then normalized and weighted summed to obtain the importance score for each device. The depth value of a device in the hierarchical association matrix is ​​defined as the number of levels a device is at in the dependency propagation path from the source device. The depth value is obtained by traversing the matrix structure from all diagonal elements, counting the number of non-zero vector elements encountered on the way from each device to the top-level device, and using this count as the depth value for that device. The connectivity value is defined as the total number of non-zero vector elements in the corresponding row of the hierarchical association matrix for a device. The connectivity value is obtained by traversing the row corresponding to a device's index in the hierarchical association matrix, counting the number of vector elements with a non-zero modulus in that row, and using this count as the connectivity value for that device. All depth values ​​in the depth value set are input into a normalization operation. The normalization operation uses maximum value normalization, which divides each depth value by the maximum depth value in the depth value set to obtain a normalized depth value. When the maximum depth value in the normalized depth value set is zero, the normalized depth value is uniformly set to 1. All connectivity values ​​in the connectivity value set are input into the same maximum value normalization operation, which divides each connectivity value by the maximum connectivity value in the connectivity value set to obtain a normalized connectivity value. When the maximum connectivity value in the normalized connectivity value set is zero, the normalized connectivity value is uniformly set to 1. The normalized depth values ​​and normalized connectivity values ​​are then weighted and summed using depth weight coefficients and connectivity weight coefficients. The depth weight coefficient is set to 0.35, and the connectivity weight coefficient is set to 0.65. The depth weighting coefficient of 0.35 and the connectivity weighting coefficient of 0.65 are set based on the following: In substation 3D modeling, the connection relationship between equipment has a more direct impact on the geometric dependency path than the depth of the equipment's layer. Equipment with high connectivity has a stronger constraint effect on surrounding equipment due to its geometric shape. Therefore, increasing the connectivity weighting coefficient makes the importance level score focus more on the degree of topological connectivity of the equipment. The weighted summation formula is calculated by multiplying the normalized depth value by the depth weighting coefficient and adding the normalized connectivity value by the connectivity weighting coefficient. The result is used as the importance level score for the corresponding equipment.

[0037] Based on the relative spatial coordinates of each device, calculate the average spatial distance between the corresponding device and all other devices in the hierarchical association matrix, and use the average spatial distance as the spatial sparsity of the corresponding device. The calculation method for the average spatial distance is as follows: For a device with index number u, obtain the relative spatial coordinates of the device with index number u from the set of relative spatial coordinates of the devices. Traverse all devices with index numbers from 1 to N and not equal to u, and calculate the Euclidean distance between the relative spatial coordinates of the device with index number u and the relative spatial coordinates of the device with index number v, where v takes all index values ​​except u, and N is the total number of devices. Sum all the Euclidean distances and divide by (N-1) to obtain the average spatial distance of the device with index number u.

[0038] The importance level score of each device is multiplied by a preset importance level weighting coefficient to obtain the first weighting coefficient. The importance level weighting coefficient is set to 0.6. The reason for setting the importance level weighting coefficient to 0.6 is that the importance level score integrates information on both device depth and connectivity, and is a major indicator for measuring the degree of device modeling needs. It should play a dominant role in determining the sampling rate adjustment coefficient. The spatial location sparsity of the corresponding device is multiplied by a preset sparsity weighting coefficient to obtain the second weighting coefficient. The sparsity weighting coefficient is set to 0.4. The reason for setting the sparsity weighting coefficient to 0.4 is that spatial location sparsity reflects the degree of dispersion of devices in space and plays an auxiliary role in adjusting the sampling density. Therefore, its value is lower than the importance level weighting coefficient. The first weighting coefficient and the second weighting coefficient are added together, and the result is used as the sampling rate adjustment coefficient for the corresponding device. The numerical range of the sampling rate adjustment coefficient is theoretically determined by the specific data. When the sampling rate adjustment coefficient is less than 0.1, it is truncated to 0.1; when the sampling rate adjustment coefficient is greater than 2.0, it is truncated to 2.0.

[0039] Based on the sampling rate adjustment coefficient, random downsampling or uniform upsampling operations are performed on the point cloud corresponding to the device in the initial 3D point cloud model to obtain a density-adjusted 3D point cloud model. In specific implementation, for the device with index number u, all points belonging to that device in the initial 3D point cloud model are obtained, and the total number of points is recorded as M. When the value of the sampling rate adjustment coefficient is greater than 1.0, uniform upsampling is performed: the target number of points M is calculated by multiplying by the sampling rate adjustment coefficient. For every two points in the current point cloud whose nearest neighbor distance is less than the distance threshold, a new point is generated through linear interpolation. The spatial coordinates of the new point are the average of the 3D coordinates of the two points, until the total number of points in the point cloud reaches the target number of points. When the value of the sampling rate adjustment coefficient is less than 1.0, random downsampling is performed: a random number is generated for each independent point in the point cloud with the sampling rate adjustment coefficient as the retention probability. The random number is uniformly distributed in the interval [0,1]. If the random number is less than the sampling rate adjustment coefficient, the point is retained; otherwise, the point is discarded. When the sampling rate adjustment coefficient is equal to 1.0, no density increase or decrease operations are performed on the point cloud. After all devices have had their sampling rates adjusted, a density-adjusted 3D point cloud model is formed.

[0040] The surface curvature, edge position, and plane normal of each device are extracted from the density-adjusted 3D point cloud model, and the surface curvature, edge position, and plane normal are combined into a geometric feature set for each device.

[0041] In practice, a local neighborhood search is performed on the point cloud corresponding to each device in the density-adjusted 3D point cloud model. The radius of the local neighborhood search is a fixed value, determined based on the average point spacing of the laser scanning devices, and set to 5 times the average point spacing. For each point in the current device point cloud, a spherical region is constructed with that point as the center and a fixed radius. All points falling within this spherical region constitute the local neighborhood point set of that point. A least-squares plane is fitted within the local neighborhood point set of each point. The fitting process involves solving the covariance matrix of the local neighborhood point set, extracting the eigenvector corresponding to the smallest eigenvalue as the plane normal, and calculating the constant term of the plane equation to obtain the locally fitted plane. Based on the fitted least-squares plane, the principal curvature and Gaussian curvature of that point are calculated. The principal curvature is calculated as follows: A local coordinate system is established on the local fitting plane. A quadratic surface is fitted to the coordinates of the local neighborhood point set to obtain the first and second fundamental form coefficients of the quadratic surface. Among the extreme values ​​of normal curvature formed by the ratio of the second to the first fundamental form coefficient, the extreme value with the larger absolute value is taken as the maximum curvature, and the extreme value with the smaller absolute value is taken as the minimum curvature. The principal curvature is the arithmetic mean of the absolute values ​​of the maximum and minimum curvatures. The Gaussian curvature is the product of the maximum and minimum curvatures. For all points within the same device, the arithmetic mean of the principal curvatures of all points is calculated as the average principal curvature; the arithmetic mean of the Gaussian curvatures of all points is calculated as the average Gaussian curvature. The average principal curvature and the average Gaussian curvature are combined into a two-dimensional vector, which is used as the surface curvature of the device.

[0042] Gradient calculations are performed on the point cloud corresponding to each device in the density-adjusted 3D point cloud model. Points with a normal change rate exceeding a preset threshold are extracted as candidate edge points. The normal vector within the local neighborhood of each point is obtained; the normal vector is the unit normal vector of the local fitted plane at that point. The normal change rate of each point in the point cloud is calculated, defined as the standard deviation of the angle between the normal vector of that point and the normal vectors of its neighbors. The angle is obtained by calculating the dot product of the two unit normal vectors using the inverse cosine function. Points with a normal change rate greater than the preset threshold are marked as candidate edge points. The preset threshold is 0.2 radians. The basis for setting the preset threshold to 0.2 radians is that, based on the statistical measurement of geometric discontinuities at the edges of common equipment in substations, the standard deviation of the angle between the normal vectors at edge positions is usually stable above 0.25 radians. Selecting 0.2 radians can retain enough candidate edge points while suppressing false detections in planar areas. Connector tracing is performed on candidate edge points. The process is as follows: The point with the largest rate of change of normal is selected from the candidate edge point set as the starting point. This starting point is added to the current edge segment, and then removed from the candidate edge point set. Using the starting point as a reference, the point in the candidate edge point set whose spatial distance from the starting point is less than a preset connection distance and whose difference in the direction of normal change is minimal is searched as the connection point. The preset connection distance is three times the average point spacing of the point cloud. This connection point is added to the current edge segment, and then removed from the candidate edge point set. This process is repeated, using the connection point as a new starting point, until no connection point meeting the criteria can be found. If the current edge segment contains fewer than a preset threshold of 5 points, the segment is discarded. This process is repeated until the candidate edge point set is empty or all points have been processed. The point sequence of all valid edge segments constitutes the edge position of the device.

[0043] Principal component analysis (PCA) is performed on the point cloud corresponding to each device in the density-adjusted 3D point cloud model. The eigenvector corresponding to the smallest eigenvalue of the point cloud covariance matrix is ​​extracted as the plane normal vector for each point of the corresponding device. The PCA process is as follows: For the coordinates of all 3D points in the device point cloud, a 3-column data matrix is ​​constructed, with each row representing the x, y, and z coordinates of a point; the mean of each column of the data matrix is ​​calculated, and the mean of the corresponding column is subtracted from the coordinates of each point to obtain a mean-free coordinate matrix; the covariance matrix of the mean-free coordinate matrix is ​​calculated, and the covariance matrix is ​​a 3×3 real symmetric matrix; eigenvalue decomposition is performed on the covariance matrix to obtain three eigenvalues ​​and their corresponding eigenvectors, all of which are unit vectors; the eigenvector corresponding to the smallest eigenvalue is used as the estimated plane normal vector for the device. Since all points participate in the calculation, a unified estimated plane normal vector is obtained for the same device, and this estimated plane normal vector is normalized and used as the plane normal vector for the device.

[0044] The surface curvature, edge position, and plane normal of each device are combined according to a preset feature arrangement order. The preset feature arrangement order is as follows: first, the two-dimensional vector of surface curvature is stored; then, the complete three-dimensional point coordinate sequence of the edge position is stored; and finally, the three-dimensional vector of the plane normal is stored. Each field is separated by a delimiter, resulting in a geometric feature set for the corresponding device. The geometric feature set is stored using the device index number as the key.

[0045] The geometric feature set of each device is grouped according to its hierarchical position in the hierarchical association matrix. The hierarchical position in the matrix is ​​determined by the device's depth value; devices with the same depth value are grouped into the same hierarchical group. For each hierarchical group, the surface curvature, edge positions, and plane normals of all devices within the group are extracted from their geometric feature sets to serve as the device's internal geometric constraints. For any two devices connected within each hierarchical group, plane normals and relative spatial coordinates are extracted from their geometric feature sets. The angle between the plane normals is calculated as the relative orientation difference, and the vector difference of the relative spatial coordinates is calculated as the relative position difference. The relative position difference and relative orientation difference are combined to form the relative geometric constraints between devices. All the device internal geometric constraints and relative geometric constraints between devices in all hierarchical groups are organized together to form a multi-scale geometric constraint network. Different levels in the multi-scale geometric constraint network correspond to different geometric constraint scales; lower levels correspond to the internal constraints of detailed devices, while higher levels correspond to the relative constraints between the overall structure. Example 4

[0046] In practice, based on a multi-scale geometric constraint network, the initial 3D point cloud model is progressively refined and reconstructed to generate a structured 3D basic model of the substation. The internal geometric constraints of each device at the current scale are extracted from the multi-scale geometric constraint network. Based on these internal geometric constraints, local surface fitting is performed on the point cloud of the corresponding device in the initial 3D point cloud model to generate a locally refined surface for the corresponding device at the current scale.

[0047] From the multi-scale geometric constraint network, a set of internal geometric constraints is extracted for the device with index number 's' at the current scale. These internal geometric constraints include surface curvature constraints, edge position constraints, and plane normal constraints. The surface curvature constraint is a two-dimensional vector composed of the principal curvature average and the Gaussian curvature average; the edge position constraint is a sequence of multiple three-dimensional point coordinates; and the plane normal constraint is a unit three-dimensional vector. The surface curvature constraints, edge position constraints, and plane normal constraints are combined to form the set of internal geometric constraints for the corresponding device at the current scale. The organization of the internal geometric constraint set is as follows: the first field stores the two-dimensional vector of the surface curvature constraint, the second field stores the three-dimensional point sequence of the edge position constraint, and the third field stores the three-dimensional vector of the plane normal constraint. Fields are separated by preset boundary markers.

[0048] Based on the current set of internal geometric constraints, all point clouds corresponding to the device with index number 's' are selected from the initial 3D point cloud model. A bicubic B-spline surface is fitted to the selected point clouds. The bicubic B-spline surface fitting uses uniform node vectors, and the surface order is set to 3rd in both parameter directions u and v. The size of the control point mesh is determined by the surface curvature constraints in the internal geometric constraint set: the number of control points R×S is the larger of the rounded-up average principal curvature multiplied by a preset mesh density factor and the rounded-up average Gaussian curvature multiplied by a preset mesh density factor. The preset mesh density factor is set to 50, ensuring a denser control point mesh in areas with higher curvature. Using the selected point cloud coordinates as fitting data, the control point positions are solved using the least squares method to obtain the preliminary fitted surface. The preliminary fitted surface is expressed as a bicubic B-spline surface represented by the product of the control point mesh and the B-spline basis functions.

[0049] Edge position constraints and plane normal constraints are applied as boundary conditions to the initially fitted surface. The energy function E is defined as follows:

[0050] Where Q is the total number of points in the 3D point sequence with edge position constraints. Let a be the spatial coordinate vector of the a-th 3D point in the edge position constraint sequence. To initially fit the surface with The corresponding parameter coordinates are The coordinate vector of a point in three-dimensional space, parameter coordinates By finding the surface with the initial fitting curve The parameter coordinates of the point with the closest Euclidean distance are determined; W is the number of sampling points on the initial fitting surface under the plane normal constraint. The sampling points are evenly distributed on the parameter domain of the initial fitting surface, and the number W is twice the product of the number of control points R and S. To initially fit the surface in parametric coordinates The unit normal vector at that point is obtained by normalizing the cross product of the partial derivatives of the surface at that point with respect to parameters u and v; It is the plane normal constraint vector, that is, the plane normal unit three-dimensional vector in the set of internal geometric constraints; Normal constraint weight coefficient, normal constraint weight coefficient The value is determined based on the average Gaussian curvature in the surface curvature constraint. When the average Gaussian curvature is greater than 0... The value is taken as the average Gaussian curvature multiplied by 10. When the average Gaussian curvature equals 0... The value is 0.1. The first term in the energy function measures the geometric deviation of the surface approximation edge position, and the second term measures the deviation of the surface normal from the target normal in terms of consistency.

[0051] The energy function E is minimized by adjusting the coordinates of the control points on the initially fitted surface, ensuring positional continuity at the edges and normal continuity on the overall surface. The Levenberg-Marquardt optimization algorithm is used to minimize the energy function E, with the optimization variable being a vector composed of the coordinates of all control points on the initially fitted surface. The iterative convergence threshold is set to a relative change in the energy function value of less than 1 × 10⁻⁻⁻⁶. 6 After optimization, the local refined surface of the device with index number s at the current scale is obtained.

[0052] In practical implementation, the relative geometric constraints between devices in the multi-scale geometric constraint network are mapped to docking boundary conditions between locally refined surfaces. Based on the docking boundary conditions, the locally refined surfaces of all devices are stitched together to form a global refined surface. Relative position constraints and relative orientation constraints between every two adjacent devices are extracted from the multi-scale geometric constraint network. The criterion for determining adjacent devices is that the magnitudes of the corresponding off-diagonal elements in the hierarchical association matrix of the two devices are not zero. For the first and second devices that satisfy the adjacency relationship, the locally refined surfaces of the first and second devices are denoted as surface F and surface G, respectively. The extracted relative position constraint is a three-dimensional vector pointing from the centroid of the first device to the centroid of the second device. The relative orientation constraint is the angle between the plane normals of the first and second devices, obtained by applying the inverse cosine function to the dot product of the two normal unit vectors. The relative position constraints and relative orientation constraints are used as the initial docking boundary conditions between surface F and surface G.

[0053] Based on the initial docking boundary conditions, calculate the first boundary point sequence on the docking boundary of surface F and the second boundary point sequence on the docking boundary of surface G. In the parametric boundary of surface F, select the boundary curve that is closest to the direction of the relative position constraint vector. This boundary curve is determined as follows: traverse the four parametric boundary curves of surface F, calculate the average outward normal of all sampling points on each boundary curve, and select the boundary with the smallest angle between the average outward normal and the relative position constraint vector as the docking boundary of surface F. Then, uniformly sample H points on this docking boundary to obtain the first boundary point sequence, where H is 50. Similarly, determine the docking boundary on surface G that is opposite to the direction of the relative position constraint vector, and uniformly sample H points on this docking boundary to obtain the second boundary point sequence. The initial docking boundary conditions are optimized by minimizing the spatial distance between the first and second boundary point sequences: An objective function is constructed, which is the sum of the squares of the Euclidean distances between the spatial coordinates of the t-th point in the first boundary point sequence and the corresponding spatial coordinates of the t-th point in the second boundary point sequence, plus an orientation penalty term corresponding to the relative orientation constraint. The orientation penalty term is the calculated value of the relative orientation angle minus the square of the relative orientation constraint angle multiplied by the orientation penalty weight of 1.0. The rigid body transformation parameters of surfaces F and G are used as optimization variables to decrease the objective function. The quasi-Newton method is used for iterative solution to obtain the optimized docking boundary conditions, which are expressed by the adjusted relative poses of surfaces F and G.

[0054] The BFGS algorithm, a quasi-Newton method, is used iteratively to obtain the optimized docking boundary conditions. These optimized boundary conditions are expressed by the adjusted relative poses of surfaces F and G. The specific implementation of the BFGS algorithm is as follows: Initialize the rigid body transformation parameter vectors of surfaces F and G (Including rotation quaternions and 3D translation vectors, a total of 7 degrees of freedom), calculate the objective function in gradient at and the inverse of the approximate Hessian matrix (initialized to) (Identity matrix); in the k-th iteration, calculate the search direction. The step size is determined by Armijo line search. Update parameters ; Calculate the gradient under the new parameters The iteration stops if any of the following termination conditions are met: Gradient norm ; Changes in objective function value ; Number of iterations .

[0055] If the process does not terminate, update the approximate Hessian inverse matrix:

[0056] in Based on the optimized docking boundary conditions, the docking boundaries of surface F and surface G are aligned and stitched. Each point in the aligned first boundary point sequence is registered with its closest spatially distant point in the second boundary point sequence as a corresponding point pair. When the spatial distance between two corresponding points is less than the distance merging threshold, the arithmetic mean of the spatial coordinates of the two corresponding points is taken as the stitching point coordinates; otherwise, their respective coordinates are retained. The distance merging threshold is set to 0.8 times the average distance between adjacent points in the first boundary point sequence. At the stitching point, mesh vertices are merged: the registered and merged corresponding points are replaced with a single point, and the topological relationships of all involved facets are updated. After merging, the polygonal regions within the stitching neighborhood are retriangulated using restricted Delaunay triangulation, with the stitching boundary points as constraint points, to generate new triangular facets. The above alignment, stitching, and retriangulation operations are repeated for all adjacent device pairs until the local refined surfaces of all devices are stitched together to obtain the overall refined surface.

[0057] The overall refined surface is meshed to generate a structured mesh model with topological connections. The meshing process employs the Delaunay triangulation algorithm, using all sampling points and stitching points on the overall refined surface as the input point set to generate a triangular mesh, removing degenerate triangles outside the cutting boundaries that are irrelevant to the surface shape. Each triangular element is assigned an index number of the equipment it belongs to as an equipment type identifier. The triangular mesh undergoes manifold verification to ensure that each edge is shared by at most two triangles, and non-manifold edges are removed. The processed triangular mesh is used as the structured mesh model, outputting a 3D foundation model of the structured substation. Example 5

[0058] In specific implementation, please refer to Figure 3The system acquires substation operating environment parameters and embeds these parameters into the spatial coordinate nodes of a structured 3D basic model of the substation, generating a digital twin 3D model containing dynamic environmental information. Temperature, humidity, and electromagnetic field strength parameters are collected from environmental sensors deployed throughout the substation. These sensors are located on transformer heat sinks, high-voltage switchgear contact chambers, and busbar connection points; humidity sensors are located in cable trenches and on the walls of the high-voltage room at a height of 1.5 meters above the ground; and electromagnetic field strength sensors are located 0.5 meters outside the circuit breaker operating mechanism enclosure and at the bottom flange of the instrument transformer insulator. Each environmental sensor, while acquiring data, simultaneously outputs the sampling time, its installation coordinates in the substation's global coordinate system, and the measured value. Temperature measurements are in degrees Celsius, humidity measurements are in percentage relative humidity, and electromagnetic field strength measurements are in microtesla (µT).

[0059] Temperature, humidity, and electromagnetic field strength parameters were time-series aligned with the spatial coordinate nodes of the structured substation's 3D foundation model according to their acquisition time. Each spatial coordinate node in the 3D foundation model carries a 3D spatial position coordinate, derived from the vertex coordinates of the node's location in the structured mesh model. For each spatial coordinate node, the sensor with the smallest Euclidean distance to that node was found among all environmental sensors. The Euclidean distance was calculated by taking the square root of the sum of the squares of the differences between the node's 3D position coordinates and the sensor's installation coordinates. This found sensor was designated as the associated sensor for that spatial coordinate node. The time series of temperature, humidity, and electromagnetic field strength parameters acquired by the associated sensors were extracted. Each time series consisted of a sampling time and a measured value pair. A dynamic display timeline for the 3D foundation model of the structured substation was determined, consisting of equally spaced model time frames, with an interval of 0.5 seconds between adjacent model time frames. For each model time frame on the dynamic display timeline, the sampling record with the smallest absolute time difference between the sampling time and the model time frame is found in the temperature parameter time series of the associated sensors. If the minimum time difference is less than a preset alignment tolerance of 0.25 seconds, the temperature measurement value of this sampling record is directly used as the temperature parameter value of the spatial coordinate node in that model time frame. If the minimum time difference is greater than or equal to the preset alignment tolerance of 0.25 seconds, the two sampling records closest in time to the model time frame are selected in the temperature parameter time series. Linear interpolation is performed between the time and temperature measurement values ​​of the two sampling records to obtain the temperature parameter value of the spatial coordinate node in that model time frame. The same method is used to obtain the humidity parameter value and electromagnetic field strength parameter value in the humidity parameter time series and electromagnetic field strength parameter time series, respectively. The above process is repeated for all spatial coordinate nodes and all model time frames to complete the time alignment and obtain the temperature parameter value, humidity parameter value, and electromagnetic field strength parameter value of each spatial coordinate node in each model time frame.

[0060] The aligned temperature, humidity, and electromagnetic field strength parameters are mapped to the attribute fields of the corresponding spatial coordinate nodes in the 3D basic model of the structured substation. Each spatial coordinate node in the 3D basic model of the structured substation stores three independent attribute fields: temperature, humidity, and electromagnetic field strength, all of which support storing floating-point values. For each spatial coordinate node, the temperature parameter value at a specific model time frame is written to the temperature attribute field of the corresponding spatial coordinate node; the humidity parameter value is written to the humidity attribute field; and the electromagnetic field strength parameter value is written to the electromagnetic field strength attribute field. The attribute field contents of all spatial coordinate nodes across all model time frames constitute a node attribute set carrying dynamic environmental information, organized using the model time frame identifier and node index number as a joint key.

[0061] Based on the node attribute set of each spatial coordinate node in the 3D foundation model of the structured substation, the display parameters of the 3D foundation model are updated to generate a digital twin 3D model containing dynamic environmental information. Display parameters include color mapping, transparency, and annotation information. The heatmap color is driven by temperature parameters: for each model time frame, the temperature parameter values ​​in the temperature attribute fields of all spatial coordinate nodes are read, and the minimum temperature value is extracted. and maximum temperature Minimum temperature The minimum and maximum temperature values ​​for all spatial coordinate nodes. This represents the highest temperature parameter value among all spatial coordinate nodes. For each spatial coordinate node, a color mapping vector is calculated based on its temperature parameter value T. The calculation method is as follows:

[0062] in, This is a three-dimensional color vector. The first component of the three-dimensional color vector corresponds to the intensity value of the red channel, the second component corresponds to the intensity value of the green channel, and the third component corresponds to the intensity value of the blue channel. The values ​​of the three intensity values ​​are all within closed intervals. T represents the temperature attribute value of the current spatial coordinate node in the current model time frame, in degrees Celsius. This represents the minimum temperature attribute value for all spatial coordinate nodes in the current model time frame, in degrees Celsius. This represents the maximum temperature attribute value for all spatial coordinate nodes in the current model time frame, in degrees Celsius. Minimum temperature With maximum temperature The arithmetic mean, The unit is Celsius. When and When they are equal, a division by zero occurs in the formula, at which point the color mapping vector is changed. Set as This corresponds to pure green. Based on the color mapping vector... The three component values ​​are set to the surface color of the corresponding spatial coordinate node and its adjacent triangular facet in the three-dimensional basic model of the structured substation. The intensity values ​​of the red channel, green channel and blue channel are multiplied by 255 and then rounded to obtain the RGB components of the rendering color, so as to realize the effect of the temperature heat map dynamically changing with the temperature distribution.

[0063] Humidity parameter adjustment for local transparency: For each model time frame, the humidity parameter value in the humidity attribute field of each spatial coordinate node is read. The humidity parameter value is divided by 100 to obtain a normalized humidity value. When the humidity parameter value is greater than 100, the normalized humidity value is set to 1.0. The normalized humidity value is used as the transparency adjustment coefficient, and the range of the transparency adjustment coefficient is... 1.0 represents completely opaque, and 0.0 represents completely transparent. The transparency adjustment coefficient is written into the material transparency channel of the corresponding spatial coordinate node and its associated triangular facet. During the rendering of the digital twin 3D model, the alpha blending factor is adjusted to make high humidity areas appear semi-transparent.

[0064] Electromagnetic field strength parameter control contour overlay display: A set of electromagnetic field strength contour level thresholds is preset, and the contour level thresholds are set to values ​​of 1000. The unit is microtesla, corresponding to five levels, each with a designated contour line color. From lowest to highest level, the colors are light yellow, orange-yellow, orange, orange-red, and dark red. For each model time frame, the electromagnetic field strength attribute value of each spatial coordinate node is read. This value is compared sequentially with the threshold values ​​for the five contour line levels to determine the highest level to which the electromagnetic field strength attribute value belongs. A closed circular loop is generated 0.1 meters directly above the 3D position coordinates of the spatial coordinate node. The plane of the loop is parallel to the local surface section of the spatial coordinate node, and the radius of the loop is 0.3 meters. The display color of the loop uses the contour line color specified for its level, and this loop is overlaid as a vector graphic in the 3D scene. When the electromagnetic field strength attribute value is below the lowest level threshold of 2.5 microtesla, no contour line loop is generated. The overlaid contour line display, combined with the temperature heatmap and humidity transparency effects, updates the 3D scene rendering output, generating a digital twin 3D model containing dynamic environmental information.

[0065] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for three-dimensional modeling and design of substations based on digital twins, characterized in that, The method includes: Spatial topology mapping is performed on multi-source heterogeneous equipment in the substation scenario, and the collected static structural data and dynamic operating parameters of the equipment are synchronously converted into an initial three-dimensional point cloud model. Based on the initial 3D point cloud model, the relative spatial coordinates of each device and the connection relationship between the devices are extracted, and a hierarchical association matrix based on spatial constraints is constructed. The association matrix is ​​used to locate the geometric dependency path between devices. The geometric dependency path is used to perform dynamic sampling and adjustment on the initial 3D point cloud model. Geometric feature sets for each device are extracted from the sampled 3D point cloud model, and a multi-scale geometric constraint network is established based on the geometric feature sets. Based on the multi-scale geometric constraint network, the initial three-dimensional point cloud model is progressively refined and reconstructed to generate a three-dimensional basic model of a structured substation. Obtain the substation operating environment parameters, embed the operating environment parameters into the spatial coordinate nodes of the structured substation three-dimensional basic model, and generate a digital twin three-dimensional model containing dynamic environmental information.

2. The method for three-dimensional modeling and design of substations based on digital twins according to claim 1, characterized in that, Spatial topology mapping is performed on multi-source heterogeneous equipment in a substation scenario, synchronously converting the collected static structural data and dynamic operating parameters of the equipment into an initial 3D point cloud model, including: The laser scanning equipment, image acquisition equipment, and operation status monitoring equipment deployed in the substation scene acquire point cloud data of equipment shape, equipment identification image data, and real-time operating parameters of the equipment, respectively. The device shape point cloud data, device identification image data, and device real-time operating parameters are aligned according to a unified timestamp. The aligned device shape point cloud data is then normalized in spatial coordinates. The normalized device shape point cloud data is mapped to the substation global coordinate system to generate the initial three-dimensional point cloud model.

3. The method for three-dimensional modeling and design of substations based on digital twins according to claim 1, characterized in that, Based on the initial 3D point cloud model, the relative spatial coordinates of each device and the connection relationships between devices are extracted. A hierarchical association matrix based on spatial constraints is constructed. This association matrix is ​​used to locate the geometric dependency paths between devices, including: Identify the point cloud cluster corresponding to each device from the initial 3D point cloud model, and extract the centroid coordinates of each point cloud cluster as the relative spatial coordinates of the device; Based on the physical layout rules of the substation, determine the cable connection relationships and support and fixing relationships between various devices, and convert the cable connection relationships and support and fixing relationships into connection relationships between devices; Using the relative spatial coordinates of each device as diagonal elements and the connection relationships between devices as off-diagonal elements, the hierarchical association matrix is ​​constructed. Spatial distance constraints and connection direction constraints are nested layer by layer in the hierarchical association matrix to form a matrix structure for locating geometrically dependent paths between devices.

4. The method for three-dimensional modeling and design of substations based on digital twins according to claim 1, characterized in that, The initial 3D point cloud model is dynamically sampled and adjusted using the geometric dependency path. A geometric feature set for each device is extracted from the sampled 3D point cloud model, and a multi-scale geometric constraint network is established based on the geometric feature set, including: Based on the importance level and spatial location of each device in the geometric dependency path, the sampling rate adjustment coefficient of each device is determined, and the density increase or decrease operation is performed on the point cloud of the corresponding device in the initial three-dimensional point cloud model according to the sampling rate adjustment coefficient to obtain the density-adjusted three-dimensional point cloud model. The surface curvature, edge position, and plane normal of each device are extracted from the density-adjusted 3D point cloud model, and the surface curvature, edge position, and plane normal are combined into a geometric feature set for each device. The geometric feature set of each device is grouped according to the hierarchical position of the device in the hierarchical association matrix, and the internal geometric constraints and relative geometric constraints between devices are established at different scales based on the grouped geometric feature sets to form a multi-scale geometric constraint network.

5. The method for three-dimensional modeling and design of substations based on digital twins according to claim 1, characterized in that, Generate a three-dimensional basic model of the structured substation, including: Extract the internal geometric constraints of each device at the current scale from the multi-scale geometric constraint network, and perform local surface fitting on the point cloud of the corresponding device in the initial 3D point cloud model according to the internal geometric constraints to generate the local refined surface of the device at the current scale. The relative geometric constraints between devices in the multi-scale geometric constraint network are mapped to the docking boundary conditions between the local refined surfaces. Based on the docking boundary conditions, the local refined surfaces of all devices are spliced ​​into an overall refined surface. The overall refined surface is meshed to generate a structured mesh model with topological connections, and the structured mesh model is used as the three-dimensional basic model of the structured substation.

6. The method for three-dimensional modeling and design of substations based on digital twins according to claim 1, characterized in that, Generate a digital twin 3D model containing dynamic environmental information, including: Temperature, humidity, and electromagnetic field strength parameters are collected from environmental sensors deployed in the substation, and these parameters are then time-sequentially aligned with the spatial coordinate nodes of the structured substation three-dimensional basic model according to the time of collection. The aligned temperature, humidity, and electromagnetic field strength parameters are mapped to the attribute fields of the corresponding spatial coordinate nodes in the three-dimensional basic model of the structured substation, forming a set of node attributes carrying dynamic environmental information. Based on the node attribute set of each spatial coordinate node in the three-dimensional basic model of the structured substation, the display parameters of the three-dimensional basic model of the structured substation are updated to generate the digital twin three-dimensional model containing dynamic environmental information.

7. The method for three-dimensional modeling and design of substations based on digital twins according to claim 4, characterized in that, The density-adjusted 3D point cloud model is obtained, including: The depth and connectivity values ​​of each device's level are extracted from the hierarchical association matrix. The depth and connectivity values ​​are then normalized and weighted and summed to obtain the importance level score of each device. Based on the relative spatial coordinates of each device, calculate the average spatial distance between the device and all other devices in the hierarchical association matrix, and use the average spatial distance as the spatial sparsity of the device. The importance level score of each device is multiplied by a preset importance level weighting coefficient to obtain the first weighting coefficient. The spatial location sparsity of the device is multiplied by a preset sparsity weighting coefficient to obtain the second weighting coefficient. The first weighting coefficient and the second weighting coefficient are added together to obtain the sampling rate adjustment coefficient of the device. Based on the sampling rate adjustment coefficient, random downsampling or uniform upsampling operations are performed on the point cloud corresponding to the device in the initial three-dimensional point cloud model to obtain a density-adjusted three-dimensional point cloud model.

8. The method for three-dimensional modeling and design of substations based on digital twins according to claim 4, characterized in that, The surface curvature, edge position, and plane normal of each device are extracted from the density-adjusted 3D point cloud model. These surface curvatures, edge positions, and plane normals are then combined into a geometric feature set for each device, including: For each device in the density-adjusted 3D point cloud model, a local neighborhood search is performed. A least squares plane is fitted in the local neighborhood of each point. The principal curvature and Gaussian curvature of the point are calculated based on the fitted least squares plane. The average principal curvature and Gaussian curvature of all points in the same device are used as the surface curvature of the device. Gradient calculation is performed on the point cloud corresponding to each device in the density-adjusted 3D point cloud model. Points with a normal change rate exceeding a preset threshold are extracted as candidate edge points. Connecting lines are traced on the candidate edge points to obtain the edge position of the device. Principal component analysis is performed on the point cloud corresponding to each device in the density-adjusted 3D point cloud model. The eigenvector corresponding to the minimum eigenvalue of the point cloud covariance matrix is ​​extracted as the plane normal of each point of the device. The average plane normal of all points in the same device is taken as the plane normal of the device. The surface curvature, edge position, and plane normal of each device are combined according to a preset feature arrangement order to obtain a geometric feature set for that device.

9. The method for three-dimensional modeling and design of substations based on digital twins according to claim 5, characterized in that, Generate a locally refined surface for the device at the current scale, including: Extract the surface curvature constraints, edge position constraints, and plane normal constraints of the current device at the current scale from the multi-scale geometric constraint network, and combine the surface curvature constraints, edge position constraints, and plane normal constraints into a set of internal geometric constraints at the current scale; Based on the set of geometric constraints within the current scale, the point cloud corresponding to the current device is selected from the initial three-dimensional point cloud model, and the selected point cloud is fitted with a bicubic B-spline surface according to the surface curvature constraints to obtain a preliminary fitted surface. The edge position constraints and plane normal constraints are applied as boundary restrictions to the initial fitted surface. By adjusting the coordinates of the control points of the initial fitted surface, the surface satisfies positional continuity and normal continuity at the edge positions, thereby generating a localized refined surface of the device at the current scale.

10. A method for three-dimensional modeling and design of substations based on digital twins according to claim 5, characterized in that, The relative geometric constraints between devices in the multi-scale geometric constraint network are mapped to the docking boundary conditions between the local refined surfaces. Based on these docking boundary conditions, the local refined surfaces of all devices are stitched together into a global refined surface, including: Extract the relative position constraints and relative orientation constraints between every two adjacent devices from the multi-scale geometric constraint network, and use the relative position constraints and relative orientation constraints as the initial docking boundary conditions between the local refined surfaces of the two adjacent devices; Based on the initial docking boundary conditions, the sequence of first boundary points on the docking boundary of the local refined surface of the first device and the sequence of second boundary points on the docking boundary of the local refined surface of the second device are calculated. By minimizing the spatial distance between the first boundary point sequence and the second boundary point sequence, the initial docking boundary conditions are optimized to obtain the optimized docking boundary conditions. Based on the optimized docking boundary conditions, the docking boundary of the local refined surface of the first device is aligned and stitched with the docking boundary of the local refined surface of the second device. At the stitching point, the mesh vertices are merged and retriangulated. The above alignment, stitching and retriangulation operations are repeated until the local refined surfaces of all devices are stitched together to obtain the overall refined surface.

Images