Substation unmanned aerial vehicle inspection route planning method and device based on real scene model

Abstract

This invention discloses a method and apparatus for planning unmanned aerial vehicle (UAV) inspection routes for substations based on a real-world model. The method includes: dividing substation point cloud data into hierarchical and uniform blocks and adaptively removing noise to obtain point cloud file groups; generating dynamic partitioning thresholds for each point cloud file based on an elevation clustering algorithm, accurately extracting ground point cloud data, and achieving multi-level ground point cloud stripping; loading the point cloud data and a real-world substation model, extracting multi-dimensional feature point coordinates of identical components, and performing precise six-degree-of-freedom calibration of the real-world model using a spatial rigid body transformation algorithm to obtain a calibration model; fusing the point cloud data and the calibration model to construct a hierarchically hidden fused data model; and planning routes based on the fused data. This invention solves the technical problems of low model calibration accuracy, static route planning, and incomplete ground point cloud removal in traditional methods.

Description

Technical Field

[0001] This invention relates to unmanned aerial vehicle (UAV) navigation technology, and more particularly to a method and apparatus for planning UAV inspection routes for substations based on a real-scene model, applicable to fully automated UAV inspection route planning for various types of substations. Background Technology

[0002] As a critical hub for power transmission, the inspection of equipment operation status in substations is of paramount importance. Unmanned aerial vehicle (UAV) inspection has become the mainstream method due to its high flexibility and efficiency, and the accuracy and efficiency of flight path planning directly determine the inspection results. Current technologies often employ a method of single-point cloud data block processing combined with simple calibration of real-world models for UAV flight path planning, but this approach suffers from numerous technical bottlenecks.

[0003] Firstly, in terms of point cloud data processing, traditional methods use fixed thresholds to extract ground point clouds, which cannot adapt to the complex terrain of substations. Ground point cloud removal is incomplete, and no layering, segmentation, or adaptive noise removal is performed, resulting in low data processing efficiency and poor accuracy.

[0004] Secondly, in terms of real-world model calibration, translation calibration is performed only through simple three-dimensional coordinate differences, lacking six-degree-of-freedom calibration such as rotation and scaling. This results in large deviations in model position and insufficient accuracy in route planning.

[0005] Thirdly, regarding data fusion, the fused data is a single, unified whole, making it impossible to flexibly view information at different equipment levels according to inspection needs, resulting in poor data usage flexibility.

[0006] Fourth, in terms of route planning, most of the planning is static and does not take into account the inspection priority of substation equipment. Furthermore, it uses a uniform safety distance threshold, which makes it impossible to achieve adaptive optimization of the route and results in problems such as missed inspections and high safety risks. Summary of the Invention

[0007] The purpose of this invention is to provide a method and device for planning the inspection route of a substation using a drone based on a real-scene model, which solves the technical problems of incomplete removal of ground point clouds, low model calibration accuracy, poor data fusion flexibility, and static route planning in the prior art, and realizes high-precision, high-efficiency, and dynamic planning of the inspection route of a substation using a drone.

[0008] To address this, the present invention provides a dynamic route planning method for substation UAV inspection based on a real-scene model, comprising: S1, performing hierarchical and uniform block processing on the substation point cloud data, and removing point cloud noise using an adaptive noise removal algorithm to obtain a point cloud file group, wherein the point cloud file group includes multiple hierarchical point cloud files; S2, obtaining a corresponding dynamic partitioning threshold for each hierarchical point cloud file based on an elevation clustering algorithm, and performing multi-level ground point cloud extraction on each hierarchical point cloud file according to the dynamic partitioning threshold to determine the ground point cloud data corresponding to each hierarchical point cloud file; S3, loading the denoised point cloud data and the substation UAV inspection data. The power station's real-world model is used to extract the coordinates of multi-dimensional feature points of the same components from the point cloud data and the real-world model. A spatial rigid body transformation algorithm is then used to calibrate the real-world model's six-degree-of-freedom spatial position based on the point cloud data, resulting in a calibrated model. S4: The denoised point cloud data and the calibrated model are fused to construct a fused data model that supports layered hiding. The ground point cloud data in the fused data model is hidden, resulting in layered fused data. S5: Initial route planning is performed based on the layered fused data. The initial route path is then adaptively optimized and replanned using the substation equipment inspection priority and dynamic safety distance algorithm to obtain the optimal route path.

[0009] According to another aspect of the present invention, a dynamic route planning device for substation UAV inspection based on a real-scene model is provided, comprising: a hierarchical block denoising module for performing hierarchical and uniform block processing on the substation point cloud data, and removing point cloud noise through an adaptive noise removal algorithm to obtain a point cloud file group; a dynamic threshold ground extraction module for generating dynamic partitioning thresholds for each hierarchical point cloud file based on an elevation clustering algorithm, and extracting multi-level ground point cloud data according to the dynamic partitioning thresholds; a six-degree-of-freedom calibration module for loading point cloud data and a real-scene model, extracting multi-dimensional feature point coordinates, and performing six-degree-of-freedom precise calibration on the real-scene model through a spatial rigid body transformation algorithm to obtain a calibration model; a hierarchical fusion module for fusing point cloud data and calibration model to construct a fused data model that supports hierarchical hiding, and obtaining hierarchical fused data after hiding the ground point cloud data; and a dynamic planning optimization module for planning an initial route based on the hierarchical fused data, and performing adaptive optimization and replanning of the route by combining equipment inspection priority and dynamic safety distance algorithms to obtain the optimal route path.

[0010] The present invention also provides an electronic device, including at least one processor and a memory communicatively connected to the at least one processor, the memory storing a computer program / instruction executable by the at least one processor, wherein when the at least one processor executes the computer program / instruction, it implements the steps of the above-described dynamic route planning method for substation UAV inspection based on a real-scene model.

[0011] The present invention also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the above-described dynamic route planning method for substation UAV inspection based on a real-scene model.

[0012] Compared with existing technologies, this invention employs a combined approach of layered uniform block division, adaptive noise removal, and dynamic thresholding for elevation clustering in the fine processing of laser point clouds. This achieves complete stripping of ground point clouds, adapting to the complex terrain of substations and effectively improving the accuracy and efficiency of point cloud processing. Regarding model calibration accuracy, it departs from traditional 3D translation calibration, using a multi-dimensional feature point + spatial rigid body transformation algorithm to achieve precise six-degree-of-freedom calibration of the real-world model. The model position deviation is controlled within 2cm, effectively improving the basic accuracy of flight path planning. In terms of data fusion flexibility, it constructs a layered hidden fusion... The integrated data model supports flexible data retrieval at the device level, solving the problems of single and inconvenient use of traditional integrated data, and adapting to the needs of different inspection scenarios. In terms of dynamic route planning, it adopts equipment inspection priority + differentiated dynamic safety distance to achieve adaptive optimization and replanning of the route, which not only ensures comprehensive inspection of core equipment, but also improves the safety of the route and reduces the inspection omission rate. In terms of practicality, this method and device achieve full-process automation without manual intervention, has a low hardware loading threshold, and can be adapted to the UAV inspection route planning of various substations, and has broad engineering application value.

[0013] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the figures. Attached Figure Description

[0014] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0015] Figure 1 is a schematic diagram of the overall process of the dynamic route planning method according to an embodiment of the present invention;

[0016] Figure 2 is a schematic diagram of the point cloud layering and noise removal process according to an embodiment of the present invention.

[0017] Figure 3 is a schematic diagram of the elevation clustering dynamic threshold ground point cloud extraction process according to an embodiment of the present invention;

[0018] Figure 4 is a schematic diagram of the six-degree-of-freedom calibration process of the real-world model according to an embodiment of the present invention;

[0019] Figure 5 is a schematic diagram of the construction of the hierarchical fusion data model according to an embodiment of the present invention;

[0020] Figure 6 is a schematic diagram of the route dynamic optimization and replanning process according to an embodiment of the present invention;

[0021] Figure 7 is a schematic diagram of point cloud data after denoising and layering according to an embodiment of the present invention;

[0022] Figure 8 is a schematic diagram of the real-world model after six degrees of freedom calibration according to an embodiment of the present invention;

[0023] Figure 9 and Figure 10 This is a schematic diagram of a hierarchical fusion data model according to an embodiment of the present invention, wherein... Figure 9 It is an unlabeled version. Figure 10 It is an annotated version;

[0024] Figure 11 This is a schematic diagram of the dynamic route planning device according to an embodiment of the present invention. Detailed Implementation

[0025] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0026] like Figure 1 As shown, the dynamic route planning method for substation UAV inspection based on real-scene 3D model of the present invention includes the following steps S1 to S5.

[0027] S1. Point Cloud Layering and Noise Removal: The 3D point cloud data of the substation acquired by LiDAR is layered and evenly divided according to the spatial height and equipment distribution of the substation, dividing the point cloud data into multiple layered point cloud files to form point cloud file groups. At the same time, an adaptive noise removal algorithm is adopted to automatically identify and remove isolated noise points and pseudo-point clouds based on the spatial density and distance characteristics of the point cloud, resulting in denoised point cloud file groups, which significantly reduces the amount of subsequent data processing.

[0028] S2. Dynamic Threshold Ground Point Cloud Extraction Based on Elevation Clustering. An elevation clustering algorithm is used to perform K-means clustering on the elevation data of each layered point cloud file to obtain multiple elevation clusters. The cluster with the smallest variance is identified as the ground elevation cluster. Based on the elevation characteristics of the ground elevation clusters and combined with the substation terrain type, a multi-level dynamic partitioning threshold is adaptively generated. Ground point clouds are extracted layer by layer from the layered point cloud files to achieve complete removal of ground point clouds. If ground point clouds still exist in the extracted point cloud data to be processed, the above steps are repeated until no ground point clouds remain.

[0029] S3. Precise Calibration of the Real-Scene Model in Six Degrees of Freedom. The denoised point cloud data is loaded and compared with the substation real-scene model constructed based on close-up photography and laser scanning. Multi-dimensional feature point coordinates (including 3D spatial coordinates, normal vector coordinates, and curvature coordinates) of the same core equipment (such as transformers, circuit breakers, current transformers, voltage transformers, and surge arresters) are extracted. The difference matrix of the multi-dimensional feature point coordinates is calculated, and the rotation matrix and translation vector are solved using a spatial rigid body transformation algorithm to achieve precise calibration of the real-scene model in six degrees of freedom (translation, rotation, scaling). A six-degree-of-freedom calibration threshold is set; if the model deviation exceeds the threshold, the position is adjusted to obtain a calibrated model, ensuring that the model position perfectly matches the actual substation.

[0030] S4. Construction of a Layered Fusion Data Model. The denoised point cloud data is deeply fused with the calibration model. The fused data is divided into layers according to the equipment levels of the substation (ground layer, basic equipment layer, core equipment layer, and high-altitude equipment layer). A fusion data model supporting layered hiding and retrieval is constructed. The ground point cloud data in the hidden model is used to obtain the layered fusion data. Independent retrieval commands are set for each layer, supporting individual or combined retrieval of point cloud data and calibration models from any layer, improving the flexibility of data use.

[0031] S5. Dynamic route planning based on inspection priority and dynamic safety distance. First, initial route planning is performed based on hierarchical fusion data, covering all equipment in the substation. Then, the substation equipment is prioritized for inspection (core energized equipment → critical connection equipment → general equipment → auxiliary equipment), and differentiated dynamic safety distance thresholds are set for different priority equipment. Each trajectory point of the initial route is compared with the point cloud data in multiple dimensions to calculate the actual safety distance. If the actual distance is less than the threshold or the route does not cover high-priority equipment, the route is replanned using a dual-objective optimization algorithm of shortest path + most complete coverage, and finally the optimal route path is obtained.

[0032] Dynamic route planning example

[0033] Combined with reference Figures 1 to 10 This embodiment is a case study of drone inspection route planning for a 500 kV substation.

[0034] S1. Point cloud layering and segmentation with noise removal.

[0035] Three-dimensional point cloud data of a 500 kV substation was acquired using LiDAR, with a point cloud density of 180 points / m². The point cloud data was divided into four layers according to the substation's spatial height: 0-1m (ground layer), 1-5m (basic equipment layer), 5-20m (core equipment layer), and above 20m (high-altitude equipment layer). Each layer was evenly divided into 10m × 10m blocks, resulting in a total of 3200 layered point cloud files. An adaptive noise removal algorithm was used, setting the point cloud spatial density threshold to 3 points / m², to remove isolated noise points and pseudo-point clouds. After noise removal, the point cloud data volume was reduced by approximately 15%.

[0036] S2. Dynamic threshold ground point cloud extraction based on elevation clustering.

[0037] K-means clustering (k=3) was performed on the elevation data of each layered point cloud file to obtain three elevation clusters. The cluster with the smallest variance was selected as the ground elevation cluster. The substation has flat terrain, and the elevation expansion coefficient was set to 0.08. Dynamic partitioning thresholds were calculated based on the minimum elevation value (0m) and the average elevation value (0.5m) of the ground elevation cluster. The partitioning threshold for the first level was 0-0.58m, and for the second level it was 0.58-1.2m. Ground point cloud was extracted layer by layer from the layered point cloud file according to the dynamic partitioning threshold. After repeated extraction twice, no ground point cloud residue was found in the point cloud data to be processed, and the ground point cloud removal rate reached 100%.

[0038] (1) K-means clustering formula:

[0039] ;

[0040] in, : Number of clusters (taken as 3 here);

[0041] : Total squared error (sum of variances) of an elevation cluster;

[0042] The i-th elevation cluster ;

[0043] : Elevation value of point cloud p;

[0044] : No. The average elevation of each cluster is calculated using the following formula: ;

[0045] : No. The number of point clouds in each cluster.

[0046] (2) Ground cluster screening formula:

[0047]

[0048] Key parameters based on ground clusters (Minimum elevation) mean ) and expansion coefficient Calculate the level threshold:

[0049] Minimum elevation of ground cluster: ;

[0050] Ground cluster mean: ;

[0051] Elevation expansion coefficient: .

[0052] (3) Upper limit of the first level threshold ( ):

[0053]

[0054] First-level scope: .

[0055] (4) Upper limit of the second level threshold ( ):

[0056] (Where 1.24 is the derived coefficient for the "0.58-1.2m" range, which can also be simplified to the business setting value for direct use).

[0057] Second-level scope:

[0058] (4) Formula for ground point cloud extraction and removal rate:

[0059] No. The extracted ground point cloud set ( )

[0060]

[0061] Cumulative ground point cloud set:

[0062]

[0063] Ground point cloud removal rate:

[0064]

[0065] Condition: After two extractions, there are no ground point clouds in the point cloud to be processed, i.e. Includes all ground points.

[0066] S3, precise calibration of the six degrees of freedom of the real-world model.

[0067] After loading the denoised point cloud data and a near-photorealistic model of the substation, multi-dimensional feature points (50 feature points in total) of core equipment such as transformers, circuit breakers, and disconnectors were extracted, including three-dimensional spatial coordinates (x, y, z) and normal vector coordinates (n). x ,n y ,n z The point cloud and the curvature coordinates (c) are calculated; the multi-dimensional difference matrix between the feature points of the point cloud and the real scene model is calculated, and the rotation matrix (rotation angle ≤ 0.5°) and translation vector (translation distance ≤ 0.08m) are obtained by solving the spatial rigid body transformation algorithm; the six degrees of freedom calibration threshold is set to rotation angle ≤ 1° and translation distance ≤ 0.1m. If the model deviation is within the threshold range, the real scene model is directly used as the calibration model.

[0068] (1) Construction of multi-dimensional difference matrix of feature points:

[0069] Let the set of feature points in the point cloud be... The set of feature points of the real-world model is Each feature point contains three-dimensional coordinates. Normal vector curvature Construct multi-dimensional feature vectors:

[0070]

[0071] Multidimensional difference matrix Defined as the residual matrix of the feature point vectors:

[0072]

[0073] (2) Spatial rigid body transformation, i.e., the core of six-degree-of-freedom calibration, is solving the rotation matrix. ( Orthogonal matrices, satisfying Translation vector To minimize the mean square error of the three-dimensional coordinates:

[0074]

[0075] Note , (The three-dimensional coordinate components of the feature points, the normal vector, and the curvature are used to justify the constraint transformation, and additional weighting is required.)

[0076] (3) Optimization objective with multi-dimensional constraints (fusion of normal vector + curvature)

[0077] To balance the characteristics of normal vectors and curvature, a weighting coefficient is introduced. (coordinate), (Normal vector) (curvature) (satisfies) ,generally The optimization objective is:

[0078]

[0079] ( , (The normal vector component is denoted by ; the curvature is a scalar, and the absolute difference constraint is directly applied.)

[0080] (4) Conversion between rotation matrix and rotation angle:

[0081] Rotation matrix It can be expressed by the axis angle (rotation axis) Rotation angle ), corresponding formula:

[0082] in Here are the antisymmetric matrices for the rotation axes:

[0083] :

[0084] Rotation angle It can be solved by the trace of the rotation matrix:

[0085]

[0086] The trace of the rotation matrix is ​​the sum of its diagonal elements; it is required that... Final calibration threshold .

[0087] (5) Calculation of translation distance

[0088] Translation vector The modulus is the translation distance. ,formula:

[0089]

[0090] in Final calibration threshold .

[0091] (6) Calibration Judgment Formula

[0092] If the following conditions are met: and If the deviation of the real-world model is within the threshold, it can be directly used as the calibration model.

[0093] S4. Construction of a hierarchical fusion data model.

[0094] The denoised point cloud data is deeply fused with the calibration model. The fused data is divided into layers according to the ground layer, basic equipment layer, core equipment layer and high-altitude equipment layer to construct a layered fused data model. The ground layer point cloud data in the model is hidden, and an independent retrieval command is set for each layer. For example, the command "01" retrieves the point cloud data of the basic equipment layer and the command "10" retrieves the calibration model of the core equipment layer. Individual or joint retrieval is supported.

[0095] S5. Dynamic route planning based on inspection priority and dynamic safety distance

[0096] Substation equipment was divided into four inspection priority levels: Level 1 includes core live equipment such as transformers and busbars; Level 2 includes critical connection equipment such as circuit breakers and disconnect switches; Level 3 includes ordinary equipment such as surge arresters and instrument transformers; and Level 4 includes auxiliary equipment such as fences and lighting. Differentiated dynamic safety distance thresholds were set: 5m for Level 1 equipment, 3m for Level 2 equipment, 2m for Level 3 equipment, and 1m for Level 4 equipment. An initial route was planned based on hierarchical fusion data, generating a total of 86 trajectory points. The trajectory points were compared with the point cloud data, and it was found that the actual distance between 3 trajectory points and Level 1 equipment was 4.2m, which was less than the 5m threshold. A dual-objective optimization algorithm of shortest path and most comprehensive coverage was used to replan the route, adjust the position of the trajectory points and optimize the path, and finally obtain the optimal route path with a total of 92 trajectory points. The actual distance between all trajectory points and their corresponding equipment met the dynamic safety distance threshold, and the route fully covered the Level 1 and Level 2 high-priority equipment.

[0097] (1) Formula for determining dynamic safety distance threshold:

[0098]

[0099] in:

[0100] : No. The trajectory point to the first The actual Euclidean distance of each device (calculated in three-dimensional space);

[0101] : No. Safety distance threshold for priority inspection equipment (Level 1) hour , hour , hour , hour );

[0102] Judgment rule: If If the trajectory point does not meet the safety requirements, the route will be replanned.

[0103] 3D Euclidean distance calculation (trajectory point - actual device distance):

[0104]

[0105] : No. The three-dimensional coordinates of the trajectory points;

[0106] : No. The device's exterior and the three-dimensional coordinates of the closest position to the trajectory point.

[0107] (2) Bi-objective optimization objective function (shortest path + most complete coverage):

[0108] The core optimization formula for route replanning must simultaneously satisfy "minimum total path length" and "maximum coverage of high-priority equipment." A weighted summation method is used to transform the dual objectives into a single objective optimization:

[0109]

[0110] in:

[0111] : Comprehensive optimization target value (the smaller the better);

[0112] Weighting coefficients ( Prioritize safety and coverage when taking action Prioritize ensuring short-term path retrieval (This can be adjusted according to the scenario)

[0113] Total route length (the sum of all path points);

[0114] Equipment coverage (when high-priority devices are fully covered) (If not covered, the reduction will be proportional).

[0115] Calculation of total route length:

[0116]

[0117] Total number of trajectory points (initially 86, optimized to 92); : No. The three-dimensional coordinates of the trajectory points.

[0118] Equipment coverage calculation (weighted, highlighting high priority):

[0119]

[0120] in Priority weight (Level 1) Level 2 Level 3 Level 4 Strengthen high-priority coverage);

[0121] : Coverage ratio of level k devices (e.g., if there are 10 level k devices, and the coverage is 10) There are 20 secondary devices, covering 18 of them. ).

[0122] (3) Constraints (Optimization Boundary)

[0123] The hard constraints that must be met during replanning to ensure that the optimization results meet safety and coverage requirements:

[0124]

[0125] Route planning device embodiment

[0126] This invention provides a route planning device 100, such as... Figure 11 As shown, this is used to implement the above-mentioned method for planning the route of UAV inspection of substations based on a real-scene model.

[0127] The route planning device 100 specifically includes: a block module 11, a division module 12, a calibration module 13, a fusion module 14, and a route planning module 15.

[0128] The segmentation module 11 is used to evenly divide the point cloud data of the substation into segments to obtain point cloud file groups, which include multiple point cloud files.

[0129] The segmentation module 12 is used to obtain the corresponding segmentation threshold for each point cloud file and determine the ground point cloud data corresponding to each point cloud file based on the segmentation threshold.

[0130] The calibration module 13 is used to simultaneously load point cloud data and the real scene model of the substation, extract the feature point coordinates of the same components based on the point cloud data and the real scene model, and calibrate the position of the real scene model based on the feature point coordinates of the same components to obtain the calibration model.

[0131] The fusion module 14 is used to fuse point cloud data and calibration model and hide ground point cloud data to obtain fused data.

[0132] The route planning module 15 is used to plan routes based on the fused data and obtain the route path.

[0133] The present invention provides a dynamic route planning scheme for substation drone inspection based on a real-scene model. This scheme achieves refined point cloud data processing, high-precision calibration of the real-scene model, hierarchical data fusion, and dynamic route planning, which significantly improves the accuracy, efficiency, and safety of drone inspection route planning for substations. It can be directly applied to fully automated drone inspection route planning for outdoor and indoor substations of various voltage levels, and can also be extended to drone inspection route planning in industrial scenarios such as wind power, photovoltaic, and chemical industries. It has broad industrial application value and market prospects.

[0134] The above description is merely an embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A dynamic route planning method for substation UAV inspection based on a real-scene model, characterized in that, include: S1. The point cloud data of the substation is processed by layering and uniformly dividing it into blocks, and point cloud noise is removed by noise adaptive elimination algorithm to obtain point cloud file group, which includes multiple layered point cloud files. S2. Based on the elevation clustering algorithm, obtain the corresponding dynamic partitioning threshold for each layered point cloud file, and perform multi-level ground point cloud extraction on each layered point cloud file according to the dynamic partitioning threshold to determine the ground point cloud data corresponding to each layered point cloud file. S3. Load the denoised point cloud data and the real scene model of the substation, extract the multi-dimensional feature point coordinates of the same components in the point cloud data and the real scene model, and perform six-degree-of-freedom spatial position calibration of the real scene model based on the point cloud data through the spatial rigid body transformation algorithm to obtain the calibration model. S4. Fuse the denoised point cloud data and the calibration model to construct a fused data model that supports hierarchical hiding, hide the ground point cloud data in the fused data model, and obtain hierarchical fused data; S5. Based on the hierarchical fusion data, perform initial route planning, and combine the substation equipment inspection priority and dynamic safety distance algorithm to adaptively optimize and replan the initial route path to obtain the optimal route path.

2. The method for dynamic route planning of UAV inspection of substations based on a real-scene model according to claim 1, characterized in that, Step S2 includes: S201. Parse each of the layered point cloud files to obtain a point cloud elevation dataset, and cluster the point cloud elevation dataset using the K-means clustering algorithm to obtain multiple elevation clusters; S202. Perform data fitting on each of the elevation clusters to obtain the dynamic partitioning threshold corresponding to each of the hierarchical point cloud files. The dynamic partitioning threshold includes a lower limit and an upper limit for multi-level partitioning. S203. Based on the multi-level dynamic division threshold, perform layer-by-layer ground point cloud extraction on each of the layered point cloud files to obtain multi-level ground point cloud data and point cloud data to be processed. S204. The point cloud data to be processed is converted back into a hierarchical point cloud file, and the elevation value clustering step is re-executed until there is no ground point cloud data in the point cloud data to be processed.

3. The dynamic route planning method for substation UAV inspection based on a real-scene model according to claim 2, characterized in that, Step S202 includes: S2021. Calculate the elevation mean and variance for each elevation cluster, and take the elevation cluster with the smallest variance as the ground elevation cluster, and use the minimum elevation value of the ground elevation cluster as the lower limit of the first level division. S2022. Calculate the initial value of the threshold based on the elevation mean of the ground elevation cluster and the preset elevation expansion coefficient. Then, modify the initial value of the threshold based on the different ground characteristics of the substation to obtain the modified threshold. S2023. The sum of the maximum elevation value of the ground elevation cluster and the correction threshold is used as the upper limit of the first level division, and so on to obtain the lower limit and upper limit of the multi-level division, thus forming a dynamic division threshold. S2024. The elevation expansion coefficient is adaptively adjusted according to the substation ground type. For example, the value is 0.01 to 0.03 in areas such as hardened roads and cable trench covers in the substation, 0.03 to 0.08 in gravel areas in areas with oil-filled equipment such as transformers and current transformers, and 0.08 to 0.2 in grass areas directly below gantry cranes and busbars.

4. The dynamic route planning method for substation UAV inspection based on a real-scene model according to claim 1, characterized in that, Step S3 includes: S301. Extract the coordinates of the first multi-dimensional feature points corresponding to the point cloud data and the coordinates of the second multi-dimensional feature points corresponding to the real scene model. The multi-dimensional feature point coordinates include three-dimensional spatial coordinates, normal vector coordinates and curvature coordinates. S302. Calculate the multidimensional difference matrix between the coordinates of the first multidimensional feature point and the coordinates of the second multidimensional feature point, and input the multidimensional difference matrix into the spatial rigid body transformation algorithm to solve for the rotation matrix and translation vector. S303. Set a six-degree-of-freedom calibration threshold. If the rotation angle of the rotation matrix or the translation distance of the translation vector exceeds the calibration threshold, adjust the spatial position of the real-world model according to the rotation matrix and the translation vector to obtain a calibrated model. S304. If the rotation angle of the rotation matrix and the translation distance of the translation vector are both within the calibration threshold range, then the real-world model is directly used as the calibration model.

5. The method for dynamic route planning of UAV inspection of substations based on a real-scene model according to claim 1, characterized in that, Step S4 includes: S401. The fused point cloud data and calibration model are divided into layers according to the substation equipment level. The layers include ground layer, basic equipment layer, core equipment layer, and high-altitude equipment layer. S402: Set independent hiding and retrieval commands for each level, supporting the individual retrieval of point cloud data, calibration models or fused data of any level, and also supporting the joint retrieval of multi-level data.

6. The dynamic route planning method for substation UAV inspection based on a real-scene model according to claim 1, characterized in that, Step S5 includes: S501. Prioritize the inspection of all equipment in the substation, with the priority from high to low as core live equipment, critical connection equipment, general equipment, and auxiliary equipment. S502. Based on the dynamic safety distance algorithm, differentiated dynamic safety distance thresholds are set for devices of different priorities, with the safety distance threshold for core energized equipment being higher than that for other devices. S503. Compare each trajectory point of the initial flight path with the point cloud data in multiple dimensions, calculate the actual spatial distance between the trajectory point and each device, and obtain dynamic safety distance data. S504. If the actual spatial distance is less than the dynamic safety distance threshold of the corresponding equipment, or the route path does not cover high-priority equipment, the route path shall be replanned according to the inspection priority and the dynamic safety distance threshold. S505: During the replanning process, a dual-objective optimization algorithm that combines the shortest path and the most comprehensive coverage is used to obtain the optimal flight path.

7. The method for dynamic route planning of UAV inspection of substations based on a real-scene model according to claim 1, characterized in that, The real-scene model is a three-dimensional real-scene model of the substation constructed based on the fusion of close-up photography and laser scanning, and the point cloud data is three-dimensional point cloud data of the substation collected by lidar.

8. A dynamic route planning device for substation UAV inspection based on a real-scene model, characterized in that, include: The layered and block-based denoising module is used to perform layered and uniform block processing on the point cloud data of the substation, and remove point cloud noise through an adaptive noise removal algorithm to obtain point cloud file groups. The dynamic threshold ground extraction module is used to generate dynamic partitioning thresholds for each hierarchical point cloud file based on the elevation clustering algorithm, and to extract multi-level ground point cloud data according to the dynamic partitioning thresholds. The six-degree-of-freedom calibration module is used to load point cloud data and real-world models, extract multi-dimensional feature point coordinates, and perform six-degree-of-freedom precise calibration on the real-world model through a spatial rigid body transformation algorithm to obtain a calibrated model. The layered fusion module is used to fuse point cloud data and calibration models to build a fused data model that supports layered hiding. After hiding the ground point cloud data, the layered fused data is obtained. The dynamic planning optimization module is used to plan the initial route based on the hierarchical fusion data, and to complete the adaptive optimization and replanning of the route by combining equipment inspection priority and dynamic safety distance algorithm to obtain the optimal route path.

9. An electronic device, characterized in that, The method includes at least one processor and a memory communicatively connected to the at least one processor. The memory stores a computer program / instruction executable by the at least one processor. When the at least one processor executes the computer program / instruction, it implements the steps of the dynamic route planning method for substation UAV inspection based on a real-scene model according to any one of claims 1-7.

10. A computer program product, characterized in that, It includes a computer program / instruction, which, when executed by a processor, implements the steps of the dynamic route planning method for substation UAV inspection based on a real-scene model according to any one of claims 1-7.

Images