Spacing determination method and device for substation equipment, medium and program product

Abstract

The invention discloses a method and device for determining the distance between substation equipment, a medium and a program product. Relates to the power system field. The method comprises the following steps: acquiring point cloud data of a transformer substation, and preprocessing the point cloud data to obtain target point cloud data; inputting the target point cloud data into a target modeling system to obtain a Gaussian three-dimensional model; processing the Gaussian three-dimensional model by adopting a target gridding strategy to obtain a processed Gaussian three-dimensional model; based on the processed Gaussian three-dimensional model, a target index structure is constructed, and the type of the target index structure comprises an index structure of a K-dimensional tree of the processed Gaussian three-dimensional model; and based on the Gaussian three-dimensional model and the target index structure, determining the shortest spatial distance between any two devices in the transformer substation. According to the invention, the technical problem of low measurement accuracy of the shortest space distance between equipment based on the 3D Gaussian object of the transformer substation is solved.

Description

Technical Field

[0001] This invention relates to the field of power systems, and more specifically, to a method, apparatus, medium, and program product for determining the spacing of substation equipment. Background Technology

[0002] As power systems continue to evolve towards intelligence and digitalization, substations, as crucial hubs of the power grid, have a vital impact on the stability of the entire power system through their equipment operation safety and maintenance efficiency. Traditional methods such as manual inspections and two-dimensional drawing checks are no longer sufficient to meet the increasingly complex structures and urgent operation and maintenance needs of modern substations.

[0003] In the 3DGS scenario of substations, accurate measurement of the shortest spatial distance between equipment is crucial for various key applications such as electrical safety assessment, assembly verification, and span detection. A certain phase-to-phase insulation distance must be maintained between busbars, and the distance between insulators and metal components must also meet design specifications; otherwise, corona discharge, partial discharge, and even flashover breakdown can easily occur, seriously threatening the safe operation of the power grid. Furthermore, the layout of substation equipment is typically spatially complex, with multiple layers spanning, intersecting horizontal and vertical pipe corridors, and interactive reflections from various metals and insulating materials, greatly increasing the difficulty of distance measurement between models.

[0004] From a data scale perspective, a complete 3DGS acquisition and reconstruction typically generates datasets containing tens of millions of data points. The sheer volume of data means that using traditional O(n²) brute-force search results in a quadratic increase in computational complexity, compromising real-time performance. While using simple geometric bounding bodies (such as convex hulls or OBBs) for approximation can easily overlook structural details, leading to significant measurement errors, convex hull-based methods simplify shortest distance calculations by constructing the convex hull of the point cloud. A convex hull is the smallest convex polyhedron containing all points, representing the outer boundary of the point cloud. Calculating the shortest distance between the convex hulls of two point clouds can approximate the shortest distance between them. However, for point clouds with complex shapes or concavities, the convex hull may not accurately describe the boundary, resulting in inaccurate distance calculations. Constructing a convex hull is computationally complex, especially in high-dimensional spaces. An OBB is a minimum-volume rectangle that tightly encloses a point cloud, and its orientation can be adjusted according to the main direction of the point cloud. Calculating the shortest distance between the OBBs of two point clouds allows for a rapid estimation of the shortest distance between them. This method is suitable for scenarios where point clouds have regular shapes and clear orientations. For point clouds with irregular shapes or inconsistent orientations, the OBB may not accurately enclose the point cloud, leading to inaccurate distance calculations. Constructing an OBB is computationally complex, requiring operations such as principal component analysis. In substations, even millimeter-level distance deviations in minute components such as busbar joints, small sensors, and insulator petals can affect equipment installation accuracy and operational safety.

[0005] There is currently no effective solution to the above problems. Summary of the Invention

[0006] This invention provides a method, apparatus, medium, and program product for determining the spacing of substation equipment, in order to at least solve the technical problem of low measurement accuracy of the shortest spatial distance between equipment based on 3D Gaussian objects in substations in related technologies.

[0007] According to one aspect of the present invention, a method for determining the spacing of substation equipment is provided, comprising: acquiring point cloud data of the substation and preprocessing the point cloud data to obtain target point cloud data, wherein multiple devices are deployed in the substation; inputting the target point cloud data into a target modeling system to obtain a Gaussian 3D model, wherein the target modeling system is used to perform 3D Gaussian splash modeling on the target point cloud data; processing the Gaussian 3D model using a target meshing strategy to obtain a processed Gaussian 3D model, wherein the target meshing strategy is used to adjust the size and distribution of voxels according to the density distribution characteristics of the devices; constructing a target index structure based on the processed Gaussian 3D model, wherein the type of the target index structure includes: an index structure of a K-dimensional tree of the processed Gaussian 3D model, where K is a positive integer; and determining the shortest spatial distance between any two devices in the substation based on the Gaussian 3D model and the target index structure.

[0008] Furthermore, the processed Gaussian 3D model includes: individual point clouds of each device in the substation, wherein each individual point cloud includes: point cloud data processed by 3D Gaussian splash modeling and the target meshing strategy. Based on the processed Gaussian 3D model, a target index structure is constructed, including: analyzing the data volume of individual point clouds of any two devices and determining the analysis results; when the analysis results indicate that the difference in data volume between the individual point clouds of any two devices exceeds a preset threshold, merging the individual point clouds of the two devices, and constructing a K-tree index structure based on the merged individual point clouds to obtain a first index structure; when the analysis results indicate that the difference in data volume between the individual point clouds of any two devices does not exceed a preset threshold, constructing K-tree index structures for the individual point clouds of the two devices respectively to obtain a second index structure; and determining the target index structure based on the first index structure and the second index structure.

[0009] Further, the data volume of individual point clouds of any two devices is analyzed to determine the analysis results, including: calculating the information entropy between the data volumes of individual point clouds of any two devices to obtain the point cloud information entropy between each pair of devices; if the point cloud information entropy of any two devices is less than a preset threshold, the analysis results indicate that the difference in data volume between the individual point clouds of the two devices exceeds a preset threshold; if the point cloud information entropy of any two devices is not less than the preset threshold, the analysis results indicate that the difference in data volume between the individual point clouds of the two devices does not exceed the preset threshold.

[0010] Further, based on the Gaussian 3D model and the target index structure, determining the shortest spatial distance between any two devices within the substation includes: for any two devices, based on the Gaussian 3D model and the target index structure, using a nearest neighbor search strategy, calculating the distance between each pair of points in the point clouds of the individual devices of the two devices, obtaining a distance set between the two devices; and determining the shortest spatial distance between the two devices based on the minimum distance in the distance set of the two devices.

[0011] Furthermore, the Gaussian 3D model is processed using a target meshing strategy to obtain a processed Gaussian 3D model, including: determining the voxel mesh size of each region based on the point cloud density distribution of different regions in the Gaussian 3D model; and dividing the Gaussian 3D model based on the voxel mesh size of each region to obtain the processed Gaussian 3D model.

[0012] Furthermore, before constructing the target index structure based on the processed Gaussian 3D model, the method further includes: segmenting the processed Gaussian 3D model to obtain an initial point cloud cluster for each equipment category; and determining multiple individual point clouds of equipment based on the initial point cloud clusters of each equipment category, wherein each individual point cloud of equipment includes the point cloud of one of the equipment in the substation.

[0013] Further, the point cloud data is preprocessed to obtain target point cloud data, including: applying a target filtering strategy to filter the point cloud data to obtain filtered point cloud data; the target filtering strategy includes at least one of the following: statistical filtering and radius filtering; applying a target filling strategy to fill missing values ​​in the filtered point cloud data to obtain filled point cloud data, wherein the target filling strategy includes at least one of the following: plane fitting and surface interpolation; in the filled point cloud data, performing a downsampling operation on regions with point cloud density greater than a preset density threshold to obtain operated point cloud data; and determining the target point cloud data based on the operated point cloud data.

[0014] According to another aspect of the present invention, a device for determining the spacing of substation equipment is also provided, comprising: a first processing unit, configured to acquire point cloud data of a substation and preprocess the point cloud data to obtain target point cloud data, wherein multiple devices are deployed in the substation; a second processing unit, configured to input the target point cloud data into a target modeling system to obtain a Gaussian 3D model, wherein the target modeling system is used to perform 3D Gaussian splash modeling on the target point cloud data; a third processing unit, configured to process the Gaussian 3D model using a target meshing strategy to obtain a processed Gaussian 3D model, wherein the target meshing strategy is used to adjust the size and distribution of voxels according to the density distribution characteristics of the devices; a construction unit, configured to construct a target index structure based on the processed Gaussian 3D model, wherein the type of the target index structure includes: an index structure of a K-dimensional tree of the processed Gaussian 3D model, where K is a positive integer; and a first determining unit, configured to determine the shortest spatial distance between any two devices in the substation based on the Gaussian 3D model and the target index structure.

[0015] Furthermore, the processed Gaussian 3D model includes: individual point clouds of each device in the substation, wherein each individual point cloud includes: point cloud data processed by 3D Gaussian splash modeling and the target meshing strategy. The construction unit includes: an analysis subunit, used to analyze the data volume of individual point clouds of any two devices and determine the analysis result; a first processing subunit, used to merge the individual point clouds of the two devices when the analysis result indicates that the difference in data volume between the individual point clouds of any two devices exceeds a preset threshold, and construct a K-dimensional tree index structure based on the merged individual point clouds to obtain a first index structure; a second processing subunit, used to construct K-dimensional tree index structures for the individual point clouds of the two devices respectively when the analysis result indicates that the difference in data volume between the individual point clouds of any two devices does not exceed a preset threshold, to obtain a second index structure; and a first determination subunit, used to determine the target index structure based on the first index structure and the second index structure.

[0016] Further, the analysis subunit includes: a calculation module, used to calculate the information entropy between the data volume of individual point clouds of any two devices, to obtain the point cloud information entropy between each pair of devices; a first determination module, used to determine that the analysis result indicates that the difference in data volume between the individual point clouds of the two devices exceeds a preset threshold when the point cloud information entropy of any two devices is less than a preset threshold; and a second determination module, used to determine that the analysis result indicates that the difference in data volume between the individual point clouds of the two devices does not exceed a preset threshold when the point cloud information entropy of any two devices is not less than a preset threshold.

[0017] Further, the first determining unit includes: a calculation subunit, used for calculating the distance between each pair of points between the device individual point clouds of the two devices based on the Gaussian 3D model and the target index structure, using a nearest neighbor search strategy, to obtain a distance set between the two devices; and a second determining subunit, used for determining the shortest spatial distance between the two devices based on the minimum distance in the distance set between the two devices.

[0018] Furthermore, the third processing unit includes: a third determining subunit, used to determine the voxel mesh size of each region based on the point cloud density distribution of different regions in the Gaussian 3D model; and a dividing subunit, used to divide the Gaussian 3D model based on the voxel mesh size of each region to obtain the processed Gaussian 3D model.

[0019] Furthermore, the substation equipment spacing determination device further includes: a segmentation unit, used to segment the processed Gaussian 3D model before constructing the target index structure based on the processed Gaussian 3D model to obtain an initial point cloud cluster for each equipment category; and a second determination unit, used to determine the individual point clouds of multiple equipment based on the initial point cloud clusters of each equipment category, wherein each individual point cloud of equipment includes the point cloud of one of the equipment in the substation.

[0020] Further, the first processing unit includes: a filtering subunit, used to filter the point cloud data using a target filtering strategy to obtain filtered point cloud data; the target filtering strategy includes at least one of the following: statistical filtering and radius filtering; a filling subunit, used to fill missing values ​​in the filtered point cloud data using a target filling strategy to obtain filled point cloud data, wherein the target filling strategy includes at least one of the following: plane fitting and surface interpolation; a sampling subunit, used to perform downsampling operations on regions with point cloud density greater than a preset density threshold in the filled point cloud data to obtain operated point cloud data; and a third determining subunit, used to determine the target point cloud data based on the operated point cloud data.

[0021] According to another aspect of the present invention, an electronic device is also provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute the spacing determination method of any of the above-described substation equipment by executing the executable instructions.

[0022] According to another aspect of the present invention, a computer-readable storage medium is also provided, which stores a computer program, wherein the computer program controls the device where the computer-readable storage medium is located to execute the substation equipment spacing determination method described above when the computer program is running.

[0023] According to another aspect of this application, a computer program product is provided, including computer instructions that, when executed by a processor, implement the steps of the method for determining the spacing of the substation equipment.

[0024] In this invention, the following methods are employed: Point cloud data of a substation is acquired and preprocessed to obtain target point cloud data, wherein multiple devices are deployed in the substation; the target point cloud data is input into a target modeling system to obtain a Gaussian 3D model, wherein the target modeling system is used to perform 3D Gaussian splash modeling on the target point cloud data; a target meshing strategy is used to process the Gaussian 3D model to obtain a processed Gaussian 3D model, wherein the target meshing strategy is used to adjust the size and distribution of voxels according to the density distribution characteristics of the devices; based on the processed Gaussian 3D model, a target index structure is constructed, wherein the type of the target index structure includes: a K-dimensional tree index structure of the processed Gaussian 3D model, where K is a positive integer; based on the Gaussian 3D model and the target index structure, the shortest spatial distance between any two devices in the substation is determined. This solves the technical problem of low measurement accuracy of the shortest spatial distance between devices based on 3D Gaussian objects of substations in related technologies. In this invention, the shortest spatial distance between any two devices in a substation is determined by using a dynamic voxel grid (i.e., the voxel grid size adjusted by the target meshing strategy) and a K-dimensional tree index structure. This avoids the low measurement accuracy of the shortest spatial distance between devices based on the 3D Gaussian object of the substation in related technologies, thereby achieving the technical effect of improving the measurement accuracy of the shortest spatial distance between devices in the substation. Attached Figure Description

[0025] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0026] Figure 1This is a flowchart of an optional method for determining the spacing of substation equipment according to an embodiment of the present invention;

[0027] Figure 2 This is a schematic diagram of an optional K-dimensional tree and its corresponding binary tree according to an embodiment of the present invention;

[0028] Figure 3 This is a flowchart illustrating an optional method for determining the shortest distance according to an embodiment of the present invention;

[0029] Figure 4 This is a flowchart of an optional substation equipment spacing determination system according to an embodiment of the present invention;

[0030] Figure 5 This is a flowchart of an optional voxelized 3D point cloud scene classification according to an embodiment of the present invention;

[0031] Figure 6 This is a schematic diagram of an optional substation equipment spacing determination device according to an embodiment of the present invention;

[0032] Figure 7 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0033] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0034] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0035] It should be noted that the user information (including but not limited to user device information, user personal information, etc.), the collected information and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of the relevant data all comply with the relevant laws, regulations, and standards of the relevant regions, have taken necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation portals for users to choose to authorize or refuse.

[0036] Example 1

[0037] According to an embodiment of the present invention, an alternative method for determining the spacing of substation equipment is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0038] Figure 1 This is a flowchart of an optional method for determining the spacing of substation equipment according to an embodiment of the present invention, such as... Figure 1 As shown, the method includes the following steps:

[0039] Step S101: Obtain point cloud data of the substation and preprocess the point cloud data to obtain target point cloud data. Multiple devices are deployed in the substation.

[0040] In this embodiment, a high-precision 3D laser scanner can be used to perform a full-range scan of the substation to obtain complete point cloud data. Due to the complexity of the substation equipment, the scanning process can be performed from multiple angles and positions to reduce obstruction and missed areas. The acquired point cloud data can also be preprocessed, including noise filtering and ground point cloud removal, to obtain target point cloud data, ensuring the accuracy and completeness of the data.

[0041] Step S102: Input the target point cloud data into the target modeling system to obtain a Gaussian 3D model, wherein the target modeling system is used to perform 3D Gaussian splash modeling on the target point cloud data.

[0042] In this embodiment, target point cloud data (i.e., preprocessed point cloud data) can be input into a 3DGS (3D Gaussian Splatting) modeling system (corresponding to a target modeling system) to generate a 3D model represented by Gaussian points. 3DGS can more effectively represent and render complex scenes by projecting Gaussian distribution points in 3D space. During the modeling process, the parameters of the Gaussian points can be adjusted according to the device's geometric features and surface reflection characteristics to obtain a high-quality model.

[0043] Step S103: The Gaussian 3D model is processed using a target meshing strategy to obtain the processed Gaussian 3D model. The target meshing strategy is used to adjust the size and distribution of voxels according to the density distribution characteristics of the device.

[0044] To simplify the data structure and improve processing efficiency, in this embodiment, the 3DGS model (corresponding to a Gaussian 3D model) can also be dynamically voxelized. For example, the size and distribution of voxels can be adaptively adjusted according to the density and distribution characteristics of the devices to more effectively represent complex 3D scenes. The voxelized data structure is beneficial for subsequent spatial indexing and distance calculation.

[0045] Step S104: Based on the processed Gaussian 3D model, construct the target index structure. The type of the target index structure includes: the index structure of the K-dimensional tree of the processed Gaussian 3D model, where K is a positive integer.

[0046] In this embodiment, a kd-tree index structure can be constructed on the processed Gaussian 3D model (e.g., point cloud data after voxel meshing) to accelerate nearest neighbor search. A kd-tree is a tree-like data structure for multidimensional spatial data, used for efficient nearest neighbor search.

[0047] In this embodiment, a kd-tree can be used to quickly find the nearest point pairs between devices, improving the efficiency of distance calculation.

[0048] Before constructing the KD-Tree, the ground of the entire Gaussian point cloud scene needs to be segmented. This can be done using SAH-based optimized segmentation to build a KD-tree of the processed Gaussian 3D model. For example, in this embodiment, the cost function can be improved by introducing a density weighting factor to optimize the selection of segmentation planes.

[0049]

[0050] Where: CSAH: the total expected cost of a certain partitioning scheme. Ct: the cost of traversing to the current node (which can be considered a constant). ρ L A L NL Ci / ρ total The cost of future searches falling into the left subspace. N L : The number of points in the left subspace. A L ρ: Surface area of ​​the left sub-boundary box. L : Density weight of the left subspace. Ci: Cost of performing one ray-object intersection test (can be considered a constant). The cost term for the right subspace is similar.

[0051] In this embodiment, principal component segmentation optimization can be performed during the construction of the K-dimensional tree. Segmentation along the direction of maximum variance typically separates the point cloud more evenly and faster, contributing to the construction of a more balanced and shallower tree structure, thereby further improving search efficiency. In this embodiment, the eigenvector directions of the covariance matrix can be calculated to construct an orthogonal base coordinate system.

[0052]

[0053] In this embodiment, the direction with the largest variance can be selected as the splitting axis to ensure the balance of the tree structure. v is the obtained unit vector, and this direction is the direction with the largest variance of the point cloud distribution. T It is the transpose of vector v.

[0054] Figure 2 This is a schematic diagram of an optional K-dimensional tree and its corresponding binary tree according to an embodiment of the present invention, as shown below. Figure 2 As shown, P represents a data point in the point cloud, and L is used to divide the space into different regions, defining the branching structure of the tree.

[0055] Step S105: Based on the Gaussian 3D model and target index structure, determine the shortest spatial distance between any two devices in the substation.

[0056] In this embodiment, each point of the device after voxel meshing (i.e., each device can correspond to a small point cloud set in the Gaussian 3D model) is traversed. For each device, a kd-tree can be used to find its nearest neighbor in the point cloud set of another device, and the Euclidean distance between the two points is calculated. All point pairs and their corresponding distances are recorded, and finally, the point pair with the smallest distance is selected as the shortest distance between the devices, which can effectively avoid measurement errors caused by uneven point cloud density or noise.

[0057] In one optional example, the calculated shortest distance and corresponding point pairs can be visualized in a 3D model to assist engineers in optimizing equipment layout and conducting safety assessments. Simultaneously, the measurement results are output in report form, including equipment name, shortest distance, and corresponding point coordinates, facilitating subsequent engineering design and management. In another optional example, a 3D early warning system can be used for alerts, for example: red alert: distance < safe value (150mm); yellow alert: 150mm ≤ distance < design value (180mm); green alert: distance ≥ 180mm.

[0058] Through the above steps, in this embodiment, the shortest spatial distance between any two devices within a substation is determined using a dynamic voxel grid (i.e., the voxel grid size adjusted by the target meshing strategy) and a K-dimensional tree index structure. This avoids the low measurement accuracy of the shortest spatial distance between devices based on 3D Gaussian objects in substations in related technologies, thereby improving the technical accuracy of the measurement of the shortest spatial distance between substation devices. This solves the technical problem of low measurement accuracy of the shortest spatial distance between devices based on 3D Gaussian objects in substations in related technologies.

[0059] Optionally, the processed Gaussian 3D model includes: individual point clouds of each piece of equipment in the substation. Each individual point cloud includes point cloud data processed using 3D Gaussian splash modeling and a target meshing strategy. Based on the processed Gaussian 3D model, a target index structure is constructed, including: analyzing the data volume of individual point clouds of any two pieces of equipment and determining the analysis results; merging the individual point clouds of the two pieces of equipment when the analysis results indicate that the difference in data volume between the individual point clouds of any two pieces of equipment exceeds a preset threshold, and constructing a K-tree index structure based on the merged individual point clouds to obtain a first index structure; constructing a K-tree index structure for each individual point cloud of the two pieces of equipment when the analysis results indicate that the difference in data volume between the individual point clouds of any two pieces of equipment does not exceed a preset threshold, and obtaining a second index structure; and determining the target index structure based on the first and second index structures.

[0060] In this embodiment, the optimal data organization structure can be dynamically selected based on the number of points in the point cloud dataset, thereby ensuring processing efficiency while avoiding resource waste or performance bottlenecks caused by data imbalance. Threshold determination: When the difference in the number of points is greater than 20%, a dynamic allocation strategy is triggered (to prevent small-scale point clouds from misusing KD-Trees). Memory optimization mechanism: Experiments show that it can reduce memory usage by 42% (compared to traditional dual KD-Tree construction).

[0061] Dynamic allocation strategy:

[0062] const kdtreePoints=point1.length>point2.length ? point1:point2; constvoxelGridPoints=point1.length>point2.length ? point2: point1.

[0063] Here, kdtreePoints represents the set of points in a kdtree, voxelGridPoints represents the set of points in a voxel grid, point1 and point2 represent the point clouds of individual devices corresponding to the two devices, and point1.length and point2.length represent the data size of the point clouds of individual devices corresponding to the two devices.

[0064] In this embodiment, a smaller number of point clouds can be voxelized first to generate several discrete units, and the number of points and center positions within each voxel can be counted, providing a basis for coarse-grained screening. For the side with more points, a balanced kd-tree index structure is constructed, organizing all points into a multi-level binary partitioning tree to support efficient nearest neighbor search. This judgment and allocation process does not require a preset fixed threshold, but can be dynamically adjusted according to the point ratio of the current ranging scenario, acceptable real-time requirements, and hardware performance, ensuring that the optimal strategy can be quickly selected under different device combinations. In extreme cases, when the number of points on both sides is similar or both are large, a hybrid strategy of "voxel coarse screening + kd-tree fine search" can be used for both sides simultaneously to further compress the search range and ensure search depth, effectively balancing efficiency and accuracy.

[0065] In an optional example, if H (information entropy, used to identify the "balance" of point cloud data volume between individual point clouds of the two devices (A, B)) ≥ 0.8: it indicates that N A and N B The quantities are relatively similar and balanced. In this case, a bidirectional systematization can be adopted, that is, efficient processing of A and B separately (i.e., constructing K-dimensional tree index structures separately).

[0066] If H < 0.8: This indicates N A and N B The number of points is severely imbalanced. This triggers asymmetric processing, where the two point clouds are first merged, then a KD-Tree is constructed and systematically processed. This avoids the need to build and manage an inefficient data structure separately for a small dataset.

[0067] The target index structure can be formed by associating all the equipment in the substation with the first index structure and the second index structure.

[0068] Optionally, a comparative analysis is performed on the data volume of individual point clouds of any two devices to determine the analysis results, including: calculating the information entropy between the data volume of individual point clouds of any two devices to obtain the point cloud information entropy between each pair of devices; if the point cloud information entropy of any two devices is less than a preset threshold, the analysis results indicate that the difference in data volume between the individual point clouds of the two devices exceeds a preset threshold; if the point cloud information entropy of any two devices is not less than a preset threshold, the analysis results indicate that the difference in data volume between the individual point clouds of the two devices does not exceed a preset threshold.

[0069] In this embodiment, for the point cloud N of two individual devices A N B By performing shortest distance calculations, the optimal data organization structure can be dynamically selected based on the quantity balance among point cloud datasets, thereby ensuring processing efficiency while avoiding resource waste or performance bottlenecks caused by data imbalance.

[0070] In this embodiment, information entropy H(N) is used. A N B To determine the "balance" of the two batches of point cloud data (corresponding to the data volume mentioned above), a dynamic threshold function is used:

[0071]

[0072] Information entropy H(N) A N B The formula calculates two sets (N) A N B The degree of disorder in the ratio of points. The higher the entropy value, the more "balanced" the two sets are.

[0073] Decision rule: If H ≥ 0.8 (corresponding to a preset threshold): This indicates N A and N B The quantities are similar and relatively balanced. At this point, a two-way systematization is adopted, that is, A and B are respectively...

[0074] Efficient processing.

[0075] If H < 0.8: This indicates N A and N B The number of points is severely imbalanced. This triggers asymmetric processing, which involves first merging the two point clouds, then constructing a KD-Tree for them and performing systematic processing. This avoids building and managing an inefficient data structure separately for a small dataset.

[0076] Exception handling mechanism: In one optional example, to ensure stable operation even in the event of unexpected situations, the following two safety measures can also be included:

[0077] Changes in point cloud size were detected using an exponentially weighted moving average (EWMA).

[0078]

[0079] Where Nt represents the number of point clouds at the current moment. This represents the predicted value at the current moment (obtained by smoothing based on historical data). This represents the predicted value at the previous moment. λ=0.2 represents the smoothing coefficient. The smaller the value, the higher the weight of historical data, and the smoother the prediction curve.

[0080] when When a data structure is restructured, it indicates that the change in the current data volume has exceeded the normal fluctuation range (which can be understood as an "abnormality"). This can trigger a data structure restructure, which means readjusting the internal data organization structure to adapt to the new data scale and prevent performance degradation or errors.

[0081] If the KD-Tree construction fails, it can switch to building a hybrid octree to ensure system robustness.

[0082] Optionally, based on the Gaussian 3D model and the target index structure, the shortest spatial distance between any two devices in the substation is determined, including: for any two devices, based on the Gaussian 3D model and the target index structure, using the nearest neighbor search strategy, calculating the distance between each pair of points between the individual point clouds of the two devices, and obtaining the distance set of the two devices; and determining the shortest spatial distance between the two devices based on the minimum distance in the distance set of the two devices.

[0083] In this embodiment, each point in the voxel mesh (each point in the device's individual point cloud) can be traversed, and a kd-tree can be used to find its nearest neighbor in another device's point cloud set, calculating the Euclidean distance between the two points. All point pairs and their corresponding distances are recorded, and finally, the point pair with the smallest distance is selected as the shortest distance between the devices, which can effectively avoid measurement errors caused by uneven point cloud density or noise.

[0084] In one optional example, the shortest distance calculation can be divided into two stages: "coarse-grained screening" and "refined search." In the coarse-grained screening stage, the spatial coverage information of the discrete cells generated by the voxel grid and the corresponding nodes in the kd-tree (K-dimensional tree) is used to quickly estimate the distance from the center of the voxel cell to the kd-tree node region. Voxel-subtree pairs that are clearly unlikely to contain the shortest point pair are pruned, thus significantly reducing the number of subsequent comparisons. Then, the refined search stage begins: for each retained candidate voxel cell, all original points are traversed, and for each query point, the kd-tree is called to perform a nearest neighbor query. Leveraging the branching and backtracking pruning mechanism of the kd-tree, the nearest point is quickly located, and the global minimum distance is updated. Throughout the search process, the current minimum distance threshold is maintained in real time to dynamically prune subtrees that guarantee a minimum distance greater than the threshold under any index node, further improving search efficiency. To enhance the robustness of the distance measurement, the system also performs multi-point verification on the returned nearest neighbor distance: a secondary dense search is performed on the surrounding area of ​​points whose distance is close to the current minimum value to eliminate occasional errors caused by noise or boundary effects. Finally, the system outputs the global minimum distance value and can optionally generate a detailed report containing the coordinates of the shortest point pair, corresponding voxel elements, or tree node information for subsequent visualization and engineering verification. Through the collaborative design of "voxel coarse screening + kd-tree fine search," the system not only ensures real-time ranging capabilities in million-level point cloud scenarios but also provides sub-centimeter-level high-precision results, meeting the stringent requirements for substation equipment safety distance monitoring and assembly verification.

[0085] In another alternative example, a three-stage pipeline strategy can significantly accelerate the process of "quickly finding the nearest neighbor of a given point in a large 3D point cloud." This can be understood as a "coarse-to-fine" intelligent search strategy. The three-stage optimization process includes:

[0086] (1) Spatial hash pre-screening:

[0087] To quickly narrow down the search area, a 3D Morton code spatial index can be constructed for each point:

[0088]

[0089] Only candidate points from the same or adjacent hash buckets are retained, where x i y i , z i These represent the i-th bit of x, y, and z, respectively, and their values ​​can only be 0 or 1.

[0090] BBF Priority Search: Determines which region is most likely to be searched first, defining the priority of each KD-Tree node:

[0091]

[0092] Where q represents the query point. cn represents the center point of the space represented by the current tree node. rn represents the radius of the bounding sphere of the current tree node (i.e., the distance from the center point to the farthest point that this node can contain). ||q-cn||-rn represents the shortest distance from the query point q to the bounding sphere of this node. The smaller this distance, the closer the node is to the query point, and the more likely it is to contain its nearest neighbor.

[0093] SIMD (Single Instruction Multiple Data) precise computation: It can use vector processing extended instruction sets to calculate the distance between 8 points in parallel.

[0094]

[0095] The computational throughput reaches 32 points / cycle, of which: (x q y q , z q ), (x i y i , z i () represents the x, y, and z coordinates of two points.

[0096] Figure 3 This is a flowchart of an optional shortest distance determination according to an embodiment of the present invention, such as... Figure 3 As shown, for two sets of point clouds, point1 and point2, corresponding to two devices, we can first determine the number of point clouds. If the number of point1 points is greater than that of point2, we can set kdtreePoints (the set of points in the kdtree) = point1 and voxelGridPoints (the set of points in the voxel grid) = point2; otherwise, we set kdtreePoints (the set of points in the kdtree) = point2 and voxelGridPoints (the set of points in the voxel grid) = point1. Then, we perform point cloud filtering preprocessing, construct a KD-Tree structure and a voxel network (leafSize (voxel grid size) = 0.3), traverse the voxel grid points, and determine the distance between point pairs through KD-Tree nearest neighbor search. We then extract the point pairs with the smallest distance and return the shortest distance result.

[0097] Optionally, a target meshing strategy is used to process the Gaussian 3D model to obtain a processed Gaussian 3D model, including: determining the voxel mesh size of each region based on the point cloud density distribution of different regions in the Gaussian 3D model; and dividing the Gaussian 3D model based on the voxel mesh size of each region to obtain the processed Gaussian 3D model.

[0098] In this embodiment, an adaptive voxel size can be assigned to each region based on the local density and shape complexity of the point cloud: smaller voxels are used in areas with rich details and large curvature changes to ensure shape accuracy; larger voxels are used in flat or sparse areas to improve computational efficiency.

[0099] For example, since 3DGS (corresponding to Gaussian 3D model) point cloud data is massive, direct processing would be computationally burdensome. The 3D space can be divided into regular grids (voxels), and multiple points within each voxel can be replaced by one or more representative points, which greatly reduces the total number of points and thus alleviates the pressure of storage, transmission and computation.

[0100] The dynamic hierarchical strategy is shown in Table 1:

[0101] Table 1

[0102]

[0103] Performance Comparison: Among related technologies, uniform voxels (10mm) require 4.2 million data units, while dynamic voxels require only 1.87 million data units (a 55.5% reduction), resulting in a 40% increase in resolution in key areas. This avoids the problems encountered in related technologies where fixed-size meshes are typically used when generating voxel meshes. In dense point cloud regions (rich in detail), the mesh is too large, leading to a loss of model details. In sparse point cloud regions (flat and open), the mesh is too small, generating numerous unnecessary triangles and wasting computational resources.

[0104] In one alternative example, a multi-scale approach can also be used to perceive density from three different scales.

[0105] This yields more robust results; for example, the average distance from point p to its kl nearest neighbors can be calculated.

[0106]

[0107] Where p represents any point in space. l represents the scale level. There are three levels: l=0 (small scale), l=1 (medium scale), and l=2 (large scale). kl represents the number of nearest neighbors to be found at different scales. pi(l) represents the i-th nearest neighbor of point p at scale l. ||p-pi^(l)|| represents the distance from point p to its i-th nearest neighbor.

[0108] Dynamic dimensional control equations:

[0109]

[0110] Where L(p) represents the final grid size used at position p. This is the output. Indicates the basic grid size. (p) represents the average distance at the most local scale (50 nearest neighbors), representing the local density. σ ref =0.5m: A preset reference density threshold. It serves as the baseline for judging "sparse" and "dense" density.

[0111] Optionally, before constructing the target index structure based on the processed Gaussian 3D model, the method further includes: segmenting the processed Gaussian 3D model to obtain an initial point cloud cluster for each equipment category; and determining multiple individual point clouds of equipment based on the initial point cloud clusters of each equipment category, wherein each individual point cloud of equipment includes: the point cloud of one of the equipment in the substation.

[0112] In this embodiment, based on the processed Gaussian 3D model, the devices in the point cloud can be individualized. This individualization is a key step in extracting various substation devices from the background and other objects in the overall scene point cloud (corresponding to the processed Gaussian 3D model).

[0113] First, a semantic label or shape feature library for equipment types can be established, including the geometric contours and reflection characteristics of common substation equipment such as circuit breakers, busbars, insulators, and switchgear. Using deep learning or traditional geometric segmentation methods, the preprocessed point cloud or the processed Gaussian 3D model is initially segmented to generate candidate point cloud clusters for each equipment category. Among these, algorithms based on voxelization clustering and region growing can merge adjacent points through point cloud density and normal vector consistency, extracting clusters with similar shapes and spatial continuity. Second, each candidate cluster is validated posteriorly. Combining the shape constraints of the equipment model or template matching on the projection plane, misclassified clusters or residual background points are eliminated to obtain more accurate individual equipment point clouds. Then, each individual equipment point cloud is numbered and named according to its category, and its spatial location is recorded for rapid indexing during subsequent ranging. To improve the robustness of individualization, multi-view feature fusion and temporal information can be introduced to avoid loss due to occlusion during a single scan or object movement. Finally, local coordinate system alignment can be performed on the point clouds of individual devices, isolating each device in a coordinate system relative to its centroid or installation reference plane, unifying scale and rotation to simplify subsequent distance calculations. After device individualization is completed, the shortest distance can be measured for any two device point cloud sets according to a unified and specified process, thereby supporting refined operational safety assessments and installation verifications.

[0114] Optionally, preprocessing the point cloud data to obtain target point cloud data includes: applying a target filtering strategy to filter the point cloud data to obtain filtered point cloud data; the target filtering strategy includes at least one of the following: statistical filtering and radius filtering; applying a target filling strategy to fill missing values ​​in the filtered point cloud data to obtain filled point cloud data; wherein the target filling strategy includes at least one of the following: plane fitting and surface interpolation; performing downsampling operations on regions with point cloud density greater than a preset density threshold in the filled point cloud data to obtain manipulated point cloud data; and determining the target point cloud data based on the manipulated point cloud data.

[0115] In this embodiment, to improve the quality and usability of point cloud data, noise removal can be performed on the original point cloud (corresponding to point cloud data) collected from the overall scene of the substation. Noise in the original point cloud can include isolated points or artifacts introduced by reflection, occlusion, or multipath propagation. In this embodiment, a combination of statistical filtering and radius filtering can be used to filter the data, resulting in filtered point cloud data. Statistical filtering can remove points with excessively large deviations from the average distance of their neighborhood by analyzing the distribution of points within each point's neighborhood; radius filtering can remove isolated points with excessively low density based on a threshold number of points within a fixed radius. Subsequently, holes in the point cloud caused by missed scans or occlusions can be filled by using algorithms based on local plane fitting or surface interpolation to appropriately fill in the missing areas, ensuring spatial continuity during subsequent segmentation and searching, resulting in filled point cloud data. For areas with excessively high point density, downsampling operations can be performed to obtain, for example, voxel mesh downsampling or uniform random sampling, simplifying excessively overlapping points into representative point cloud data, while retaining sufficient details at edges and where curvature changes significantly, in order to balance computational efficiency and shape accuracy.

[0116] In this embodiment, point cloud coordinate system registration and fusion can also be performed. Multi-view scanning results are registered to a unified coordinate system using feature matching or iterative nearest point (ICP) to eliminate pose differences and ensure consistency across the entire point cloud. Finally, the preprocessed point cloud can be coarsely segmented to separate the ground and device point clouds. Background points irrelevant to ranging are removed using plane fitting or semantic segmentation methods to obtain target point cloud data. This provides a clean and efficient input foundation for subsequent device individualization and shortest distance calculation. This preprocessing significantly reduces the sensitivity of subsequent algorithms to noise and holes, and controls the point cloud size while preserving key structural details, providing strong support for the real-time performance and accuracy of the entire ranging process.

[0117] In one optional example, a 3D laser scanning device can be used to acquire point cloud data of a substation scene, obtaining a comprehensive point cloud including equipment such as busbars, circuit breakers, and insulators. Then, noise removal is performed on the point cloud data. For example, point clouds often contain "outliers" or "noise points," such as isolated points caused by reflection, occlusion, or errors during scanning. To remove these invalid points, it is necessary to determine whether a point is an "anomaly." In this embodiment, an adaptive filtering strategy based on a local density field can be used. This involves finding the k surrounding points of point pi, calculating the average distance (μk) and standard deviation (σk) between these neighbors, and defining the density confidence interval of point pi as:

[0118]

[0119] If the actual distance from pi to its neighbors exceeds this range, it indicates that its local density differs too much from its surroundings, and it is considered an outlier (noise) that needs to be removed.

[0120] Where μk represents the mean of the distances between all points in the neighborhood of k surrounding points of point pi, σk represents the standard deviation of the neighborhood distance, and α=1.2 represents the empirical adjustment coefficient, which is determined through Monte Carlo simulation optimization.

[0121] In the point cloud preprocessing stage, a neighborhood search acceleration method based on spherical spatial hashing can be adopted. By controlling the search range with an adaptive radius, the complexity of neighborhood queries is significantly reduced, and the efficiency of noise removal is improved. At the same time, the point cloud is divided into several sub-regions for parallel processing, and atomic operations are combined to ensure thread safety. The overall preprocessing efficiency is improved by about 3.8 times in actual tests, thus ensuring real-time performance even with a point cloud scale of millions.

[0122] Implementation details: Accelerated neighborhood search: Construct a spherical spatial hash index, with the query radius r adaptively calculated based on the point cloud density.

[0123]

[0124] Where V is the bounding box volume of the point cloud, k is the average number of points to be included in the neighborhood, and N is the total number of points in the point cloud.

[0125] In this embodiment, parallel computing optimization can also be performed. For example, the point cloud can be divided into 8 sub-regions according to a fixed bounding box size and processed in parallel. Thread safety is ensured through atomic operations, and the measured speedup is 3.8 times.

[0126] Because 3DGS point cloud data is massive, direct processing would be computationally burdensome. A better approach is to divide the 3D space into a regular grid (voxels), and replace multiple points within each voxel with one or more representative points. This significantly reduces the total number of points, thus alleviating the pressure on storage, transmission, and computation. The voxel grid partitioning can include:

[0127] Step 1: Dynamic Voxel Raster Generation:

[0128] Traversing the point cloud data, for any point P(x, y, z) in space, dynamically calculate the size L(x, y, z) of its corresponding voxel based on its local geometric features. The voxel size is controlled by the following function:

[0129]

[0130] Where L(x, y, z) represents the voxel size used at spatial coordinates (x, y, z). L0 is the basic voxel size (e.g., 0.8 m). γ is the curvature sensitivity coefficient (e.g., 0.5), used to adjust the sensitivity of the voxel size to geometric features. p is the local density gradient at point P, used to quantify the severity of geometric changes at that point. σp is the global density standard deviation of the entire point cloud, used to normalize the gradient values.

[0131] Step 2: Differentiated Sampling Strategy

[0132] Different sampling strategies can be adopted based on the geometric features within the voxels defined in step one:

[0133] Centroid sampling for flat regions:

[0134] Judgment condition: When the normal vectors of points within a voxel are aligned at the same height (e.g., the cosine of the angle between the normal vectors is greater than 0.9), the region is determined to be a flat region.

[0135] Sampling operation: Weighted centroid sampling is used within the voxel, using a representative point to replace all points within the voxel.

[0136] Weight Calculation: The position of a point is obtained by calculating the weighted center of gravity of each point. The formula for calculating the weight wi of a point is:

[0137]

[0138] Where pi is a point within a voxel, p is the initial centroid of that voxel, and σp is the global density standard deviation. This weighting formula ensures that the influence of outliers on the centroid calculation is reduced, making the position of representative points more stable.

[0139] Conservative sampling for edge / feature regions:

[0140] Judgment criteria: When the difference in the normal vector direction of points within a voxel is large (e.g., cosθ≤0.9), the region is determined to be an edge or feature region.

[0141] Sampling operation: In this type of voxel, all original data points are retained, or a second fine sampling is performed using a significantly reduced voxel size to preserve its geometric features to the maximum extent.

[0142] In this embodiment, combining 3DGS and dynamic voxel meshing technology enables a more accurate representation of the device's geometric features, improving measurement precision. The kd-tree index structure accelerates nearest neighbor search, significantly improving distance calculation efficiency. The entire measurement process can be automated, reducing manual intervention and human error. Dynamic voxel meshing technology adaptively adjusts voxel sizes based on the device's density and distribution characteristics, adapting to the needs of different scenarios. Measurement results can also be visually displayed in a 3D model, assisting engineers in decision-making.

[0143] This embodiment effectively improves the automation and accuracy of measurements, and is suitable for equipment layout optimization and safety assessment in complex scenarios such as substations. Verification through real-world cases demonstrates the feasibility and practicality of this embodiment.

[0144] Example 2

[0145] Figure 4 This is a flowchart of an optional substation equipment spacing determination system according to an embodiment of the present invention, such as... Figure 4 As shown, it includes: a data acquisition layer, a point cloud preprocessing module, a 3DGS Gaussian modeling module, a dynamic voxel mesh module, a KD-Tree index construction module, a shortest distance calculation engine, and a visualization and output module.

[0146] The following example illustrates the processing flow of this spacing determination system:

[0147] In the expansion project of a 500kV smart substation, it is necessary to accurately measure the minimum safe distance between the newly added GIS equipment (gas-insulated switchgear) and adjacent surge arresters. Manual measurement in related technologies suffers from three major drawbacks: Structural complexity: The busbar corridor of the GIS equipment and the surge arrester skirt structure intersect, and the closest point may exist in a non-directly visible area. Dynamic deformation impact: Equipment is subject to millimeter-level deformation due to temperature changes, requiring consideration of dynamic safety margins. Efficiency bottleneck: Measuring the distance between all equipment in the substation requires 72 man-hours, which is difficult to meet the project schedule requirements. In this embodiment, to avoid the drawbacks of existing technologies, the following process is adopted:

[0148] 1. Data Acquisition and Preprocessing:

[0149] Equipment configuration: Employs a 3D laser scanning device. Scanning performance: Distance measurement accuracy: ±3cm; Scanning rate: 320,000 points / second; Continuous scanning time: 1.5 hours.

[0150] Core Functions and Advantages: Real-time Point Cloud Generation: Through its self-developed Multi-SLAM (Simultaneous Localization and Mapping) algorithm, the 3D laser scanning equipment can achieve high-precision real-time localization and mapping in complex environments. The quality of the generated real-time point cloud data is comparable to the post-processing effect of previous generation products, with an absolute accuracy of up to 3 centimeters.

[0151] Seamless Point Cloud Algorithm: This algorithm can generate dense point clouds with a density of up to 1 million points per square meter and a point spacing as low as 1 millimeter, bringing photo-quality color point clouds and accurately capturing every detail.

[0152] High-precision fusion positioning: Even in environments with weak GNSS signals, it can ensure data continuity and accuracy within a range of 100 meters.

[0153] Comprehensive supporting software: The supporting software supports multiple mainstream point cloud data formats and optimizes core functions such as one-click engineering processing, map fusion, cloud map overlay, registration, noise reduction, measurement and cropping, thereby improving work efficiency.

[0154] The preprocessing process is as follows:

[0155] (1) Data synchronization and timestamp calibration:

[0156] After the device is powered on, the LiDAR, IMU, and binocular cameras begin to acquire data synchronously. The system timestamps the data streams from each sensor to ensure consistency between spatial points acquired at the same time and image information. This step provides the temporal basis for subsequent spatial-visual fusion.

[0157] (2) IMU data pre-integration:

[0158] The IMU module continuously records raw data from the accelerometer and gyroscope. The system performs pre-integration calculations on this data to estimate the device's displacement and attitude changes between two time points. This result serves as the initial input for laser and vision data fusion positioning.

[0159] (3) LiDAR data filtering:

[0160] The original point cloud may contain outliers, echo anomalies, or background noise. The system first applies primary filtering rules such as distance thresholds and echo intensity thresholds to remove obviously invalid points. Subsequently, a voxel grid is used to simplify the data density and improve real-time processing efficiency.

[0161] (4) Visual feature extraction and matching:

[0162] SIFT (Scale-Invariant Feature Transform) and ORB (Oriented Fast and Rotated BRIEF) or custom feature points are extracted from synchronously acquired color images to construct matching relationships between image frames, assisting in device pose estimation. Image features also participate in the point cloud colorization stage, serving as the basis for the projection relationship between points and the image.

[0163] (5) Multi-sensor fusion positioning:

[0164] By combining the inter-frame matching of LiDAR, visual feature registration, and IMU calculation results, the system constructs a fusion localization framework (e.g., factor map or extended Kalman filter) to calculate the device's pose in 3D space in real time. This pose information is used for spatial registration of the point cloud to ensure its continuity. It should be noted that by using optimization algorithms (such as factor map), the inter-frame matching of LiDAR, the feature tracking of vision, and the pre-integration results of IMU are mutually corrected to fuse and calculate an optimal and smooth device pose sequence.

[0165] (6) Initial point cloud stitching and registration:

[0166] The system stitches multiple frames of point cloud data into a point cloud set in a unified coordinate system based on the device pose. This stage employs fast rigid body transformation, eliminating the need for iterative optimization to ensure real-time performance. The stitched result forms a preview point cloud with basic geometric structure. Specifically, when stitching multiple frames of point cloud data, each pose calculated by fusion positioning can be directly utilized to unify the corresponding point cloud frames to the global coordinate system through rapid coordinate transformation, directly merging them to form a complete initial 3D scene.

[0167] (7) Preparation for point cloud coloring and mapping:

[0168] To achieve real-time point cloud coloring, the system records the original acquisition viewpoint and image frame number for each point, and calculates its image projection position based on camera intrinsic parameters. This information will be used in the coloring module to accurately match image pixels with 3D points.

[0169] 2. 3DGS Gaussian point cloud modeling:

[0170] By optimizing millions of 3D Gaussian ellipsoids with customizable size, shape, and color, they are "grown" from the initial point cloud and fitted to the perspective of multiple photographs, thus realistically reproducing the entire 3D scene.

[0171] To meet the Gaussianization requirements of large point clouds in substations, a server architecture must be adopted that meets specific CPU, GPU, and memory requirements.

[0172] (1) Generation of Gaussian points:

[0173] Density control: based on curvature adaptive distribution (500 points / m³ in flat areas, 3000 points / m³ in high curvature areas).

[0174] Covariance matrix: Sets the anisotropic ratio (maximum axial ratio 1:5) for slender structures such as insulator strings.

[0175] Figure 5 This is a flowchart of an optional voxelized 3D point cloud scene classification according to an embodiment of the present invention, such as... Figure 5 As shown, the process includes: step (1) point cloud input and point cloud voxelization; step (2) sampling rate spectral convolution and concatenation to obtain point feature vectors; step (3) max pooling and concatenation to obtain voxel feature vectors; and step (4) fully connected layer to obtain the classification result. Figure 5 In this context, 'rate' represents the weight between nodes, and 'node' represents a voxel node in the voxelized 3D space.

[0176] (2) Color blending:

[0177] Introducing laser reflection intensity as an additional feature enhances the distinction between metals and non-metals;

[0178] RGB-I four-channel input avoids confusion between the materials of the insulator porcelain bottle and the metal flange.

[0179] 3. Dynamic voxel meshing:

[0180] Since 3DGS point cloud data is massive, direct processing would be computationally burdensome. A dynamic hierarchical strategy (as shown in Table 1 of Example 1) can be adopted to divide the three-dimensional space into regular grids (voxels) and replace multiple points in each voxel with one or more representative points, which greatly reduces the total number of points and thus alleviates the pressure of storage, transmission and computation.

[0181] 4. kd-tree index construction:

[0182] Based on the number of points in the cloud dataset, the optimal data organization structure is dynamically selected, thereby ensuring processing efficiency while avoiding resource waste or performance bottlenecks caused by data imbalance.

[0183] Dynamic allocation strategy:

[0184] const kdtreePoints=point1.length>point2.length ? point1: point2;

[0185] const voxelGridPoints=point1.length>point2.length ? point2: point1.

[0186] Quantity threshold determination: When the difference in the number of points is greater than 20%, allocation is triggered (to avoid misuse of KD-Tree for small-scale point clouds).

[0187] Memory optimization mechanism: Experiments show that it can reduce memory usage by 42% (compared to traditional dual KD-Tree construction).

[0188] 5. Shortest distance parallel computation:

[0189] 6. Visualization and decision support:

[0190] 3D early warning system: Red warning: distance < safe value (150mm); Yellow warning: 150mm ≤ distance < design value (180mm); Green: distance ≥ 180mm.

[0191] The measurement accuracy and single-device time in this embodiment and related technologies are shown in Table 2.

[0192] Table 2

[0193]

[0194] In this embodiment, combining 3DGS and dynamic voxel meshing technology enables a more accurate representation of the device's geometric features, improving measurement precision. The kd-tree index structure accelerates nearest neighbor search, significantly improving distance calculation efficiency. The entire measurement process can be automated, reducing manual intervention and human error. Dynamic voxel meshing technology adaptively adjusts voxel sizes based on the device's density and distribution characteristics, adapting to the needs of different scenarios. Measurement results can also be visually displayed in a 3D model, assisting engineers in decision-making.

[0195] This embodiment effectively improves the automation and accuracy of measurements, and is suitable for equipment layout optimization and safety assessment in complex scenarios such as substations. Verification through real-world cases demonstrates the feasibility and practicality of this embodiment.

[0196] Example 3

[0197] Embodiment 3 of the present invention provides an optional substation equipment spacing determination device, wherein each implementation unit in the spacing determination device corresponds to each implementation step in Embodiment 1.

[0198] Figure 6 This is a schematic diagram of an optional substation equipment spacing determination device according to an embodiment of the present invention, such as... Figure 6As shown, the device includes: a first processing unit 61, a second processing unit 62, a third processing unit 63, a construction unit 64, and a first determination unit 65.

[0199] The first processing unit 61 is used to acquire point cloud data of the substation and preprocess the point cloud data to obtain target point cloud data. Multiple devices are deployed in the substation.

[0200] The second processing unit 62 is used to input the target point cloud data into the target modeling system to obtain a Gaussian 3D model, wherein the target modeling system is used to perform 3D Gaussian splash modeling on the target point cloud data.

[0201] The third processing unit 63 is used to process the Gaussian 3D model using a target meshing strategy to obtain the processed Gaussian 3D model. The target meshing strategy is used to adjust the size and distribution of voxels according to the density distribution characteristics of the device.

[0202] Construction unit 64 is used to construct a target index structure based on the processed Gaussian 3D model. The type of the target index structure includes: an index structure of a K-dimensional tree of the processed Gaussian 3D model, where K is a positive integer.

[0203] The first determining unit 65 is used to determine the shortest spatial distance between any two devices in the substation based on the Gaussian three-dimensional model and the target index structure.

[0204] In the substation equipment spacing determination device provided in this embodiment, the first processing unit 61 acquires point cloud data of the substation and preprocesses the point cloud data to obtain target point cloud data. Since multiple devices are deployed in the substation, the second processing unit 62 inputs the target point cloud data into a target modeling system to obtain a Gaussian 3D model. The target modeling system performs 3D Gaussian splash modeling on the target point cloud data. The third processing unit 63 processes the Gaussian 3D model using a target meshing strategy to obtain a processed Gaussian 3D model. The target meshing strategy adjusts the size and distribution of voxels based on the density distribution characteristics of the devices. The construction unit 64 constructs a target index structure based on the processed Gaussian 3D model. The target index structure includes a K-dimensional tree index structure of the processed Gaussian 3D model, where K is a positive integer. The first determination unit 65 determines the shortest spatial distance between any two devices in the substation based on the Gaussian 3D model and the target index structure. This solves the technical problem of low measurement accuracy of the shortest spatial distance between devices based on 3D Gaussian objects in substations in related technologies. In this embodiment, the shortest spatial distance between any two devices in a substation is determined by using dynamic voxel grids (i.e., voxel grid size adjusted by the target meshing strategy) and a K-dimensional tree index structure. This avoids the low measurement accuracy of the shortest spatial distance between devices based on 3D Gaussian objects in substations in related technologies, thereby achieving the technical effect of improving the measurement accuracy of the shortest spatial distance between substation devices.

[0205] Optionally, in the substation equipment spacing determination device provided in this embodiment, the processed Gaussian 3D model includes: equipment individual point clouds of each piece of equipment in the substation. The equipment individual point clouds include: point cloud data processed by 3D Gaussian splash modeling and target meshing strategy. The construction unit includes: an analysis subunit, used to analyze the data volume of equipment individual point clouds of any two pieces of equipment and determine the analysis result; a first processing subunit, used to merge the equipment individual point clouds of the two pieces of equipment when the analysis result indicates that the difference in data volume between the equipment individual point clouds of any two pieces of equipment exceeds a preset threshold, and construct a K-dimensional tree index structure based on the merged equipment individual point clouds to obtain a first index structure; a second processing subunit, used to construct K-dimensional tree index structures for the equipment individual point clouds of the two pieces of equipment respectively when the analysis result indicates that the difference in data volume between the equipment individual point clouds of any two pieces of equipment does not exceed a preset threshold, to obtain a second index structure; and a first determination subunit, used to determine the target index structure based on the first index structure and the second index structure.

[0206] Optionally, in the substation equipment spacing determination device provided in this embodiment, the analysis subunit includes: a calculation module, used to calculate the information entropy between the data volume of individual point clouds of any two devices, to obtain the point cloud information entropy between each pair of devices; a first determination module, used to determine that the analysis result indicates that the difference in data volume between the individual point clouds of the two devices exceeds a preset threshold when the point cloud information entropy of any two devices is less than a preset threshold; and a second determination module, used to determine that the analysis result indicates that the difference in data volume between the individual point clouds of the two devices does not exceed a preset threshold when the point cloud information entropy of any two devices is not less than a preset threshold.

[0207] Optionally, in the substation equipment spacing determination device provided in this embodiment, the first determination unit includes: a calculation subunit, used to calculate the distance between each pair of points between the individual point clouds of the two devices based on the Gaussian three-dimensional model and the target index structure, using the nearest neighbor search strategy, to obtain the distance set of the two devices; and a second determination subunit, used to determine the shortest spatial distance between the two devices based on the minimum distance in the distance set of the two devices.

[0208] Optionally, in the substation equipment spacing determination device provided in this embodiment, the third processing unit includes: a third determination subunit, used to determine the voxel grid size of each region based on the point cloud density distribution of different regions in the Gaussian 3D model; and a division subunit, used to divide the Gaussian 3D model based on the voxel grid size of each region to obtain the processed Gaussian 3D model.

[0209] Optionally, in the substation equipment spacing determination device provided in this embodiment, the substation equipment spacing determination device further includes: a segmentation unit, used to segment the processed Gaussian 3D model before constructing the target index structure based on the processed Gaussian 3D model to obtain an initial point cloud cluster for each equipment category; and a second determination unit, used to determine the individual point clouds of multiple equipment based on the initial point cloud clusters of each equipment category, wherein each individual point cloud of equipment includes: the point cloud of one of the equipment in the substation.

[0210] Optionally, in the substation equipment spacing determination device provided in this embodiment, the first processing unit includes: a filtering subunit, used to filter point cloud data using a target filtering strategy to obtain filtered point cloud data; the target filtering strategy includes at least one of the following: statistical filtering and radius filtering; a filling subunit, used to fill missing values ​​in the filtered point cloud data using a target filling strategy to obtain filled point cloud data; wherein the target filling strategy includes at least one of the following: plane fitting and surface interpolation; a sampling subunit, used to perform downsampling operations on regions with point cloud density greater than a preset density threshold in the filled point cloud data to obtain operated point cloud data; and a third determining subunit, used to determine target point cloud data based on the operated point cloud data.

[0211] The aforementioned substation equipment spacing determination device may further include a processor and a memory. The first processing unit 61, the second processing unit 62, the third processing unit 63, the construction unit 64, and the first determination unit 65 are all stored in the memory as program units, and the processor executes the aforementioned program units stored in the memory to realize the corresponding functions.

[0212] The aforementioned processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured. By adjusting kernel parameters, the shortest spatial distance between any two devices within a substation can be determined using a dynamic voxel grid (i.e., the voxel grid size adjusted by the target meshing strategy) and a K-dimensional tree index structure. This avoids the low accuracy of measuring the shortest spatial distance between devices based on 3D Gaussian objects in substations in related technologies, thus achieving the technical effect of improving the measurement accuracy of the shortest spatial distance between substation devices.

[0213] The aforementioned memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0214] According to another aspect of the present invention, an electronic device is also provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute the spacing determination method of any of the above-described substation equipment by executing the executable instructions.

[0215] According to another aspect of the present invention, a computer-readable storage medium is also provided, which stores a computer program, wherein the computer program controls the device where the computer-readable storage medium is located to execute the substation equipment spacing determination method described above when the computer program is running.

[0216] Figure 7 This is a schematic diagram of an electronic device according to an embodiment of the present invention, such as... Figure 7 As shown, an embodiment of the present invention provides an electronic device 70, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the method for determining the spacing of substation equipment as described above.

[0217] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0218] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0219] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0220] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0221] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0222] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0223] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for determining the spacing of substation equipment, characterized in that, include: The point cloud data of the substation is acquired and preprocessed to obtain the target point cloud data, wherein multiple devices are deployed in the substation. The target point cloud data is input into the target modeling system to obtain a Gaussian 3D model, wherein the target modeling system is used to perform 3D Gaussian splash modeling on the target point cloud data; The Gaussian 3D model is processed using a target meshing strategy to obtain a processed Gaussian 3D model. The target meshing strategy is used to adjust the size and distribution of voxels according to the density distribution characteristics of the device. Based on the processed Gaussian 3D model, a target index structure is constructed, wherein the type of the target index structure includes: an index structure of a K-dimensional tree of the processed Gaussian 3D model, where K is a positive integer; Based on the Gaussian 3D model and the target index structure, the shortest spatial distance between any two devices in the substation is determined.

2. The spacing determination method according to claim 1, characterized in that, The processed Gaussian 3D model includes: individual point clouds of each piece of equipment in the substation, wherein the individual point clouds include: point cloud data processed by 3D Gaussian splash modeling and the target meshing strategy. Based on the processed Gaussian 3D model, a target index structure is constructed, including: Analyze the data volume of point clouds of any two of the aforementioned devices and determine the analysis results; If the analysis results indicate that the difference in data volume between any two device point clouds exceeds a preset threshold, the device point clouds of the two devices are merged, and a K-dimensional tree index structure is constructed based on the merged device point clouds to obtain the first index structure. If the analysis results indicate that the difference in data volume between the individual point clouds of any two devices does not exceed a preset threshold, a K-dimensional tree index structure is constructed for the individual point clouds of the two devices respectively to obtain a second index structure. The target index structure is determined based on the first index structure and the second index structure.

3. The spacing determination method according to claim 1, characterized in that, Analyze the data volume of the point cloud of any two of the aforementioned devices to determine the analysis results, including: Calculate the information entropy between the data volume of individual point clouds of any two devices to obtain the point cloud information entropy between each pair of devices; If the entropy of the point cloud information of any two of the devices is less than a preset threshold, the analysis result indicates that the difference in the amount of data between the individual point clouds of the two devices exceeds a preset threshold. If the entropy of the point cloud information of any two of the devices is not less than a preset threshold, the analysis result indicates that the difference in the amount of data between the individual point clouds of the two devices does not exceed a preset threshold.

4. The spacing determination method according to claim 2, characterized in that, Based on the Gaussian 3D model and the target index structure, the shortest spatial distance between any two devices within the substation is determined, including: For any two devices, based on the Gaussian 3D model and the target index structure, the nearest neighbor search strategy is used to calculate the distance between each pair of points in the point clouds of individual devices of the two devices, and the distance set between the two devices is obtained. The shortest spatial distance between the two devices is determined based on the minimum distance in the set of distances between the two devices.

5. The spacing determination method according to claim 1, characterized in that, The Gaussian 3D model is processed using a target meshing strategy to obtain a processed Gaussian 3D model, including: Based on the point cloud density distribution in different regions of the Gaussian 3D model, the voxel mesh size of each region is determined. The Gaussian 3D model is divided based on the voxel mesh size of each region to obtain the processed Gaussian 3D model.

6. The spacing determination method according to claim 1, characterized in that, Before constructing the target index structure based on the processed Gaussian 3D model, the following steps are also included: The processed Gaussian 3D model is segmented to obtain initial point cloud clusters for each device category; Based on the initial point cloud clusters for each equipment category, multiple individual point clouds of equipment are determined, wherein each individual point cloud of equipment includes the point cloud of one of the equipment in the substation.

7. The spacing determination method according to claim 1, characterized in that, The point cloud data is preprocessed to obtain target point cloud data, including: The point cloud data is filtered using a target filtering strategy to obtain filtered point cloud data. The target filtering strategy includes at least one of the following: statistical filtering and radius filtering. The missing values ​​in the filtered point cloud data are filled using a target filling strategy to obtain filled point cloud data. The target filling strategy includes at least one of the following: plane fitting and surface interpolation. In the filled point cloud data, the regions with point cloud density greater than a preset density threshold are downsampled to obtain the processed point cloud data. Based on the point cloud data after the operation, the target point cloud data is determined.

8. A device for determining the spacing of substation equipment, characterized in that, include: The first processing unit is used to acquire point cloud data of the substation and preprocess the point cloud data to obtain target point cloud data, wherein multiple devices are deployed in the substation. The second processing unit is used to input the target point cloud data into the target modeling system to obtain a Gaussian 3D model, wherein the target modeling system is used to perform 3D Gaussian splash modeling on the target point cloud data. The third processing unit is used to process the Gaussian 3D model using a target meshing strategy to obtain a processed Gaussian 3D model. The target meshing strategy is used to adjust the size and distribution of voxels according to the density distribution characteristics of the device. A construction unit is used to construct a target index structure based on the processed Gaussian 3D model, wherein the type of the target index structure includes: an index structure of a K-dimensional tree of the processed Gaussian 3D model, where K is a positive integer; The determining unit is used to determine the shortest spatial distance between any two devices in the substation based on the Gaussian 3D model and the target index structure.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored executable program, wherein, when the executable program is executed, it controls the device containing the computer-readable storage medium to perform the substation equipment spacing determination method according to any one of claims 1 to 7.

10. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the method for determining the spacing of substation equipment as described in any one of claims 1 to 7.

Images