Mountain wind power road intelligent route selection method and system based on unmanned aerial vehicle three-dimensional modeling

Abstract

The application discloses a mountainous wind power road intelligent route selection method based on a three-dimensional modeling of a unmanned aerial vehicle, and relates to the technical field of route site selection. The method comprises the following steps: collecting tilt photography data, road planning data and fan transportation data and registering to generate road route selection basic data; obtaining a three-dimensional model of a real scene of a mountainous wind power area based on three-dimensional reconstruction of the tilt photography data; analyzing spatial features and extracting environmental feature data; simulating a fan transportation process to generate a fan transportation trajectory envelope area; performing Gaussian space expression by using a 3D Gaussian Splatting algorithm to generate a transportation space Gaussian distribution model; fusing the transportation space Gaussian distribution model and the environmental feature data to generate a road feasible region; and performing path planning based on the road feasible region to output a road route selection result. The application realizes intelligent route selection of a mountainous wind power road and accurate planning of a transportation space, and improves the reliability of road planning.

Description

Technical Field

[0001] This invention relates to the field of route selection technology, and in particular to a method and system for intelligent route selection for mountain wind power roads based on UAV 3D modeling. Background Technology

[0002] Current road planning technologies for mountain wind farms primarily rely on topographic survey data, manual survey data, and two-dimensional map information for road design, determining road routes by analyzing regional topographic conditions, slope variations, and wind turbine placement. With the development of UAV oblique photogrammetry and 3D modeling technologies, some road planning methods utilize UAVs to collect regional imagery data, construct realistic 3D models, and combine this with spatial analysis techniques to assist in road alignment selection, improving planning accuracy in complex mountainous environments.

[0003] Existing technologies still have certain shortcomings in the planning of roads for mountainous wind power projects. On the one hand, existing methods mainly focus on terrain conditions for road planning, lacking analysis of the spatial occupancy characteristics during the transportation of large wind turbines. This makes it difficult to accurately reflect the dynamic spatial requirements of ultra-long components such as turbine blades and towers during transportation, turning, and attitude changes, resulting in planned roads that meet terrain conditions but fail to meet actual transportation requirements. On the other hand, existing 3D modeling methods are mainly used for terrain visualization and spatial measurement, lacking in-depth representation of the spatial distribution of transportation trajectories, making it difficult to convert the continuous spatial occupancy range formed during transportation into a basis for road planning.

[0004] Existing route planning methods mostly rely on grid maps or terrain cost models for route search. The planning results are easily affected by local terrain features and it is difficult to comprehensively consider the matching relationship between wind turbine transportation space requirements and mountainous environmental conditions. Therefore, how to provide an intelligent route selection method and system for mountainous wind power roads based on UAV 3D modeling is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose an intelligent route selection method and system for mountain wind power roads based on UAV 3D modeling. This invention utilizes UAV oblique photography, 3D spatial analysis, 3D Gaussian Splatting, and intelligent path planning technology to achieve reverse modeling of wind turbine transportation space and generation of road feasible regions. It has the advantages of strong transportation adaptability, high spatial planning accuracy, and high degree of road route selection intelligence.

[0006] The intelligent route selection method for mountain wind power roads based on UAV 3D modeling according to an embodiment of the present invention includes the following steps:

[0007] Collect oblique photography data, road planning data, and wind turbine transportation data of mountainous wind power areas, perform spatial registration, and generate basic data for road alignment.

[0008] Based on the oblique photography data in the road alignment basic data, a 3D reconstruction is performed to generate a real-scene 3D model of the mountain wind power area.

[0009] Spatial feature analysis was performed on a real-world 3D model of a mountainous wind power area to extract environmental feature data;

[0010] A transportation motion model is established based on wind turbine transportation data in the road alignment basic data. The spatial simulation of the wind turbine transportation process is performed to generate the wind turbine transportation trajectory envelope area.

[0011] The wind turbine transportation trajectory envelope region is converted into a set of spatial sampling points. The set of spatial sampling points is then expressed in Gaussian space based on the 3D Gaussian Splatting algorithm to generate a Gaussian distribution model of the transportation space.

[0012] The Gaussian distribution model of transportation space is spatially fused with environmental feature data. Based on the spatial occupancy relationship of the Gaussian distribution model of transportation space, the feasible area of ​​the road is determined, and the feasible region of the road is generated.

[0013] Based on the road feasible region, road alignment evaluation data is constructed, and path planning is performed on the road alignment evaluation data to generate road alignment results.

[0014] Optionally, the spatial registration process includes:

[0015] Spatial coordinate calibration is performed on the oblique photogrammetry data to establish spatial reference relationships;

[0016] By mapping road planning data and wind turbine transportation data to the corresponding spatial coordinates of oblique photogrammetry data, a correlation is established to generate basic data for road alignment.

[0017] Optionally, the process of generating a realistic 3D model of the mountain wind power area includes:

[0018] Image features are extracted from the oblique photogrammetry data in the basic data for road alignment, and image correspondence is established based on the feature matching relationship between oblique photogrammetry data from different perspectives.

[0019] Spatial point cloud data is generated by matching spatial points based on image correspondence.

[0020] Spatial structure reconstruction is performed based on spatial point cloud data, and coordinate correction is performed in conjunction with spatial reference relationships to generate a real-world 3D model of the mountain wind power area.

[0021] Optionally, the process of performing spatial feature parsing includes:

[0022] Spatial grid division was performed on the 3D model of the mountain wind power area, and a spatial location index was established.

[0023] Extract the spatial location relationship of the corresponding spatial region based on the spatial location index, and calculate the elevation change relationship and surface continuity change relationship between adjacent spatial regions;

[0024] Based on the texture and spatial structure information in the real-world 3D model of the mountain wind power area, ground features are identified and classified to generate ground feature distribution information;

[0025] Spatial feature data is generated by associating spatial location relationships, elevation changes, surface continuity changes, and ground feature distribution information.

[0026] Optionally, the process of generating the wind turbine transport trajectory envelope region includes:

[0027] A transportation coordinate system is established based on the wind turbine transportation data in the road alignment basic data. The wind turbine transportation process is divided into continuous transportation states, and the spatial position corresponding to each transportation state is recorded according to the transportation sequence.

[0028] Based on the transportation coordinate system, update the spatial position of the transported object under each transportation state, transform the local coordinates corresponding to the transported object to the spatial coordinates of the mountain wind power area, and obtain the spatial occupancy points corresponding to each transportation state.

[0029] Fill the three-dimensional space with the spatial occupancy points corresponding to each transportation status to generate the spatial occupancy area corresponding to the transportation status.

[0030] By superimposing the occupied areas of each space according to the transportation status sequence, merging the overlapping space areas, extracting the spatial range covered by the continuous transportation process, and generating the wind turbine transportation trajectory envelope area.

[0031] Optionally, the transformation process of the spatial sampling point set is as follows: voxel boundary extraction is performed on the envelope area of ​​the wind turbine transportation trajectory, three-dimensional sampling coordinate points are generated according to the spatial position of the voxels, and the three-dimensional sampling coordinate points are combined in order of spatial position to generate a spatial sampling point set.

[0032] Optionally, the process of generating the Gaussian distribution model of the transportation space includes:

[0033] The set of spatial sampling points is divided into spatial sampling regions. The distance relationship between spatial sampling points is calculated, and spatial sampling points whose distance is less than a preset distance threshold are assigned to the same spatial sampling region.

[0034] The 3D Gaussian Splatting algorithm is used to initialize each spatial sampling region with Gaussian elements. The center position of the spatial sampling region is set as the center position of the Gaussian element, the boundary range of the spatial sampling region is converted into the scale parameter of the Gaussian element, and the extension direction of the spatial sampling region is converted into the rotation parameter of the Gaussian element.

[0035] According to the location of the wind turbine transport trajectory envelope region corresponding to the spatial sampling region, update the Gaussian element scale parameter, expand the Gaussian element scale range corresponding to the turning region, and keep the Gaussian element scale range corresponding to the straight path region.

[0036] Based on the spatial connectivity between adjacent spatial sampling regions, the overlapping region between adjacent Gaussian elements is calculated, and the Gaussian elements corresponding to the overlapping region are fused. The center position, scale parameter and covariance parameter of the fused Gaussian elements are then updated.

[0037] The updated Gaussian elements are spatially combined to generate a Gaussian distribution model for the transportation space.

[0038] Optionally, the process of generating the road feasible region includes:

[0039] The Gaussian elements in the Gaussian distribution model of the transportation space are projected onto a three-dimensional spatial grid. The spatial distance between each spatial grid and the Gaussian element is calculated. The Gaussian occupancy probability corresponding to the spatial grid is calculated based on the spatial distance to generate the transportation space occupancy probability field.

[0040] The spatial location corresponding to the environmental feature data is mapped to the probability field of transportation space occupancy. Based on the distribution information of ground features and the relationship of continuous surface changes in the environmental feature data, the spatial grid is filtered and the spatial grid that does not meet the transportation space requirements of the wind turbine is deleted.

[0041] A connected component search is performed on the filtered spatial grids to aggregate spatial grids with continuous spatial connections, extract continuous transportation spatial regions, and generate road feasible regions.

[0042] Optionally, the process of generating road alignment results includes:

[0043] The feasible region of the road is converted into a set of spatial nodes, and a road search graph is built based on the connection relationship between the spatial nodes.

[0044] Calculate the passage cost corresponding to each spatial node in the road search map to generate road alignment evaluation data;

[0045] Based on the road alignment evaluation data, a path search is performed on the road search map. The cumulative path cost between the starting node and the target node is calculated, and the node connection path with the minimum cumulative path cost is selected.

[0046] Spatial smoothing is performed on the filtered node connection paths to convert continuous spatial nodes into road centerline data and generate road alignment results.

[0047] The intelligent route selection system for mountain wind power roads based on UAV 3D modeling according to an embodiment of the present invention includes the following modules:

[0048] The data acquisition module is used to collect oblique photography data, road planning data, and wind turbine transportation data in mountainous wind power areas, perform spatial registration, and generate basic data for road alignment.

[0049] The 3D reconstruction module is used to perform 3D reconstruction based on oblique photogrammetry data in the road alignment base data, and generate a real-world 3D model of the mountain wind power area.

[0050] The feature parsing module is used to perform spatial feature parsing on the real-world 3D model of the mountain wind power area and extract environmental feature data;

[0051] The transportation simulation module is used to build a transportation motion model based on wind turbine transportation data in the road alignment basic data, perform spatial simulation of the transportation process, and generate the wind turbine transportation trajectory envelope area.

[0052] The Gaussian representation module is used to convert the envelope region of the wind turbine transportation trajectory into a set of spatial sampling points, and generate a Gaussian distribution model of the transportation space based on the 3DGaussian Splatting algorithm.

[0053] The feasible region generation module is used to spatially fuse the Gaussian distribution model of transportation space with environmental feature data to determine the feasible area of ​​the road and generate the road feasible region.

[0054] The route planning module is used to construct road route evaluation data based on the road feasible region, perform path planning, and generate road route results.

[0055] The beneficial effects of this invention are:

[0056] (1) This invention uses UAV oblique photography data to perform three-dimensional reconstruction, and combines spatial feature analysis to extract environmental feature data, thereby realizing the digital expression of the terrain structure, land feature distribution and spatial change status of the mountain wind power area, and improving the spatial perception ability of road planning in complex mountain environments;

[0057] (2) This invention establishes a transportation motion model based on wind turbine transportation data, performs spatial simulation of wind turbine transportation process, and constructs a Gaussian distribution model of transportation space by combining 3D Gaussian Splatting algorithm, thereby realizing the reverse expression of wind turbine transportation space demand and reducing the problem of low road adaptability caused by insufficient consideration of transportation space in traditional road planning.

[0058] (3) This invention spatially integrates the Gaussian distribution model of transportation space with environmental feature data, determines the feasible region of roads based on the occupancy relationship of transportation space, and generates road selection results by combining path planning, thereby improving the rationality, accuracy and engineering applicability of road selection for mountain wind power. Attached Figure Description

[0059] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0060] Figure 1 The flowchart shows the intelligent route selection method for mountain wind power roads based on UAV 3D modeling proposed in this invention.

[0061] Figure 2 This is a module connection diagram of the intelligent route selection system for mountain wind power roads based on UAV 3D modeling proposed in this invention;

[0062] Figure 3 This is a voxel representation of the transportation trajectory of the intelligent route selection method for mountain wind power roads based on UAV 3D modeling proposed in this invention.

[0063] Figure 4 This is a spatial sampling and Gaussian initialization diagram of the intelligent route selection method for mountain wind power roads based on UAV 3D modeling proposed in this invention;

[0064] Figure 5 This is a Gaussian element modulation fusion image of the intelligent route selection method for mountain wind power roads based on UAV 3D modeling proposed in this invention.

[0065] Figure 6 The transportation space Gaussian model diagram of the intelligent route selection method for mountain wind power roads based on UAV 3D modeling proposed in this invention;

[0066] Figure 7 This is a road feasible region generation map for the intelligent route selection method for mountain wind power roads based on UAV 3D modeling proposed in this invention. Detailed Implementation

[0067] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0068] Example 1: Reference Figures 1-2 The intelligent route selection method for mountain wind power roads based on UAV 3D modeling includes the following steps:

[0069] Collect oblique photography data, road planning data, and wind turbine transportation data of mountainous wind power areas, perform spatial registration, and generate basic data for road alignment.

[0070] Based on the oblique photography data in the road alignment basic data, a 3D reconstruction is performed to generate a real-scene 3D model of the mountain wind power area.

[0071] Spatial feature analysis was performed on a real-world 3D model of a mountainous wind power area to extract environmental feature data;

[0072] A transportation motion model is established based on wind turbine transportation data in the road alignment basic data. The spatial simulation of the wind turbine transportation process is performed to generate the wind turbine transportation trajectory envelope area.

[0073] The wind turbine transportation trajectory envelope region is converted into a set of spatial sampling points. The set of spatial sampling points is then expressed in Gaussian space based on the 3D Gaussian Splatting algorithm to generate a Gaussian distribution model of the transportation space.

[0074] The Gaussian distribution model of transportation space is spatially fused with environmental feature data. Based on the spatial occupancy relationship of the Gaussian distribution model of transportation space, the feasible area of ​​the road is determined, and the feasible region of the road is generated.

[0075] Based on the road feasible region, road alignment evaluation data is constructed, and path planning is performed on the road alignment evaluation data to generate road alignment results.

[0076] In this embodiment, the spatial registration process includes:

[0077] Spatial coordinate calibration is performed on the oblique photogrammetry data to establish spatial reference relationships;

[0078] In this embodiment, the oblique photography data of the mountain wind power area is used to construct the expression of the mountain spatial environment. The oblique photography data includes multi-view image data and image acquisition location data. The multi-view image data is obtained by a drone equipped with an oblique photography camera taking pictures of the mountain wind power area from multiple angles according to a preset flight path. The image acquisition location data is collected synchronously by the drone positioning device. The multi-view image data is stored in JPEG format, and the image acquisition location data is stored in latitude and longitude coordinate format.

[0079] Road planning data is used to describe the locational relationships of target areas for road construction. The road planning data includes road starting location data, road connection location data, and road planning area data. The road starting location data and road connection location data are represented in latitude and longitude coordinate format, while the road planning area data is represented in vector surface data format. The road planning data is obtained through wind farm planning information.

[0080] Wind turbine transportation data is used to describe the spatial requirements corresponding to the transportation process of wind turbine equipment. Wind turbine transportation data includes transportation start location data, transportation destination location data, and transportation spatial parameter data. The transportation start location data and transportation destination location data are represented in latitude and longitude coordinate format. The transportation spatial parameter data includes transportation vehicle size parameters and wind turbine component size parameters. The transportation spatial parameter data is obtained through the wind turbine transportation plan.

[0081] By mapping road planning data and wind turbine transportation data to the corresponding spatial coordinates of oblique photogrammetry data, a correlation is established to generate basic data for road alignment.

[0082] The specific process for spatial coordinate calibration of oblique photogrammetry data is as follows: First, determine the initial spatial position corresponding to each multi-view image based on the image acquisition location data. Then, transform the latitude and longitude coordinates to a unified spatial coordinate system, establishing the transformation relationship between the image acquisition location and the spatial coordinates. Second, select ground control points with fixed spatial positions within the mountainous wind power area. Match the pixel positions of the ground control points in the multi-view image data with the corresponding spatial coordinates of the ground control points, calculating the transformation parameters between the image coordinates and the spatial coordinates. Third, use the transformation parameters to perform spatial correction on the multi-view image data, mapping the image data acquired from different perspectives to a unified spatial coordinate system, thus establishing a spatial reference relationship for the oblique photogrammetry data.

[0083] The specific process of spatial mapping is as follows: The road starting position data, road connection position data, and road planning area data in the road planning data are transformed to a unified spatial coordinate system corresponding to the oblique photography data. Based on the coordinate transformation relationship, the location of the road planning data in the real-world space of the mountainous wind power area is determined. The transportation starting position data and transportation target position data in the wind turbine transportation data are transformed to a unified spatial coordinate system, and the spatial range corresponding to the transportation demand is established based on the transportation spatial parameter data. The road planning location, transportation location, and the corresponding spatial location in the oblique photography data are matched to establish the correlation between the road planning data, wind turbine transportation data, and the spatial location of the mountainous wind power area.

[0084] Based on the above spatial relationships, the spatial environment information corresponding to the oblique photography data, the road location relationship corresponding to the road planning data, and the transportation space demand corresponding to the wind turbine transportation data are bound together, and basic data for road alignment is established according to a unified spatial coordinate index.

[0085] In this embodiment, the process of generating a realistic 3D model of a mountainous wind power area includes:

[0086] Image features are extracted from the oblique photogrammetry data in the basic data for road alignment, and image correspondence is established based on the feature matching relationship between oblique photogrammetry data from different perspectives.

[0087] Specifically, grayscale normalization and image enhancement processing are performed on images from different perspectives in the oblique photogrammetry data. Based on the pixel changes in local areas of the image, image feature points with spatial distinguishability are extracted. Image feature points include corner points, edge points, and feature points corresponding to areas with obvious texture changes. Feature descriptions are performed on the extracted image feature points to generate feature description information corresponding to each image feature point.

[0088] Similarity calculations are performed on feature description information in images from different perspectives, and corresponding feature points are determined based on the degree of matching between feature description information. Feature point pairs with mismatches are removed, and feature point pairs with consistent spatial locations and geometric relationships are retained to establish image correspondence between oblique photogrammetric data from different perspectives. Image correspondence is used to describe the corresponding mapping relationship of the same spatial location in images from different perspectives.

[0089] Spatial point cloud data is generated by matching spatial points based on image correspondence.

[0090] Based on the image correspondence, multiple image observation points corresponding to the same spatial location are determined. The spatial coordinates of the image observation points are calculated by combining the acquisition position relationship of images from different perspectives. The multiple image observation points are converted into spatial three-dimensional coordinate points. Error screening is performed on the spatial three-dimensional coordinate points to delete abnormal points whose spatial position deviation exceeds the preset range, retain the valid spatial three-dimensional coordinate points, and generate spatial point cloud data.

[0091] Spatial structure reconstruction is performed based on spatial point cloud data, and coordinate correction is performed in conjunction with spatial reference relationships to generate a real-world 3D model of the mountain wind power area.

[0092] Spatial connectivity analysis is performed on adjacent spatial points in spatial point cloud data, and a continuous spatial structure is established based on the positional relationship between spatial points. Spatial interpolation is performed on the missing areas in the spatial structure to supplement the structural information corresponding to the missing spatial locations. Based on the spatial point cloud data and spatial connectivity, the continuous surface morphology of the mountain wind power area is restored to form a complete spatial representation.

[0093] Based on the spatial reference relationship in the road alignment base data, the coordinate transformation parameters corresponding to the oblique photography data are obtained, and the spatial data after spatial structure reconstruction is transformed into a unified spatial coordinate system. The spatial reference relationship is used to correct the spatial position offset, so that the position of ground features in the spatial data is consistent with the road planning position and the wind turbine transportation position. The spatial data after coordinate correction is textured, and the image texture in the oblique photography data is mapped to the corresponding spatial position, generating a real-scene 3D model of the mountain wind power area containing spatial position information and surface texture information.

[0094] The real-world 3D model of the mountain wind power area includes spatial location data, surface morphology data, and image texture data. Spatial location data is used to represent the positional relationship of various spatial points within the mountain wind power area, surface morphology data is used to represent the elevation changes and spatial structure of the mountain wind power area, and image texture data is used to represent the surface visual features of the mountain wind power area.

[0095] In this embodiment, the process of spatial feature parsing includes:

[0096] Spatial grid division was performed on the 3D model of the mountain wind power area, and a spatial location index was established.

[0097] Obtain the spatial coordinate range corresponding to the real-world 3D model of the mountain wind power area, and determine the 3D spatial division boundary based on the spatial coordinate range; divide the 3D space according to the preset spatial size, and divide the real-world 3D model of the mountain wind power area into multiple spatial regions; the spatial regions are represented in the form of cubic mesh, and each spatial region corresponds to a unique spatial number; establish a spatial location index based on the center coordinates of the spatial regions, and map the spatial number to the spatial point coordinates in the real-world 3D model of the mountain wind power area.

[0098] In this embodiment, the spatial division size is set according to the road alignment accuracy requirements. The spatial division size can be set to 5 meters × 5 meters × 5 meters. Each spatial area stores the spatial point coordinates, texture information and surface structure information within the corresponding range.

[0099] Extract the spatial location relationship of the corresponding spatial region based on the spatial location index, and calculate the elevation change relationship and surface continuity change relationship between adjacent spatial regions;

[0100] The spatial numbers corresponding to adjacent spatial regions are obtained based on the spatial location index. Adjacent spatial regions are determined according to the spatial connection relationship between them. When there is a common boundary between two spatial regions or the distance between their center points is less than a preset distance threshold, the two spatial regions are determined to be adjacent spatial regions. Spatial connection relationship is established based on the spatial numbers of the adjacent spatial regions to generate spatial location relationship.

[0101] The elevation values ​​corresponding to spatial points within each spatial region are extracted. The elevation values ​​of multiple spatial points within the spatial region are statistically calculated to obtain the average elevation value corresponding to the spatial region. The average elevation difference between adjacent spatial regions is calculated and used as the elevation change relationship between adjacent spatial regions. When the average elevation difference between adjacent spatial regions increases, it indicates that there is a significant change in terrain height between adjacent spatial regions. When the average elevation difference decreases, it indicates that there is a continuous elevation change between adjacent spatial regions.

[0102] The three-dimensional coordinate information of spatial points in adjacent spatial regions is obtained, and the continuity of spatial point distribution is calculated based on the positional relationship between spatial points. The surface continuity relationship is determined based on the changes in the number of spatial points, the continuity of spatial point connections, and the changes in spatial structure in adjacent spatial regions. When the spatial point distribution in adjacent spatial regions is continuous and the changes in spatial structure are gradual, the surface continuity relationship is determined to be high. When there are broken areas in the spatial point distribution in adjacent spatial regions or significant changes in spatial structure, the surface continuity relationship is determined to be low.

[0103] Based on the texture and spatial structure information in the real-world 3D model of the mountain wind power area, ground features are identified and classified to generate ground feature distribution information;

[0104] The textural information in the 3D model of the mountain wind power area is converted into 2D texture feature data, and the spatial structure information is converted into spatial geometric feature data. For example, the 2D texture feature data and spatial geometric feature data are input into the DeepLabv3+ semantic segmentation network. The encoder extracts land cover features at different scales, and the hollow spatial pyramid pooling structure extracts spatial semantic information under different receptive ranges. The decoder restores the spatial resolution and outputs the land cover category probabilities corresponding to different spatial locations. Based on the land cover category probabilities, the land cover categories corresponding to each spatial region are determined, generating land cover distribution information. The land cover distribution information includes the spatial location relationships of vegetation areas, bare land areas, rock mass areas, gully areas, and steep slope areas.

[0105] Spatial feature data is generated by associating spatial location relationships, elevation changes, surface continuity changes, and ground feature distribution information.

[0106] Using spatial location indexes as association identifiers, spatial location relationships, elevation changes, surface continuity changes, and feature distribution information corresponding to the same spatial region are bound together. The spatial ID, spatial location relationships, elevation changes, surface continuity changes, and feature category information corresponding to each spatial region are combined to form a spatial region environmental feature record. All spatial region environmental feature records are then summarized to generate environmental feature data. This environmental feature data is used to describe the spatial connectivity, topographic changes, surface continuity, and feature distribution at different spatial locations within the mountainous wind power area.

[0107] In this embodiment, the process of generating the wind turbine transport trajectory envelope region includes:

[0108] A transportation coordinate system is established based on the wind turbine transportation data in the road alignment basic data. The wind turbine transportation process is divided into continuous transportation states, and the spatial position corresponding to each transportation state is recorded according to the transportation sequence.

[0109] The starting position of the transportation is obtained from the wind turbine transportation data, and the spatial position corresponding to the starting position is used as the origin of the transportation coordinate system. The direction of the first coordinate axis of the transportation coordinate system is determined according to the transportation direction, so that the direction of the first coordinate axis is consistent with the initial movement direction of the transported object. The direction of the third coordinate axis is determined according to the vertical direction in the spatial coordinates of the mountain wind power area, so that the direction of the third coordinate axis is consistent with the direction of gravity. The direction of the second coordinate axis is determined by cross product calculation based on the directions of the first and third coordinate axes, forming a mutually perpendicular three-dimensional transportation coordinate system.

[0110] The transportation coordinate system is used to describe the spatial state of the transported object. The origin of the transportation coordinate system is determined by the starting position of the transportation. The first coordinate axis is used to represent the forward direction of transportation, the second coordinate axis is used to represent the lateral direction of transportation, and the third coordinate axis is used to represent the height direction of the transported object.

[0111] After establishing the transportation coordinate system, the spatial reference relationship in the road alignment data is used to map the transportation coordinate system to the spatial coordinates of the mountain wind power area, so that the spatial position in the transportation coordinate system can be transformed into the corresponding spatial range of the real-world 3D model of the mountain wind power area.

[0112] The transportation process is divided into states based on changes in the transportation path. A transportation process with continuous changes in transportation direction and stable changes in spatial position is classified as a transportation state. When the transportation direction changes, the transportation position changes significantly, or the posture of the transported object changes, a new transportation state is generated. The transportation coordinate system position corresponding to each transportation state is recorded in the order of transportation time. For example, the transportation process is sampled at fixed spatial intervals, with the sampling interval set to 1 meter to 5 meters. Each sampling position corresponds to a transportation state, which is used to describe the spatial change process of the transported object under different spatial positions.

[0113] Based on the transportation coordinate system, update the spatial position of the transported object under each transportation state, transform the local coordinates corresponding to the transported object to the spatial coordinates of the mountain wind power area, and obtain the spatial occupancy points corresponding to each transportation state.

[0114] In each transportation state, the position coordinates of each component of the transportation object are updated according to the initial spatial position of the transportation object in the transportation coordinate system; when the transportation object undergoes a turning motion, the spatial orientation of the transportation object is updated according to the turning angle corresponding to the transportation state; when the transportation object undergoes an attitude change, the spatial position of the transportation object is updated according to the attitude change amount corresponding to the transportation state, so that the transportation object maintains the spatial state corresponding to the actual transportation process in each transportation state.

[0115] Based on the transformation relationship between the transportation coordinate system and the spatial coordinates of the mountain wind power area, the spatial position in the transportation coordinate system is converted into the position coordinate in the unified spatial coordinate system; the three-dimensional spatial coordinate points corresponding to the transportation objects are output in the order of transportation status, forming the spatial occupancy points corresponding to each transportation status.

[0116] Fill the three-dimensional space with the spatial occupancy points corresponding to each transportation status to generate the spatial occupancy area corresponding to the transportation status.

[0117] A three-dimensional voxel space covering the space occupancy points corresponding to the transportation status is established. The spatial range corresponding to the transportation status is divided into multiple regular voxel units, and each voxel unit has a unique spatial number. The voxel unit to which the space occupancy point belongs is determined based on the three-dimensional coordinates of the space occupancy point. The voxel unit containing the space occupancy point is marked as occupied, and the voxel unit not containing the space occupancy point is marked as idle.

[0118] Specifically, the voxel dimensions are set according to the size of the wind turbine transport object and the accuracy of road alignment, with voxel side lengths ranging from 0.2 meters to 1 meter, enabling the voxel units to represent the spatial changes of the wind turbine blades and tower during transportation. After determining the voxel occupancy state, the spatial continuity of adjacent occupied voxels is judged. Based on the spatial adjacency relationship between voxel units, occupied voxels with common faces, common edges, or common vertices are determined as continuously occupied voxels; these continuously occupied voxels are then merged to form a continuous three-dimensional spatial region. For areas where blade swaying and vehicle turning cause spatial expansion during transportation, the sets of occupied voxels corresponding to different transportation states are superimposed. Voxel regions repeatedly occupied under multiple transportation states are retained, and voxel regions occupied under any state during transportation are merged to generate a complete transportation state spatial occupancy region.

[0119] For example, in a mountainous wind power road planning scenario, a transportation coordinate system is established based on wind turbine transportation data, dividing the transportation process into multiple transportation states. When the transport vehicle is in a straight transportation state, a corresponding set of spatial occupancy points is generated, mainly distributed in the area along the length of the vehicle; when the transport vehicle enters a turning state, the wind turbine component attitude is updated according to the transportation coordinate system, the spatial position of the blades changes, and a set of spatial occupancy points with an expanded coverage area is generated.

[0120] The spatial occupancy points corresponding to the straight-line transportation state and the turning transportation state are mapped to voxel space, with each voxel having a side length of 0.5 meters. The spatial occupancy points generated in the straight-line transportation state occupy 12,000 voxel units, while those generated in the turning transportation state occupy 18,000 voxel units, with 5,000 voxel units overlapping with those in the straight-line transportation state. After merging the occupied voxels corresponding to the two transportation states, 25,000 consecutive occupied voxels are obtained, generating the spatial occupancy region corresponding to this transportation process.

[0121] Through the above process, the vehicle body, blade swing range, and turning expansion space are all converted into a continuous three-dimensional spatial region during the transportation state change.

[0122] By superimposing the occupied areas of each space according to the transportation status sequence, merging the overlapping space areas, extracting the spatial range covered by the continuous transportation process, and generating the wind turbine transportation trajectory envelope area.

[0123] According to the time sequence of the wind turbine transportation process, the occupied voxel sets corresponding to each transportation state are sorted, the occupied voxel set corresponding to the first transportation state is obtained, and the occupied voxel set is initialized as a spatial cover set; the occupied voxel sets corresponding to subsequent transportation states are obtained in sequence, and the occupied voxel set corresponding to the current transportation state is spatially superimposed with the spatial cover set.

[0124] The spatial overlay process is achieved by comparing the spatial numbers corresponding to the occupying voxels. When the spatial number of the occupying voxel corresponding to the current transportation state does not exist in the spatial overlay set, the occupying voxel is added to the spatial overlay set. When the spatial number of the occupying voxel corresponding to the current transportation state already exists in the spatial overlay set, the existing occupying state is retained and no new spatial region is generated.

[0125] Read the spatial numbers of all occupying voxels in the spatial coverage set, and determine the adjacent occupying voxels based on the three-dimensional positional relationship corresponding to the spatial numbers; divide the occupying voxels with continuous spatial connection relationship into the same spatial region, and uniformly number the occupying voxels belonging to the same continuous region to form a continuous occupying voxel set.

[0126] When extracting the spatial range covered by the continuous transportation process, the spatial coverage boundary is determined according to the voxel boundary coordinates in the continuously occupied voxel set. The continuous spatial range occupied under all transportation states is taken as the spatial coverage area corresponding to the wind turbine transportation process, and the wind turbine transportation trajectory envelope area is generated.

[0127] For example, in a mountainous wind power road planning scenario, the wind turbine transportation process is divided into 5 consecutive transportation states. Each transportation state, after being filled with voxel space, yields a set of occupied voxels. The first transportation state corresponds to 12,000 occupied voxels, the second transportation state to 13,500 occupied voxels, the third transportation state to 15,000 occupied voxels, the fourth transportation state to 17,000 occupied voxels, and the fifth transportation state to 16,000 occupied voxels.

[0128] When superimposing data according to the time sequence of transport states, the voxels occupied by the first transport state are used as the initial spatial coverage set, and the voxels corresponding to subsequent transport states are added sequentially. The second transport state adds 1,500 voxels with different spatial numbers, of which 12,000 voxels overlap with existing spatial regions and are retained only once. After accumulating data from 5 transport states, 42,500 voxels with unique spatial numbers are obtained.

[0129] Spatial continuity is determined for 42,500 occupied voxels, and adjacent connected occupied voxels are merged into a continuous spatial region to obtain a continuous spatial range covering the straight transport region, the turning extension region, and the attitude change region, thus generating the wind turbine transport trajectory envelope region.

[0130] In this embodiment, the transformation process of the spatial sampling point set is as follows: voxel boundaries are extracted from the envelope area of ​​the wind turbine transportation trajectory, three-dimensional sampling coordinate points are generated according to the spatial position of the voxels, and the three-dimensional sampling coordinate points are combined in order of spatial position to generate a spatial sampling point set.

[0131] When performing voxel boundary extraction, the set of occupied voxels corresponding to the envelope region of the wind turbine transport trajectory is obtained, and the spatial number and three-dimensional coordinate information corresponding to each occupied voxel are read. For each occupied voxel, the voxel state corresponding to its six adjacent directions is detected. The six adjacent directions include the positive X direction, the negative X direction, the positive Y direction, the negative Y direction, the positive Z direction, and the negative Z direction.

[0132] When there is no occupying voxel in any adjacent direction of an occupying voxel or the corresponding position belongs to an empty voxel, the occupying voxel is determined as a boundary occupying voxel; when there are adjacent occupying voxels in all six directions of an occupying voxel, the occupying voxel is determined as an internal occupying voxel and is not involved in boundary extraction.

[0133] After determining the boundary occupying voxels, the position of the outer surface of the voxels is obtained according to the spatial number corresponding to the boundary occupying voxels. The boundary surfaces with continuous connection between adjacent boundary occupying voxels are merged, and the external spatial boundary corresponding to the wind turbine transport trajectory envelope area is extracted.

[0134] In this embodiment, a six-neighbor search method is used to determine voxel boundaries. Each occupying voxel only needs to check the states of six adjacent voxels to determine whether it belongs to the boundary region. When an occupying voxel is located at the outer contour of the transport trajectory envelope region, at least one of its adjacent directions corresponds to free space, and this occupying voxel is marked as a boundary occupying voxel.

[0135] In this embodiment, the process of generating the Gaussian distribution model of the transportation space includes:

[0136] After obtaining the set of spatial sampling points, this implementation method uses the 3D Gaussian Splatting algorithm to represent the set of spatial sampling points in Gaussian space. Traditional 3D Gaussian Splatting mainly initializes Gaussian elements based on the spatial point positions and achieves static scene representation by optimizing the position, scale, and rotation parameters of the Gaussian elements. This implementation method addresses the problems of continuous occupation of transportation space, spatial expansion of turning areas, and spatial changes under different transportation states during wind turbine transportation. It improves the 3D Gaussian Splatting algorithm by converting the set of spatial sampling points into a Gaussian distribution model of transportation space that can describe the extent of space occupied by wind turbine transportation.

[0137] The spatial sampling point set is divided into spatial sampling regions, the distance relationship between spatial sampling points is calculated, and spatial sampling points whose distance is less than a preset distance threshold are assigned to the same spatial sampling region; the specific process is as follows:

[0138] Obtain the three-dimensional coordinate information corresponding to each spatial sampling point in the spatial sampling point set, and represent the spatial sampling point in three-dimensional coordinate form; for any two spatial sampling points, calculate the spatial distance between the two spatial sampling points, and determine the spatial proximity relationship between the spatial sampling points based on the spatial distance.

[0139] When the spatial distance between two spatial sampling points is less than a preset distance threshold, the two spatial sampling points are assigned to the same spatial sampling region; when the spatial distance between two spatial sampling points is greater than or equal to the preset distance threshold, the two spatial sampling points are assigned to different spatial sampling regions.

[0140] The spatial sampling area represents a local continuous spatial range within the envelope region of the wind turbine transportation trajectory, with different spatial sampling areas corresponding to different transportation spatial locations. The spatial distance threshold is set based on the size of the wind turbine transportation object and the road alignment accuracy. In this embodiment, the spatial distance threshold is set to 1 meter to ensure that the spatial sampling point division results maintain the continuity of the wind turbine transportation space.

[0141] For example, in a mountainous wind power road planning scenario, the wind turbine transportation trajectory envelope area is transformed into 50,000 spatial sampling points. By calculating the distance between the spatial sampling points, spatial sampling points with a spatial distance of less than 1 meter are divided into the same spatial sampling area, ultimately forming multiple continuous spatial sampling areas. Among them, straight road sections correspond to spatial sampling areas that are continuous in length, while turning areas correspond to spatial sampling areas that are horizontally expanded, used to reflect different space occupancy states during wind turbine transportation.

[0142] The 3D Gaussian Splatting algorithm is used to initialize each spatial sampling region with Gaussian elements. The center position of the spatial sampling region is set as the center position of the Gaussian element, the boundary range of the spatial sampling region is converted into the scale parameter of the Gaussian element, and the extension direction of the spatial sampling region is converted into the rotation parameter of the Gaussian element.

[0143] Specifically, based on the traditional 3D Gaussian Splatting algorithm, this invention does not directly use a single spatial sampling point as a Gaussian element, but uses the spatial sampling region as the initialization object of the Gaussian element, so that each Gaussian element corresponds to the local spatial area occupied during the wind turbine transportation process.

[0144] For each spatial sampling region, the three-dimensional coordinates of all spatial sampling points within that region are obtained. The X, Y, and Z coordinates of all spatial sampling points are averaged and calculated. The calculation results are used as the center coordinates of the spatial sampling region, and these center coordinates are set as the center position parameters of the corresponding Gaussian element.

[0145] The specific calculation process is as follows:

[0146] The center coordinate in the X direction is equal to the average value of the X-direction coordinates of all spatial sampling points within the spatial sampling area;

[0147] The center coordinate in the Y direction is equal to the average value of the Y-direction coordinates of all spatial sampling points within the spatial sampling area;

[0148] The center coordinate in the Z direction is equal to the average value of the Z-direction coordinates of all spatial sampling points within the spatial sampling area.

[0149] After determining the center position of the Gaussian element, the boundary range of the spatial sampling region is calculated. The maximum and minimum coordinate values ​​of the spatial sampling points within the spatial sampling region in the X, Y, and Z directions are obtained, and the spatial ranges corresponding to the three directions are calculated respectively. The spatial ranges in the three directions are then converted into Gaussian element scale parameters.

[0150] Among them, the X-direction scale parameter represents the extension range of the transportation space along the road direction, the Y-direction scale parameter represents the lateral swing range of the wind turbine component, and the Z-direction scale parameter represents the vertical occupancy range of the wind turbine component.

[0151] After determining the scale parameters, the main extension direction of the spatial sampling points within the spatial sampling region is calculated. Specifically, the direction of positional change between adjacent spatial sampling points is calculated, all directional change directions are statistically analyzed, and the direction with the highest frequency is selected as the extension direction of the spatial sampling region. This extension direction is then converted into Gaussian element rotation parameters.

[0152] In the above manner, the Gaussian elements generated by the present invention no longer represent only a single spatial location, but rather the local spatial occupancy area during the wind turbine transportation process.

[0153] For example, in the spatial sampling area of ​​a straight road segment, since the change in the transportation direction is small, the spatial sampling points are mainly distributed along the road direction, so the generated Gaussian elements have a longer road direction scale parameter; in the turning area, since the blades expand laterally as the vehicle turns, the lateral range of the corresponding spatial sampling area increases, so the generated Gaussian elements have a larger lateral scale parameter.

[0154] According to the location of the wind turbine transport trajectory envelope region corresponding to the spatial sampling region, update the Gaussian element scale parameter, expand the Gaussian element scale range corresponding to the turning region, and keep the Gaussian element scale range corresponding to the straight path region.

[0155] Obtain the location of the wind turbine transport trajectory envelope region corresponding to the spatial sampling area, and determine the transport state type corresponding to the current spatial sampling area based on the spatial direction changes between adjacent transport states.

[0156] When the directional change angle between consecutive transport states is less than a preset angle threshold, the current spatial sampling area is determined as a straight path area, while keeping the corresponding Gaussian element scale parameter unchanged.

[0157] When the angle of change of direction between consecutive transport states is greater than or equal to a preset angle threshold, the current spatial sampling area is determined as a turning area, and the corresponding Gaussian element scale parameter is adjusted according to the degree of change of transport direction.

[0158] The specific adjustment process is as follows: read the lateral diffusion range of the corresponding spatial sampling point in the turning area, convert the change in lateral diffusion range into a scale adjustment amount, and superimpose the scale adjustment amount onto the lateral scale parameter of the Gaussian element so that the coverage range of the Gaussian element is consistent with the spatial requirement range during the wind turbine turning process.

[0159] When a transport vehicle enters a curved section of road, the vehicle's main body changes direction, and the blades rotate with the vehicle, causing the lateral range of the transport trajectory envelope to expand. At this point, a corresponding Gaussian element in the Y-direction scale parameter is added so that the Gaussian element can cover the spatial occupancy range of the blades after they turn.

[0160] Based on the spatial connectivity between adjacent spatial sampling regions, the overlapping region between adjacent Gaussian elements is calculated, and the Gaussian elements corresponding to the overlapping region are fused. The center position, scale parameter and covariance parameter of the fused Gaussian elements are then updated.

[0161] Obtain the Gaussian elements corresponding to adjacent spatial sampling regions, and calculate the spatial proximity based on the distance between the centers of the Gaussian elements; obtain the scale ranges corresponding to two Gaussian elements, and calculate the spatial overlap region between the two scale ranges.

[0162] When the center distance between two Gaussian elements is less than a preset distance threshold, and there is an overlapping area in the scale range of the two Gaussian elements, the two Gaussian elements are identified as related Gaussian elements.

[0163] When fusing associated Gaussian elements, the fusion ratio is determined based on the number of spatial sampling points corresponding to the two Gaussian elements; the weighted average of the center positions of the two Gaussian elements is calculated using the fusion ratio, and the calculation result is used as the new center position of the fused Gaussian elements.

[0164] New scale parameters are calculated based on the original scale parameters of the two Gaussian elements and the fusion ratio, so that the fused Gaussian elements maintain the original transportation space coverage.

[0165] The covariance parameter is updated based on the relationship between the rotation direction and spatial orientation of the two Gaussian elements, so that the fused Gaussian elements can describe the orientation changes in the continuous transport space.

[0166] For example, during wind turbine transportation, if there is a large amount of overlapping space in the spatial regions corresponding to two consecutive transportation states S2 and S3, and the Gaussian elements corresponding to the two states have a large overlapping area, then the two Gaussian elements will be merged; for areas where the space changes significantly before and after the transportation vehicle turns, the two Gaussian elements have a low degree of overlap, so they will be expressed as independent Gaussians.

[0167] The updated Gaussian elements are spatially combined to generate a Gaussian distribution model for the transportation space.

[0168] Obtain all updated Gaussian elements, and arrange the center position, scale parameter, rotation parameter, and covariance parameter of each Gaussian element according to their spatial position relationship to form a set of Gaussian elements.

[0169] Based on the spatial adjacency relationship between Gaussian elements, all Gaussian elements are mapped to a unified three-dimensional space, so that each Gaussian element can jointly describe the spatial occupancy range corresponding to the wind turbine transportation trajectory envelope region, thus generating a Gaussian distribution model of the transportation space.

[0170] The input to the Gaussian distribution model of transportation space is a set of spatial sampling points. After spatial sampling region division, Gaussian element initialization, transportation state modulation, and Gaussian element fusion processing, the output is a three-dimensional spatial representation composed of multiple Gaussian elements.

[0171] Each Gaussian element contains a center location parameter, a scale parameter, a rotation parameter, and a covariance parameter. The center location parameter represents the location of the transportation space, the scale parameter represents the coverage area of ​​the transportation space, the rotation parameter represents the direction of the transportation space extension, and the covariance parameter represents the distribution variation of the transportation space.

[0172] Unlike traditional 3D Gaussian Splatting used for static scene visual reconstruction, the Gaussian distribution model of the transportation space generated in this invention is used to describe the spatial occupancy range during the transportation of large wind turbines.

[0173] In this embodiment, the process of generating the road feasible region includes:

[0174] The Gaussian elements in the Gaussian distribution model of the transportation space are projected onto a three-dimensional spatial grid. The spatial distance between each spatial grid and the Gaussian element is calculated. The Gaussian occupancy probability corresponding to the spatial grid is calculated based on the spatial distance to generate the transportation space occupancy probability field.

[0175] A three-dimensional spatial grid is established based on the spatial range corresponding to the real-world three-dimensional model of the mountain wind power area, dividing the mountain wind power area into multiple regular spatial units. Each spatial unit has a unique spatial number and three-dimensional coordinate position. The three-dimensional spatial grid is used to uniformly carry the Gaussian distribution model of transportation space and environmental feature data, so that transportation space information and mountain environment information are mapped to the same spatial location.

[0176] The three-dimensional spatial grid differs from the voxel space. The voxel space is used to describe the occupancy status of the wind turbine transportation trajectory envelope area, while the three-dimensional spatial grid is used for spatial evaluation in the road feasible domain generation stage, transforming the transportation space Gaussian distribution model and environmental feature data to a unified spatial scale.

[0177] Specifically, the Gaussian element parameters in the transportation space Gaussian distribution model are obtained, including the Gaussian center position, scale parameter, and covariance parameter. Each spatial grid cell in the 3D spatial grid is traversed to obtain the coordinates of the spatial grid center. The Euclidean distance between the spatial grid center coordinates and the Gaussian element center position is calculated. The spatial grid center coordinates are then substituted into the spatial distribution range of the corresponding Gaussian element to obtain the occupancy probability of the corresponding Gaussian element in that spatial grid.

[0178] When a spatial grid cell corresponds to multiple Gaussian elements, the occupancy probability between the spatial grid cell and each of the multiple Gaussian elements is calculated separately. These multiple occupancy probabilities are then fused to obtain the final occupancy probability of the spatial grid cell. The occupancy probabilities corresponding to all spatial grid cells are arranged according to their spatial numbers to form a transportation space occupancy probability field. The transportation space occupancy probability field is a three-dimensional spatial representation of all three-dimensional spatial grid cells and their corresponding occupancy probabilities. Each spatial grid cell corresponds to an occupancy probability value, which indicates the degree to which the spatial location meets the transportation space requirements of the wind turbine.

[0179] The spatial location corresponding to the environmental feature data is mapped to the probability field of transportation space occupancy. Based on the distribution information of ground features and the relationship of continuous surface changes in the environmental feature data, the spatial grid is filtered and the spatial grid that does not meet the transportation space requirements of the wind turbine is deleted.

[0180] The spatial requirements for wind turbine transportation refer to the spatial conditions that ensure continuous passage between the transported object and the mountainous environment during wind turbine transportation. Specifically, this includes: the spatial grid location being within the coverage area of ​​the transportation space occupancy probability field; the absence of any terrain features affecting transportation passage in the area corresponding to the spatial grid; and the maintenance of continuous spatial connectivity between spatial grids.

[0181] In practice, the occupancy probability of the transportation space corresponding to the three-dimensional spatial grid is obtained, and the occupancy probability of the transportation space is used to determine whether the current spatial grid meets the transportation space coverage requirements of the wind turbine. When the occupancy probability of the spatial grid is lower than the preset occupancy probability threshold, it is determined that the spatial grid cannot meet the transportation space requirements of the wind turbine.

[0182] The system acquires information on the distribution of ground features and the relationship between continuous surface changes in environmental feature data. When the area corresponding to the spatial grid belongs to a steep slope area, a gully area, or the degree of continuous change of the spatial surface is lower than the preset continuity threshold, it is determined that the spatial grid does not meet the space requirements for wind turbine transportation.

[0183] Retain spatial grids that simultaneously meet the needs of transportation space coverage and environmental access, and delete spatial grids that do not meet the transportation space requirements of wind turbines.

[0184] A connected component search is performed on the filtered spatial grids to aggregate spatial grids with continuous spatial connections, extract continuous transportation spatial regions, and generate road feasible regions.

[0185] Specifically, the process involves obtaining the retained spatial grids after filtering, establishing three-dimensional adjacency relationships based on the spatial number of each spatial grid, and for each spatial grid, detecting the six adjacent spatial positions corresponding to the positive X direction, negative X direction, positive Y direction, negative Y direction, positive Z direction, and negative Z direction to determine whether there are retained spatial grids at the adjacent positions.

[0186] When two spatial grids have a common surface connection, the two spatial grids are defined as continuous spatial grids and a spatial connection is established; when two spatial grids do not have a common surface connection, a spatial connection is not established.

[0187] After establishing spatial connectivity, all retained spatial grids are labeled as regions. The specific process is as follows: select an unlabeled spatial grid as the starting grid, perform a neighborhood search on the starting grid, and add adjacent spatial grids that have a direct connection with the starting grid to the same region; continue to perform a neighborhood search on newly added spatial grids, and continuously add spatial grids that have a connection with existing regions to that region until no new connected spatial grids are added, thus completing the labeling of a continuous spatial region.

[0188] Repeat the above process to continue region labeling for the remaining unlabeled spatial grids until all retained spatial grids have been divided into regions, resulting in multiple continuous spatial regions.

[0189] After completing the division of continuous spatial regions, the spatial extent of each continuous spatial region is statistically analyzed, and the number of spatial grids, spatial boundary range, and spatial connection length of each continuous spatial region are calculated. Isolated regions with fewer than a preset threshold of spatial grids are deleted, and continuous spatial regions that meet the requirements for continuous wind turbine transportation space are retained.

[0190] The preserved continuous spatial regions are combined with corresponding spatial grids to extract spatial boundary contours, forming the road feasible domain.

[0191] For example, in a mountainous wind power road planning scenario, after screening for transportation space occupancy probability and environmental features, 80,000 spatial grids are retained. The adjacency relationships in six directions are checked for each spatial grid, and a connection is established when two spatial grids have a common surface connection. After connected component marking, the 80,000 spatial grids are divided into three continuous spatial regions. The largest continuous region contains 76,000 spatial grids, corresponding to the continuous passage area for wind turbine transportation; the other two regions contain 2,000 and 2,000 spatial grids respectively, and are deleted because their spatial range is small and they cannot connect to the main transportation area. The finally retained continuous spatial regions are converted into road feasible regions.

[0192] In this embodiment, the process of generating road alignment results includes:

[0193] The feasible region of the road is converted into a set of spatial nodes, and a road search graph is built based on the connection relationship between the spatial nodes.

[0194] Calculate the spatial distance between any two spatial nodes. If the distance between two spatial nodes is less than a preset connection distance threshold, and the connecting line between the two spatial nodes is entirely within the road feasible region, then establish a connection between the two spatial nodes. The spatial nodes are treated as nodes in the road search graph, and the connections between spatial nodes are treated as edges in the road search graph. The road search graph represents the reachability relationships between different spatial locations within the road feasible region; nodes represent candidate road locations, and edges represent traversable paths between adjacent spatial locations.

[0195] Calculate the passage cost corresponding to each spatial node in the road search map to generate road alignment evaluation data;

[0196] The spatial location of each spatial node in the road search map is obtained, and the terrain and feature status corresponding to that location are queried in the environmental feature data. The passage cost of the spatial node is calculated based on the degree of terrain change and the degree of feature influence in the area corresponding to the spatial node. The degree of terrain change is determined based on the elevation change relationship in the environmental feature data; the degree of feature influence is determined based on the feature distribution information in the environmental feature data.

[0197] When the elevation change of the area corresponding to a spatial node is small and there are no features that affect road construction, the passage value corresponding to that spatial node is reduced; when the elevation change of the area corresponding to a spatial node is significant or there are features that affect road construction, the passage value corresponding to that spatial node is increased.

[0198] The specific calculation process is as follows:

[0199] Obtain the elevation change value of the area corresponding to the spatial node, and convert the elevation change value into terrain cost; obtain the land feature category of the area corresponding to the spatial node, and set the land feature impact cost according to different land feature categories; combine the terrain cost and land feature impact cost to obtain the access cost corresponding to the spatial node.

[0200] For example, if the elevation change of a certain spatial node A is small and the land feature category is bare land, the corresponding terrain cost is 0.2 and the land feature impact cost is 0.1, then the passage cost of this spatial node is low.

[0201] A certain spatial node B corresponds to an area with steep slopes and significant elevation changes. The corresponding terrain cost is 0.8, and the feature impact cost is 0.7. Therefore, the passage cost of this spatial node is relatively high.

[0202] When generating road alignment evaluation data, the accessibility value corresponding to all spatial nodes in the road search map is first read. The spatial node number, spatial location coordinates, node connection relationships, and accessibility value are then stored to form node evaluation records. All node evaluation records are organized according to the connection relationships in the road search map to generate road alignment evaluation data. This data describes the accessibility difficulty of different candidate paths at their corresponding spatial locations in the road search map. The data includes spatial node information, node connection information, and node accessibility value. Spatial node information is used to determine the location of candidate roads, node connection information is used to determine the path search direction, and node accessibility value is used to evaluate the construction difficulty of the areas traversed by the candidate roads.

[0203] Based on the road alignment evaluation data, a path search is performed on the road search map. The cumulative path cost between the starting node and the target node is calculated, and the node connection path with the minimum cumulative path cost is selected.

[0204] Obtain the starting position node and the target position node in the road search graph, set the starting position node as the current search node, and add the starting position node to the set of nodes to be searched; use the A* path search algorithm to traverse the nodes in the road search graph, and obtain the adjacent spatial nodes corresponding to the current search node according to the node connection relationship in the road search graph.

[0205] For each adjacent spatial node, obtain the connection edge between the current search node and the adjacent spatial node, and calculate the node movement cost based on the spatial distance corresponding to the connection edge and the travel cost corresponding to the adjacent spatial node; add the cumulative path cost corresponding to the current search node to the node movement cost to obtain the cumulative path cost corresponding to the adjacent spatial node.

[0206] Among them, spatial distance is used to represent the actual length of the road, and passage cost is used to represent the difficulty of road construction in the area corresponding to the spatial node; the larger the spatial distance, the higher the cost of moving the node; the higher the passage cost, the less suitable the corresponding spatial area is as a road passage area.

[0207] After calculating the cumulative path cost of adjacent spatial nodes, the estimated path cost is calculated based on the spatial distance between the target location node and the current spatial node. The cumulative path cost and the estimated path cost are combined to obtain the spatial node search evaluation value. The next search node is selected in ascending order of the spatial node search evaluation value, and the road search map is continuously expanded.

[0208] When the search process reaches the target node, node expansion stops, and all spatial nodes traversed between the starting node and the target node are connected in the search order to form a candidate node connection path.

[0209] The cumulative path cost calculation process is as follows: Obtain all connecting edges between adjacent spatial nodes in the candidate node connection path, and read the node movement cost corresponding to each connecting edge; according to the connection order from the starting node to the target node, accumulate the node movement costs corresponding to all connecting edges, and use the accumulated result as the cumulative path cost of the candidate node connection path. For example, if a candidate node connection path includes starting node A, spatial node B, spatial node C, and target node D, and the node movement cost from A to B is 5, the node movement cost from B to C is 8, and the node movement cost from C to D is 6, then the cumulative path cost of this candidate node connection path is 19.

[0210] Calculate the cumulative path cost for all candidate node connection paths in the road search graph, and determine the candidate node connection path with the minimum cumulative path cost as the target node connection path.

[0211] Spatial smoothing is performed on the filtered node connection paths to convert continuous spatial nodes into road centerline data and generate road alignment results.

[0212] Obtain the coordinates of all spatial nodes in the filtered node connection path, generate a discrete spatial node sequence according to the node connection order; calculate the directional change angle between adjacent spatial nodes, and identify the directional change locations in the node connection path.

[0213] For continuous spatial nodes whose direction change angle is less than a preset angle threshold, the original node coordinates are kept unchanged; for spatial nodes whose direction change angle is greater than or equal to the preset angle threshold, a cubic B-spline curve fitting method is used for curve interpolation.

[0214] The specific process is as follows: the spatial nodes in the selected node connection path are used as curve control points, the curve interpolation points are calculated according to the order of the control points, and the direction change between adjacent road segments is reduced by increasing the transition points between nodes, so that the discrete node connection path is transformed into a continuous spatial curve path.

[0215] In this embodiment, curve fitting parameters are set according to the turning radius requirements of mountain roads to keep the generated curve path within the road's feasible region and prevent the smoothed path from deviating from the road's feasible region.

[0216] Obtain the continuous spatial curve path after spatial smoothing, sort the curve interpolation points according to the road extension direction, and use the sorted three-dimensional spatial coordinates as the set of road centerline points.

[0217] The road centerline data is generated by combining the 3D coordinates, node connection order, and path length information from the set of road centerline points. This data represents the spatial location and direction of the planned road within the mountainous wind power area, including the road's starting and ending points, and the corresponding 3D coordinate sequence of the areas the road traverses. The road centerline data is then spatially validated against the road's feasible region. If the entire road centerline lies within the feasible region, the road centerline data is determined as the road alignment result.

[0218] For example, in a mountainous wind power road planning scenario, 12,000 spatial nodes are generated based on the feasible road domain, and a road search map is established based on the distance relationships between the spatial nodes. The spatial node corresponding to the starting position of wind turbine transportation is used as the starting position node, and the spatial node corresponding to the wind turbine installation position is used as the target position node.

[0219] The A* path search algorithm is used to search the road search graph, generating 36 candidate node connection paths during the search process. The cumulative path cost is calculated for each candidate node connection path, where the path with the lowest cumulative path cost contains 420 spatial nodes.

[0220] Spatial smoothing is performed on the node connection path containing 420 spatial nodes. The 420 discrete spatial nodes are then input into a cubic B-spline curve fitting process to generate a continuous spatial curve path. Curve interpolation points are extracted according to the road extension direction to generate road centerline data, forming the final road alignment result.

[0221] The intelligent route selection system for mountain wind power roads based on UAV 3D modeling according to an embodiment of the present invention includes the following modules:

[0222] The data acquisition module is used to collect oblique photography data, road planning data, and wind turbine transportation data in mountainous wind power areas, perform spatial registration, and generate basic data for road alignment.

[0223] The 3D reconstruction module is used to perform 3D reconstruction based on oblique photogrammetry data in the road alignment base data, and generate a real-world 3D model of the mountain wind power area.

[0224] The feature parsing module is used to perform spatial feature parsing on the real-world 3D model of the mountain wind power area and extract environmental feature data;

[0225] The transportation simulation module is used to build a transportation motion model based on wind turbine transportation data in the road alignment basic data, perform spatial simulation of the transportation process, and generate the wind turbine transportation trajectory envelope area.

[0226] The Gaussian representation module is used to convert the envelope region of the wind turbine transportation trajectory into a set of spatial sampling points, and generate a Gaussian distribution model of the transportation space based on the 3DGaussian Splatting algorithm.

[0227] The feasible region generation module is used to spatially fuse the Gaussian distribution model of transportation space with environmental feature data to determine the feasible area of ​​the road and generate the road feasible region.

[0228] The route planning module is used to construct road route evaluation data based on the road feasible region, perform path planning, and generate road route results.

[0229] Example 2: Reference Figures 3-7 To verify the feasibility of this invention in practice, it was applied to a mountainous wind power road planning scenario to intelligently select routes for transporting large wind turbines.

[0230] In the planning of roads for mountainous wind power projects, wind turbine transport vehicles need to carry long blades, towers, and other equipment through complex mountainous environments. Traditional road planning mainly relies on topographic surveying results to determine road locations, which makes it difficult to fully consider vehicle turning, equipment swaying, and changes in space occupation during wind turbine transportation. This results in problems such as insufficient turning space, impassable sections, and subsequent road adjustments. This embodiment utilizes the method proposed in this invention to comprehensively process oblique photogrammetry data collected by UAVs, road planning data, and wind turbine transportation data to generate road alignment results that meet the requirements of wind turbine transportation.

[0231] In the application process, oblique photogrammetry data, road planning data, and wind turbine transportation data of the mountainous wind power area were collected. Spatial coordinates from different data sources were unified, and the spatial locations in the oblique photogrammetry data were matched with corresponding locations in the road planning and wind turbine transportation data to generate basic road alignment data. 3D reconstruction was then performed using the oblique photogrammetry data from the basic road alignment data. Through feature matching between images from different perspectives, spatial point calculation, and coordinate correction, a realistic 3D model of the mountainous wind power area was constructed. (Reference) Figure 3 This study utilizes a real-world 3D model of a mountainous wind power area to represent the space surrounding the wind turbine transportation road, converting the transportation vehicles, road space, and surrounding environment into spatial voxel representations, thus providing a foundation for subsequent transportation space analysis.

[0232] After completing the 3D reconstruction, spatial feature analysis was performed on the 3D model of the mountainous wind power area. Spatial units were obtained through spatial grid division, and the elevation and surface continuity relationships between adjacent spatial areas were calculated. Based on model texture and spatial structure information, land cover classification was performed, identifying vegetated areas, bare land areas, rocky areas, gully areas, and steep slope areas, thus forming environmental feature data. This environmental feature data is used to describe the spatial state of the mountainous environment and provides a basis for selecting feasible road zones.

[0233] A transportation motion model is established based on wind turbine transportation data from the road alignment baseline data to simulate the wind turbine transportation process spatially. The transportation motion model calculates the spatial occupancy range under different states during transportation based on the vehicle's trajectory, vehicle attitude changes, and changes in the spatial dimensions of wind turbine components, generating the wind turbine transportation trajectory envelope region. (Reference) Figure 3 The wind turbine transportation trajectory envelope area is spatially represented in the form of voxels, which transforms the vehicle driving area and the wind turbine component swing area into a continuous spatial occupation range, enabling the road planning process to take into account the actual transportation space requirements.

[0234] The wind turbine transport trajectory envelope is converted into a set of spatial sampling points, and these spatial sampling points are then divided into spatial regions. (Reference) Figure 4 The sampling points within a continuous spatial region are divided according to spatial distance relationships. Gaussian elements are initialized for each spatial sampling region. The center position of the spatial sampling region is converted into the center position of the Gaussian element, the spatial range is converted into a scale parameter, and the spatial extension direction is converted into a rotation parameter, thus forming a Gaussian element for the transportation space.

[0235] To address the issue that traditional 3D Gaussian Splatting algorithms are primarily used for static scene reconstruction and cannot describe the dynamic space occupancy during transportation, this embodiment adapts and improves the 3D Gaussian Splatting algorithm. (Reference) Figure 5The scale range of Gaussian elements is adjusted according to changes in the wind turbine's transportation status. The transportation space scale remains stable on straight sections, while the coverage of Gaussian elements is expanded in turning areas, enabling the Gaussian elements to represent the spatial expansion state of wind turbine components during turning. For overlapping areas in the computational space of adjacent Gaussian elements, Gaussian elements with continuous spatial occupancy relationships are merged, and the fused center position, scale parameters, and covariance parameters are updated to form a continuous transportation space representation.

[0236] refer to Figure 6 The updated Gaussian elements are combined according to their spatial relationships to generate a Gaussian distribution model for the transportation space. This model uses multiple Gaussian elements to describe the spatial occupancy during wind turbine transportation. Each Gaussian element corresponds to a local transportation space region, and the transportation space extent is expressed through its center position, scale parameter, rotation parameter, and covariance parameter.

[0237] A Gaussian distribution model of transportation space is spatially fused with environmental feature data to establish a unified spatial evaluation area. The spatial occupancy probability is calculated based on the Gaussian elements in the transportation space Gaussian distribution model, and the Gaussian spatial expression is converted into a transportation space occupancy probability field. Combining the information on land cover distribution and surface continuity in the environmental feature data, areas with steep slopes, gullies, and discontinuous terrain are filtered out, retaining continuous spatial areas that meet the transportation space requirements of wind turbines. (Reference) Figure 7 A three-dimensional adjacency relationship is established for the filtered spatial grid, and continuous spatial regions are extracted by the connected component search method to generate the road feasible region.

[0238] Road alignment evaluation data is constructed based on the road feasible region, which is converted into a set of spatial nodes. A road search graph is then established based on the connectivity between these nodes. The passage cost corresponding to each spatial node is calculated, and the construction difficulty corresponding to spatial distance and environmental characteristics is converted into path evaluation indicators. The A* path search algorithm is used to search for paths on the road search graph, calculating the cumulative path cost for different node connection paths, and selecting the node connection path with the minimum cumulative path cost. The selected node connection paths are then smoothed using cubic B-spline curves, converting discrete spatial nodes into continuous road centerline data, and generating road alignment results.

[0239] Through the above process, this embodiment completes the entire processing flow from UAV 3D spatial data acquisition, wind turbine transportation space simulation, Gaussian space representation, road feasible region generation to road path planning. Compared with methods that rely solely on terrain data for road planning, this invention can identify spatial conflict areas in advance during wind turbine transportation, improving the matching degree between road plans and transportation needs.

[0240] To verify the effectiveness of the method of the present invention, a comparative test was conducted between the method of the present invention and the traditional road planning method based on terrain analysis. The test results are shown in Table 1 below.

[0241] Table 1: Performance Comparison of Different Road Alignment Methods Traditional terrain analysis methods 0.85 76.4 81.2 8.6 7 Three-dimensional model-based planning method 0.52 86.7 89.5 5.1 4 Method of the present invention 0.31 96.8 95.6 2.3 1

[0242] Table 1 shows the performance comparison of different road alignment methods. The method of this invention achieves good results in 3D spatial modeling, transportation space representation, road feasible region identification, and route planning optimization. Traditional terrain analysis methods have a 3D spatial modeling error of 0.85m, a transportation space coverage rate of 76.4%, a road feasible region identification accuracy of 81.2%, a route planning length deviation of 8.6%, and require 7 planning adjustments. This indicates that relying solely on terrain data for road alignment does not adequately consider the spatial occupancy during wind turbine transportation, leading to discrepancies between planned roads and actual transportation needs.

[0243] The 3D model-based planning method improves terrain representation by constructing a regional 3D spatial model. The 3D spatial modeling error is reduced to 0.52m, the transportation space coverage completeness rate is increased to 86.7%, the road feasible domain identification accuracy reaches 89.5%, the path planning length deviation is reduced to 5.1%, and the number of planning adjustments is reduced to 4. While this method can utilize 3D spatial information to assist road planning, its ability to represent dynamic spatial changes during wind turbine transportation remains limited, making it difficult to fully reflect the spatial requirements formed by the turning and swaying processes of large wind turbine components.

[0244] This invention utilizes UAV oblique photogrammetry for 3D reconstruction, wind turbine transportation space simulation, and the construction of a Gaussian distribution model of the transportation space based on the 3D Gaussian Splatting algorithm to achieve joint analysis of transportation space and mountainous environmental space. Test results show that the 3D spatial modeling error of this invention is reduced to 0.31m, the transportation space coverage completeness rate is increased to 96.8%, the road feasible region identification accuracy reaches 95.6%, the path planning length deviation is reduced to 2.3%, and the number of planning adjustments is reduced to 1. These data indicate that this invention improves the expressive power of wind turbine transportation space requirements through a Gaussian representation of transportation space, making the road feasible region generation process more consistent with actual transportation scenarios, and improving the accuracy and engineering applicability of road selection results for mountainous wind power projects.

[0245] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent route selection for mountain wind power roads based on UAV 3D modeling, characterized in that, Includes the following steps: Collect oblique photography data, road planning data, and wind turbine transportation data of mountainous wind power areas, perform spatial registration, and generate basic data for road alignment. Based on the oblique photography data in the road alignment basic data, a 3D reconstruction is performed to generate a real-scene 3D model of the mountain wind power area. Spatial feature analysis was performed on a real-world 3D model of a mountainous wind power area to extract environmental feature data; A transportation motion model is established based on wind turbine transportation data in the road alignment basic data. The spatial simulation of the wind turbine transportation process is performed to generate the wind turbine transportation trajectory envelope area. The wind turbine transportation trajectory envelope region is converted into a set of spatial sampling points. The set of spatial sampling points is then expressed in Gaussian space based on the 3D Gaussian Splatting algorithm to generate a Gaussian distribution model of the transportation space. The Gaussian distribution model of transportation space is spatially fused with environmental feature data. Based on the spatial occupancy relationship of the Gaussian distribution model of transportation space, the feasible area of ​​the road is determined, and the feasible region of the road is generated. Based on the road feasible region, road alignment evaluation data is constructed, and path planning is performed on the road alignment evaluation data to generate road alignment results.

2. The intelligent route selection method for mountain wind power roads based on UAV 3D modeling according to claim 1, characterized in that, The spatial registration process includes: Spatial coordinate calibration is performed on the oblique photogrammetry data to establish spatial reference relationships; By mapping road planning data and wind turbine transportation data to the corresponding spatial coordinates of oblique photogrammetry data, a correlation is established to generate basic data for road alignment.

3. The intelligent route selection method for mountain wind power roads based on UAV 3D modeling according to claim 1, characterized in that, The process of generating a realistic 3D model of the mountain wind power area includes: Image features are extracted from the oblique photogrammetry data in the basic data for road alignment, and image correspondence is established based on the feature matching relationship between oblique photogrammetry data from different perspectives. Spatial point cloud data is generated by matching spatial points based on image correspondence. Spatial structure reconstruction is performed based on spatial point cloud data, and coordinate correction is performed in conjunction with spatial reference relationships to generate a real-world 3D model of the mountain wind power area.

4. The intelligent route selection method for mountain wind power roads based on UAV 3D modeling according to claim 1, characterized in that, The process of performing spatial feature analysis includes: Spatial grid division was performed on the 3D model of the mountain wind power area, and a spatial location index was established. Extract the spatial location relationship of the corresponding spatial region based on the spatial location index, and calculate the elevation change relationship and surface continuity change relationship between adjacent spatial regions; Based on the texture and spatial structure information in the real-world 3D model of the mountain wind power area, ground features are identified and classified to generate ground feature distribution information; Spatial feature data is generated by associating spatial location relationships, elevation changes, surface continuity changes, and ground feature distribution information.

5. The intelligent route selection method for mountain wind power roads based on UAV 3D modeling according to claim 1, characterized in that, The process of generating the wind turbine transport trajectory envelope region includes: A transportation coordinate system is established based on the wind turbine transportation data in the road alignment basic data. The wind turbine transportation process is divided into continuous transportation states, and the spatial position corresponding to each transportation state is recorded according to the transportation sequence. Based on the transportation coordinate system, update the spatial position of the transported object under each transportation state, transform the local coordinates corresponding to the transported object to the spatial coordinates of the mountain wind power area, and obtain the spatial occupancy points corresponding to each transportation state. Fill the three-dimensional space with the spatial occupancy points corresponding to each transportation status to generate the spatial occupancy area corresponding to the transportation status. By superimposing the occupied areas of each space according to the transportation status sequence, merging the overlapping space areas, extracting the spatial range covered by the continuous transportation process, and generating the wind turbine transportation trajectory envelope area.

6. The intelligent route selection method for mountain wind power roads based on UAV 3D modeling according to claim 1, characterized in that, The transformation process of the spatial sampling point set is as follows: voxel boundaries are extracted from the envelope area of ​​the wind turbine transportation trajectory, three-dimensional sampling coordinate points are generated according to the spatial position of the voxels, and the three-dimensional sampling coordinate points are combined in order of spatial position to generate a spatial sampling point set.

7. The intelligent route selection method for mountain wind power roads based on UAV 3D modeling according to claim 1, characterized in that, The process of generating the Gaussian distribution model of the transportation space includes: The set of spatial sampling points is divided into spatial sampling regions. The distance relationship between spatial sampling points is calculated, and spatial sampling points whose distance is less than a preset distance threshold are assigned to the same spatial sampling region. The 3D Gaussian Splatting algorithm is used to initialize each spatial sampling region with Gaussian elements. The center position of the spatial sampling region is set as the center position of the Gaussian element, the boundary range of the spatial sampling region is converted into the scale parameter of the Gaussian element, and the extension direction of the spatial sampling region is converted into the rotation parameter of the Gaussian element. According to the location of the wind turbine transport trajectory envelope region corresponding to the spatial sampling region, update the Gaussian element scale parameter, expand the Gaussian element scale range corresponding to the turning region, and keep the Gaussian element scale range corresponding to the straight path region. Based on the spatial connectivity between adjacent spatial sampling regions, the overlapping region between adjacent Gaussian elements is calculated, and the Gaussian elements corresponding to the overlapping region are fused. The center position, scale parameter and covariance parameter of the fused Gaussian elements are then updated. The updated Gaussian elements are spatially combined to generate a Gaussian distribution model for the transportation space.

8. The intelligent route selection method for mountain wind power roads based on UAV 3D modeling according to claim 1, characterized in that, The process of generating the road feasible region includes: The Gaussian elements in the Gaussian distribution model of the transportation space are projected onto a three-dimensional spatial grid. The spatial distance between each spatial grid and the Gaussian element is calculated. The Gaussian occupancy probability corresponding to the spatial grid is calculated based on the spatial distance to generate the transportation space occupancy probability field. The spatial location corresponding to the environmental feature data is mapped to the probability field of transportation space occupancy. Based on the distribution information of ground features and the relationship of continuous surface changes in the environmental feature data, the spatial grid is filtered and the spatial grid that does not meet the transportation space requirements of the wind turbine is deleted. A connected component search is performed on the filtered spatial grids to aggregate spatial grids with continuous spatial connections, extract continuous transportation spatial regions, and generate road feasible regions.

9. The intelligent route selection method for mountain wind power roads based on UAV 3D modeling according to claim 1, characterized in that, The process of generating road alignment results includes: The feasible region of the road is converted into a set of spatial nodes, and a road search graph is built based on the connection relationship between the spatial nodes. Calculate the passage cost corresponding to each spatial node in the road search map to generate road alignment evaluation data; Based on the road alignment evaluation data, a path search is performed on the road search map. The cumulative path cost between the starting node and the target node is calculated, and the node connection path with the minimum cumulative path cost is selected. Spatial smoothing is performed on the filtered node connection paths to convert continuous spatial nodes into road centerline data and generate road alignment results.

10. A mountain wind power road intelligent route selection system based on UAV 3D modeling, applied to the mountain wind power road intelligent route selection method based on UAV 3D modeling as described in any one of claims 1 to 9, characterized in that, Includes the following modules: The data acquisition module is used to collect oblique photography data, road planning data, and wind turbine transportation data in mountainous wind power areas, perform spatial registration, and generate basic data for road alignment. The 3D reconstruction module is used to perform 3D reconstruction based on oblique photogrammetry data in the road alignment base data, and generate a real-world 3D model of the mountain wind power area. The feature parsing module is used to perform spatial feature parsing on the real-world 3D model of the mountain wind power area and extract environmental feature data; The transportation simulation module is used to build a transportation motion model based on wind turbine transportation data in the road alignment basic data, perform spatial simulation of the transportation process, and generate the wind turbine transportation trajectory envelope area. The Gaussian representation module is used to convert the envelope region of the wind turbine transportation trajectory into a set of spatial sampling points, and generate a Gaussian distribution model of the transportation space based on the 3DGaussian Splatting algorithm. The feasible region generation module is used to spatially fuse the Gaussian distribution model of transportation space with environmental feature data to determine the feasible area of ​​the road and generate the road feasible region. The route planning module is used to construct road route evaluation data based on the road feasible region, perform path planning, and generate road route results.

Images