Building information model and city information model cross-scale consistency three-dimensional modeling method based on space intelligence

Abstract

The invention provides a building information model and city information model cross-scale consistency three-dimensional modeling method based on space intelligence. The method comprises the following steps: extracting a three-dimensional geometric patch set from an original building information model, constructing an initial modeling object and a component space diagram, and calculating a transformation operator through footprint matching to realize automatic alignment; the user-defined family attributes are analyzed to execute cross-scale semantic mapping, semantic classification information is generated, and a building shell model is generated through geometric generalization processing; opening features are extracted and mapped to the shell model, three-dimensional topology reconstruction is executed in combination with the component space diagram, and city information model target components with topology closing performance are generated. And finally, performing incremental updating based on the alignment deviation. According to the method, the problems of inaccurate geographic alignment, geometric degradation and semantic loss in cross-scale model fusion are solved, and the logic leakproofness and data maintenance efficiency of urban digital twin base modeling are effectively improved.

Description

Technical Field

[0001] This application relates to the field of spatial intelligence and urban digital twin data fusion technology, specifically, to a cross-scale consistent 3D modeling method for building information models and urban information models based on spatial intelligence. Background Technology

[0002] With the deepening of urban digital twin construction, the cross-scale integration of Building Information Modeling (BIM) and City Information Modeling (CIM) has become a core element in building a refined urban management foundation. BIM typically contains highly detailed component geometry and attributes, while the city-level foundation requires a lightweight, geolocated, and topologically consistent 3D model to support application needs such as urban planning approval, energy consumption analysis, and intelligent scheduling.

[0003] In existing 3D model conversion schemes, direct export and manual translation and alignment are commonly used. First, the building information model is exported to a standard exchange format using modeling software. Then, in the 3D geographic information system environment, technicians manually specify the insertion point and adjust the rotation angle based on the site boundary coordinates. Finally, a general simplification operator is used to downsample and thin the component geometry to reduce the data volume and generate a block model.

[0004] However, this traditional model processing approach has significant technical drawbacks. Because Building Information Modeling (BIM) typically uses a local engineering coordinate system and often lacks northward information, directly deriving the solution can lead to model misalignment, rotation, or mirroring in geographic space. Furthermore, general simplification algorithms often lack semantic association and topological constraints when handling detailed level generalizations, easily causing geometric degradation problems such as roof shape distortion, door and window opening drift, and non-manifold surface gaps. In addition, the semantic conversion accuracy for non-standard attributes such as "custom families" is low, and the lack of object-level differential recognition and incremental maintenance mechanisms means that any minor design change requires a full data reconstruction, significantly increasing operational costs. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides a cross-scale consistency 3D modeling method for building information models and city information models based on spatial intelligence, which can at least alleviate the aforementioned technical problems.

[0006] A cross-scale consistency 3D modeling method for building information models and city information models based on spatial intelligence includes: Step 1: Extract the set of 3D geometric facets of building components from the original building information model, and construct an initial 3D modeling object and a spatial diagram of the building information model components; determine the projection footprint of the initial 3D modeling object, calculate the transformation operator, and perform a 3D geometric transformation on the initial 3D modeling object to obtain the aligned initial 3D modeling object; Step 2: Parse the custom family attributes from the aligned initial 3D modeling object and perform cross-scale semantic mapping on them to obtain the semantic classification information of the city information model; extract the outer contour surface of the aligned initial 3D modeling object and perform geometric generalization processing on it to obtain the city information building shell model; Step 3: Extract opening features from the set of three-dimensional geometric faces and map them onto the urban information building shell model to obtain a set of geometric faces with opening features; based on the semantic classification information of the urban information model, the urban information building shell model and the set of geometric faces with opening features, and combined with the spatial map of the building information model components, perform three-dimensional topological reconstruction to generate urban information model target components with topological closure. Step 4: Calculate the alignment deviation of the target component of the city information model relative to the city information model base, and perform incremental updates on the city information model base based on the alignment deviation.

[0007] Optionally, in step 1, constructing the initial 3D modeling object includes: Project metadata is extracted from the original building information model. If the corresponding modeling unit is determined to be millimeters, a scaling factor is applied to correct the model dimensions of the original building information model to obtain a scale-normalized model. Extract the normal vector distribution of building components in the scale-normalized model, rotate the scale-normalized model to the geographic coordinate system to perform skew angle compensation, and obtain the normalized three-dimensional modeling object; Calculate the envelope volume of each building component in the normalized 3D modeling object, remove unstructured connectors whose envelope volume is less than a preset volume threshold, and obtain a simplified 3D modeling object as the initial 3D modeling object.

[0008] Optionally, in step 1, constructing the spatial diagram of the building information model components includes: The building information model extracts building component attributes to generate a set of component nodes. Based on the set of component nodes, it identifies the spatial connection relationships and opening filling constraints between building components to generate a set of topological edges. Then, it constructs a spatial graph of the building information model that represents the inclusion, adjacency and constraint relationships between components.

[0009] Optionally, in step 1, calculating the transformation operator includes: Using the affine transformation solution logic based on the feature-fast matching algorithm, the transformation operator parameter set consisting of scale coefficients, rotation matrices, and translation vectors is calculated; Extract urban road line constraint information from the geographic information system to perform angle correction on the projected footprint, so as to correct the transformation operator parameter set; A three-dimensional affine transformation is performed on the initial three-dimensional modeling object based on the modified transformation operator parameter set, and the terrain elevation data in the urban information model base is retrieved to perform vertical height alignment, so as to obtain the height-calibrated initial three-dimensional modeling object as the aligned initial three-dimensional modeling object.

[0010] Optionally, in step 2, parsing the custom family attributes and performing cross-scale semantic mapping includes: The custom family attributes in the aligned initial 3D modeling object are parsed to extract attribute keyword evidence, and the classification probability of the attribute keyword evidence is calculated using a probability mapping model to obtain semantic mapping feature values. Based on the semantic mapping feature values, a classification determination is performed, mapping the component type based on the building information model data exchange standard to the classification type based on the city information model standard, thereby obtaining the corresponding city information model semantic classification information.

[0011] Optionally, in step 2, obtaining the urban information building shell model includes: The aligned initial 3D modeling object is subjected to columnar pixelation to identify the top surface of the building, and probe rays are emitted upward along the height axis to extract roof feature patches; The roof feature patches are generalized at a level of detail using a roof extraction operator to simplify their topology, thus obtaining the urban information building shell model.

[0012] Optionally, step 3 includes: The opening filling constraint relationship is extracted from the component space diagram of the building information model to identify the opening components belonging to the door and window category, and the normal projection position of the opening component relative to the urban information building shell model is calculated using the opening projection operator to generate the opening position parameters. Boolean operations are performed on the surface of the urban information building shell model based on the opening location parameters to map and generate the set of geometric patches with opening features.

[0013] Optionally, in step 3, generating the target component of the city information model with topological closure includes: The semantic classification information of the urban information model is used to identify the component hierarchy relationship between the urban information building shell model and the set of geometric facets with opening features; Based on the component hierarchy and the opening filling constraint, the city information building shell model and the set of geometric facets with opening features are subjected to facet stitching to construct a closed geometry that matches the manifold constraint. Perform watertightness testing on the closed geometry to identify and close geometric gaps, thereby obtaining the target component of the city information model with topological closure.

[0014] Optionally, in step 4, calculating the alignment deviation of the target component of the city information model relative to the base of the city information model includes: Calculate the coordinates of the geometric center point and the envelope area of ​​the projected footprint, and the spatial deviation and area error value between them and the coordinates of the center point of the determined red line boundary and the corresponding legal area; The classification rate of the semantic classification information of the city information model is statistically analyzed, and a confidence score representing the alignment degree is generated based on the spatial deviation, the area error value, the classification rate, and the verification results of the watertightness test.

[0015] Optionally, step 4 further includes: in response to the confidence score being greater than a preset threshold, using a geometric hash algorithm to compare the original building information model under different version timestamps to identify the set of changed components; establishing a cross-scale association mapping table containing the unique identifier of the set of changed components and the urban information model base object, and performing local remapping and data replacement on the urban information model base based on the cross-scale association mapping table.

[0016] Optionally, the local remapping includes: for the changed component set, regenerating the corresponding urban information model semantic classification information using the parsed custom family attributes, and re-performing the geometric generalization processing and the mapping processing of the opening features on the geometric outer contour corresponding to the changed component set based on the updated semantic classification information, so as to generate an updated urban information model target component with topological closure.

[0017] Optionally, the data replacement includes: extracting the unique identifier of each component in the set of changed components based on the cross-scale association mapping table, retrieving the corresponding geometric data node in the urban information model base, and performing overlay data writing on the geometric data node using the updated urban information model target component with topological closure.

[0018] This application presents a spatially intelligent method for cross-scale consistency 3D modeling of Building Information Models (BIM) and City Information Models (CIM). Addressing the technical shortcomings of traditional methods, such as alignment uncertainty due to lack of geographic reference, geometric degradation caused by generalization of detail levels, and semantic loss, this method constructs initial modeling objects and component spatial maps, and performs automatic alignment using transformation operators. This solves the problems of reliance on manual intervention and low alignment efficiency and accuracy inherent in traditional methods. Compared to the manual specification of insertion points in traditional methods, this application utilizes a fast feature matching algorithm based on projection footprints and urban road lines / red lines as constraints. This not only automatically infers the scale and rotation orientation but also keeps the planar alignment deviation within a low range, providing a high geospatial reference for subsequent cross-scale fusion.

[0019] Based on the aligned model, this application addresses the problems of incomplete non-standard attribute mapping and geometric generalization distortion in traditional solutions by parsing custom family attributes, performing cross-scale semantic mapping, and generating urban information building shell models. Traditional solutions typically employ a single thinning operator, while this application improves semantic classification by extracting attribute keyword evidence through a probabilistic mapping model. Simultaneously, it utilizes columnar pixelation and roof extraction operators for detailed-level generalization, reducing geometric complexity while better preserving the building's macroscopic form. Compared to the random thinning of traditional solutions, the generated model exhibits higher robustness in both semantic expression and geometric representation.

[0020] By mapping opening features to the shell model and performing 3D topological reconstruction, and combining this with the building information model component spatial diagram to generate target components with topological closure, the shortcomings of traditional solutions, such as door and window opening drift and model non-closure, are resolved. This application explicitly introduces opening filling and adjacency constraints between components during the generalization process, improving the consistency of opening positions with the original logic after cross-scale transformations. Through panel stitching and watertightness testing, the generated model meets manifold constraint requirements. Compared to the roof fragmentation and panel cantilever problems easily generated by traditional solutions, this solution has superior performance in meeting the requirements of physical simulation and urban analysis applications.

[0021] Finally, by calculating alignment deviations, generating consistency verification reports, and performing incremental updates, a closed-loop mechanism of "modeling-verification-feedback-maintenance" is formed, solving the problems of traditional solutions lacking auditing mechanisms and having high maintenance costs. Traditional solutions require a full reconstruction when the source model changes, while this application uses a geometric hash algorithm to identify changed components and perform local remapping and replacement, which significantly reduces the operation and maintenance update pressure of the urban information model base while ensuring long-term data consistency. Attached Figure Description

[0022] Figure 1This is a flowchart illustrating a cross-scale consistency 3D modeling method for building information models and city information models based on spatial intelligence, according to an embodiment of this application. Figure 2 This is a structural diagram of a three-dimensional modeling device for cross-scale consistency between building information model and city information model based on spatial intelligence, according to an embodiment of this application. Figure 3 This is a hardware architecture diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0023] like Figure 1 As shown in the figure, this application provides an embodiment of a cross-scale consistency 3D modeling method for building information models and city information models based on spatial intelligence, comprising: Step 1: Extract the set of 3D geometric facets of building components from the original building information model, and construct an initial 3D modeling object and a spatial diagram of the building information model components; determine the projection footprint of the initial 3D modeling object, calculate the transformation operator, and perform a 3D geometric transformation on the initial 3D modeling object to obtain the aligned initial 3D modeling object; Step 2: Parse the custom family attributes from the aligned initial 3D modeling object and perform cross-scale semantic mapping on them to obtain the semantic classification information of the city information model; extract the outer contour surface of the aligned initial 3D modeling object and perform geometric generalization processing on it to obtain the city information building shell model; Step 3: Extract opening features from the set of three-dimensional geometric faces and map them onto the urban information building shell model to obtain a set of geometric faces with opening features; based on the semantic classification information of the urban information model, the urban information building shell model and the set of geometric faces with opening features, and combined with the spatial map of the building information model components, perform three-dimensional topological reconstruction to generate urban information model target components with topological closure. Step 4: Calculate the alignment deviation of the target component of the city information model relative to the city information model base, and perform incremental updates on the city information model base based on the alignment deviation.

[0024] In this application, the Building Information Modeling (BIM) serves as a refined data source containing microscopic geometric patches and spatial topological relationships. By extracting its building components and constructing a spatial diagram of the BIM components, it provides the basic data carrier for the 3D geometric alignment in step 1. The City Information Modeling (CIM), on the other hand, serves as a macroscopic digital twin foundation, receiving the model data after semantic mapping and geometric generalization processing in step 2. This application extracts opening features from the BIM and maps them to the CIM building shell model. Using topological reconstruction in step 3, it generates physically closed CIM target components. Finally, through deviation calculation and incremental updates in step 4, it achieves cross-scale 3D modeling and consistent placement from BIM microscopic components to the CIM macroscopic scene.

[0025] Optionally, in step 1, constructing the initial 3D modeling object includes: extracting project metadata from the original building information model; if the corresponding modeling unit is determined to be millimeters, applying a scaling factor to correct the model dimensions of the original building information model to obtain a scale-normalized model; extracting the normal vector distribution of building components in the scale-normalized model to rotate the scale-normalized model to the geographic coordinate system to perform skew angle compensation, thereby obtaining a normalized 3D modeling object; calculating the envelope volume of each building component in the normalized 3D modeling object, and removing unstructured connectors whose envelope volume is less than a preset volume threshold to obtain a simplified 3D modeling object as the initial 3D modeling object.

[0026] Preferably, the specific implementation process of step 1 is as follows: Low-level data parsing processing is performed on the loaded original building information model to extract project metadata representing the global settings of the project. Logical judgment is performed on the length unit attribute recorded in the extracted project metadata. If the corresponding modeling unit is determined to be millimeters (i.e., a numerical order of magnitude of one-thousandth of a meter), then a scaling operator is called and a preset scaling factor (such as 0.001) is applied to perform numerical normalization scaling processing on the three-dimensional spatial coordinates of all geometric vertices in the original building information model to correct the model dimensions of the original building information model and thus obtain a scale-normalized model. This processing step alleviates the differences in the order of magnitude of low-level numerical representations between different modeling software and improves the execution of subsequent geometric operations within a unified metric length space.

[0027] Preferably, in the specific technical implementation of step 1, component normal sampling and spatial attitude analysis are performed on the obtained scale-normalized model. For example, building components of a selected scale are randomly selected or sequentially traversed from the scale-normalized model, and the normal vector distribution characteristics of these building components in the current local coordinate system are extracted. By statistically analyzing the projection intensity of the normal vector distribution characteristics on different coordinate axes, the upward axis of the current model is identified. If the upward axis is determined to be the vertical axis (i.e., an axis inconsistent with the horizontal geographic plane standard), an axis correction algorithm is called to rotate the scale-normalized model to a geographic coordinate system conforming to the vertical axis. Subsequently, the deviation angle values ​​of the geographic north angle and the project north angle recorded in the project metadata are further analyzed, and rotation compensation processing is performed on the building components using a rotation transformation matrix to generate the corresponding deviation angle compensation result, thereby obtaining the normalized 3D modeling object. This step, through the alignment and orientation correction of the spatial coordinate system, enables the model's orientation in virtual space to achieve a high degree of consistency with the real geographic space.

[0028] Preferably, in one scenario, step 1 is specifically implemented as follows: Component-level geometric filtering and redundancy removal are performed on the obtained normalized 3D modeling object. For example, the geometric topological descriptions of each building component contained in the normalized 3D modeling object are traversed and parsed, and the envelope volume corresponding to each building component is calculated based on a 3D bounding box algorithm. The calculated envelope volume is logically compared with a preset volume threshold (e.g., 0.001 cubic meters or less) to identify unstructured connectors (such as screws, washers, small brackets, etc.) that have a very low volume percentage and do not carry core structural information of the building. The action of removing unstructured connectors whose envelope volume is smaller than the preset volume threshold is performed, thereby reducing the data scale of the model and simplifying the topological complexity to obtain a simplified 3D modeling object, which is then used as the initial 3D modeling object. This simplification process preserves the macroscopic form and key features of the building, while effectively reducing the computational overhead of the urban digital twin base during loading and rendering, providing a lightweight and logically rigorous data base for subsequent alignment verification in a massive data environment.

[0029] Preferably, in the specific technical implementation of the 3D bounding box algorithm in step 1, a geometric bottom-level vertex traversal process is performed on the obtained normalized 3D modeling object. For example, the geometric topology description file (such as a set of faces containing a 3D mesh) of each building component in the normalized 3D modeling object is parsed, and the 3D spatial coordinate information of all geometric vertices is extracted from it. By establishing a vertex coordinate index for each independent component, the component vertex coordinate sequence is obtained. This process abstracts the complex geometric shape of building components into a series of discrete spatial coordinate points, solving the problem of irregular shapes being difficult to directly quantify in volume from the data level, and providing a clear underlying processing object for subsequently establishing geometric envelope relationships through extreme value search.

[0030] Preferably, in the specific implementation logic of the three-dimensional bounding box algorithm, the obtained sequence of component vertex coordinates is used to perform axial extreme value search processing in a preset rectangular coordinate system. For example, point-by-point comparison is performed from the full set of three-dimensional spatial coordinate information for the three mutually perpendicular projection directions: the horizontal axis, the vertical axis, and the height axis, thereby locking the maximum and minimum coordinate values ​​in each axis. By performing a difference operation on the maximum and minimum coordinate values ​​in the same axis, the spatial span length of the component in each axis (such as length span, width span, and height span) is calculated, thus obtaining the component's axial span set. This processing method establishes a virtual axially aligned bounding box that can tightly enclose the original building component entity, simplifying the originally complex topological structure of the component into a regular cuboid geometric envelope with clear three-dimensional dimensional characteristics.

[0031] Preferably, envelope space quantization is performed on the obtained set of axial spans of the components. For example, the extracted length span, width span, and height span are multiplied logically to obtain the maximum virtual volume value representing the component's position in three-dimensional space, thus obtaining the envelope volume corresponding to the building component. This quantization process transforms multi-dimensional dimensional features into a single physical attribute scalar, providing a standardized physical criterion for efficient selection of large-scale components through logical comparison. Compared to directly calculating the mathematical volume of non-manifold complex geometric entities, the envelope volume calculated using the three-dimensional bounding box algorithm has lower computational overhead and higher logical stability, and can better reflect the spatial weight of components under cross-scale mapping perspectives.

[0032] Preferably, after calculating the volume of the envelope, filtering and redundancy removal based on logical thresholds are performed on the building components. For example, each obtained envelope volume is compared logically with a preset volume threshold (e.g., 0.001 cubic meters or a smaller range). If the envelope volume of the current component is determined to be lower than the preset volume threshold, the component is identified as a non-structured connector (e.g., redundant screw entities, micro-wafers, or non-load-bearing supports in the model) that does not bear macroscopic visual features and does not affect the overall topological logic of the building, and a geometric removal operation is performed. After performing the geometric removal operation, the remaining high-volume components undergo geometric merging and topological reorganization to generate a processing result and define it as a simplified 3D modeling object. This step optimizes the data storage load of the urban information model base by eliminating a massive amount of meaningless micro-geometric details in the early stages of data processing, achieving the technical benefit of allowing subsequent cross-scale alignment and verification tasks to focus on the core skeleton structure of the building.

[0033] Optionally, in step 1, constructing the building information model component space diagram includes: extracting building component attributes from the original building information model to generate a set of component nodes, identifying the spatial connection relationships and opening filling constraint relationships between each building component based on the set of component nodes to generate a set of topological edges, and then constructing the building information model component space diagram that represents the inclusion, adjacency and constraint relationships between components.

[0034] Preferably, the specific implementation process of constructing the building information model component spatial diagram in step 1 is as follows: For the loaded original building information model, perform data exchange standard parsing processing (such as performing low-level data parsing on exchange format files containing architectural design information), extracting all component instances from top to bottom according to the global spatial hierarchy, including project components, building components, floor components, and detailed wall components, window components, door components, and floor components. For each extracted component instance, by reading its low-level attribute dictionary, obtain the corresponding globally unique identifier, semantic entity classification description, 3D geometric envelope box feature parameters, and business data from the custom attribute set, and convert them into digital graph nodes with semantic labels. Aggregate and encapsulate the converted digital graph nodes to obtain a component node set. This processing step realizes the transformation from unstructured model files to structured node data, providing atomic description objects for the subsequent construction of a topological network with spatial logic.

[0035] Preferably, in the specific technical implementation of step 1, which constructs the spatial diagram of building information model components, spatial topology link identification processing is performed using the obtained set of component nodes and the hierarchical association description in the original building information model. For example, the decomposition relationship description (used to determine the aggregation logic between site and building, and building and floor), spatial inclusion relationship description (used to determine the attribution logic between floor and interior components), and adjacency connection relationship description (used to determine the physical contact logic between different wall components or different spatial areas) defined in the building information model data exchange standard are retrieved. Topological mapping is performed on the identified relationship descriptions to generate hierarchical attribution links and horizontal adjacency links connecting different digitized diagram nodes, thereby obtaining a basic topology relationship edge subset. This step establishes the basic spatial skeleton structure of the building's interior by logically locking the macroscopic attribution and microscopic adjacency relationships between components.

[0036] Preferably, in a scenario, when constructing the building information model component space diagram in step 1, the following steps are performed: Opening filling logic extraction processing is executed for the host objects marked as wall components and the filling objects marked as door / window components in the component node set. For example, the filling relationship descriptions defined in the original building information model are retrieved, and the physical openings and positional occupancy relationships between the filling objects and their corresponding host objects in three-dimensional space are identified. This filling logic with clear master-slave characteristics is defined as a topological connection link with constraint directions, and the corresponding processing results are generated to obtain the edge subset of opening filling constraint relationships. Therefore, by recording the dynamic opening coupling relationship between door / window components and wall components at the logical level, the topological characteristics of the building facade openings are maintained during subsequent cross-scale geometric generalization and multi-level detail simplification required by the urban information model.

[0037] Preferably, the specific implementation process of constructing the building information model component spatial graph in step 1 is as follows: The generated subset of basic topological relationship edges and the subset of opening-filling constraint relationship edges are merged to generate a topological edge set. Then, the component node set and the topological edge set are synthesized using graph theory. By using each node in the component node set as a vertex and the connections in the topological edge set as topological edges, an association network representing global semantics and local geometric constraints is constructed in the computational space, thus obtaining the building information model component spatial graph. This processing result achieves a deep digital representation of the physical entity relationships within the building, not only completely recording the component attribute information but also defining the inclusion, adjacency, and constraint relationships between components through complex edge sets. This provides crucial logical guidance for subsequent projection footprint extraction and adaptive topological reconstruction of target components in the urban information model under conditions lacking global geographic reference.

[0038] Preferably, the specific implementation process of constructing the building information model component spatial graph in step 1 is as follows: Graph theory structure synthesis processing is performed on the obtained set of component nodes and the generated set of topological edges. For example, each digital node representing a specific building entity (such as a column, beam, or door / window with a specific number) in the set of component nodes is mapped to a vertex in the graph theory model, and topological connection arcs connecting the corresponding vertices are established according to the connection relationships defined in the set of topological edges (such as spatial affiliation or physical contact between components). After performing the graph theory structure synthesis processing, a graph structure with multi-dimensional association attributes is constructed in the computational memory space, thereby obtaining the association network of global semantics and local geometric constraints. This processing step transforms the originally isolated, unstructured geometric patch data into a topological network with rigorous mathematical association relationships, providing the underlying data carrier for subsequent complex spatial logic retrieval in multi-dimensional scenarios.

[0039] Preferably, in the specific technical implementation of the association network of global semantics and local geometric constraints in step 1, global semantic hierarchical mapping processing is performed on the obtained graph structure. For example, the hierarchical metadata carried in the digital nodes (such as the encoded information of building projects, individual buildings, specific floors, and room entities) is parsed, and a top-down tree-like logical chain is established using the set of topological edges. By performing deep coupling between macroscopic building space units and microscopic component instances at the logical level, a global semantic association topology is obtained. This step realizes the extraction of the logical skeleton of the entire building lifecycle information, enabling each local component to trace its position and business role in the global building space. This achieves the technical benefit of allowing the system to deduce the macroscopic distribution range of the target area through hierarchical inclusion relationships even in scenarios lacking absolute geographical coordinates.

[0040] Preferably, in a specific implementation of the network linking global semantics and local geometric constraints in a scenario: geometric spatial constraint identification processing is performed on local nodes in the global semantic linking topology. For example, constraint parameters representing physical contact and opening filling are retrieved from the topological edge set and converted into geometric offset vectors connecting adjacent digital nodes. Specifically, the boundary coplanar constraint between two adjacent wall nodes and the occupancy filling constraint of a specific window node within its corresponding wall node are identified. By solidifying this microscopic geometric topological constraint in the connecting arc segments between nodes, a local geometric constraint network is obtained. This step improves the simplification and generalization of geometric details during subsequent 3D reconstruction tasks for City Information Modeling (CIM) requirements, ensuring adherence to the original physical spatial principles and effectively avoiding geometric anomalies such as overlapping patches, gaps, or logical position drift.

[0041] Preferably, the specific implementation process of constructing the building information model component spatial graph in step 1 is as follows: Multi-dimensional relationship fusion processing is performed on the generated global semantic association topology and the local geometric constraint network. By logically superimposing the semantic inclusion relationships representing the macro-level and the local geometric occupancy constraints representing the micro-components within a unified computational data structure, a complex association body capable of comprehensively describing the interactions of building physical entities is generated, thus obtaining the building information model component spatial graph. This processing result achieves a deep digital representation of the relationships between physical entities within the building, not only fully recording the component attribute information but also defining the inclusion, adjacency, and constraint relationships between components through complex edge sets. This deeply associated network model provides crucial logical guidance for subsequent projection footprint extraction under extreme conditions of incomplete geographic reference and adaptive topology reconstruction of target components in the urban information model, greatly improving the data logic rigor during the model's cross-scale transformation process.

[0042] Optionally, in step 1, calculating the transformation operator includes: using the affine transformation solution logic based on a feature-fast matching algorithm to calculate a transformation operator parameter set consisting of scale coefficients, rotation matrices, and translation vectors; extracting urban road line constraint information from the geographic information system to perform angle correction on the projected footprint to correct the transformation operator parameter set; performing a three-dimensional affine transformation on the initial three-dimensional modeling object based on the corrected transformation operator parameter set, and retrieving terrain elevation data from the urban information model base to perform vertical height alignment, obtaining a height-calibrated initial three-dimensional modeling object as the aligned initial three-dimensional modeling object.

[0043] Preferably, the specific implementation process of step 1 is as follows: Orthogonal projection processing of the horizontal geographic plane is performed on the obtained initial 3D modeling object. By extracting the geometric contour boundary of the 3D entity on the 2D bottom surface, a projection footprint representing the planar outline of the model is obtained. Subsequently, redline boundary data associated with the geographical location of the project are retrieved from the relevant geographic information system database, and a feature-based fast matching algorithm is used to perform a spatial geometric feature similarity comparison between the projection footprint and the redline boundary data. The feature-based fast matching algorithm is equivalent to performing a large number of rigid transformation random samplings in the search space, calculating the distance cost function when the sample point set in the projection footprint is transformed to the redline boundary data, and obtaining the correspondence with the lower cost function value through optimization iteration to obtain the initial alignment mapping features. This processing step establishes a logical connection between the 3D model, which was originally in an isolated engineering coordinate system, and the land parcel boundary with geographic attributes, providing a data foundation for subsequent quantification of spatial displacement.

[0044] Preferably, the specific technical implementation process for extracting the projection footprint in step 1 is as follows: Orthogonal projection processing based on the spatial envelope surface is performed on the obtained initial 3D modeling object. For example, the geometric facet vertex coordinates of each component entity in the initial 3D modeling object are analyzed, and the vertical dimension height information is masked, thereby mapping the spatial distribution of the 3D entities to the two-dimensional coordinate space of the horizontal geographic plane. A convex hull search algorithm (i.e., the logic of finding the smallest convex polygon boundary that can contain all discrete point sets) is performed on the mapped full set of coordinate points to extract the closed boundary line representing the outermost geometric contour of the building shape on the bottom surface, thus obtaining the projection footprint. This processing step realizes the abstract transformation from high-dimensional fine component information to low-dimensional planar geometric features, providing a standardized comparison base map for subsequent feature matching with large-scale land parcel red lines in the geographic information system without losing the building's planar orientation features.

[0045] Preferably, in step 1, when performing feature comparison between the projected footprint and the redline boundary data: the designed feature fast matching logic (i.e., the corresponding full name in English is Fast-Match algorithm) is used to perform geometric feature sampling processing on the projected footprint. For example, according to a preset sampling step size (such as a sampling interval range of 0.1 meters to 0.5 meters), a discretization point sampling operation is performed along the closed boundary line of the projected footprint to obtain a series of discrete spatial coordinate points characterizing the shape of the projected footprint, thus obtaining the projected footprint sampling point set. This step effectively improves the computational efficiency of calculating pose offset relationships in the complex geographic base search space by transforming continuous geometric lines into discrete point matrices with clear numerical indices, supporting the association recognition of cross-scale data models at the geometric topology level.

[0046] Preferably, in a specific implementation of step 1, calculating the distance cost function, a spatial pose simulation based on rigid transformation random sampling is performed on the obtained projection footprint sampling point set. For example, within a preset geographic search area (e.g., within a specific radius around the center point of the target parcel), candidate transformation matrices containing different rotation angles and translational displacements are randomly generated, and these candidate transformation matrices are used to drive the spatial placement simulation of the projection footprint sampling point set. Subsequently, the vertical distance value of each sampling point after the simulated placement relative to the nearest geometric segment in the redline boundary data is calculated. By performing a summation or weighted average operation on the above vertical distance values ​​of all points in the projection footprint sampling point set, an evaluation scalar characterizing the alignment degree of the model under the current pose is generated, thus obtaining the distance cost function. This quantitative evaluation mechanism objectively reflects the degree of "mismatch" of the digital model in the geographic base, providing a key performance indicator for finding the alignment position through mathematical optimization.

[0047] Preferably, the specific implementation process of step 1 is as follows: Multiple rounds of optimization iteration and pose update processing are performed on the obtained distance cost function. For example, using gradient descent logic or a heuristic optimization algorithm, the rotation and translation components in the candidate transformation matrix are continuously adjusted to drive the projected footprint sampling point set to perform dynamic displacement in geographic space, thereby finding a transformation state that makes the distance cost function achieve a local minimum or a lower numerical range. When it is determined that the distance cost function tends to be stable and lower than a preset alignment quality threshold, the current optimal transformation matrix parameters are extracted and used as a logical link to establish a mapping relationship between the building coordinate system and the geographic coordinate system, thus obtaining the initial alignment mapping features. This step, through automated iterative calculation, alleviates the risk of orientation deviation caused by the lack of geographic reference, realizes the logical stitching between the microscopic building shape and the macroscopic geographic red line, and achieves the technical benefit of improving the initial alignment efficiency of cross-scale modeling tasks.

[0048] Preferably, in the specific technical implementation of step 1, an affine transformation operator solution process is performed on the obtained initial alignment mapping features. For example, based on the least squares optimization criterion, analytical calculations are performed on the successfully matched feature point pairs to separate a transformation operator parameter set composed of a scale coefficient, a rotation matrix, and a translation vector. The scale coefficient is used to correct the unit-order deviation (such as the millimeter-to-meter conversion ratio) between the Building Information Model (BIM) and the geographic base; the rotation matrix is ​​used to characterize the angular deviation of the model relative to the geographic north direction in the National 2000 Coordinate System or the World Geodetic Coordinate System; and the translation vector represents the horizontal displacement component of the model's centroid on the geographic plane, thus obtaining the transformation operator parameter set. This step realizes the transformation from geometric feature matching to vectorized spatial transformation matrix, enabling the digital model to possess the mathematical description for performing placement operations within a macroscopic geographic base.

[0049] Preferably, step 1 performs affine transformation operator solving based on the least squares optimization criterion on the obtained initial alignment mapping features. The specific implementation process is as follows: Multiple feature point pairs that establish a correspondence between the projected footprint and the redline boundary data are extracted from the initial alignment mapping features. The sampling point coordinates of each feature point in the local coordinate system of the Building Information Modeling (BIM) and the target reference coordinates in the base coordinate system of the Geographic Information System are obtained. By establishing a numerical correlation mapping between each sampling point coordinate and the corresponding target reference coordinate, a sequence of feature point pairs to be optimized is obtained. This processing step provides quantified observation samples for subsequent analytical calculations of spatial transformation parameters, improving the optimization calculation by providing a clear digital processing object.

[0050] Preferably, in the specific technical implementation of the least squares optimization criterion, the obtained feature points to be optimized are used to construct a global alignment error function for the sequence. The least squares optimization criterion is equivalent to setting an objective function characterizing the degree of geometric deviation, requiring the sum of the squared distances between the coordinates of all sampled points after spatial transformation and their corresponding target reference coordinates to reach a minimum value range. For example, a set of affine transformation parameters containing unknowns (involving scaling, rotation, and translation components) is preset, and a simulated transformation operation is performed on each of the sampled point coordinates using the current affine transformation parameters, calculating the coordinate differences between the simulated transformed coordinates and the corresponding target reference coordinates along each coordinate axis. A squaring operation is performed on each coordinate difference to eliminate negative sign interference and amplify the deviation characteristics, and the squared values ​​of all points are summed to obtain the global alignment residual sum of squares function.

[0051] Preferably, in one scenario, parameter separation processing based on extremum analysis is performed on the global aligned residual sum of squares function. For example, using partial derivative operation logic, differentiation is performed on each variable to be solved in the global aligned residual sum of squares function (such as parameters representing lateral translation, parameters representing longitudinal translation, sine and cosine parameters representing rotation angles, and parameters representing scaling) to obtain the gradient change trend of the error surface in the multidimensional parameter space. By setting all partial derivative values ​​to zero, a set of linear equations about the transformation variables is constructed, and algebraic solutions are performed using matrix inversion or decomposition algorithms, thereby separating the optimal parameter combination that satisfies the requirement of minimizing global error. By structurally encapsulating the solved values ​​according to their physical meaning, the transformation operator parameter set is obtained. This step realizes the essential transformation from massive discrete geometric matching to simplified mathematical transformation rules, achieving the technical benefit of providing a rigorous mathematical description for subsequent model placement.

[0052] Preferably, in the specific implementation of step 1, a full spatial affine transformation is performed on the initial 3D modeling object using the obtained transformation operator parameter set. For example, the scale coefficients in the transformation operator parameter set are extracted to perform dimensional correction processing on the initial 3D modeling object (e.g., automatic conversion from millimeter to meter units), the rotation matrix is ​​used to perform skew angle correction processing on the building's geographic orientation, and the translation vector is used to perform a positional translation of the model's absolute position on the geographic base. By performing the above composite geometric transformation, the micro-model originally in an isolated engineering coordinate system is transformed into a City Information Modeling (CIM) spatial environment with a unified geographic benchmark, thus obtaining a geographically aligned initial 3D modeling object, which is then used as the aligned initial 3D modeling object. This process utilizes the global error balancing capability brought by the least squares criterion to effectively weaken the impact of local feature point sampling errors on the overall positioning, resulting in a model with high spatial distribution rationality within the macro-city base.

[0053] Preferably, in one scenario, step 1 is specifically implemented as follows: if the alignment confidence of the transformation operator parameter set is determined to be lower than a preset threshold, then the urban road line constraint information of the target area is further extracted from the underlying geographic information system. Extraction processing is performed on the main direction of the facade of the initial 3D modeling object, and the angular deviation between this main direction and the axial direction of the road centerline in the adjacent urban road line constraint information is analyzed, thereby performing angular correction processing on the original rotation matrix components. By using the linear topological direction contained in the urban road line constraint information as the attitude calibration benchmark, local compensation and fine-tuning are performed on the orientation information in the transformation operator parameter set to obtain the corrected transformation operator parameter set. This correction process cleverly utilizes the parallel or perpendicular constraint logic that buildings and streets typically possess in urban scenes, effectively solving the ambiguity that may exist in the orientation dimension of single redline matching, and improving the spatial distribution rationality of the model at the street scale.

[0054] Preferably, the specific technical implementation process for extracting the urban road line constraint information in step 1 is as follows: Topological feature retrieval processing is performed on the loaded geographic information system base data. By identifying the municipal road vector layers surrounding the target plot, the central axis segment representing the road direction is extracted. Spatial vectorization processing is performed on the extracted central axis segment to obtain its direction vector description in a unified geographic coordinate system (such as the national 2000 coordinate system), thereby obtaining the urban road line constraint information. This processing step locks in the inherent traffic network skeleton from the macroscopic scene, providing a linear topological benchmark with global semantic reference for building models in isolated coordinate systems, and solving the problem of insufficient directional guidance force that may exist when aligning the red line of a single plot over long distances.

[0055] Preferably, in the specific implementation logic for extracting the main orientation of the facade, the obtained initial 3D modeling object (whose full English name is Building Information Modeling) is used to perform component normal statistical analysis. For example, the normal distribution of all vertical facade components (such as exterior wall panels) in the analytical model is traversed, and the distribution intensity of the normals at different azimuth angles in the horizontal plane is statistically analyzed based on the angle histogram algorithm. The azimuth angle interval with the highest distribution intensity is locked as the orientation of the main building, and the average geometric axial characteristics of the building are calculated based on this to obtain the main orientation of the facade. This step, through the regularity analysis of massive micro-geometric panels, extracts the inherent geometric order of the architectural design, thereby obtaining a macroscopic directional scalar that can represent the overall three-dimensional posture of the building.

[0056] Preferably, in one scenario, the specific technical implementation of angle correction processing for the original rotation matrix components is as follows: Spatial angular deviation calculation processing is performed using the obtained main orientation of the facade and the constraint information of the urban road line. Specifically, the direction vector of the main orientation of the facade and the central axis segment of the adjacent road is projected and compared, and the original rotation matrix components (i.e., the rotation components representing the initial azimuth angle of the building in geographic space) obtained in the previous feature matching calculation are retrieved. For example, it is analyzed whether the main orientation of the facade and the road orientation satisfy the geometric constraint logic of parallelism (e.g., the included angle is within a very small error range of 0 degrees or 180 degrees) or perpendicularity (e.g., the included angle is within a very small error range of 90 degrees or 270 degrees). By calculating the residual angular deviation value of the current building posture relative to this geometric constraint logic, the azimuth calibration correction amount is obtained.

[0057] Preferably, the specific implementation process of step 1, performing local compensation and fine-tuning, is as follows: For the obtained orientation calibration correction amount, incremental update processing is performed on the original rotation matrix components in the transformation operator parameter set. For example, the orientation calibration correction amount is used as a small angle compensation operator in the current rotation transformation description. By correcting the parameters of the original rotation matrix components, the three-dimensional posture of the building can follow the typical linear spatial layout rules in urban scenes, thus obtaining the corrected transformation operator parameter set. This local compensation and fine-tuning process effectively corrects the small rotational drift that may occur when relying solely on footprint matching, improves the logical rationality of the spatial distribution of the generated urban information model target components at the street scale, and achieves the technical benefit of strengthening the overall spatial consistency of cross-scale modeling tasks.

[0058] Preferably, the specific implementation process of step 1 is as follows: Using the obtained modified transformation operator parameter set, perform a three-dimensional coordinate affine transformation on the initial three-dimensional modeling object, and simultaneously retrieve terrain elevation data of the target overlapping area from the corresponding urban information model base. Analyze the indoor floor height attribute recorded in the initial three-dimensional modeling object, and calculate the numerical deviation of the indoor floor height in the vertical dimension relative to the terrain elevation data. Perform a vertical height alignment operation on the model based on this numerical deviation, thereby eliminating the "floating" or "deeply buried" phenomenon of the model in the three-dimensional geographic scene caused by inconsistent local elevation origins, thus obtaining a height-calibrated initial three-dimensional modeling object, which is then used as the aligned initial three-dimensional modeling object. This achieves coordinated calibration of horizontal coordinates, vertical orientation, and vertical height, improving the ability of the micro-building model to be effectively integrated into the macro-level urban digital base, and supporting the closed-loop execution of cross-scale consistent modeling tasks.

[0059] Preferably, the specific technical implementation process of performing coordinate affine transformation in step 1 is as follows: For the obtained initial 3D modeling object (i.e., the corresponding full English name is Building Information Modeling), a full spatial coordinate recalculation of all geometric vertices is performed. For example, the scale coefficient, rotation matrix components, and horizontal translation vector are extracted from the corrected transformation operator parameter set, and an augmented matrix representing the linear transformation of 3D space is constructed. By performing matrix multiplication on the vertex coordinates of each geometric facet in the initial 3D modeling object, the component is mapped from the original local engineering coordinate system to the unified geographic spatial coordinate system of the target area, thus obtaining a preliminary aligned modeling object. This processing step, while maintaining the topological characteristics of the building's internal geometric structure, achieves numerical alignment between the micro-model and the macro-base in terms of scale, azimuth, and horizontal pose, providing a unified spatial reference for subsequent vertical dimension stitching.

[0060] Preferably, in the specific implementation logic of the coordinate affine transformation process, the geographical features of the regions overlapping with the initially aligned modeling object in the horizontal projection are retrieved synchronously from the base of the city information model (i.e., the corresponding full English name is City Information Modeling). Specifically, the action of retrieving the topographic elevation data of the target overlapping area is performed to obtain the set of vertical elevation values ​​of the real surface at that geographical location, thereby obtaining the base elevation reference surface. This step, through real-time collection of macro-geographical elevation information, introduces real ground physical constraints to the initially aligned modeling object, which is in a "floating" state, improving the cross-scale modeling task by providing not only the rationality of the planar dimension but also the positioning basis of the vertical spatial dimension.

[0061] Preferably, a consistency correction process based on a height reference is performed for vertical alignment. For example, the indoor floor height attributes (such as the zero-zero elevation position defined in engineering drawings) recorded in the initial alignment modeling object are traversed and parsed, and the elevation difference of the spatial point corresponding to this attribute relative to the corresponding coordinate point in the base elevation reference plane is calculated in the vertical dimension. By performing a subtraction operation, a value representing the height difference between the building base and the actual terrain is generated, and the resulting value is defined as the vertical numerical deviation. Subsequently, the vertical numerical deviation is applied as a vertical displacement compensation parameter to all geometric vertices of the initial alignment modeling object. By performing a vertical alignment action along the height axis, the building's indoor floor can smoothly connect with the terrain surface of the urban base, thus obtaining the initial 3D modeling object after height calibration. This effectively alleviates the defects of model "floating" or "deeply buried" caused by inconsistent local elevation origin definitions, improving the visual realism and spatial logic rigor of the 3D scene.

[0062] Preferably, in one scenario, a closed-loop verification process in 3D space is performed on the initial 3D modeling object after height calibration. For example, the residual spatial deviation of the calibrated model from adjacent urban elements (such as red-line boundaries and road centerlines) in geographic space is calculated, and it is verified whether this deviation is within a preset calibration threshold range (such as 0.1 to 0.3 meters). If the consistency requirement is met, the initial 3D modeling object after height calibration is determined as the final aligned initial 3D modeling object, and it is synchronized to the data cache for subsequent semantic mapping and topology reconstruction tasks. This process achieves coordinated calibration of horizontal coordinates, vertical orientation, and vertical height, improving the ability of the micro-building model to be seamlessly and tightly integrated into the macro-level digital urban foundation, supporting high-quality closed-loop execution of cross-scale consistency modeling tasks. Through this coordinated processing action based on affine transformation and elevation compensation, the technical benefit of enabling the digital model to have extremely high placement reliability within the macro-level urban foundation is achieved.

[0063] Optionally, in step 2, parsing the custom family attributes and performing cross-scale semantic mapping includes: parsing the custom family attributes in the aligned initial 3D modeling object to extract attribute keyword evidence, and using a probability mapping model to calculate the classification probability of the attribute keyword evidence to obtain semantic mapping feature values; performing classification determination based on the semantic mapping feature values, mapping the component type based on the building information model data exchange standard to the classification type based on the city information model standard, and obtaining the corresponding city information model semantic classification information.

[0064] Preferably, the specific implementation process of step 2 is as follows: For the aligned initial 3D modeling object, perform component-level attribute space traversal processing. Retrieve the component attribute set metadata defined by the Building Information Modeling Data Exchange Standard (such as Industry Foundation Classes) to obtain non-standard definition information or custom parameter descriptions contained therein. Perform text feature extraction and semantic segmentation processing on the retrieved non-standard definition information. By identifying functional definition fields (such as functional description fields) or construction process description fields, extract attribute keyword evidence that can characterize the physical use or logical belonging of the component, thereby obtaining the text evidence set to be mapped. This processing step achieves in-depth mining of the underlying metadata of non-standard components, providing original evidence for subsequently eliminating the expression differences between building modeling standards and urban management standards at the semantic level.

[0065] Preferably, in the specific technical implementation of step 2, a pre-constructed probability mapping model is used to perform semantic association strength evaluation processing on the text evidence set to be mapped. The probability mapping model is equivalent to establishing a prior association probability mapping matrix between an attribute keyword lexicon and each semantic classification label in the city information model standard. In this matrix, rows represent different text feature words, columns represent target semantic classification labels, and the element at the intersection of rows and columns represents the association probability weight of that feature word pointing to that label. By inputting the attribute keyword evidence into the probability mapping model, a weighted summation operation is performed between the evidence vector and the prior association probability mapping matrix to calculate the probability distribution vector of the attribute keyword evidence belonging to each target semantic classification label, thereby obtaining the semantic mapping feature value. This calculation method based on probability weight distribution effectively solves the problem of recognition omissions that easily occur in traditional hard matching methods when processing custom family components.

[0066] Preferably, in one scenario, step 2 is specifically implemented as follows: Taking the identified special glass panel (such as a building component with custom family attributes) as the processing object, classification and attribution reasoning is performed on the external covering (such as functional attribute words characterizing the component's purpose) extracted from its attribute set. If the probability weight score pointing to the wall surface classification (such as the target classification label) in the semantic mapping feature values ​​is the highest and meets the preset classification judgment threshold (such as a value range above 0.85), then the action of mapping the component type under the Building Information Model Data Exchange Standard to the wall surface classification under the Urban Information Model Standard is executed. By performing the above cross-standard logical mapping, the business identity of the custom component in the macro-urban foundation is established, thereby obtaining the classification judgment result. This process utilizes probabilistic inference technology to achieve an adaptive conversion from local fine-grained design semantics to global urban management semantics, supporting the consistency of cross-scale data models at the logical layer.

[0067] Preferably, in the specific technical implementation of step 2, structured feature encapsulation and globally unique identifier allocation are performed on the obtained classification judgment result. For example, based on the classification judgment result, a coded identifier conforming to the city information model standard (such as city-level asset management specifications) is assigned to the corresponding component instance, and its original attribute data in the original building information model is mapped and bound to the newly generated classification label. By synchronously associating the mapped semantic description information with the spatial pose features in the aligned initial 3D modeling object, semantic classification information of the city information model that can be called by the city digital twin platform is generated. This processing step alleviates the semantic fragmentation caused by non-standard naming of custom components, achieves "name-real alignment" of micro-components in the macro-geographic scene, and realizes the technical benefit of enabling subsequent city-level analysis applications to identify and access each micro-building component.

[0068] Optionally, in step 2, obtaining the urban information building shell model includes: performing columnar pixelation on the aligned initial 3D modeling object to identify the top surface of the building, and emitting probe rays upward along the height axis to extract roof feature patches; using the roof extraction operator to perform detail level generalization processing on the roof feature patches to simplify the topological structure of the roof feature patches, thereby obtaining the urban information building shell model.

[0069] Preferably, step 2 is implemented as follows: Spatial discretization is performed on the obtained aligned initial 3D modeling object. A columnar pixelation algorithm (i.e., voxelization based on vertical columns) is used to transform the complex geometric boundaries of the building information model into a set of columnar volumes with regular grid features. During processing, the spatial occupancy of each vertical column voxel along the height axis is analyzed, identifying the voxel node at the highest elevation in each vertical column, and extracting the geometric patch information associated with the voxel node to generate a candidate set of the topmost surface representing the building's roof coverage area. This processing step simplifies the traversal logic of massive geometric patches using a discretized spatial index structure, providing a data filtering mechanism at the computational level for locking the macroscopic boundaries of the building roof.

[0070] Preferably, in the specific implementation of step 2, a ray projection extraction process based on spatial visibility is performed on the generated candidate set of topmost surfaces. For example, using the geometric centroid of each geometric facet in the candidate set of topmost surfaces as the emission starting point, a probe ray is emitted along the height axis towards the zenith, and the spatial intersection state of the probe ray with the remaining component faces in the aligned initial 3D modeling object is monitored in real time. The action of retaining the geometric faces without occlusion interference as core roof components is performed, thereby effectively eliminating non-exposed faces located inside the building, in mezzanine floors, or covered by other obstructions, generating preliminary roof feature faces with global spatial visibility. This process, through physical visibility simulation logic, ensures that the extraction results conform to the observation characteristics from a macro-urban modeling perspective, mitigating the interference of redundant internal geometry on the building shell construction.

[0071] Preferably, in one scenario, step 2 is specifically implemented as follows: The designed roof extraction operator is used to perform geometric generalization and patch fitting processing on the obtained preliminary roof feature patches. The roof extraction operator is equivalent to performing spatial clustering and coplanarity constraint operations on a discrete set of patches. By identifying adjacent patches whose normal vector deviation is within a preset threshold range (e.g., 0.1 to 0.3 radians), they are merged into a unified roof planar topological unit. Based on this, boundary simplification and topological void filling processing are performed on the merged roof planar topological unit to repair geometric discontinuities caused by seams or minor structural protrusions in the original building components, generating a fitted roof geometry with semantic integrity. This achieves a cross-scale transformation from microscopic fine component geometry to macroscopic regular geometric forms, improving the model's ability to maintain robust roof contour features even when scaling at the detail level.

[0072] Preferably, in the specific technical implementation of step 2, the generated fitted roof geometry and the vertical facade boundary extracted from the aligned initial 3D modeling object are subjected to 3D stitching processing, and multi-level detail generalization processing (i.e., geometric simplification based on the display accuracy requirements of the city base) is performed on the stitched overall structure. By reducing the vertex distribution density of the fitted roof geometry and the vertical facade boundary, and removing small-scale bump textures that do not affect macroscopic visual features, a 3D envelope model of the building that meets the accuracy requirements of specific geographic information is generated, thereby obtaining the building shell model of the city information model. This processing result not only achieves lightweight storage of massive data at the geometric level, but also maintains the spatial connectivity and closure of the building body at the topological level, achieving the technical benefit of enabling efficient and smooth execution of subsequent city-level large-scale scene rendering and spatial analysis tasks.

[0073] Optionally, step 3 includes: extracting the opening filling constraint relationship from the building information model component space diagram to identify opening components belonging to the door and window category, and using the opening projection operator to calculate the normal projection position of the opening component relative to the urban information building shell model to generate opening position parameters; performing Boolean operations on the surface of the urban information building shell model based on the opening position parameters to map and generate the set of geometric patches with opening features.

[0074] Preferably, the specific implementation process of step 3 is as follows: Topological attribute parsing is performed on the pre-constructed building information model component spatial diagram to retrieve the opening filling constraint relationship, which represents the physical nesting and functional occupancy relationship between components, and uses this as a key logical index. Semantic feature comparison and object filtering are performed on the three-dimensional geometric patch set of the building components using the opening filling constraint relationship. By identifying component type labels that satisfy the filling attribute definition (such as labels marked as door components or window components), micro-component entities belonging to the door and window categories are separated from the original massive patch data to obtain the target opening component set. This processing step utilizes the edge constraint attribute in graph theory to establish a logical association between micro-functional components and the macro-host wall, providing deterministic data guidance for subsequent feature placement on the generalized building shell.

[0075] Preferably, in the specific technical implementation of step 3, the designed opening projection operator is used to perform spatial coordinate mapping processing on the obtained set of target opening components and the generated urban information building shell model. For example, using the geometric centroid and its bounding rectangle in three-dimensional space of each target opening component as the projection source, the coordinates of the intersection point of the projection source along its normal vector direction pointing to the corresponding outer surface of the urban information building shell model are calculated. By performing a normalized mapping operation on the intersection point coordinates, the relative position information of the target opening component on the simplified shell surface is determined, thereby obtaining the opening position parameters. This mapping process realizes the dimensionality reduction transfer of geometric features from complex and detailed components to abstract shell surfaces, effectively solving the problem of opening feature loss or positional deviation that easily occurs during cross-scale processes (such as conversion from millimeter-level detail to meter-level detail).

[0076] Preferably, in one scenario, step 3 is specifically implemented as follows: Three-dimensional Boolean operations are performed on the geometric plane corresponding to the generated urban information building shell model using the obtained opening location parameters. For example, based on the opening width, opening height, and center offset (such as the longitudinal and transverse coordinates of the center point in the wall coordinate system) described by the opening location parameters, geometric clipping and region culling are performed on the wall or roof panel corresponding to the urban information building shell model. After performing the geometric clipping and region culling, the originally continuous and closed geometric panels are transformed into geometric structures with physical cavity characteristics, thus obtaining a set of geometric panels with opening characteristics. This processing step, through subtraction operations performed on the generalized geometry, objectively restores the key visual and functional opening features of the building while maintaining the model's lightweight nature, improving the high fidelity of the model within the urban-level base.

[0077] Preferably, in the specific technical implementation of step 3, boundary topology alignment and geometric error compensation processing are performed on the obtained set of geometric facets with opening features. By retrieving the adjacency relationship description in the spatial diagram of the building information model component, the normal consistency of each opening edge in the set of geometric facets with opening features is checked. If it is determined that a small non-orthogonal deviation (such as an angle deviation within 0.02 radians) has occurred due to coordinate transformation or geometric generalization, angle correction compensation is performed to maintain the 90-degree corner feature and edge alignment feature of the set of geometric facets with opening features in three-dimensional space, thereby obtaining a topology-corrected set of geometric facets with opening features. The implementation of this step improves the model after cross-scale mapping, ensuring that it not only maintains consistency with the original model in appearance but also possesses rigorous topological rules in its underlying geometric description, achieving the technical benefit of providing a high-quality geometric base for generating target components of the urban information model with topological closure in subsequent steps.

[0078] Optionally, in step 3, generating the target component of the city information model with topological closure includes: using the semantic classification information of the city information model to identify the component hierarchy relationship between the city information building shell model and the set of geometric facets with opening features; based on the component hierarchy relationship and combined with the opening filling constraint relationship, performing facet stitching processing on the city information building shell model and the set of geometric facets with opening features to construct a closed geometry that matches the manifold constraint; performing watertightness detection on the closed geometry to identify and close geometric gaps, thereby obtaining the target component of the city information model with topological closure.

[0079] Preferably, the specific implementation process for generating the target component of the urban information model with topological closure in step 3 is as follows: The obtained semantic classification information of the urban information model is parsed and used as an association index to identify the component hierarchy relationship between the urban information building shell model and the set of geometric facets with opening features. By matching the attribute labels carried in each geometric facet node with the semantic encoding of the main shell model, the logical subordinate relationship of geometric objects at different detail levels is determined (e.g., determining that window opening feature facets belong to wall shell units in three-dimensional space), thereby obtaining the component hierarchy relationship. This processing step establishes a logical link between the macroscopic model envelope and the microscopic opening details in the semantic dimension, providing a clear basis and technical support for subsequent cross-scale geometric stitching.

[0080] Preferably, in the specific technical implementation of step 3, which generates the target component of the urban information model with topological closure, the opening-filling constraint relationship in the spatial diagram of the building information model component is retrieved, and combined with the component hierarchical belonging relationship, patch stitching and Boolean operation processing are performed on the urban information building shell model and the set of geometric patches with opening features. For example, using the adjacency logic defined in the topological edge set, the boundary vertices of the set of geometric patches with opening features are aligned with the corresponding opening edge vertices on the urban information building shell model using co-position alignment and geometric welding operations. By performing the patch stitching process, the originally discrete geometric description is integrated into a continuous three-dimensional surface structure, thereby constructing a closed geometry that matches manifold constraints. This step achieves physical unification of geometric form, effectively alleviating the common phenomenon of suspended or logically isolated geometric patches during cross-scale generalization, resulting in a model with high structural robustness.

[0081] Preferably, in one scenario, step 3 is specifically implemented as follows: Manifold constraint verification and watertightness detection are performed on the generated closed geometry. For example, the number of shared faces of each topological edge in the closed geometry is checked to verify whether the criterion that each edge in the manifold structure is shared by only two geometric faces is met. Simultaneously, watertightness detection is performed to identify minute geometric gaps on the surface of the closed geometry caused by floating-point operation errors or defects in the original Building Information Model Data Exchange Standard Model. If a non-closed region is detected, a local topology repair operator is invoked to perform face completion and automatic gap closure. After performing the repair and closure actions, a physically logically rigorous, non-self-intersecting, and completely closed geometric entity is generated, thus obtaining a target component of the urban information model with topological closure. This improves the physical realism of the generated model, enabling it to support subsequent complex spatial Boolean queries, volume calculations, or solar simulation analysis within the urban information model base.

[0082] Preferably, the specific implementation process of performing the manifold constraint verification and watertightness detection in step 3 is as follows: A depth-first traversal of the geometric topological connections is performed on the generated closed geometry. By parsing the edge index of each 3D facet, the number of geometric faces associated with each topological edge in the closed geometry is counted. If it is determined that the number of shared faces with a specific topological edge is not equal to two (e.g., a boundary edge shared by only one facet), then a manifold missing defect is identified in that region, and the normal discontinuity of the boundary edge is calculated simultaneously to determine the spatial distribution of geometric gaps and thus obtain a sequence of non-closed boundary edges. This processing step, starting from the connectivity of the geometric bottom layer, locks down logical breakpoints that are physically imprecise in the 3D entity, providing a clear computational object for subsequent invocation of targeted repair mechanisms.

[0083] Preferably, in the specific technical implementation of calling the local topology repair operator in step 3, the obtained non-closed boundary edge sequence is used to perform the tracking and extraction of the stitching path. For example, starting from any boundary edge in the sequence, a cyclic search is performed in three-dimensional space according to the vertex sharing criterion. By sequentially connecting the first and last non-closed boundary edges, a closed contour line that can surround the geometric gap is constructed, thus obtaining the gap boundary topological loop. This step transforms discrete boundary anomalies into a spatially ordered ring structure, realizing the macroscopic logical definition of the "damaged area" inside the three-dimensional entity, and improving the ability of subsequent geometric completion actions to act on the target damaged location.

[0084] Preferably, in one scenario, the specific implementation process of the local topology repair operator is as follows: Geometric completion processing based on the minimum area criterion is performed on the obtained gap boundary topological ring. For example, the centroid coordinates of the vertex set contained in the gap boundary topological ring are calculated, and a new set of geometric patches is generated within the ring-shaped region using a triangulation algorithm (such as Delaunay triangulation). By geometrically stitching the generated new geometric patches with the gap boundary topological ring, the original tiny geometric gaps are physically filled, thus obtaining a topology repair patch set. This processing step uses geometric interpolation technology to repair geometric holes caused by floating-point operation errors or defects in the original Building Information Modeling (BIM) data export, alleviating logical omissions on the model surface.

[0085] Preferably, in the specific technical implementation of step 3, which generates the target component of the city information model with topological closure, the generated topological repair patch set and the original closed geometry are subjected to topological fusion and secondary verification processing. For example, the manifold constraint verification is performed again on the fused 3D entity to verify whether the number of shared faces of all topological edges is equal to two, and to confirm whether there is any facet self-intersection. If it is determined that the manifold structure criterion is met and the watertightness detection is passed, then the current geometric entity is determined to be a completely closed and physically rigorous structure, thus obtaining the target component of the city information model with topological closure (its corresponding English full name is City Information Modeling, abbreviated as city information model). Therefore, by automatically identifying and high-fidelity repairing local topological defects, the technical benefit of improving the robustness of the calculation results when the model performs volume calculation, spatial conflict detection, and physical analysis tasks within the macroscopic city base is achieved.

[0086] Preferably, in the specific technical implementation of step 3, which generates the target component of the city information model with topological closure, the adjacency relationship description in the spatial graph of the building information model component is used to establish a spatial association between the target component of the city information model with topological closure and external urban elements (such as land parcel boundaries, adjacent road centerlines, or terrain features) in the city information model base to which it belongs. Simultaneously, a mapping record is executed between the globally unique identifier of the target component of the city information model with topological closure and the identity identifier of the corresponding component in the original building information model, thereby obtaining a cross-scale association mapping index. By establishing this association mapping index, not only is the topological pose of the building component locked in the macro-city base, but also, by opening up the identity recognition link between the micro-model and the macro-base, a reliable path guide is provided for the subsequent step 4, which performs object-level incremental updates and data maintenance based on the consistency verification report, achieving the technical benefit of improving the long-term data operation and maintenance efficiency.

[0087] Optionally, in step 4, calculating the alignment deviation of the target component of the city information model relative to the base of the city information model includes: calculating the spatial deviation and area error value between the geometric center point coordinates and envelope area of ​​the projected footprint and the center point coordinates and corresponding legal area of ​​the determined red line boundary; calculating the classification rate of the semantic classification information of the city information model; and generating a confidence score characterizing the alignment degree based on the spatial deviation, the area error value, the classification rate and the verification results of the watertightness test.

[0088] Preferably, the specific implementation process of step 4 is as follows: A comparative analysis of the geometrical alignment consistency is performed on the generated target components of the urban information model. For example, the geometrical center coordinates of the projected footprint of the initial 3D modeling object on the horizontal geographic plane, and the center point coordinates of the redline boundary retrieved from the underlying geographic information system are extracted. The Euclidean distance between the geometrical center coordinates of the projected footprint and the center point coordinates of the redline boundary is calculated to generate a value representing the horizontal displacement deviation, thus obtaining the spatial deviation. Simultaneously, the envelope area of ​​the projected footprint is calculated and compared with the legal area value recorded in the redline boundary data to generate a value representing the degree of area fitting, thus obtaining the area error value. This process, through dual quantitative analysis of planar location and coverage scale, provides a highly accurate geometric basis for evaluating the spatial rationality of the digital model within the land parcel redline.

[0089] Preferably, in the specific technical implementation of step 4, the obtained classification judgment result is used to perform a statistical evaluation of the semantic reliability of the semantic classification information of the urban information model. For example, the semantic labels carried by the target components of the urban information model are logically compared with the predefined standard classification library in the urban information model standard, and the degree of matching between attribute keyword evidence and geometric feature evidence extracted during the semantic mapping process is combined. By statistically analyzing the proportion of components conforming to the standard semantic definition in the mapping results to the total number of components, a numerical value representing the quality of semantic transformation is generated, thus obtaining the classification rate. This step achieves a quantitative evaluation of the quality of semantic transformation from micro-level architectural design semantics to macro-level urban management semantics, which helps to identify possible semantic conflicts or classification errors in the cross-scale mapping process and improves the semantic consistency of the urban foundation data.

[0090] Preferably, in one scenario, step 4 is specifically implemented as follows: using the watertightness verification value generated in the topology reconstruction stage, the topological rigor of the target component of the city information model is verified. For example, it verifies whether the closed geometry corresponding to the target component of the city information model with topological closure satisfies manifold constraints (e.g., each topological edge is shared by only two geometric faces and there is no self-intersection), and outputs a Boolean value representing the verification passed state, thus obtaining the watertightness verification result. If the watertightness verification result indicates that the geometric entity is a completely closed and physically rigorous structure, it is determined that it possesses the topological foundation for performing subsequent city-level analysis tasks such as spatial queries and volume calculations, effectively avoiding the impact of non-closed geometric defects on the robustness of the digital twin base.

[0091] Preferably, the specific technical implementation process of step 4 is as follows: Based on the obtained spatial deviation, area error value, classification rate, and water tightness detection verification results, multi-dimensional weighted fusion calculation processing is performed. For example, normalization processing is performed on geometric deviation indicators and semantic topology indicators, and a weighted summation operation is performed using preset weight allocation parameters (e.g., 0.4 weight for geometric deviation, 0.3 weight for semantic rate, and 0.3 weight for topological tightness) to obtain a confidence score. Logical judgment is performed between the confidence score and a preset alignment quality threshold (e.g., a value range above 0.85). If the confidence score is determined to be higher than the alignment quality threshold, an incremental update action is triggered for the city information model base, synchronizing the city information model target components and their associated cross-scale association mapping tables to the base database. This closed-loop verification mechanism ensures that only high-quality, highly consistent modeling results can enter the city base, reducing data maintenance risks and improving the automation reliability of city-level modeling tasks.

[0092] Optionally, step 4 further includes: in response to the confidence score being greater than a preset threshold, using a geometric hash algorithm to compare the original building information model under different version timestamps to identify the set of changed components; establishing a cross-scale association mapping table containing the unique identifier of the set of changed components and the urban information model base object, and performing local remapping and data replacement on the urban information model base based on the cross-scale association mapping table.

[0093] Specifically, in this application, the identification process for the set of changed components is as follows: When the obtained confidence score is higher than a preset quality criterion threshold (e.g., above 0.85), component-level geometric feature extraction is performed on the original building information model under different version timestamps. For example, a designed geometric hash algorithm (i.e., an algorithm that transforms a three-dimensional geometric topology into a fixed-length numerical fingerprint) is used to extract the three-dimensional facet connection relationships and curvature distribution features of each building component, and the curvature distribution features are transformed into hash fingerprints representing the uniqueness of the geometric shape. By performing logical comparison operations on the hash fingerprints in different time series, specific component instances whose geometric shapes have changed due to engineering changes are identified, thus obtaining the set of changed components. This process bypasses the cumbersome facet matching logic and achieves rapid location of the change target with low computational overhead in a massive component environment.

[0094] Preferably, in this application, the specific implementation process of component-level feature extraction for the geometric hash algorithm is as follows: Low-level geometric conformation analysis is performed on the original building information model loaded at different time series. By traversing the three-dimensional mesh data of each building component, the connection relationships of the three-dimensional facets representing the essential geometric attributes and the curvature distribution features of the vertices of each facet are extracted. During the processing, the changes in the normal angle between adjacent facets and the local curvature of the vertices are analyzed and transformed into a numerical sequence that can describe the surface undulation features of the component, thereby obtaining the original geometric feature descriptor. This processing step abstracts the cumbersome set of three-dimensional facets into feature vectors with physical meaning. While preserving the essential form of the component, it alleviates the interference caused by coordinate offset or redundant vertices on form recognition, providing a low-level feature foundation for subsequent cross-time series form consistency verification.

[0095] Preferably, in the specific technical implementation of the geometric hash algorithm, the extracted original geometric feature descriptors are used to perform spatial transformation-independent fingerprint mapping processing. A mathematical mapping function with translation and rotation invariance is constructed based on the geometric hash algorithm to map the high-dimensional geometric patch connections and curvature distribution features to a fixed numerical space (such as a binary vector space with a preset bit width). For example, spatial histogram statistical operations are performed on the original geometric feature descriptors to capture the frequency of component distribution within a specific curvature range, and a hash function is used to transform the statistically obtained distribution features into a fixed-length numerical sequence to obtain the hash fingerprint. This step achieves the essential compression from massive geometric data to compact digital identification, enabling each component with a unique geometric shape to possess a unique and quickly searchable digital identity label.

[0096] Preferably, in one scenario, the specific technical implementation of the geometric hash algorithm for cross-temporal morphological comparison is as follows: For the hash fingerprint extracted at the first moment and the hash fingerprint extracted at the second moment, logical comparison operations are performed in the corresponding computational memory regions. For example, bitwise XOR operations or numerical subtraction logic are used to check the numerical differences between the hash fingerprints at two different timestamps bit by bit. If the values ​​are determined to be completely identical, it is identified that the geometric shape of the component has not changed during the engineering change process; if a numerical difference is determined and the numerical difference exceeds the range of small disturbances caused by floating-point operations, it is identified that the physical conformation of the component has changed, thus obtaining a component morphological change identifier. This process bypasses the extremely time-consuming point cloud registration or full-scale surface comparison logic, greatly improving the processing efficiency of locating changed targets in large-scale urban scenarios while ensuring recognition reliability.

[0097] Preferably, in this application, the identification process for the set of changed components is as follows: The component morphology change identifiers corresponding to all the digital nodes are integrated, and associated aggregation processing is performed in conjunction with the unique identifiers of the components recorded in the building information model component space diagram. By extracting all component instances marked as having morphological changes, and performing structured encapsulation of their corresponding globally unique identifiers, floor attributes, and mapped semantic classification information, the set of changed components is obtained. This step transforms the micro-level differences at the geometric level into a business change list at the object level, providing a target guide for subsequent steps of targeted local remapping and data replacement of the city information model base. This geometric fingerprint-based identification mechanism improves the ability of the city's digital twin base to keenly capture subtle evolutions in architectural design, achieving the technical benefits of reducing data maintenance costs and improving real-time operation and maintenance.

[0098] Specifically, in this application, the construction technology for the cross-scale association mapping table is implemented as follows: For the identified set of changed components, the globally unique identifier (i.e., a globally unique identifier under the Building Information Modeling (BIM) data exchange standard) carried in each component node is parsed. Simultaneously, object mapping backtracking processing is performed in the corresponding city information model base database. By searching for city-level 3D spatial objects corresponding to the globally unique identifier, a one-to-one mapping relationship between micro-level building components and macro-level city model objects is established. The generated mapping relationship is recorded in a structured data table, thus obtaining the cross-scale association mapping table. By making this relationship explicit, the ability to hit target data nodes is improved when performing subsequent local data replacement, mitigating the risk of index loss due to cross-scale data conversion.

[0099] Preferably, the specific implementation process of object mapping backtracking is as follows: For the loaded set of modified components, perform low-level data parsing processing. By calling the attribute parsing logic corresponding to the Building Information Modeling (BIM) data exchange standard (such as Industry Foundation Classes), extract the globally unique identifier carried in each component node. For example, retrieve the unique encoding field from the component attribute dictionary and convert it into a character constant that can be quickly addressed in computing memory to obtain the component identifier index sequence. This processing step utilizes the native and permanent "identity" of the digital component as a logical anchor, providing a unique data credential for subsequent cross-dimensional object retrieval in the massive data of the macro-level foundation.

[0100] Preferably, in the specific technical implementation of the object mapping backtracking process in step 4, the obtained component identifier index sequence is used to perform a deep association retrieval process with the corresponding City Information Modeling (City Information Modeling) base database. The object mapping backtracking process utilizes a pre-built object association index table in the City Information Modeling base database (where the index key is the micro-component identifier and the corresponding content is the physical storage path of the macro-base object) to perform a storage address backtracking operation based on identifier matching. By performing a global search in the logical tree structure of the database, macro-level 3D model objects that perfectly match each globally unique identifier at the logical level are identified, thus obtaining the base target association node. This step achieves the redirection from the micro-change list to the macro-base storage location, mitigating the logical index breakage defect that may be caused by cross-scale generalization processing.

[0101] Preferably, in one scenario, the specific implementation process of constructing the association mapping is as follows: A cross-scale one-to-one mapping association process is performed on the identified base target association node and its corresponding component instance in the original building information model. For example, a logical binding relationship representing the consistency of physical entities is established, and the engineering attribute metadata of the micro-component is paired with the geometric data pointer of the macro-base object. Specifically, a record item with topological association characteristics is created in the intermediate scheduling layer, recording the correspondence between the two, spatial envelope consistency parameters, and the latest version timestamp attribute to obtain the cross-scale object association body. Therefore, at the logical level, efficient stitching of the micro-entities inside the building and the macro-objects of the urban base is achieved, improving the execution context of subsequent update actions.

[0102] Preferably, in the specific technical implementation of step 4, constructing the cross-scale association mapping table, structured encapsulation and persistent writing processing are performed on the generated full set of cross-scale object associations. For example, the cross-scale object associations are written into a structured data table according to a preset relational storage specification, and a unique hash index key is configured for each mapping entry to accelerate access, thus obtaining the cross-scale association mapping table. This processing result realizes an explicit digital representation of the physical entity relationships within a building in a cross-scale scenario. Through the logical guidance provided by this table, the ability to hit the corresponding target data node is improved when performing subsequent local data replacement actions. Therefore, this processing method alleviates the risk of index loss caused by cross-scale data conversion, achieving the technical benefit of improving the response speed and data placement reliability of the urban information model base when performing incremental update tasks.

[0103] Specifically, in this application, the implementation of local remapping is as follows: For each component instance locked by the cross-scale association mapping table, the processing logic described in steps 2 and 3 is retrieved again to perform localized iteration. For example, the latest custom family attributes in the changed component set are parsed, the corresponding semantic labels are updated using the designed probabilistic mapping model, and simultaneously, columnar pixelation and roof extraction operations are performed on the latest geometric contour of the component to obtain the updated urban information model building shell model. Subsequently, the latest opening features are mapped to the urban information model building shell model using the opening projection operator, and watertight topology reconstruction is performed in conjunction with the building information model component spatial map to obtain the updated urban information model target component with topological closure. This on-demand reconstruction processing mode avoids redundant calculation of the entire dataset and improves the timeliness of dynamic maintenance of the urban base.

[0104] Preferably, in the specific technical implementation of the probability mapping model, semantic feature words are parsed and evidence weight quantification is performed on the custom family attributes extracted from the set of changed components. For example, unstructured text descriptions (such as component names, type names, material descriptions, etc.) contained in the custom family attributes are traversed and parsed, and keyword evidence representing the core functions of the components is identified using a preset professional terminology dictionary. Subsequently, based on the prior frequency of each keyword evidence pointing to a specific component type in the field of architectural engineering, a numerical intelligence contribution weight is assigned to each keyword evidence to construct a discrete weight sequence representing the semantic features of the changed component, thereby obtaining a semantic feature evidence vector. This processing step transforms natural language attributes, which are originally difficult for the system to directly compare, into a quantified vector with mathematical operation characteristics, providing an objective computational benchmark for subsequent cross-scale automated classification and judgment.

[0105] Preferably, in the specific implementation logic of the probability mapping model, the obtained semantic feature evidence vector is used to perform multi-dimensional probability density calculation processing within a pre-set classification knowledge base space. The probability mapping model is equivalent to constructing a full probability matrix representing the correspondence between the building information model (BIM) component standards and the city information model (CIM) standards. For example, the posterior probability distribution values ​​of the semantic feature evidence vector under multiple candidate city information model classification labels are calculated using Bayesian inference logic or conditional probability distribution algorithms. By comprehensively measuring the joint support strength of different keyword evidence combinations for a specific classification target, a classification probability distribution curve is generated. This step achieves a statistical approximation from fuzzy engineering descriptions to urban semantic classification, alleviating the attribute recognition conflict problem caused by inconsistent modeling standards.

[0106] Preferably, in one scenario, a decision-making process based on the maximum likelihood criterion is performed on the classification probability distribution curve. For example, the candidate classification label with the highest probability value is searched from the classification probability distribution curve, and it is verified whether this highest probability value exceeds a preset classification confidence threshold (e.g., a value range of 0.75 to 0.95). If the determination meets the classification confidence threshold requirement, the original semantic description of the currently modified component is replaced with the corresponding standard urban information model semantic label, and the determination result is simultaneously fed back to the topology reconstruction logic in step 3 to obtain the updated urban information model semantic classification information. This process, by introducing a probabilistic statistical mechanism to replace the traditional rigid rule comparison, improves the system's robustness in recognizing non-standard custom attributes and enhances the transmission and updating of semantic information during cross-scale mapping.

[0107] Preferably, based on the obtained updated semantic classification information of the urban information model, the columnar pixelation processing and the roof extraction operation are coordinated to perform localized geometric and semantic fusion processing for the changed object. For example, the updated semantic classification information is used as a logical constraint for topology reconstruction, driving the system to perform refined repair and patch stitching operations on the latest geometric contour of the changed component. During the processing, the latest opening features (such as the geometric boundaries of window or door openings) are mapped onto the generalized model shell using an opening projection operator, and watertightness detection is performed to eliminate geometric gaps, thereby obtaining an updated urban information model target component with topological closure. This processing result achieves synchronization of the changed component from underlying semantics to macroscopic form, achieving the technical benefits of performing local model evolution on demand, avoiding redundant calculation of all data, and improving the timeliness of dynamic maintenance of the urban base.

[0108] Specifically, the data replacement process in this application is as follows: Using the generated cross-scale association mapping table as the data routing basis, the updated target components of the urban information model with topological closure are sent to the target storage area of ​​the urban information model base. The globally unique identifiers of each component in the changed component set are parsed, and the corresponding geometric data nodes are retrieved in the data tree structure of the urban information model base. An overlay update is performed on the geometric data nodes using the newly generated geometric data, thereby enabling the urban digital base to achieve consistent synchronization of local building forms and attributes while maintaining the stability of historical data, thus obtaining an incrementally synchronized urban information model base. This solution improves the ability of the urban digital twin model to evolve in real time as the project progresses, achieving the technical benefit of reducing the overall system operation and maintenance pressure.

[0109] Optionally, the local remapping includes: for the changed component set, regenerating the corresponding urban information model semantic classification information using the parsed custom family attributes, and re-performing the geometric generalization processing and the mapping processing of the opening features on the geometric outer contour corresponding to the changed component set based on the updated semantic classification information, so as to generate an updated urban information model target component with topological closure.

[0110] Preferably, the specific implementation process of the local remapping is as follows: For the identified set of changed components, by parsing the globally unique identifier carried by each component instance, the latest version of the custom family attribute corresponding to it is retrieved from the original database. For the extracted attribute feature words (such as text fields describing the function of the component), semantic reclassification calculation is performed using a pre-built probability mapping model. By comparing the semantic association probability matrix, the set of changed components is assigned the latest semantic labels that conform to urban management standards, thereby obtaining the updated semantic classification information of the urban information model. This improves the ability of the corresponding macroscopic semantic description to evolve synchronously when the physical function or business attributes of internal building components undergo design changes, mitigating the risk of semantic consistency loss due to attribute update lag.

[0111] Preferably, in the specific technical implementation of the local remapping, spatial dimensionality reduction and geometric generalization processing are performed on the latest three-dimensional geometric description of each component in the changed component set. For example, the columnar pixelation processing logic is restarted, transforming the updated component outline into a set of columnar bodies with discrete features. Subsequently, a ray projection extraction operation is performed along the height axis to lock the latest top surface of the building, and the roof extraction operator is invoked to perform patch fitting and topology simplification processing. After performing the fitting and simplification processing, the updated urban information model building shell model is obtained. Therefore, by generalizing the changed target on demand, lightweight reconstruction of local geometric changes is achieved while ensuring the fidelity of the macroscopic outline, supporting efficient synchronization of cross-scale data models at the geometric layer.

[0112] Preferably, the specific implementation process of the roof extraction operator is as follows: Vertical spatial occupancy analysis is performed on the obtained set of columnar bodies with discrete characteristics. For example, within the horizontal projection grid of a three-dimensional Cartesian coordinate system, the vertical columnar region corresponding to each grid cell is traversed, and the voxel node with the maximum elevation value along the height axis within that region is locked. By identifying the original geometric patch information carried by the voxel node, non-exposed geometric features such as floor slabs, beams, and columns inside the building are filtered out to obtain a candidate roof patch set. This processing step uses spatial discretization to abstract the extremely complex building top-level structure into a discrete lattice with regular topological patterns, providing a data index foundation for subsequent cross-scale geometric patch fitting.

[0113] Preferably, in the specific technical implementation of the roof extraction operator, the ray projection extraction action is used to perform spatial visibility and occlusion culling processing on the obtained candidate roof patch set. Specifically, starting from the geometric centroid of each candidate patch, a probe ray is emitted along the height axis towards the zenith. If it is determined that the probe ray does not collide with other components in the Building Information Model (BIM) within a preset detection distance, the patch is retained as a valid roof component, thus obtaining the exposed roof feature surface. This step, through physical collision detection logic, extracts the feature geometry that represents the macroscopic top outline of the building from the microscopic model, alleviating the problem of identifying "false roofs" caused by model overlap or nesting.

[0114] Preferably, in one scenario, the specific implementation process of the roof extraction operator is as follows: Geometric patch fitting based on coplanar constraints is performed on the obtained exposed roof feature surfaces. The roof extraction operator clusters scattered feature surfaces into planar or curved surface descriptions with unified algebraic equations by identifying the normal vector deviation between adjacent geometric patches. For example, the dihedral angle value of each group of adjacent patches is calculated. If the value is within a preset coplanar threshold range (e.g., from zero degrees to five radians), a patch merging operation is performed, and the boundary contour of the region is recalculated using the least squares fitting method to obtain the fitted roof surface. This achieves an essential transformation from discrete, fragmented, fine-grained component geometry to continuous, regular macroscopic geometric units, greatly simplifying the geometric complexity of the model.

[0115] Preferably, in the specific technical implementation of the roof extraction operator, topological simplification and boundary normalization are performed on the obtained fitted roof surface. For example, a polygon simplification algorithm is used to remove non-critical vertices from the edges of the fitted roof surface, and geometric stitching and topological closure operations are performed on the vertical facade of the building shell. By eliminating minor jagged edges and gaps caused by differences in the dimensions of local components, a shell structure with smooth geometric surfaces, logical coherence, and macroscopic recognizability is produced, thus obtaining the updated City Information Model (CIM) building shell model. Therefore, by generalizing the geometric features of changed components on demand, lightweight reconstruction of local geometric changes is achieved while ensuring the fidelity of the macroscopic outline, achieving the technical benefits of reducing the rendering load of the city digital twin base and supporting cross-scale data synchronization.

[0116] Preferably, in one scenario, the local remapping is specifically implemented as follows: The updated city information model building shell model is used as the bearing reference, and synchronous remapping processing of opening features is performed. For example, the relationships associated with the currently changed components are retrieved from the building information model component space diagram, and the geometric distribution of window or door components in the new version model is identified. The opening projection operator is invoked to calculate the normal projection trajectory of the window or door component relative to the updated city information model building shell model, thereby generating the latest opening position parameters on the shell surface. Based on the opening position parameters, a geometric trimming operation is performed to obtain an updated set of geometric patches with opening features. This step, through a geometric transfer mechanism, improves the real-time positioning of the opening logic on the building facade as the design changes.

[0117] Preferably, in the specific technical implementation of the local remapping, the opening filling constraint performs joint topological reorganization processing on the updated urban information model semantic classification information, the updated urban information model building shell model, and the updated set of geometric faces with opening features. For example, combining the adjacency constraint logic defined in the building information model component space graph, the updated urban information model building shell model and the corresponding updated set of geometric faces with opening features are subjected to edge welding and face stitching operations. Simultaneously, watertightness detection and manifold constraint verification are performed. By automatically closing minute gaps, physically logically rigorous and completely closed geometric entities are generated, thus obtaining the updated urban information model target component with topological closure. Therefore, a local full-closed-loop iteration from semantic mapping to geometric generation is achieved, improving the ability of incrementally updated components to be tightly stitched into the city-level base scenario, achieving the technical benefit of improving the long-term operation and maintenance quality of the digital twin base.

[0118] Optionally, the data replacement includes: extracting the unique identifier of each component in the set of changed components based on the cross-scale association mapping table, retrieving the corresponding geometric data node in the urban information model base, and performing overlay data writing on the geometric data node using the updated urban information model target component with topological closure.

[0119] Preferably, the specific implementation process of the data replacement is as follows: Low-level index parsing is performed on the generated cross-scale association mapping table. A globally unique identifier (i.e., a logically unique code under the Building Information Modeling (BIM) data exchange standard) carried by each component instance in the changed component set is extracted as a key retrieval element for locating the target object in the macro-level base. The distributed spatial database corresponding to the city information model base is accessed synchronously, and an object-level path search is performed in the logical tree structure of the database using the globally unique identifier. After performing the path search, the data storage unit that completely corresponds to the changed component set in terms of geospatial and business attributes is located in the massive city-level data, thus obtaining the target geometric data node. This processing step establishes a mapping path between micro-level change information and the macro-level base storage architecture, providing a definite operational target for subsequent local coverage.

[0120] Preferably, in the specific technical implementation of the data replacement, the obtained target geometric data node is used to perform logical unloading processing of the old version of geometric description and attribute load on the urban information model base. For example, the memory address or disk storage offset of the target geometric data node is parsed, and according to a preset data replacement protocol, cleanup or invalidation actions are performed on the historical 3D patch topology relationships, detail level model data, and associated semantic tags under that node to obtain a storage node in a write-ready state. This step, through the orderly cleanup of redundant historical data, mitigates the risk of rendering ghosting or topology conflicts that may be caused by overlapping data versions, improves the purity of the base data environment, and provides a physical basis for the subsequent migration of high-quality data.

[0121] Preferably, in the specific technical implementation of step 4, the historical 3D patch topological relationships are parsed and decoupled for the locked target geometric data node. The historical 3D patch topological relationships are equivalent to storing the connectivity indexes between vertices, edges, and patches in the 3D mesh model, typically represented as a topological association matrix describing the adjacency state of patches. For example, the topological association matrix stored in the target geometric data node is traversed to identify the shared edge constraints and vertex index mapping relationships between various old version patches. The logical links between elements in the topological association matrix are severed, logically restoring the originally interconnected geometric patches to isolated discrete point sets, thus obtaining topologically decoupled geometric fragments. This processing step dismantles the old 3D structural skeleton from the underlying logic, alleviating structural interference for the subsequent migration of new geometric topological logic.

[0122] Preferably, in the specific implementation logic of the logic unloading process, for the obtained topologically decoupled geometric fragments, a multi-scale cache cleanup process is performed on the Level of Detail (LOD) model data. The LOD model data is equivalent to constructing a geometric precision pyramid that dynamically changes with the observation distance, containing multiple discrete precision levels from macroscopic blocky outlines to microscopic fine components. For example, sub-data blocks corresponding to different precision levels in the target geometric data node are parsed, and the vertex density parameters and texture map indexes corresponding to each precision level are obtained. The sub-data blocks are removed from the current display list and preloading queue according to the data replacement protocol, thereby obtaining geometric nodes with empty visual loads. This step, through the orderly cleanup of models at different resolutions, effectively solves the risk of rendering ghosting and memory overflow caused by the coexistence of multiple LOD models in large urban scene rendering, improving display stability during model updates.

[0123] Preferably, in one scenario, step 4 is specifically implemented as follows: For the geometric nodes with empty visual loads, metadata tagging invalidation processing is performed on the associated semantic tags. The associated semantic tags are equivalent to establishing a dictionary of mapping relationships between 3D geometric entities and business attributes, where the key value is usually a globally unique identifier of the component, and the value is a semantic classification code conforming to the City Information Modeling (CIM) standard. For example, the attribute description list attached to the target geometric data node is retrieved, and all historical semantic codes representing the component's function are identified. The logical status field of the historical semantic code is modified to "invalid" or "expired," thereby shielding the interference of old version semantics on the system query task at the logical level and obtaining the geometric nodes with invalid semantic load tags. This achieves synchronous unbinding of geometry and business semantics, improving the model from semantic confusion or attribute attachment errors during updates.

[0124] Preferably, in the specific technical implementation of step 4, which generates the storage node in the write-ready state, the memory address or disk storage offset of the parsed target geometric data node is used to perform physical reset and verification processing of the storage space for the geometric node whose semantic payload has been marked as invalid. For example, based on the memory address or disk storage offset, a zeroing operation or space reclamation instruction is performed on the storage cluster occupied by the unloaded data, and the addressing index table of the database is updated synchronously. Subsequently, it is verified whether the storage space meets the capacity quota required to store the updated target component of the city information model with topological closure. If it is determined that the storage space has been completely released and the logical state is idle, the current storage location is marked as available, thus obtaining the storage node in the write-ready state. Therefore, by thoroughly unloading historical topology, multi-precision models, and semantic labels, the technical benefit of providing a clean physical foundation for the subsequent migration of high-quality, cross-scale consistent data is achieved.

[0125] Preferably, in one scenario, the data replacement is specifically implemented as follows: Using the newly generated urban information model target component with topological closure, an overlay data writing process is performed on the storage node in the write-ready state. For example, the vertex coordinate array, facet connection table, and updated semantic attribute features contained in the urban information model target component with topological closure are subjected to structured serialization operations according to the standard format of the urban digital twin platform. By sequentially filling the serialized data stream into the storage node in the write-ready state and re-establishing the three-dimensional spatial index (such as an octree or dynamic spatial hierarchical index), real-time replacement of the geometric entities and attribute descriptions at the target location is achieved, thereby obtaining the incrementally synchronized local three-dimensional model node. This overlay writing mechanism effectively solves the problems of excessive data volume and long synchronization time in traditional full-scale reconstruction schemes, achieving the technical benefit of high-fidelity data updates with lower network bandwidth consumption.

[0126] Preferably, in the specific technical implementation of the data replacement, spatial connectivity verification processing is performed on the generated incrementally synchronized local 3D model nodes and their adjacent existing city objects. By retrieving the topological connection description of adjacent nodes in the city information model base, the geometric consistency between the newly written city information model target component with topological closure and the surrounding terrain, roads, or other building models at the edge stitching is verified. If it is determined that there are tiny floating-point cracks (such as discontinuities in millimeter-level values) at the stitching edge, real-time topology correction logic is invoked to perform edge alignment and vertex fusion processing on the facets at the intersection. After performing the connectivity verification and correction actions, a city information model base with spatial logical coherence, rigorous physical representation, and the latest engineering status is generated, thus obtaining the updated city information model base. This scheme utilizes local incremental replacement technology to improve the ability of the city-level digital base to dynamically evolve with changes in building engineering, thereby improving the long-term operation and maintenance efficiency and data authenticity of large-scale digital twin scenarios.

[0127] Preferably, the specific implementation process of executing the real-time topology correction logic is as follows: Automatic detection of spatial connection gaps is performed on the generated incrementally synchronized local 3D model nodes and their adjacent existing city objects. For example, the geometric vertex coordinates at the boundary of the local 3D model node are extracted, and the edge vertex indices of adjacent city model base objects (such as surrounding terrain patches or adjacent building bases) within the same spatial area are retrieved simultaneously. The spatial Euclidean distance between the two sets of edge vertices is calculated and compared logically with a preset geometric tolerance threshold (such as a numerical range of 0.001 meters to 0.005 meters). If the spatial Euclidean distance is determined to be within the geometric tolerance threshold but not zero, then a small spatial crack due to limited floating-point arithmetic precision or cross-scale generalization is identified in the area, thus obtaining a set of boundary geometric deviations. This processing step captures subtle defects at cross-scale model seams, providing a quantified correction target for subsequent physical-level geometric alignment.

[0128] Preferably, vertex fusion processing based on geometric attraction is performed on the obtained set of boundary geometric deviations. The real-time topology correction logic is equivalent to eliminating physical gaps between geometric entities by changing the topological connection direction of the boundary vertices. For example, taking the boundary vertices of the local 3D model nodes as the active source points, the nearest anchored vertex at the edge of the adjacent existing city object is searched. The action of forcibly updating the spatial coordinates of the active source point to the spatial coordinates of the anchored vertex is performed, thereby achieving the overlap of two independent geometric entities at the edge in the underlying data structure and thus obtaining the topology fused boundary point set. This step, through coordinate recalculation and vertex merging, alleviates the "floating-point crack" phenomenon at the physical level, enhances the continuity of the 3D scene, and effectively solves the common problems of geometric light leakage or structural breakage in large-scale rendering.

[0129] Preferably, in one scenario, patch edge alignment and normal smoothing are performed on the obtained topology fusion boundary point set. For example, the geometric patch indices associated with the topology fusion boundary point set are parsed, and the normal vectors of these patches in the stitching region are recalculated. To avoid surface rendering black spots or lighting distortion caused by forced vertex stretching, the Laplacian smoothing algorithm is invoked to perform local geometric optimization on the triangular patches at the stitching edges, enabling a smooth transition between the newly written component edges and the surface slope of the original base object, thus obtaining a geometrically continuous stitched surface structure. This achieves a technological evolution from "physical alignment" to "visual smoothness," improving the logically consistent physical texture features at the junction of the building and the base when performing high dynamic range lighting rendering.

[0130] Preferably, in the specific technical implementation of the data replacement, the generated geometrically continuous stitched surface structure and the city information model base undergo global topological consistency verification and data persistence processing. For example, manifold constraint verification is performed on the corrected stitched area to verify whether each newly added topological edge satisfies the physical criterion shared by two faces, and to eliminate possible geometric self-intersection anomalies. By performing incremental updates to the spatial index of the city information model base, the locally corrected geometric topological relationship is permanently solidified in the base database, thus obtaining the updated city information model base (its full English name is City Information Modeling, abbreviated as city information model). Therefore, through real-time local topological repair, the technical benefit of enabling the macro-level city base to rigorously support micro-level changes in components is achieved, supporting the professional requirement for the digital twin base to evolve in real-time and robustly with engineering changes.

[0131] like Figure 2 The diagram shown is a structural diagram of a three-dimensional modeling device for cross-scale consistency between building information model and city information model based on spatial intelligence, according to an embodiment of this application. The device includes: The initial modeling construction module is used to construct an initial 3D modeling object and a spatial diagram of the building information model components based on the original building information model; determine the projection footprint of the initial 3D modeling object, calculate the transformation operator, and perform automatic alignment on the initial 3D modeling object to obtain the aligned initial 3D modeling object; The semantic mapping generalization module is used to parse custom family attributes from the aligned initial 3D modeling object and perform cross-scale semantic mapping on them to obtain semantic classification information of the urban information model; determine the outer contour surface of the aligned initial 3D modeling object and perform geometric generalization processing on it to obtain the urban information building shell model; The 3D topology reconstruction module is used to determine the opening features of the 3D geometric patch set and map them to the urban information building shell model to obtain a geometric patch set with opening features; based on the semantic classification information of the urban information model, the urban information building shell model and the geometric patch set with opening features, and combined with the building information model component space map, topology reconstruction is performed to generate urban information model target components with topological closure. The incremental update module is used to calculate the alignment deviation of the target component of the city information model relative to the city information model base, so as to perform incremental updates on the city information model base based on the alignment deviation.

[0132] like Figure 3The diagram shown illustrates the hardware architecture of an electronic device according to an embodiment of this application. The device includes a processor, a memory, an input / output interface, and a communication bus. The memory stores computer programs; the processor executes the programs stored in the memory to implement the steps of the aforementioned cross-scale consistent 3D modeling of the building information model and city information model based on spatial intelligence.

[0133] Figure 2-3 For exemplary descriptions of each step, please refer to the above. Figure 1 The details of that record will not be repeated here.

Claims

1. A cross-scale consistency 3D modeling method for building information models and city information models based on spatial intelligence, characterized in that, include: Step 1: Extract the set of 3D geometric facets of building components from the original building information model, and construct the initial 3D modeling object and the spatial diagram of the building information model components; The projection footprint of the initial 3D modeling object is determined, and the transformation operator is calculated and a 3D geometric transformation is performed on the initial 3D modeling object to obtain the aligned initial 3D modeling object; Step 2: Parse the custom family attributes from the aligned initial 3D modeling object and perform cross-scale semantic mapping on them to obtain the semantic classification information of the city information model; extract the outer contour surface of the aligned initial 3D modeling object and perform geometric generalization processing on it to obtain the city information building shell model; Step 3: Extract opening features from the set of three-dimensional geometric faces and map them onto the urban information building shell model to obtain a set of geometric faces with opening features; based on the semantic classification information of the urban information model, the urban information building shell model and the set of geometric faces with opening features, and combined with the spatial map of the building information model components, perform three-dimensional topological reconstruction to generate urban information model target components with topological closure. Step 4: Calculate the alignment deviation of the target component of the city information model relative to the city information model base, and perform incremental updates on the city information model base based on the alignment deviation.

2. The method for cross-scale consistency 3D modeling of building information model and city information model based on spatial intelligence according to claim 1, characterized in that, In step 1, constructing the initial 3D modeling object includes: Project metadata is extracted from the original building information model. If the corresponding modeling unit is determined to be millimeters, a scaling factor is applied to correct the model dimensions of the original building information model to obtain a scale-normalized model. Extract the normal vector distribution of building components in the scale-normalized model, rotate the scale-normalized model to the geographic coordinate system to perform skew angle compensation, and obtain the normalized three-dimensional modeling object; Calculate the envelope volume of each building component in the normalized 3D modeling object, remove unstructured connectors whose envelope volume is less than a preset volume threshold, and obtain a simplified 3D modeling object as the initial 3D modeling object.

3. The method for cross-scale consistency 3D modeling of building information model and city information model based on spatial intelligence according to claim 1, characterized in that, Step 1, constructing the spatial diagram of the building information model components, includes: The building information model extracts building component attributes to generate a set of component nodes. Based on the set of component nodes, it identifies the spatial connection relationships and opening filling constraints between building components to generate a set of topological edges. Then, it constructs a spatial graph of the building information model that represents the inclusion, adjacency and constraint relationships between components.

4. The method for cross-scale consistency 3D modeling of building information model and city information model based on spatial intelligence according to claim 1, characterized in that, In step 1, calculating the transformation operator includes: Using the affine transformation solution logic based on the feature-fast matching algorithm, the transformation operator parameter set consisting of scale coefficients, rotation matrices, and translation vectors is calculated; Extract urban road line constraint information from the geographic information system to perform angle correction on the projected footprint, so as to correct the transformation operator parameter set; A three-dimensional affine transformation is performed on the initial three-dimensional modeling object based on the modified transformation operator parameter set, and the terrain elevation data in the urban information model base is retrieved to perform vertical height alignment, so as to obtain the height-calibrated initial three-dimensional modeling object as the aligned initial three-dimensional modeling object.

5. The method for cross-scale consistency 3D modeling of building information model and city information model based on spatial intelligence according to claim 1, characterized in that, Step 2, parsing the custom family attribute and performing cross-scale semantic mapping, includes: The custom family attributes in the aligned initial 3D modeling object are parsed to extract attribute keyword evidence, and the classification probability of the attribute keyword evidence is calculated using a probability mapping model to obtain semantic mapping feature values. Based on the semantic mapping feature values, a classification determination is performed, mapping the component type based on the building information model data exchange standard to the classification type based on the city information model standard, thereby obtaining the corresponding city information model semantic classification information.

6. The method for cross-scale consistency 3D modeling of building information model and city information model based on spatial intelligence according to claim 1, characterized in that, In step 2, obtaining the urban information building shell model includes: The aligned initial 3D modeling object is subjected to columnar pixelation to identify the top surface of the building, and probe rays are emitted upward along the height axis to extract roof feature patches; The roof feature patches are generalized at a level of detail using a roof extraction operator to simplify their topology, thus obtaining the urban information building shell model.

7. The method for cross-scale consistency 3D modeling of building information model and city information model based on spatial intelligence according to claim 1, characterized in that, Step 3 includes: The opening filling constraint relationship is extracted from the component space diagram of the building information model to identify the opening components belonging to the door and window category, and the normal projection position of the opening component relative to the urban information building shell model is calculated using the opening projection operator to generate the opening position parameters. Boolean operations are performed on the surface of the urban information building shell model based on the opening location parameters to map and generate the set of geometric patches with opening features.

8. The method for cross-scale consistency 3D modeling of building information model and city information model based on spatial intelligence according to claim 1, characterized in that, In step 3, generating the target component of the city information model with topological closure includes: The semantic classification information of the urban information model is used to identify the component hierarchy relationship between the urban information building shell model and the set of geometric facets with opening features; Based on the component hierarchy and the opening filling constraint, the city information building shell model and the set of geometric facets with opening features are subjected to facet stitching to construct a closed geometry that matches the manifold constraint. Perform watertightness testing on the closed geometry to identify and close geometric gaps, thereby obtaining the target component of the city information model with topological closure.

9. A cross-scale consistency 3D modeling method for building information model and city information model based on spatial intelligence according to claim 8, characterized in that, In step 4, calculating the alignment deviation of the target component of the city information model relative to the base of the city information model includes: Calculate the coordinates of the geometric center point and the envelope area of ​​the projected footprint, and the spatial deviation and area error value between them and the coordinates of the center point of the determined red line boundary and the corresponding legal area; The classification rate of the semantic classification information of the city information model is statistically analyzed, and a confidence score representing the alignment degree is generated based on the spatial deviation, the area error value, the classification rate, and the verification results of the watertightness test.

10. A cross-scale consistency 3D modeling method for building information models and city information models based on spatial intelligence according to claim 1, characterized in that, Step 4 further includes: in response to the confidence score being greater than a preset threshold, using a geometric hash algorithm to compare the original building information model under different version timestamps to identify the set of changed components; establishing a cross-scale association mapping table containing the unique identifier of the set of changed components and the urban information model base object, and performing local remapping and data replacement on the urban information model base based on the cross-scale association mapping table.

Images