A digital twin-based urban and rural planning method and system

Abstract

This invention relates to the field of architectural planning technology, specifically to a method and system for urban and rural planning based on digital twins. The method includes acquiring building information models and geographic information system (GIS) data of the area to be planned; constructing a three-dimensional digital twin basic scene through spatiotemporal registration and semantic alignment; identifying planning entities and multidimensional constraints to construct a topological hypergraph model; calculating topological centrality and constraint coupling weights to generate a non-uniform computational potential energy field; constructing a hierarchical scheduling strategy based on the potential energy field gradient; loading a LOD3 level fine model to extract topological parameters for high-potential-energy sensitive areas; loading a LOD1 level block model to extract filling data for low-potential-energy background areas; inputting a generative layout algorithm to generate intermediate data; detecting and repairing geometric overlaps; and outputting a planning scheme. This invention, by constructing a non-uniform computational potential energy field and a hierarchical scheduling strategy, achieves on-demand loading and differentiated processing of large-scale scene data, effectively solving the computational bottleneck problem of processing massive heterogeneous data.

Description

Technical Field

[0001] This invention relates to the field of architectural planning technology, specifically to a method and system for urban and rural planning based on digital twins. Background Technology

[0002] With the advancement of smart city and urban-rural integration, urban and rural planning has gradually shifted from traditional two-dimensional drawings and static data analysis to a full-cycle dynamic governance model based on digital twin technology. Through urban information models, it deeply integrates multi-source heterogeneous data such as building information models, geographic information systems, and the Internet of Things to achieve full-element digital expression and real-time interaction of urban and rural physical entities. In existing methods for digital modeling of urban and rural spaces, to balance rendering frame rate and geometric accuracy in large-scale scenes, multi-level detail algorithms based on geometric error metrics are typically used for static simplification and scheduling of the model. However, this preprocessing algorithm, which relies solely on the geometric perspective, has significant computational efficiency limitations when dealing with massive amounts of heterogeneous data from multiple sources. On the one hand, existing systems cannot automatically remove redundant geometric features such as high-frequency textures or non-fluid boundaries when performing computational fluid dynamics mesh generation or physics field solving, making it difficult to generate meshes in the computational domain. At the same time, when rendering macroscopic scenes, the attempt to concurrently process all high-precision geometric topology and discrete sensor data leads to GPU memory overflow and rendering pipeline blockage, failing to meet the computational power requirements for real-time interaction. On the other hand, the fusion of heterogeneous data from Building Information Modeling (BIM) and Geographic Information Systems (GIS) largely relies on static file format conversion and lacks the ability to dynamically reconstruct geometric topology based on parameterization. When geometric model parameters change, the cumbersome data cleaning and mesh generation process must be re-executed, resulting in delayed simulation results and failing to support an automated closed loop of parameterized modeling, physical simulation, and result feedback. In summary, how to overcome the limitations of traditional static preprocessing methods and achieve dynamic and precise scheduling of data and computing power according to the real-time needs of planning tasks is a technical challenge that urgently needs to be solved in the field of urban and rural planning technology based on digital twins.

[0003] To address this, a digital twin-based urban and rural planning method and system are proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for urban and rural planning based on digital twins, and to carry out urban and rural planning through digital twins.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A digital twin-based urban and rural planning method includes: Obtain building information model data and geographic information system data of the urban and rural areas to be planned, and construct a three-dimensional digital twin basic scene through spatiotemporal registration and data semantic alignment; The planning entities and multidimensional constraint relationships in the three-dimensional digital twin basic scene are identified and a topological hypergraph model is constructed. The topological centrality and constraint coupling weight of the topological hypergraph model are calculated. The three-dimensional digital twin basic scene is discretized into a planning space grid, and the topological centrality and constraint coupling weight are mapped to the planning space grid to generate a non-uniform computational potential field. A hierarchical scheduling strategy is constructed based on the gradient distribution of the non-uniform computation potential energy field. For high potential energy sensitive areas, a LOD3 level fine model is loaded to extract geometric features and generate topological parameters. For low potential energy background areas, a LOD1 level block model is loaded to extract boundary features and generate filling data. The data is then input into a generative layout algorithm to generate intermediate planning layout data. The system detects and automatically repairs geometric overlaps in the intermediate planning layout data, converts the repaired data into a solid model, and outputs it as an urban and rural planning layout scheme.

[0006] Preferably, the spatiotemporal registration involves extracting the outer contour corner points of the building information model and the building base corner points of the geographic information system data as pairs of corresponding feature points, calculating the coordinate transformation matrix based on the least squares method, and mapping the local coordinate system of the building information model data to the geographic coordinate system of the geographic information system data. The data semantic alignment involves determining the spatial inclusion relationship between the mapped building information model data and the geographic information system data, and associating and mapping the land parcel attribute information of the geographic information system data to the geometric objects corresponding to the building information model data.

[0007] Preferably, the construction of the three-dimensional digital twin basic scene is based on the bottom contour of the mapped building information model data, and the topographic triangulation of the geographic information system data is reconstructed by meshing, and redundant terrain meshes that overlap with the building base are removed. The building information model data is embedded into the reconstructed terrain grid, an octree-based spatial index structure is established, and the embedded building information model data and the terrain grid are assigned a globally unique identifier.

[0008] Preferably, the specific process of constructing the topological hypergraph model includes parsing the globally unique identifier in the three-dimensional digital twin basic scene and extracting individual buildings, road intersections and land parcels as hypergraph node sets; Based on a pre-defined set of spatial layout rules, spatial adjacency, road network accessibility, and public service coverage are defined as multi-dimensional constraint types. Through spatial buffer analysis and attribute association retrieval, multiple hypergraph nodes that satisfy the same dimension constraint are connected as the same hyperedge, thus constructing a hypergraph structure that associates nodes with hyperedges. Establish an association matrix describing the connection relationship between the hypergraph nodes and the hyperedges, and assign an initial weight value to each hyperedge according to the preset priority of each constraint type.

[0009] Preferably, the process of generating a non-uniform computational potential energy field is based on the correlation matrix, calculating the hyperdegree centrality of the hypergraph nodes as the topological centrality, and accumulating the weight values ​​of all hyperedges associated with each node to obtain the constraint coupling weight. A two-dimensional grid matrix is ​​constructed with a preset spatial resolution. A potential energy decay model centered on each hypergraph node is constructed using a Gaussian kernel function. The spatial radiation intensity of each hypergraph node to the grid cell is calculated. Traverse all grid cells, linearly superimpose the spatial radiation intensities received by all hypergraph nodes in that cell, and normalize the superposition results to generate a non-uniform computational potential field.

[0010] Preferably, the hierarchical scheduling strategy involves setting a gradient threshold for the non-uniform computational potential energy field and dividing the planning spatial grid into a high potential energy sensitive area and a low potential energy background area. For the high potential energy sensitive area, a LOD3 level fine model is dynamically loaded, the geometric vector information of building components and functional entrance and exit nodes are analyzed, and topological parameters containing precise spatial positioning and connection relationships are generated. For the low potential energy background area, a LOD1 level block model is loaded in a reduced dimension, the convex hull boundary of the building bottom is extracted, and filling data for space occupancy detection is generated. The topology parameters are used as strong constraints for the generative layout algorithm, and the filling data is used as boundary avoidance conditions. The driving force-oriented layout algorithm performs iterative calculations to generate intermediate planning layout data.

[0011] Preferably, the force-directed layout algorithm sets the geometric objects corresponding to the topology parameters as fixed anchor points and the geometric objects corresponding to the fill data as active nodes; Construct a position update rule based on spatial constraints, and calculate the non-overlapping repulsion vector between active nodes and other nodes, as well as the gradient pulling vector of active nodes pointing to high potential energy sensitive regions. The non-overlapping repulsion vector and the gradient traction vector are weighted and superimposed to obtain the comprehensive displacement vector of the active node, and the planar coordinates of the active node are iteratively updated based on the comprehensive displacement vector.

[0012] A digital twin-based urban and rural planning system includes: Scene construction module: Acquire building information model data and geographic information system data of the urban and rural areas to be planned, and construct a three-dimensional digital twin basic scene through spatiotemporal registration and data semantic alignment; Potential energy calculation module: Identifies the planning entities and multidimensional constraint relationships in the three-dimensional digital twin basic scene and constructs a topological hypergraph model, and calculates the topological centrality and constraint coupling weight of the topological hypergraph model; discretizes the three-dimensional digital twin basic scene into a planning space grid, maps the topological centrality and constraint coupling weight to the planning space grid, and generates a non-uniform computational potential energy field; Layout generation module: Based on the gradient distribution of the non-uniform computational potential energy field, a hierarchical scheduling strategy is constructed. For high potential energy sensitive areas, a LOD3 level fine model is loaded to extract geometric features and generate topological parameters. For low potential energy background areas, a LOD1 level block model is loaded to extract boundary features and generate filling data. The data is then input into the generative layout algorithm to generate intermediate planning layout data. Reverse correction module: detects geometric overlaps in the intermediate planning layout data and automatically repairs them, converts the repaired data into a solid model and outputs it as an urban and rural planning layout scheme.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention effectively solves the computational bottleneck problem in rendering and computing large-scale urban and rural scenes by constructing a non-uniform computational potential energy field and a hierarchical scheduling strategy. Traditional methods often rely solely on static simplification from a geometric perspective, which can easily lead to memory overflow or rendering blockage. This invention generates a non-uniform potential energy field that reflects the importance of a region by calculating topological centrality and constraint coupling weights. The system allocates computing power on demand accordingly, loading LOD3 level fine models to high potential energy sensitive areas to obtain accurate topological parameters, and loading only LOD1 level block models for low potential energy background areas for placeholder detection. This dynamic and precise scheduling mechanism significantly reduces the computational load of non-critical areas and achieves efficient processing and real-time interaction of massive multi-source heterogeneous data while ensuring the accuracy of core planning elements.

[0014] 2. To address the challenges of integrating BIM and GIS data and generating grids, this invention utilizes the least squares method to calculate the coordinate transformation matrix for high-precision spatiotemporal registration and achieves semantic alignment through spatial inclusion relationship determination. Simultaneously, the system can reconstruct the terrain triangulation network based on the building's base outline and automatically eliminate redundant grids overlapping with the building's base. This process not only eliminates geometric interpenetration and dangling errors between models but also provides a high-quality, clean base for subsequent computational fluid dynamics mesh generation or physics field solution, thus paving the way for a seamless technical path from modeling to simulation.

[0015] 3. This invention establishes a generative planning mechanism based on a topological hypergraph and a force-directed algorithm, significantly improving the automation level and decision-making response speed of urban and rural planning. Unlike traditional static drawing planning, this invention can identify multi-dimensional constraint relationships between planning entities and construct a topological hypergraph model. By transforming planning constraints into repulsion and traction vectors in the force-directed algorithm, it can automatically drive layout iteration calculations and has the ability to automatically repair geometric overlaps. This parameter-driven dynamic reconstruction capability enables planning schemes to be generated and adjusted in real time as parameters change, supporting an automated closed loop of parameter modeling-physical simulation-result feedback, and meeting the needs of dynamic governance throughout the entire urban and rural lifecycle. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a digital twin-based urban and rural planning method. Figure 2 This is a flowchart illustrating a hierarchical scheduling strategy based on a non-uniform computational potential field. Figure 3 This is a schematic diagram of the structure of an urban and rural planning system based on digital twins. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1: Please see Figure 1 This invention provides a digital twin-based urban and rural planning method, the technical solution of which is as follows: A digital twin-based urban and rural planning method includes: Obtain building information model data and geographic information system data of the urban and rural areas to be planned, and construct a three-dimensional digital twin basic scene through spatiotemporal registration and data semantic alignment; The planning entities and multidimensional constraint relationships in the three-dimensional digital twin basic scene are identified and a topological hypergraph model is constructed. The topological centrality and constraint coupling weight of the topological hypergraph model are calculated. The three-dimensional digital twin basic scene is discretized into a planning space grid, and the topological centrality and constraint coupling weight are mapped to the planning space grid to generate a non-uniform computational potential field. A hierarchical scheduling strategy is constructed based on the gradient distribution of the non-uniform computation potential energy field. For high potential energy sensitive areas, a LOD3 level fine model is loaded to extract geometric features and generate topological parameters. For low potential energy background areas, a LOD1 level block model is loaded to extract boundary features and generate filling data. The data is then input into a generative layout algorithm to generate intermediate planning layout data. The system detects and automatically repairs geometric overlaps in the intermediate planning layout data, converts the repaired data into a solid model, and outputs it as an urban and rural planning layout scheme.

[0019] The building information model (BIM) data targets existing buildings within the region. High-density point cloud data of building surfaces is acquired using a ground-based 3D laser scanner. A 3D geometric model is constructed based on reverse engineering technology, and building material textures and functional attribute information are added according to the LOD3 accuracy standard. For new buildings in the planning and design phase, digital design models generated by architectural design software are directly imported. The existing and new building models are uniformly exported to industrial basic standard formats or general 3D data exchange formats. The BIM parsing engine is used to extract the geometric vertex coordinates, normal vectors, and metadata including the number of floors, functional zoning, and floor area ratio of each building component as micro-level planning entity inputs. The geographic information system data is obtained by using a multi-rotor UAV equipped with a five-lens oblique photography camera to perform low-altitude aerial surveys of the target urban and rural areas. The forward overlap rate is set to 80% and the lateral overlap rate to collect raw image data from multiple perspectives. The collected images are then processed by aerial triangulation and texture mapping using real-scene modeling software (ContextCapture is used in this embodiment) to generate a real-scene 3D mesh model, a digital elevation model, and a digital orthophoto containing high-precision texture information. At the same time, the corresponding vector geographic information data for the urban and rural areas is retrieved from the urban surveying and mapping database, including road network vector line maps, land parcel boundaries, water system boundaries, and land use attribute tables, thereby constructing a macro-level urban and rural geographic environment base. The spatiotemporal registration involves extracting the outer contour corner points of the building information model and the building base corner points of the geographic information system data as pairs of corresponding feature points, and calculating the coordinate transformation matrix based on the least squares method to map the local coordinate system of the building information model data to the geographic coordinate system of the geographic information system data. The data semantic alignment involves determining the spatial inclusion relationship between the mapped building information model data and the geographic information system data, and associating and mapping the land parcel attribute information of the geographic information system data to the geometric objects corresponding to the building information model data.

[0020] Specifically, the geometric vertex information of the Building Information Model (BIM) data is read and orthographically projected onto a horizontal projection plane. The inflection points on the projected contour line are extracted as the outer contour corner points of the BIM model, and their three-dimensional coordinate values ​​in the local coordinate system are recorded. Simultaneously, the building vector layer in the Geographic Information System (GIS) data is read, the geometric nodes of the building base polygon are parsed, and the key nodes corresponding to the geometric shape of the outer contour corner points are extracted as the building base corner points of the GIS data. Through manual interaction or automatic feature matching algorithms, the outer contour corner points are matched one-to-one with the building base corner points to construct a set of no less than three pairs of identically named feature points. The least squares method for solving the coordinate transformation matrix involves constructing a coordinate transformation model that includes rotation, translation, and scaling parameters. Using the set of corresponding feature point pairs as input data, an error equation is established. Utilizing the least squares adjustment principle, with the constraint of minimizing the sum of squared coordinate residuals of all corresponding feature point pairs after transformation, the optimal rotation matrix, translation vector, and scaling factor are iteratively calculated and combined to generate a 4×4 coordinate transformation matrix. All geometric vertices in the building information model are traversed, and their local coordinates are multiplied by the coordinate transformation matrix to complete the rigid body transformation from the local coordinate system to the geodetic coordinate system of the geographic information system. The spatiotemporal registration involves extracting the convex hull vertices of the horizontal projection of the Building Information Model (BIM) and the geometric inflection points of the building base polygons in the Geographic Information System (GIS) data. A set of corresponding feature point pairs is constructed using a geometric similarity matching algorithm. Based on the least squares principle, a coordinate transformation model including rotation, translation, and uniform scaling parameters is iteratively calculated to minimize the Euclidean distance error of all feature point pairs after transformation. If the registration residual of any feature point exceeds a preset tolerance threshold (0.1 meters in this embodiment), the point will be marked as an outlier and removed. Subsequently, the parameter calculation is re-executed. The spatial inclusion relationship determination process in the data semantic alignment adopts the ray method or polygon inclusion algorithm in computational geometry. After completing the spatiotemporal registration, the geometric center point or bottom projection polygon of the building information model in the geographic coordinate system is extracted. The land parcels in the geographic information system data are traversed, and it is automatically determined whether the geometric center point of the building information model is located inside the polygon boundary of a certain land parcel. If the determination result is true, it is confirmed that there is a spatial inclusion relationship between the building information model and the land parcel data in the geographic information system. The spatial inclusion relationship is automatically determined by the computational geometry algorithm, and the matching of building entities and land parcel attributes can be achieved without manual intervention. The macro planning indicators are automatically mapped to the micro building model, realizing the deep binding of geometric information and semantic attributes. The process involves associating and mapping the parcel attribute information from the geographic information system (GIS) data. After confirming the spatial inclusion relationship, the attribute table of the parcel in the GIS database is read, and text attribute fields, including land use, floor area ratio limit, building height limit, green space ratio, and parcel ownership, are extracted. A key-value mapping table is constructed in memory, using the aforementioned attribute fields as values ​​and the globally unique identifier of the building information model as the key for binding. The attribute information is then written into the extended attribute set file of the building information model, ensuring that each 3D building model entity carries planning constraint attributes from the GIS. By employing least squares registration technology, high-precision coordinate unification between the building micro-model and the geographic macro-environment was achieved, effectively eliminating geometric deviations in the fusion of heterogeneous data. Simultaneously, based on a semantic mapping mechanism of spatial inclusion relationships, the planning and control indicators at the plot level were automatically assigned to the three-dimensional building entity, realizing a deep fusion of geometric topology and semantic attributes.

[0021] The construction of the three-dimensional digital twin basic scene is based on the bottom contour of the mapped building information model data, and the terrain triangulation of the geographic information system data is reconstructed by meshing, and redundant terrain meshes that overlap with the building base are removed. The building information model data is embedded into the reconstructed terrain grid, an octree-based spatial index structure is established, and the embedded building information model data and the terrain grid are assigned a globally unique identifier.

[0022] The construction of the 3D digital twin base scene resolves the geometrical intersection and conflict between the building model base and the terrain surface through adaptive reconstruction of the terrain mesh. The bottom contour line of the spatiotemporally registered building information model data is read; this contour line consists of a series of closed, ordered coordinate points. The bottom contour line is vertically projected onto the plane containing the terrain triangulation network of the geographic information system data, serving as the constraint boundary for terrain clipping. Each triangular facet in the terrain triangulation network is traversed, and based on the point-plane position relationship algorithm in computational geometry, the triangular faces are divided into three categories: internal faces completely within the constraint boundary, external faces completely outside the constraint boundary, and boundary-crossing faces intersecting with the constraint boundary. All triangular meshes marked as internal faces are deleted, i.e., redundant terrain meshes overlapping with the building base are removed. For boundary-crossing faces, the intersection points of the constraint boundary line segments and the edge lines of the triangular facet are calculated. These intersection points are used to divide the boundary-crossing facet into several smaller sub-meshes, retaining the sub-meshes located outside the constraint boundary. This creates precise holes on the terrain triangulation network that perfectly fit the building's bottom contour, completing the terrain mesh reconstruction. The process of embedding the building information model data into the reconstructed terrain mesh involves seamlessly integrating the building entities with the terrain environment at the data structure level. The building information model data is placed into the precise hole locations generated in the preceding steps. By capturing the coordinates of the building's bottom contour vertices and the terrain hole boundary vertices, vertex welding or coincidence point capture operations are performed to eliminate geometric gaps between the building and the terrain, ensuring the continuity of the scene's geometric topology and forming a manifold or near-manifold overall 3D scene mesh. During this process, the elevation of the building model's bottom is fine-tuned to ensure it matches the average elevation of the terrain hole edges, preventing the model from appearing suspended or unexpectedly deeply buried, thus completing the embedding of the physical model into the geographical environment. The establishment of an octree-based spatial index structure involves calculating the minimum bounding box of the merged 3D digital twin scene, determining the maximum extent of the scene in 3D space, and using the minimum bounding box as the root node of the octree. A spatial partitioning threshold is set, which includes the maximum number of geometric primitives contained in a node. In this embodiment, the maximum number of geometric primitives contained in a single node is set to 5000 triangular faces. The spatial cube represented by the root node is uniformly divided into eight sub-cubes according to recursive logic, and it is determined whether the number of geometric primitives in each sub-cube exceeds the set threshold. If it does, the sub-cube is further partitioned at the next level until the number of primitives in all leaf nodes meets the threshold requirement. In this way, a hierarchical octree index is constructed, distributing massive amounts of scene data discretely across different leaf nodes. By constructing an octree spatial index based on geometric primitive density, massive amounts of 3D scene data are discretely distributed across hierarchical leaf nodes, achieving efficient organization and rapid retrieval of large-scale complex scene data.

[0023] By using a hash algorithm combined with the current timestamp and machine signature, a unique string or long integer value is generated for each building unit model, each independent terrain mesh block, and related ancillary facility object in the scene, serving as a globally unique identifier. A key-value mapping table is established in memory, using the globally unique identifier as the primary key and binding the memory address, attribute data pointer, and octree node index of the corresponding 3D geometric object as the value. When the system needs to query or highlight a planning object, it can quickly locate the corresponding geometric primitive and attribute information simply by using this globally unique identifier.

[0024] By using terrain mesh reconstruction and seamless embedding technology, redundant meshes are eliminated and geometric gaps are removed, constructing a topologically continuous manifold scene that provides a high-quality computing foundation for physical simulation. At the same time, by combining octree indexes and globally unique identifiers, the discretization management of massive amounts of data is realized.

[0025] The specific process of constructing the topological hypergraph model includes parsing the globally unique identifier in the three-dimensional digital twin basic scene and extracting individual buildings, road intersections, and land parcel units into a set of hypergraph nodes; Based on a pre-defined set of spatial layout rules, spatial adjacency, road network accessibility, and public service coverage are defined as multi-dimensional constraint types. Through spatial buffer analysis and attribute association retrieval, multiple hypergraph nodes that satisfy the same dimension constraint are connected as the same hyperedge, thus constructing a hypergraph structure that associates nodes with hyperedges. Establish an association matrix describing the connection relationship between the hypergraph nodes and the hyperedges, and assign an initial weight value to each hyperedge according to the preset priority of each constraint type.

[0026] The data structure of the basic 3D digital twin scene is traversed to read the globally unique identifiers of all objects in the scene and their associated attribute information. Based on the category field in the attributes, entities with planning significance are selected, including various building units (such as residential and commercial buildings), road intersection nodes, and land parcel units (such as independent land parcels within the planning red line). Each selected entity is treated as an independent vertex and aggregated to form a set of nodes in the hypergraph. During this process, a consecutive integer index starting from zero is allocated in memory for each node for row positioning in subsequent matrix operations. At the same time, the mapping relationship between the node and the globally unique identifier of the original 3D model is preserved to ensure that the calculation results can be traced back to the 3D scene. To standardize the processing of planning elements, a standardized attribute table structure was constructed for the extracted hypermap nodes. Specifically, building unit nodes are strictly defined as single, closed building outlines with independent physical entrances and exits; road intersection nodes are identified as the convergence point of three or more road segments in the road network and their adjacent radiating areas; and land parcel unit nodes are directly mapped from cadastral patch data in the geographic information system. The attribute table explicitly records the node type identifier, two-dimensional plane centroid coordinates, and functional attribute set. This structured data will serve as the basic input for subsequent potential energy calculations. The construction of the hyperedge structure based on the preset spatial layout rule set is a step in using hypergraph theory to handle complex urban-rural relationships. The system defines three types of multidimensional constraints in the preset rule base: the first type is spatial adjacency constraints, which perform geometric buffer analysis on each plot unit and building unit, setting a buffer distance of 5 meters. If the buffer areas overlap, these nodes are determined to meet the spatial adjacency condition, and these mutually adjacent nodes are combined to create a hyperedge describing the neighborhood relationship; the second type is road network accessibility constraints, which, based on road network topology data, use road intersections as the search center and perform a breadth-first search along the road network to find... The system identifies all buildings reachable within a 500-meter path from the intersection and connects the intersection to all building nodes within its radius to create a hyperedge describing the traffic service domain. The third type of constraint is public service coverage. The system identifies public service facility nodes such as schools and hospitals, and performs spatial inclusion retrieval using their service radius (1000 meters in this example) as a threshold. All residential building nodes falling within this radius are bound to the public service facility node, creating a hyperedge describing the shared relationship of public resources. Through the above process, a hypergraph structure with many-to-many relationships between nodes and hyperedges is constructed. The process of establishing an association matrix and assigning initial weight values ​​transforms the topological structure into a computable algebraic model. A two-dimensional array is constructed as the association matrix, with the number of rows corresponding to the total number of hypergraph nodes and the number of columns corresponding to the total number of generated hyperedges. The node list contained in each hyperedge is traversed. If a node belongs to a certain hyperedge, the value at the intersection of the corresponding row and column in the association matrix is ​​set to one; otherwise, it is set to zero. Based on the rigid and flexible requirements of urban and rural planning, preset priorities for each constraint type are established. For example, spatial adjacency constraints involving basic farmland or ecological red lines are set to the highest priority, public service coverage is set to the medium priority, and general road network accessibility is set to the low priority. The preset weight table is searched according to the priority. In this embodiment, the highest, medium, and low priorities are assigned values ​​of 1.0, 0.8, and 0.5, respectively, and directly assigned to the corresponding hyperedge objects as the initial weight basis for subsequent calculation of the potential energy field.

[0027] By constructing a topological hypergraph model, the limitations of traditional binary graph structures are overcome, enabling the accurate expression of complex multi-dimensional, many-to-many constraint relationships in urban and rural planning, such as spatial adjacency, transportation accessibility, and service coverage. At the same time, by establishing an association matrix and assigning weights according to constraint priority, qualitative planning rules are transformed into a computable quantitative model, providing a rigorous mathematical foundation for the subsequent generation of non-uniform potential energy fields. This ensures that the planning scheme achieves the scientific layout of functional elements while meeting rigid control requirements. refer to Figure 2The process of generating a non-uniform computational potential energy field is based on the correlation matrix, calculating the superdegree centrality of the hypergraph nodes as the topological centrality, and accumulating the weight values ​​of all hyperedges associated with each node to obtain the constraint coupling weight. A two-dimensional grid matrix is ​​constructed with a preset spatial resolution. A potential energy decay model centered on each hypergraph node is constructed using a Gaussian kernel function. The spatial radiation intensity of each hypergraph node to the grid cell is calculated. Traverse all grid cells, linearly superimpose the spatial radiation intensities received by all hypergraph nodes in that cell, and normalize the superposition results to generate a non-uniform computational potential field.

[0028] The association matrix stored in memory is read, and the row vectors representing each hypergraph node are traversed row by row. The number of non-zero elements in each row is counted, and this count is defined as the hyperdegree centrality, representing the breadth of the node's participation in spatial constraint association. At the same time, the hyperedge column corresponding to each non-zero element in the row is indexed, and the initial weight value is read. The weight values ​​of all hyperedges associated with the node are summed to obtain the constraint coupling weight of the node, which is used to represent the sum of the planning constraint strengths of the node, thereby completing the quantification and assignment of node attributes. The process of constructing a two-dimensional grid matrix with a preset spatial resolution is to discretize the continuous urban and rural space into computing units; obtain the maximum boundary range of the three-dimensional digital twin basic scene projected on the horizontal plane, and set the spatial resolution parameters (in this embodiment, the side length of each grid unit is set to ten meters); based on the maximum boundary range and resolution parameters, divide the two-dimensional grid into several rows and several columns, initialize a two-dimensional floating-point array as the grid matrix, and set the initial values ​​of all grid units to zero; at the same time, calculate the planar coordinates of the geometric center point of each grid unit in the planned coordinate system; The method of constructing a potential energy attenuation model using a Gaussian kernel function and calculating spatial radiation intensity is to simulate the physical process of the diffusion of the influence of planning constraints with distance. Each hypergraph node is traversed, and its planar coordinates are obtained as potential energy points. The bandwidth parameter of the Gaussian kernel function is set to determine the effective radius of the potential energy influence (in this embodiment, it is set to 500 meters). For each hypergraph node, the Euclidean distance between it and the center point of each grid cell in the two-dimensional grid matrix is ​​calculated. Using the mathematical logic of the Gaussian function, the negative square of the distance is divided by twice the square of the bandwidth parameter, and the natural exponent is calculated to obtain an attenuation coefficient between zero and one. The constraint coupling weight of the node is multiplied by this attenuation coefficient to obtain the spatial radiation intensity of the hypergraph node on a specific grid cell. The process of traversing all grid cells to generate a non-uniform computational potential energy field is a process of superimposing and standardizing fields based on multiple constraints. For each grid cell, the spatial radiation intensity values ​​generated by all hypergraph nodes for that cell are linearly accumulated to obtain the absolute potential energy value of that cell. After completing the calculation of the entire grid, the maximum and minimum potential energy values ​​in the entire grid matrix are counted. The minimum and maximum value standardization method is used to map the values ​​of all grid cells to a closed interval between zero and one. The normalized two-dimensional matrix generated is the non-uniform computational potential energy field. The region with values ​​close to one represents a high potential energy sensitive region with high constraint coupling, and the region with values ​​close to zero represents a low potential energy background region with low constraint coupling. When generating a non-uniform computational potential energy field, the bandwidth parameter of the Gaussian kernel function is physically matched according to the scale of the pedestrian living circle in urban planning (in this embodiment, the bandwidth is set so that the influence decays to an extremely low level at a walking distance of 500 meters). For each cell in the planning grid, all hypergraph nodes are traversed, and the spatial radiation intensity of each node to the cell is calculated. The spatial radiation intensity is proportional to the node weight and decays exponentially with the square of the distance. The gradient calculation adopts the central difference method, calculating the potential energy change rate of each grid point in the horizontal and vertical directions respectively, and synthesizing the gradient vector indicating the direction of the fastest potential energy growth.

[0029] A non-uniform computational potential field construction method based on Gaussian kernel function is adopted. By calculating the supercentrality and constraint coupling weight, the abstract topological connection relationship is transformed into a continuous spatial potential energy distribution. The spatial radiation effect of planning elements is simulated using a Gaussian decay model. Through linear superposition and normalization processing, high potential energy key areas with dense constraints and significant impacts are accurately identified in urban and rural areas, providing scientific and quantitative spatial guidance for subsequent hierarchical computing power scheduling. The hierarchical scheduling strategy involves setting a gradient threshold for the non-uniform computation potential energy field and dividing the planning spatial grid into a high potential energy sensitive area and a low potential energy background area. For the high potential energy sensitive area, a LOD3 level fine model is dynamically loaded, the geometric vector information of building components and functional entrance and exit nodes are analyzed, and topological parameters containing spatial positioning and connection relationships are generated. For the low potential energy background area, a LOD1 level block model is loaded in a reduced dimension, the convex hull boundary of the building bottom is extracted, and filling data for space occupancy detection is generated. The topology parameters are used as strong constraints for the generative layout algorithm, and the filling data is used as boundary avoidance conditions. The driving force-oriented layout algorithm performs iterative calculations to generate intermediate planning layout data.

[0030] The construction process of the hierarchical scheduling strategy involves reading the normalized non-uniform computation potential energy field matrix and setting a value between zero and one as the gradient threshold (0.7 in this embodiment); traversing each grid cell in the matrix, grid cells with potential energy values ​​greater than or equal to 0.7 are marked as high potential energy sensitive areas, which typically correspond to core areas with dense planning constraints and requiring refined design; grid cells with potential energy values ​​less than 0.7 are marked as low potential energy background areas, which typically represent the surrounding environment or non-major interference areas with less planning influence; through this binarization or multi-level threshold division, the continuous potential energy field is mapped into a discrete computational scheduling mask. The process for handling high-potential-sensitive areas involves calling the model database interface to dynamically load a detailed LOD3 model for each planned entity within the area; parsing the geometric data structure of the detailed model to extract the coordinates of key corner points of the building facade as geometric vector information; simultaneously identifying nodes in the model marked with semantic tags such as main entrance and fire escape routes, and extracting their three-dimensional spatial coordinates as functional entrance / exit nodes; encapsulating the above vector information and functional nodes to construct topological parameters containing precise spatial positioning and connection relationships. The LOD3 level detailed model is a high-precision 3D model containing geometric details of the building's exterior and main components. Its specific structure consists of a high-density triangular mesh, high-resolution texture maps, and semantic attribute nodes. In terms of parameter settings, the geometric positional accuracy of the model is controlled within 0.05 meters, that is, the spatial deviation between the model vertices and the real physical objects does not exceed 5 cm. The model retains all building facade components with dimensions greater than 0.5 m, including door and window openings, balcony railings, roof eaves, and air conditioning unit locations. At the same time, a semantic hierarchy tree is embedded in the model's data structure, dividing the building into three logical layers: base, walls, and roof. Each door and window component is assigned an independent sub-object index and normal vector to accurately identify the entrance location and view occlusion relationship in the planning calculation. In the process of processing the low potential energy background area, in order to reduce the computational load of non-core areas, only LOD1 block models are loaded for entities in this area, which only contain simplified geometric shapes and do not contain textures or internal details; the bottom vertex data of the block model is read and projected onto a two-dimensional plane; the convex hull algorithm in computational geometry (Graham scan method is used in this embodiment) is used to calculate the minimum convex polygon boundary that surrounds all projected vertices, i.e., the convex hull boundary; the convex hull boundary is defined as an impassable rigid body region, and filling data for spatial occupancy detection is generated, thereby expressing the obstacle features of the environment with the minimum amount of data; The LOD1 level block model is a conceptual volume model that only expresses the macroscopic spatial occupancy of the building; its specific structure is a closed convex polyhedron generated by stretching the building base outline vertically; in terms of parameter settings, the construction process ignores all geometric concave and convex changes with horizontal projection dimensions less than 10m, and only retains the main structural shape of the building; its top height parameter is taken from the average elevation of the roof plane in the original building information model, and the bottom parameter is taken from the maximum projected polygon at the plinth of the building's exterior wall; the data volume of the model is strictly limited, the number of triangular facets of a single building block is controlled within 50, and it does not contain any material texture information or internal structure, and is specifically used for rigid body bounding box calculation in the fast collision detection algorithm in low potential energy background areas; The process of generating intermediate planning layout data by the driving force-oriented layout algorithm involves converting the above parameters into a physical and mechanical model for solution; initializing the force-oriented layout solver by setting the topological parameters from the high potential energy sensitive area as strong constraints in the algorithm, i.e., as anchor points with fixed positions or strong attraction, to ensure the positional accuracy of the core planning elements; setting the filling data from the low potential energy background area as avoidance boundary conditions, i.e., defining a repulsive force region in the solution space; and initializing the random or preset positions of the objects to be planned based on these conditions. By adopting a hierarchical scheduling strategy based on potential energy gradient, on-demand loading and differentiated processing of large-scale scene data are achieved. While preserving the fine topological features of high potential energy sensitive areas, the geometric complexity of low potential energy background areas is significantly reduced, effectively solving the memory overflow and computational lag problems caused by full loading of high-precision models.

[0031] The force-directed layout algorithm sets the geometric objects corresponding to the topology parameters as active nodes and the geometric objects corresponding to the fill data as active nodes. Construct a position update rule based on spatial constraints, and calculate the non-overlapping repulsion vector between active nodes and other nodes, as well as the gradient pulling vector of active nodes pointing to high potential energy sensitive regions. The non-overlapping repulsion vector and the gradient traction vector are weighted and superimposed to obtain the comprehensive displacement vector of the active node, and the planar coordinates of the active node are iteratively updated based on the comprehensive displacement vector.

[0032] The force-oriented layout algorithm establishes a two-dimensional physical simulation plane with dimensions consistent with the urban and rural planning area. It reads the generated topological parameters, maps the corresponding fine model geometric center coordinates onto the simulation plane, and sets these objects as active nodes, assigning them unit virtual mass, allowing them to freely change their planar coordinate positions within the simulation plane according to the force conditions. Simultaneously, it reads the filling data, maps the corresponding block model geometric center coordinates onto the simulation plane, and similarly sets these objects as active nodes, assigning them unit virtual mass, allowing them to move freely. By constructing the physical simulation plane, all planning elements are treated as elastic nodes with dynamic adjustment capabilities, undergoing global self-organizing evolution and layout optimization under the guidance of a non-uniform potential energy field.

[0033] The non-overlapping repulsion vector calculation in the spatial constraint-based position update rule is performed by traversing each active node in each iteration and calculating the Euclidean distance between that node and **all other active nodes in the scene (including topology parameter nodes and filling data nodes)**. A safety distance threshold is set (10 meters in this embodiment). If the calculated distance is less than the safety distance threshold, it is determined that there is a potential risk of overlap between the two. At this time, a non-overlapping repulsion vector is generated. The direction of the vector is from the adjacent node to the current active node. Its magnitude (i.e., the strength) is inversely proportional to the square of the distance between the two. That is, the closer the distance, the more exponentially the repulsion force increases, thereby simulating the physical collision repulsion effect between objects and forcibly pushing the active node away from the crowded area. The gradient traction vector pointing the active node to the high potential energy sensitive area is obtained by acquiring the potential energy value corresponding to the current coordinate of the active node in the non-uniform computational potential energy field; the rate of change of potential energy at the coordinate point in the four directions of east, south, west, and north is detected by the difference calculation method, and the direction of the fastest potential energy growth, i.e., the gradient direction, is identified; a gradient traction vector is generated, pointing to the area with the highest local potential energy, and its magnitude is proportional to the potential energy gradient value at that location; the vector acts like gravity, pulling all active nodes to move towards the high potential energy sensitive area with well-developed infrastructure and high constraint coupling. The process of weighted superposition and coordinate update of non-overlapping repulsion vectors and gradient traction vectors involves assigning weight coefficients to the two vectors. In this embodiment, the weight of the repulsion vector is set to 0.6 to prioritize spatial safety, and the weight of the traction vector is set to 0.4 to balance layout efficiency. The two weighted vectors are then vector synthesized to obtain the comprehensive displacement vector acting on the active node. A simulation time step parameter (0.1s in this embodiment) is introduced, and the comprehensive displacement vector is multiplied by this time step to calculate the displacement increment of the node in the current iteration step. This displacement increment is accumulated onto the original planar coordinates of the active node to complete one position update. The above calculation process is repeated until the sum of the magnitudes of the comprehensive displacement vectors of all active nodes is less than a preset convergence threshold (0.01m in this embodiment). At this point, the output coordinates are the optimized intermediate state planning layout data. The force-oriented algorithm, which uses physical field simulation, transforms rigid specifications and functional requirements into repulsion and attraction vectors. It uses a dynamic balance mechanism to drive the automatic optimization of the placement of planning elements, achieving a conflict-free layout under complex conditions and maximizing proximity to high-potential-sensitive areas while ensuring safe spacing.

[0034] The system detects and automatically repairs geometric overlaps in the intermediate planning layout data, converts the repaired data into a solid model, and outputs it as an urban and rural planning layout scheme. The detection and automatic repair of geometric overlap involves traversing the planar contours of all planned objects and using a sweep line algorithm to quickly filter potential collision pairs. For each collision pair, the polygon intersection is calculated. If the intersection area is greater than zero, the minimum separation vector is calculated. This vector is oriented along the collision normal direction and its magnitude is slightly greater than the maximum penetration depth. This separation vector is then directly superimposed onto the object's coordinates for forced displacement. This detection and displacement process is repeated until the overlap detection result of all objects in the scene is zero. The process involves converting the model into a solid model and outputting the final coordinate points and rotation angle parameters after repair, calling the system model library; based on the object's globally unique identifier, using geometric instantiation technology, placing the corresponding LOD3 level fine model or LOD1 level block model in the corresponding position in the 3D scene, and restoring its attribute data; serializing and storing the reconstructed conflict-free 3D scene, generating a standard 3D model file containing complete geometric topology and attribute information, which is then output as an urban and rural planning layout scheme.

[0035] This invention effectively solves the computational bottleneck problem of processing massive heterogeneous data in large-scale urban and rural planning scenarios by constructing a non-uniform computational potential energy field combined with a hierarchical scheduling strategy. It uses a topological hypergraph model to identify planning entities and multidimensional constraints, calculates topological centrality and constraint coupling weights to quantify spatial importance, and then drives the dynamic differentiated loading of LOD3 level fine model and LOD1 level block model. This on-demand computing mechanism ensures the geometric and topological accuracy of the core high potential energy sensitive area while significantly reducing the resource occupation of the low potential energy background area, avoiding memory overflow and computational lag. Combined with generative layout and automatic repair algorithms, it significantly improves the level of planning automation and scheme iteration efficiency in complex urban and rural environments.

[0036] Example 2: This embodiment applies the digital twin-based urban and rural planning system proposed in this invention to an urban and rural expansion project. The scenario is characterized by high building density, complex underground pipe network, and strict height and sunlight requirements for cultural heritage buildings in the surrounding area. See Figure 3 In the renovation and expansion project, the scene construction module first retrieves the existing high-precision oblique photogrammetry real-scene model of the core area and the digital design scheme of the proposed commercial complex. The system extracts the base corner points of the existing cultural heritage buildings and the outer contour corner points of the digital design scheme model for spatiotemporal registration. The least squares method is used to solve the coordinate transformation matrix to ensure that the new and old buildings are accurately matched in the same geographic coordinate system. The module automatically performs semantic alignment between the mapped digital design scheme model data and the plot attributes of the oblique photogrammetry real-scene model. The plot's maximum floor area ratio, green space ratio and land use nature are written into the building model. Redundant grids that overlap with the base of the new building in the terrain of the oblique photogrammetry real-scene model are removed to construct a geometrically continuous three-dimensional digital twin basic scene. The potential energy calculation module identifies key planning entities (such as proposed towers, subway entrances / exits, and surrounding cultural heritage buildings) and multidimensional constraint relationships in the basic scene. The system constructs a topological hypergraph model, defining "500-meter walking coverage of the subway station," "visual corridors of cultural heritage buildings," and "sunlight shading red line" as multidimensional constraint types. It calculates the topological centrality and constraint coupling weights of each node, discretizes the core area into a 10m×10m planning space grid, and uses a Gaussian kernel function to map the above weights onto the grid, generating a non-uniform potential energy field. In this field, the area adjacent to the subway station and subject to strict height restrictions for cultural heritage buildings is calculated as an extremely high potential energy sensitive area, while the surrounding ordinary streets are presented as a low potential energy background area. The layout generation module executes a hierarchical scheduling strategy based on the gradient distribution of the potential energy field. For core plots marked as high-potential-energy-sensitive areas, the system dynamically loads a LOD3 level refined model to accurately analyze the functional entrances and exits of the proposed commercial building and the fine geometric features of the surrounding cultural heritage buildings, generating topological parameters containing precise connection relationships. For low-potential-energy background areas (such as surrounding ordinary residential areas), the system only loads a LOD1 level block model and extracts its convex hull boundary to generate filling data. The module uses these topological parameters as strong constraints and the filling data as avoidance boundaries, inputting them into a generative layout algorithm. Using a force-oriented mechanism, the algorithm drives the new building volume to automatically find the optimal location while meeting sunlight and traffic constraints, generating intermediate planning layout data. The reverse correction module performs geometric conflict detection on the generated intermediate layout data. The system traverses all planning objects, and if it detects that the annex of the new building has geometric overlap with the surrounding existing buildings, or infringes on the protection scope of the cultural heritage building, the module will automatically calculate the minimum separation vector for forced displacement repair. After the repair is completed, the system uses geometric instantiation technology to convert the optimized coordinate parameters into a solid model with complete texture and attributes, and finally outputs a commercial complex renovation and expansion planning layout scheme that meets various complex control requirements.

[0037] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A digital-twin-based urban and rural planning method, characterized in that, include: Obtain building information model data and geographic information system data of the urban and rural areas to be planned, and construct a three-dimensional digital twin basic scene through spatiotemporal registration and data semantic alignment; Identify the planning entities and multidimensional constraint relationships in the three-dimensional digital twin basic scene and construct a topological hypergraph model, and calculate the topological centrality and constraint coupling weight of the topological hypergraph model; The three-dimensional digital twin basic scene is discretized into a planning space grid, and the topological centrality and constraint coupling weight are mapped to the planning space grid to generate a non-uniform computational potential energy field. A hierarchical scheduling strategy is constructed based on the gradient distribution of the non-uniform computation potential energy field. For high potential energy sensitive areas, a LOD3 level fine model is loaded to extract geometric features and generate topological parameters. For low potential energy background areas, a LOD1 level block model is loaded to extract boundary features and generate filling data. The data is then input into a generative layout algorithm to generate intermediate planning layout data. The system detects and automatically repairs geometric overlaps in the intermediate planning layout data, converts the repaired data into a solid model, and outputs it as an urban and rural planning layout scheme.

2. The urban and rural planning method based on digital twins according to claim 1, characterized in that, The spatiotemporal registration involves extracting the outer contour corner points of the building information model and the building base corner points of the geographic information system data as pairs of corresponding feature points, and calculating the coordinate transformation matrix based on the least squares method to map the local coordinate system of the building information model data to the geographic coordinate system of the geographic information system data. The data semantic alignment involves determining the spatial inclusion relationship between the mapped building information model data and the geographic information system data, and associating and mapping the land parcel attribute information of the geographic information system data to the geometric objects corresponding to the building information model data.

3. The urban and rural planning method based on digital twins according to claim 1, characterized in that, The construction of the three-dimensional digital twin basic scene is based on the bottom contour of the mapped building information model data, and the terrain triangulation of the geographic information system data is reconstructed by meshing, and redundant terrain meshes that overlap with the building base are removed. The building information model data is embedded into the reconstructed terrain grid, an octree-based spatial index structure is established, and the embedded building information model data and the terrain grid are assigned a globally unique identifier.

4. The urban and rural planning method based on digital twins according to claim 1, characterized in that, The specific process of constructing the topological hypergraph model includes parsing the globally unique identifier in the three-dimensional digital twin basic scene and extracting individual buildings, road intersections, and land parcel units into a set of hypergraph nodes; Based on a pre-defined set of spatial layout rules, spatial adjacency, road network accessibility, and public service coverage are defined as multi-dimensional constraint types. Through spatial buffer analysis and attribute association retrieval, multiple hypergraph nodes that satisfy the same dimension constraint are connected as the same hyperedge, thus constructing a hypergraph structure that associates nodes with hyperedges. Establish an association matrix describing the connection relationship between the nodes and hyperedges of the hypergraph, and assign an initial weight value to each hyperedge according to the preset priority of each constraint type.

5. The urban and rural planning method based on digital twins according to claim 1, characterized in that, The process of generating a non-uniform computational potential energy field is to calculate the superdegree centrality of the hypergraph nodes based on the correlation matrix as the topological centrality, and to accumulate the weight values ​​of all hyperedges associated with each node to obtain the constraint coupling weight. A two-dimensional grid matrix is ​​constructed with a preset spatial resolution. A potential energy decay model centered on each hypergraph node is constructed using a Gaussian kernel function. The spatial radiation intensity of each hypergraph node to the grid cell is calculated. Traverse all grid cells, linearly superimpose the spatial radiation intensities of all hypergraph nodes received by the grid cells, and normalize the superposition results to generate a non-uniform computational potential field.

6. The urban and rural planning method based on digital twins according to claim 1, characterized in that, The hierarchical scheduling strategy involves setting a gradient threshold for the non-uniform computation potential energy field and dividing the planning spatial grid into a high potential energy sensitive area and a low potential energy background area. For the high potential energy sensitive area, a LOD3 level fine model is dynamically loaded, the geometric vector information of building components and functional entrance and exit nodes are analyzed, and topological parameters containing spatial positioning and connection relationships are generated. For the low potential energy background area, a LOD1 level block model is loaded in a reduced dimension, the convex hull boundary of the building bottom is extracted, and filling data for space occupancy detection is generated. The topology parameters are used as strong constraints for the generative layout algorithm, and the filling data is used as boundary avoidance conditions. The driving force-oriented layout algorithm performs iterative calculations to generate intermediate planning layout data.

7. A digital twin-based urban and rural planning method according to claim 6, characterized in that, The force-directed layout algorithm sets the geometric objects corresponding to the topology parameters as active nodes and the geometric objects corresponding to the fill data as active nodes. Construct a position update rule based on spatial constraints, and calculate the non-overlapping repulsion vector between active nodes and other nodes, as well as the gradient pulling vector of active nodes pointing to high potential energy sensitive regions. The non-overlapping repulsion vector and the gradient traction vector are weighted and superimposed to obtain the comprehensive displacement vector of the active node, and the planar coordinates of the active node are iteratively updated based on the comprehensive displacement vector.

8. A digital twin-based urban and rural planning system, characterized in that, include: Scene construction module: Acquire building information model data and geographic information system data of the urban and rural areas to be planned, and construct a three-dimensional digital twin basic scene through spatiotemporal registration and data semantic alignment; Potential energy calculation module: Identifies the planning entities and multidimensional constraint relationships in the three-dimensional digital twin basic scene and constructs a topological hypergraph model, and calculates the topological centrality and constraint coupling weight of the topological hypergraph model; The three-dimensional digital twin basic scene is discretized into a planning space grid, and the topological centrality and constraint coupling weight are mapped to the planning space grid to generate a non-uniform computational potential energy field. Layout generation module: Based on the gradient distribution of the non-uniform computational potential energy field, a hierarchical scheduling strategy is constructed. For high potential energy sensitive areas, a LOD3 level fine model is loaded to extract geometric features and generate topological parameters. For low potential energy background areas, a LOD1 level block model is loaded to extract boundary features and generate filling data. The data is then input into the generative layout algorithm to generate intermediate planning layout data. Reverse correction module: detects geometric overlaps in the intermediate planning layout data and automatically repairs them, converts the repaired data into a solid model and outputs it as an urban and rural planning layout scheme.

Images