Building structure information model generation method and system based on large language model, computer device and computer program product

Abstract

The application relates to the technical field of computers, and provides a building structure information model generation method and system based on a large language model, computer equipment and a computer program product, wherein parameter automatic linkage updating and level penetration are realized through construction of a multi-level structure information model and a parameter dependency graph, the problems of parameter transmission distortion, modification path fracture and information island in traditional modeling are solved, the building structure parameter level management and linkage updating can be automatically realized, and the modeling efficiency and accuracy are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method, system, computer equipment, and computer program product for generating building structure information models based on large language models. Background Technology

[0002] As the global construction industry transforms towards industrialization, greening, and intelligence, modular construction has become an important development direction due to its advantages such as high efficiency, environmental friendliness, and controllable quality. However, traditional modeling methods suffer from serious bottlenecks in efficiency, accuracy, and collaboration. Existing technologies, such as the intelligent 3D modeling method in SolidWorks disclosed in patent CN117688632A and the method for automated modeling in Revit by combining a large language model and the MCP protocol proposed in CN120409302A, have significant limitations, including heavy reliance on specific software, poor cross-platform reusability, extremely high migration and maintenance costs, difficulty in generating complex or non-standard component codes, and fragmented information hierarchy.

[0003] The current mainstream technologies suffer from the following systemic defects: Inefficient manual modeling and limitations of parametric methods: Modular building modeling in traditional CAD / BIM software can account for a significant proportion of the total project timeline. While parametric tools can establish geometric relationships, they lack a deep understanding of design intent and physical rules. Parametric methods are essentially geometry-driven, rather than physics- or logic-driven, and maintaining the topological relationships between modular units still relies heavily on manual judgment and adjustment. When design changes occur, modifications cannot be automatically and consistently propagated across different levels of the model, resulting in broken modification paths. Each design change requires a significant amount of time for manual coordination.

[0004] Hierarchical information silos and data mapping gaps: In existing technologies, geometric models and physical analysis models are independent of each other, leading to distortion in the transmission of key parameters within this architecture. For example, the mechanical properties of node domains cannot be dynamically correlated with geometric dimensions, resulting in significant errors; the cross-sectional properties of members cannot be automatically updated according to changes in the combined form of their modules, often leading to over-design and increased costs.

[0005] Multi-scale model conversion barriers: Structural analysis requires cross-scale conversion between rod models, shell models, and solid models. With current technology, a single conversion requires several hours of manual intervention, and the generated finite element mesh often suffers from mismatch after geometric model modifications, severely impacting analysis accuracy. Furthermore, mainstream topology optimization algorithms struggle to directly incorporate manufacturing constraints such as prefabrication, transportation, and hoisting of modular elements, rendering optimization solutions impractical in actual manufacturing.

[0006] Lack of cross-disciplinary collaboration mechanisms: Significant data silos exist between models from multiple disciplines, including architecture, structure, and MEP (Mechanical, Electrical, and Plumbing). Each discipline's models only perform geometric collision checks, failing to achieve intelligent co-creation of deeper information such as physical and functional details. For example, the false alarm rate for MEP pipeline conflict detection in BIM models is high, parameter transfer is delayed during forward design, and version conflicts frequently occur during reverse modification, severely slowing down the design cycle. Summary of the Invention

[0007] In response to the technical problems of information silos and parameter linkage failures in modular building modeling mentioned in the background, this invention provides a method, system, computer equipment, and computer program product for generating building structural information models based on a large language model.

[0008] In a first aspect, the present invention proposes a method for generating a building structure information model based on a large language model, the method being applied to a computer device, comprising: Step S1: In response to the input design text, the large language model performs semantic parsing on the design input text to extract structural design parameters from the design input text; Step S2: Based on the structural design parameters, create a multi-level structural information model in the database; and construct a parameter dependency graph between the various levels of the multi-level structural information model. The parameter dependency graph includes parameter nodes and directed edges connecting the parameter nodes. The directed edges are used to represent the dependency relationship between one parameter and another parameter. The path in the parameter dependency graph from one parameter node to another parameter node via a directed edge constitutes a dependency path. The multi-level structural information model, according to the progressive combination hierarchy from local components to the overall system, includes at least the node layer, section layer, connection layer, rod layer, module unit layer, and structural system layer in sequence. Step S3: Generate a geometric model based on the multi-level structural information model, perform topological identification on the geometric model to identify each geometric entity in the geometric model, the geometric entity including rods, node domains and module units; for each identified geometric entity, establish a mapping relationship between each geometric entity and the corresponding physical analysis object, and generate a mapping relationship table; Step S4: When the computer receives a modification instruction for any parameter in the multi-level structure information model, according to the parameter dependency graph, all target parameters that have direct or indirect dependencies with the modified parameter are updated sequentially along the dependency path to obtain the updated multi-level structure information model. Step S5: Generate updated geometric model and physical analysis model based on the updated multi-level structural information model and the mapping relationship table.

[0009] In one embodiment, the specific steps of performing topological identification on the geometric model in step S3 include: Step S31: Obtain the geometric identifier and boundary topology information of each candidate geometric entity from the geometric model, and treat each candidate geometric entity as a graph node; Step S32: Based on the comparison results of the geometric intersection relationship, shared boundary relationship or minimum distance with the preset geometric tolerance between the candidate geometric entities, determine the connection relationship between the candidate geometric entities, and use the connection relationship as graph edges to construct a geometric topology graph; Step S33: Traverse the geometric topology graph and identify and determine each rod, node domain and module unit in the geometric model based on at least one of the object identifier, geometric features and connection features of each candidate geometric entity.

[0010] In one embodiment, step S5, which involves generating an updated geometric model and a physical analysis model based on the updated multi-level structural information model and the mapping relationship table, includes: Step S51: Based on the updated multi-level structural information model, read the updated geometric parameters of each geometric entity and the updated physical parameters of each physical analysis object; Step S52: Based on the mapping relationship table, map the updated geometric parameters of each geometric entity to the corresponding physical analysis object; Step S53: Based on the updated geometric parameters and updated physical parameters of the physical analysis object, automatically generate or update the corresponding physical analysis model according to the preset analysis model generation rules.

[0011] In one embodiment, step S2, the step of creating a multi-level structured information model in the database, includes: Step S201: Register the project in the database and create a structural system record at the structural system layer; Step S202: Based on the structural design parameters, create each module unit instance in the module unit layer. Each module unit instance is recorded by referencing the module unit template ID and its spatial location parameters. Step S203: In each of the module unit instances, create each member instance in the member layer, and record each member instance by referencing the member template ID and its spatial position parameters; Step S204: At the connection point of each of the aforementioned member instances, create a node instance in the node layer, determine the connection attribute in the connection layer, and associate the connection attribute with the corresponding node instance; Step S205: Based on the cross-sectional requirements of each member instance, generate or reference a cross-sectional template in the cross-sectional layer, and associate the cross-sectional template with the corresponding member instance.

[0012] In one embodiment, the method further includes a step of generating irregularly shaped cross-section components: Step A1: Receive the strip sequence input, wherein each strip unit in the strip sequence includes a width parameter, a thickness parameter, and a bending angle parameter; Step A2: Based on the bending angle and width of each strip unit in the strip sequence, connect each strip unit sequentially to generate the geometric profile of the irregular cross section; Step A3: Based on the geometric profile of the irregular cross-section, calculate the physical parameters of the cross-section, including the cross-sectional area, centroid position, and principal moment of inertia; Step A4: Store the geometric contour and physical parameters of the irregular cross-section as a cross-section template in the database.

[0013] In one embodiment, when generating the updated physical analysis model in step S5, the corresponding multi-scale analysis model can be generated according to the target analysis scale type, specifically including: Step S61: Obtain the target analysis scale type, wherein the target analysis scale type includes at least one of the following: rod model, shell model, and solid model; Step S62: Based on the target analysis scale type, determine the physical analysis object corresponding to the target analysis scale type from the mapping relationship table, and perform analysis idealization and discretization processing on the updated geometric model corresponding to the target analysis scale type to generate a physical analysis model of the corresponding scale.

[0014] In one embodiment, the specific steps of constructing a parameter dependency graph between the various levels of the multi-level structured information model in step S2 include: Step S21: Identify the dependencies between parameters in the multi-level structural information model, treat each parameter as a parameter node, and use the dependencies as directed edges connecting the parameter nodes to construct a directed acyclic graph to obtain the parameter dependency graph; wherein, the direction of the directed edges is from the dependent parameter to the dependent parameter. Step S22: Sort each parameter node in the directed acyclic graph according to the topological order of the dependency relationship to generate the parameter update sequence of the parameter dependency graph.

[0015] Secondly, this application also provides a system for generating building structure information models based on large language models, including: The interaction module is used to receive design input text; A large language model engine is used to perform semantic parsing on the design input text and extract structural design parameters; The database is used to store multi-level structural information models, which are arranged in a progressive combination hierarchy from local components to the overall system, including node layer, section layer, connection layer, rod layer, module unit layer and structural system layer in sequence. The parameter dependency graph module is used to construct parameter dependency graphs between the levels of the multi-level structure information model. The parameter dependency graph includes parameter nodes and directed edges connecting the parameter nodes. The directed edges are used to represent the dependency relationship between one parameter and another parameter. The path in the parameter dependency graph from one parameter node to another parameter node via a directed edge constitutes the dependency path. The topology identification and mapping module is used to perform topology identification based on the geometric model generated by the multi-level structural information model, identify each geometric entity in the geometric model, the geometric entity includes rods, node domains and module units, and establish a mapping relationship between each identified geometric entity and the corresponding physical analysis object, and generate a mapping relationship table. The linkage update module is used to update all target parameters that have direct or indirect dependencies with the modified parameter in sequence along the dependency path according to the parameter dependency graph when a modification instruction for any parameter in the multi-level structure information model is received, so as to obtain the updated multi-level structure information model. The model generation module is used to automatically generate updated geometric and physical analysis models based on the updated multi-level structural information model and the mapping relationship table.

[0016] Thirdly, this application also proposes a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method for generating a building structure information model based on a large language model as described in the first aspect above.

[0017] Fourthly, this application provides a computer program product storing a computer program that, when executed by a computer device, implements the steps of the method for generating a building structure information model based on a large language model as described in the first aspect.

[0018] This application provides a method, system, computer equipment, and computer program product for generating building structural information models. By constructing a multi-level structural information model and parameter dependency graph, it achieves automatic parameter linkage updates and hierarchical connectivity, solving the problems of parameter transmission distortion, broken modification paths, and information silos in traditional modeling. It has the advantages of automatically realizing hierarchical management and linkage updates of building structural parameters, significantly improving modeling efficiency and accuracy. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the main process of a method for generating a building structure information model based on a large language model provided by the present invention; Figure 2 This is a schematic flowchart of an embodiment of performing topology recognition on a geometric model provided by an embodiment of the present invention; Figure 3 This is a schematic diagram of the process of generating an updated geometric model and a physical analysis model based on the updated multi-level structure information model and the mapping relationship table provided in this embodiment of the invention. Figure 4 This is a schematic diagram of an embodiment of the present invention for creating a multi-level structured information model in a database; Figure 5 This is a flowchart illustrating the steps for generating irregularly shaped cross-section components according to an embodiment of the present invention; Figure 6 This is a schematic diagram of an embodiment of a computer device for executing a method for generating a building structure information model based on a large language model, as provided in this application.

[0021] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0023] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0024] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0025] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0026] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0027] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0028] It is understandable that traditional modular building modeling suffers from problems such as low manual efficiency, limitations of parametric methods, fragmented information hierarchy, discontinuous mapping between geometric and physical models, difficulties in multi-scale model conversion, and a lack of cross-disciplinary collaboration mechanisms. When design changes occur, modifications cannot be automatically and consistently propagated across all levels of the model, leading to broken modification paths and severely impacting design efficiency and accuracy. In response, [reference needed]. Figure 1 This application proposes a method for generating a building structure information model based on a large language model. The method is applied to computer equipment and includes: Step S1: The design input text is responded to and semantically parsed by a large language model to extract structural design parameters from the design input text; It should be noted that the large language model refers to a natural language processing model based on deep learning, which is trained on large-scale text data and has the ability to understand, generate, and process human language. In this embodiment, the large language model is used to perform semantic parsing on the design input text, thereby extracting structural design parameters.

[0029] In its implementation, firstly, in step S1, the system's human-computer interface responds to the user's input design text. The large language model performs semantic parsing on the design input text to extract structural design parameters. Specifically, the design input text can be entered by the user through a graphical user interface or read by the system from a pre-set text file. After receiving the design input text, the large language model uses its natural language processing capabilities to identify design conditions such as structural type, geometric dimensions, load conditions, material grade, and seismic fortification information in the text. For example, for text information such as "three-story modular steel frame structure," "plan dimensions 18m × 36m," and "story height 3.6m," the corresponding structural type, plan dimensions, and story height parameters are extracted, and the extraction results are converted into structured data.

[0030] In one embodiment, the system organizes the extracted results according to a preset structured parameter data pattern. Each structured parameter record includes at least a parameter identifier, parameter name, parameter category, parameter value, unit of measurement, and source text fragment; it may also include the model level or object identifier associated with the parameter and the parsing status. The parameter identifier is used to uniquely identify the corresponding parameter, and the source text fragment records the corresponding content of the parameter in the design input text, establishing a traceable relationship between the extracted results and the original design input. Before writing the structured parameter record into the database, field integrity and data format verification can be performed, including checking whether the parameter name or parameter category conforms to the preset field definition, unifying or converting the unit of measurement, and checking whether the data type of the parameter value matches the corresponding parameter definition. For parameters not explicitly given in the design input text, or those that are ambiguous or cannot be uniquely determined, the corresponding parameter record is marked as pending confirmation, and the undetermined parameter value is not written into the multi-level structured information model as a determined design condition. For example, given the input text "floor height 3.6m", the system can generate a corresponding structured parameter record, where the parameter name is "floor height", the parameter value is "3.6", the unit of measurement is "m", the source text fragment is "floor height 3.6m", and the parsing status is "determined".

[0031] It should be noted that the structural design parameters extracted by the large language model in this step are mainly design conditions explicitly stated in the design input text or that can be directly determined based on the semantics of the text. For cross-sectional physical properties, component self-weight, structural response, or other derived parameters that need to be determined based on structural mechanics relationships, they are not directly generated by the large language model. Instead, they are calculated based on the extracted design conditions using preset engineering calculation rules, and the calculation results are used as parameters in subsequent multi-level structural information models.

[0032] For derived parameters that require engineering calculations, the corresponding preset calculation rules can be invoked based on the parameter type. For example, the cross-sectional area, centroid location, and moment of inertia of the component can be calculated based on the component's cross-sectional geometric parameters, and the component's self-weight can be calculated based on the component's dimensions and material parameters. For parameters that need to be judged based on design constraints, the corresponding design input parameters and calculation results are compared with the preset structural design rules to determine whether the corresponding design constraints are met and to obtain the corresponding verification status.

[0033] The derived parameters obtained through the engineering calculation rules can be stored in a data format compatible with the structured parameter records described above, and their source can be marked as rule calculation. At the same time, the corresponding parameter calculation rules can be associated with them to distinguish them from the parameters directly extracted from the design input text.

[0034] Step S2: Based on the structural design parameters, create a multi-level structural information model in the database; and construct a parameter dependency graph between the various levels of the multi-level structural information model. The parameter dependency graph includes parameter nodes and directed edges connecting the parameter nodes. The directed edges are used to represent the dependency relationship between one parameter and another parameter. The path from one parameter node to another parameter node in the parameter dependency graph constitutes the dependency path. The multi-level structural information model is arranged in a progressive combination hierarchy from local components to the overall system, and includes at least the node layer, section layer, connection layer, rod layer, module unit layer, and structural system layer in sequence. It should be noted that the aforementioned multi-level structural information model refers to a structural information model constructed according to a progressive combination hierarchy from local components to the overall system. This model includes at least, in sequence, a node layer, a section layer, a connection layer, a member layer, a module unit layer, and a structural system layer, used to uniformly express the geometric and physical information of the building structure.

[0035] The “hierarchy” described in this application is used to characterize the semantic organization and association hierarchy of different types of structural information in the data model. The hierarchies are associated with each other through object references and parameter dependencies, and it is not required that adjacent hierarchies have strict spatial inclusion relationships.

[0036] It should be noted that the parameter dependency graph is a graph structure used to represent the dependencies between parameters in a multi-level information model. The parameter dependency graph includes parameter nodes and directed edges connecting them, where directed edges represent the dependency of one parameter on another. A dependency path in the parameter dependency graph is formed by the path from one parameter node to another via a directed edge. The directed edges in the parameter dependency graph can connect parameter nodes within the same level or parameter nodes belonging to different levels. A dependency path formed by multiple consecutive directed edges can span two or more levels in the multi-level information model.

[0037] In a specific implementation, step S2, as an example, creates a six-level structural information model in the database based on the aforementioned structural design parameters. A parameter dependency graph is then constructed between the various levels of this six-level model. This graph includes parameter nodes and directed edges connecting them. Directed edges represent the dependency relationship between one parameter and another. The dependency path in the graph is formed by the path from one parameter node to another via a directed edge. Specifically, when creating the six-level structural information model in the database, the system first initializes the structural system layer based on the extracted design parameters, for example, by registering the project and defining the overall structural type. Subsequently, the system creates multiple module unit instances at the module unit layer based on the structural system layer definition, for example, by defining the geometric layout and quantity of the modules. Within each module unit instance, the system can create beam, column, and other member instances at the member layer, specifying their initial spatial position and member type. Then, at the connection points of the member instances, the system can create node instances at the node layer and determine the connection attributes at the connection layer, for example, specifying the connection type as rigid or hinged. Finally, based on the cross-sectional requirements of each member instance, the system can generate or reference cross-sectional templates at the cross-sectional layer. For example, specifying an H-section for a beam and a square steel tube for a column. When constructing the parameter dependency graph, the dependencies between parameters can be determined according to preset parameter calculation rules. For example, when the calculation rules for cross-sectional area and moment of inertia use geometric parameters such as cross-sectional width and height as input, directed edges pointing from the corresponding cross-sectional geometric parameters to the cross-sectional area and moment of inertia parameters are established. Thus, the direction of the directed edges points from the input parameters in the calculation rules to the output parameters determined by those input parameters, thereby representing the transitive relationship between parameters.

[0038] Furthermore, the parameter dependencies can span different information levels. For example, for a uniform cross-section member, the cross-sectional area parameter A in the cross-section layer and the member length parameter L in the member layer jointly determine the member volume parameter V in the member layer; the member volume parameter V and the corresponding material unit weight parameter jointly determine the member self-weight parameter G; the module unit self-weight parameter in the module unit layer can be determined according to a preset aggregation rule based on the self-weight of each component belonging to that module unit and other defined self-weight components; the total structural self-weight parameter in the structural system layer can be further determined according to a preset aggregation rule based on the self-weight parameters of each module unit. Thus, a cross-level dependency path can be formed, propagating from the cross-section layer parameters through the member layer and module unit layer to the structural system layer parameters.

[0039] Step S3: Generate a geometric model based on the multi-level structural information model, perform topology recognition on the geometric model to identify each geometric entity in the geometric model, including rods, node domains and module units; for each identified geometric entity, establish a mapping relationship between each geometric entity and the corresponding physical analysis object, and generate a mapping relationship table. It should be noted that the geometric entities can be understood as specific components or regions identified in the geometric model. In this embodiment, the geometric entities include rods, node domains, and modular units, which are the basic units constituting the geometric form of the building structure.

[0040] The physical analysis object can be understood as a logical analysis object corresponding to a geometric entity, used to record the semantics of structural analysis and related physical properties. It serves as the basis for subsequent idealization and discretization processing and is not equivalent to the finite element units formed after discretization. For example, a rod entity can correspond to a rod system physical analysis object, and a node domain can correspond to a node analysis object, a connection object, or a rigid region object. After determining the target analysis scale, analysis idealization and discretization processing are performed based on the corresponding physical analysis object and its associated geometric entity to generate the corresponding beam element, shell element, or solid element.

[0041] The mapping table in this embodiment is used to record the correspondence between geometric entities and physical analysis objects. Considering that the same geometric entity can have different physical analysis expressions at different analysis scales, the mapping table can record one or more mapping records for the same geometric entity. Each mapping record includes at least a geometric entity identifier, a physical analysis object identifier, and an analysis scale type. The analysis scale type is used to indicate whether the corresponding physical analysis object belongs to the rod analysis scale, the shell analysis scale, or the solid analysis scale.

[0042] Furthermore, each mapping record can be associated with corresponding analysis idealization rules and attribute mapping rules. The analysis idealization rules are used to instruct the corresponding geometric entities to undergo corresponding processing such as reference axis extraction, reference surface or mid-surface extraction, or 3D volume preservation when generating physical analysis models at the corresponding scale. The attribute mapping rules are used to transfer the cross-sectional parameters, thickness parameters, material parameters, or connection attributes associated with the geometric entity to the corresponding physical analysis object. For geometric entities that do not meet the corresponding analysis idealization conditions, mapping records for the corresponding analysis scale may not be established. When a mapping record corresponding to the target analysis scale does not exist, the system marks the corresponding geometric entity as unavailable for that analysis scale and adopts other applicable analysis scales according to preset scale substitution rules, or outputs the corresponding status for user confirmation.

[0043] As an example, in step S3, a geometric model is generated based on the aforementioned multi-level structural information model. Topological identification is then performed on the geometric model to identify the geometric entities, including members, node domains, and modular units. For each identified geometric entity, a mapping relationship is established between each geometric entity and its corresponding physical analysis object, generating a mapping relationship table. Specifically, the system can generate a 3D geometric model using a geometric modeling engine based on the geometric parameters stored in the multi-level structural information model, such as the coordinates, dimensions, and shape information of components. This geometric model can be stored in a standard CAD file format or a BIM model format. When performing topological identification on the geometric model, the system can use a method based on geometric feature matching, for example, identifying members such as beams and columns by analyzing the length, cross-sectional shape, and connection method of components. Furthermore, the system can identify node domains based on the intersection areas of multiple members. Modular units can be identified based on predefined combination rules, such as standard units composed of multiple members and node domains. For each identified geometric entity, the system establishes a mapping relationship between it and its corresponding physical analysis object. For structural members, corresponding physical analysis objects can be established based on the available analysis scales. For example, for the analysis scale of a truss system, a corresponding truss physical analysis object can be established, and the analysis idealization rules used to extract the reference axis of the member, as well as the corresponding section parameters and material parameters, can be associated with this truss physical analysis object. For node domains, a correspondence can be established with analysis nodes, connection objects, rigid regions, or local analysis objects of the corresponding scale based on the members connected to them and the connection attributes recorded in the connection layer. For modular elements, a correspondence can be established with a set of analysis objects composed of the physical analysis objects corresponding to the internal members and node domains.

[0044] The above mapping relationships are recorded in the mapping relationship table. For the same geometric entity, the mapping relationship table can record one or more mapping records according to different analysis scales, so that the system can locate the corresponding physical analysis object according to the geometric entity identifier and analysis scale type, and perform subsequent model generation and updating according to the analysis idealization rules and attribute mapping rules associated with the mapping record.

[0045] Step S4: When the computer receives a modification instruction for any parameter in the multi-level structure information model, according to the parameter dependency graph, update all target parameters that have direct or indirect dependencies with the parameter being modified in sequence along the dependency path to obtain the updated multi-level structure information model. In the example above, in step S4, when the computer receives a modification instruction for any parameter in the multi-level structural information model, it updates all target parameters that have direct or indirect dependencies with the modified parameter sequentially along the dependency path according to the parameter dependency graph, to obtain the updated multi-level structural information model. Specifically, when a user directly modifies the value of a parameter through the graphical user interface, for example, changing the cross-sectional height of a beam from 400mm to 450mm, the system will receive this modification instruction. As an alternative, the modification instruction can also be received from other design software through an external interface. After receiving the modification instruction, the system will start from the parameter node corresponding to the modified parameter and traverse all outgoing edges in the parameter dependency graph to find other parameters that directly depend on that parameter. Subsequently, the system will recursively propagate downstream along the dependency path, sequentially calculating and updating all affected target parameters. For example, when the beam cross-sectional height is modified while other cross-sectional geometric parameters remain unchanged, the system redetermines the cross-sectional area and moment of inertia according to the corresponding parameter calculation rules, and redetermines the self-weight of the component according to the updated cross-sectional area, member length, and material parameters. The update process can proceed according to the topological order of the parameter dependency graph, ensuring that all dependent parameters are updated correctly and consistently. The updated parameter values ​​are then written into the corresponding multi-level structural information model record in the database, resulting in the updated multi-level structural information model.

[0046] Step S5: Generate updated geometric and physical analysis models based on the updated multi-level structural information model and mapping relationship table.

[0047] In the example above, in step S5, the updated geometric model and physical analysis model are generated based on the updated multi-level structural information model and the mapping table. Specifically, the system can re-invoke the geometric modeling engine to generate a new geometric model based on the geometric parameters contained in the updated multi-level structural information model, such as updated component dimensions and location information. This updated geometric model can be immediately visualized for user review. Simultaneously, the system can automatically construct or update the physical analysis model based on the physical parameters contained in the updated multi-level structural information model, such as updated material properties, section characteristics, and connection stiffness, as well as the mapping table. For example, in the beam analysis scale, if the beam's section dimensions are modified, the section properties of the beam elements in the corresponding physical analysis model will also be automatically updated. The physical analysis model can be a finite element model, containing updated element types, node coordinates, material properties, and section properties, and can be exported in a format recognizable by structural analysis software for subsequent mechanical analysis.

[0048] To further illustrate the linkage between steps S1 to S5 above, let's take a beam member in the aforementioned three-story modular steel frame structure as an example. In step S1, the large language model extracts design conditions such as structural system type, planar dimensions, story height, and material grade from the design input text, and forms corresponding structured parameter records according to a preset structural parameter data pattern. In step S2, corresponding structural system records, module unit instances, and member instances are created based on the design conditions, and a corresponding section template is associated with one of the beam member instances. For ease of explanation, this beam member instance can be denoted as Beam_01, and its associated section template can be denoted as Section_01. Section_01 records the section height, section width, and other section geometric parameters, and establishes the parameter dependency relationship between the section geometric parameters and the section area and section moment of inertia according to preset parameter calculation rules.

[0049] In step S3, a corresponding beam geometric entity is generated based on the spatial location of Beam_01 and the cross-sectional parameters of Section_01, and the object identification association between the beam geometric entity and Beam_01 is retained. At the same time, a truss physical analysis object corresponding to the beam geometric entity is established under the truss analysis scale, and the beam geometric entity identifier, truss physical analysis object identifier, analysis scale type, and corresponding reference axis extraction rules are recorded in the mapping relationship table.

[0050] When a modification instruction is received in step S4 to change the cross-sectional height in Section_01 from 400mm to 450mm, while keeping other cross-sectional geometric parameters unchanged, the system determines the affected downstream parameters based on the parameter dependency graph and calls the corresponding parameter calculation rules according to the parameter update sequence to redetermine the cross-sectional area, cross-sectional moment of inertia, and the component self-weight determined by the cross-sectional area. When the self-weight of the corresponding module unit and the total structural self-weight are set as the downstream summary parameters of the component self-weight, the system further updates the corresponding module unit self-weight and total structural self-weight according to the corresponding summary rules. Subsequently, in step S5, the cross-sectional geometry of the beam geometric entity corresponding to Beam_01 is regenerated or updated based on the updated Section_01, and the corresponding truss physical analysis object is located according to the mapping relationship table. The updated cross-sectional parameters and cross-sectional physical properties are synchronized to the truss physical analysis object, thereby ensuring consistency between the multi-level structural information model, geometric model, and physical analysis model.

[0051] The technical advantages of this application's embodiments are as follows: By introducing a large language model for semantic parsing, automated parameter extraction of design requirements is achieved, improving modeling efficiency. By constructing a multi-level structural information model and establishing a parameter dependency graph, consistent expression and automatic linkage of structural information dependencies between levels are achieved, solving the problems of information silos and parameter linkage failures. The mapping relationship between the geometric model and the physical analysis model enables automatic model reconstruction and synchronous updating of the analysis model after design changes, eliminating the drawbacks of the disconnect between the geometric and analysis models, thereby effectively improving the efficiency and accuracy of modular building design.

[0052] In some embodiments, the specific implementation of constructing the parameter dependency graph in step S2 above may include: Step S21: Identify the dependencies between parameters in the multi-level structure information model, treat each parameter as a parameter node, and use the dependencies as directed edges connecting the parameter nodes to construct a directed acyclic graph, thus obtaining the parameter dependency graph; wherein, the direction of the directed edges is from the dependent parameter to the dependent parameter. Step S22: Sort the parameter nodes in the directed acyclic graph according to the topological order of their dependencies to generate the parameter update sequence of the parameter dependency graph.

[0053] Specifically, in step S21, the dependencies between parameters in the multi-level structural information model are first identified according to preset parameter calculation rules. Each parameter calculation rule records at least the input parameters, output parameters, and the calculation relationship used to determine the output parameters based on the input parameters; the calculation relationship can be a calculation formula, a geometric transformation relationship, or a preset parameter conversion rule.

[0054] When constructing the parameter dependency graph, the system reads the calculation rules for each parameter. When the input and output parameters involved in a parameter calculation rule exist in the current multi-level structural information model, the input and output parameters are determined as parameter nodes respectively. For each input parameter in the parameter calculation rule, a directed edge is established from the parameter node corresponding to the input parameter to the parameter node corresponding to the output parameter, and the directed edge is associated with the corresponding parameter calculation rule.

[0055] For example, when the cross-sectional area A is determined by the cross-sectional width b and the cross-sectional height h, directed edges are established pointing from the cross-sectional width b to the cross-sectional area A, and from the cross-sectional height h to the cross-sectional area A, respectively. When the cross-sectional moment of inertia I is determined by the corresponding cross-sectional geometric parameters, the corresponding directed edges are established in the same way. For cross-level parameter dependencies, taking parameter transfer between the section layer, member layer, module unit layer, and structural system layer as an example, for members with uniform cross-sections, when the member volume V is determined by the cross-sectional area A and the member length L in the section layer, directed edges are established from the cross-sectional area A and the member length L to the member volume V. When the member self-weight G is determined by the member volume V and the corresponding material unit weight, directed edges are established from the member volume V and the material unit weight parameters to the member self-weight G. When the self-weight of a module unit is determined by the self-weight of each component belonging to that module unit and other defined self-weight components according to a preset aggregation rule, directed edges are established from the self-weight parameters of each component and the corresponding other self-weight component parameters to the self-weight parameters of the corresponding module unit. When the total structural self-weight is determined by the self-weight of each module unit according to a preset aggregation rule, directed edges are established from the self-weight parameters of each module unit to the total structural self-weight parameters. Through these directed edges, a cross-level dependency path can be formed extending from the section layer through the member layer, module unit layer, and structural system layer.

[0056] Therefore, the directed edges in the parameter dependency graph not only represent the direction of dependency between parameters, but also can be associated with the calculation rules used to determine the target parameters, so that when the source parameters change, the downstream target parameters can be re-determined according to the corresponding calculation rules.

[0057] After identifying dependencies, each parameter in the multi-level structural information model is abstracted into a parameter node. These parameter nodes represent all independently definable or computable parameters in the model, such as geometric dimensions (e.g., beam width, column height), material properties (e.g., steel strength grade), load values ​​(e.g., floor live load), and connection properties (e.g., connection stiffness). This abstraction process allows the complex structural information model to exist in the form of discrete, operable units, facilitating subsequent management and analysis of the constructed structural diagram.

[0058] Subsequently, the identified parameter dependencies are represented as directed edges connecting parameter nodes. A directed edge represents a unidirectional dependency between parameters; for example, if the calculation or definition of parameter A requires the value of parameter B, then there exists a directed edge from parameter B to parameter A. This setting of directed edges clarifies the direction of influence between parameters, providing a foundation for subsequent update path tracing and topology sorting. Based on this, a directed acyclic graph (DAG), i.e., a parameter dependency graph, is constructed. A DAG is a special type of directed graph where there are no paths originating from a node and ultimately returning to that node; that is, there are no cycles. During the graph construction process, cycle acyclicity can be ensured by performing cycle checks each time a directed edge is added. If adding an edge would lead to a cycle, the edge is rejected or the user is prompted to adjust the dependency. The parameter dependency graph records unidirectional parameter dependencies that can be executed in a determined computation order. For candidate relationships that are determined by cycle checks to form cyclic dependencies, they are not directly written as directed edges into the parameter dependency graph, and the cyclic dependency is marked to avoid cyclic computation during parameter linkage updates. Alternatively, after the graph is constructed, an algorithm (such as Depth-First Search, DFS) can be used to detect whether there are cycles in the graph. If so, corrections or errors are made. The characteristics of a directed acyclic graph (DAG) ensure that the parameter update process can terminate, avoiding infinite loop updates and thus guaranteeing the stability and predictability of model updates. It is worth noting that the direction of directed edges is explicitly defined as pointing from the dependent parameter to the dependent parameter. For example, when the volume of a rod is determined by both the cross-sectional area and the rod length, directed edges are created pointing from the parameter nodes corresponding to "cross-sectional area" and "rod length" to the parameter node corresponding to "rod volume". This directional convention ensures that during the update process, the dependent parameter is always updated before the parameters that depend on it, providing a clear logical basis for subsequent topological sorting and guaranteeing the correctness of the update.

[0059] In step S22, the parameter nodes in the directed acyclic graph are sorted according to the topological order of their dependencies, generating a parameter update sequence for the parameter dependency graph. Topological sorting is a method for linearly sorting the vertices of a directed acyclic graph, ensuring that for any directed edge (u, v), node u always appears before node v in the sort. Topological sorting can be implemented using the Kahn algorithm or the depth-first search (DFS) algorithm. For example, the Kahn algorithm first finds all nodes with an in-degree of 0 and adds them to a queue. Then, it removes a node from the queue one by one, decrements the in-degree of all its adjacent nodes by 1, and adds any adjacent node to the queue if its in-degree becomes 0. This process is repeated until the queue is empty. Topological sorting generates a clear parameter update sequence, which is an ordered list of parameters that defines the order in which other affected parameters are updated when a parameter in the multi-level structural information model is modified. This sequence provides precise execution instructions for the linked update module, enabling the complex parameter linked update process to be automated, efficient, and accurate.

[0060] Through the above technical solutions, this application provides a rigorous logical framework and execution mechanism for parameter linkage updates in multi-level structural information models. By explicitly representing the dependencies between parameters as directed edges and constructing a directed acyclic graph (DAG), the dependencies and influence directions between parameters can be clearly defined, effectively avoiding logical confusion and update path conflicts caused by ambiguous dependencies in traditional modeling. Furthermore, by performing topological sorting on the parameter nodes in the DAG, a parameter update sequence is generated. This ensures that when any parameter is modified, the system can update all affected parameters in a predetermined and correct order. This means that when each parameter is updated, all its dependent parameters have already been calculated and updated, thus fundamentally solving the data inconsistency problem caused by improper calculation order and greatly improving the accuracy, stability, and efficiency of model updates. This topological logic-based update mechanism enables complex building structural information models to evolve in a controllable and predictable manner, effectively supporting the frequent change requirements in building structural design and providing a solid data foundation for the automatic generation of subsequent geometric and physical analysis models.

[0061] refer to Figure 2 In some embodiments, this application further proposes specific steps for performing topological identification on the geometric model in step S3, including: Step S31: Obtain the geometric identifier and boundary topology information of each candidate geometric entity from the geometric model, and treat each candidate geometric entity as a graph node; Step S32: Based on the comparison results of the geometric intersection relationship, shared boundary relationship or minimum distance with the preset geometric tolerance between the candidate geometric entities, determine the connection relationship between the candidate geometric entities, and use the connection relationship as graph edges to construct a geometric topology graph; Step S33: Traverse the geometric topology graph and identify and determine each rod, node domain and module unit in the geometric model based on at least one of the object identifier, geometric features and connection features of each candidate geometric entity.

[0062] Specifically, in step S31, a component refers to a basic unit constituting a building structure, such as beams, columns, walls, and slabs. In the geometric topology diagram, these components are abstracted into independent graph nodes, each representing a specific structural component. This can be achieved by parsing geometric model data, such as extracting component IDs from a BIM model or identifying layer or entity attributes from a CAD model, recognizing each independent geometric entity as a component, and assigning it a unique node identifier. For geometric models generated from the multi-level structural information model, the object identification association between the geometric entity and the corresponding rod instance, node instance, or module unit instance in the multi-level structural information model can be retained when generating geometric entities to assist in the subsequent identification of geometric entity types. Another implementation method is that the system automatically identifies geometric bodies that conform to a specific component definition based on preset geometric features, such as slenderness ratio, volume, or shape, and uses them as graph nodes.

[0063] In step S32, the connection relationship between components refers to the spatial interaction or geometric adjacency between different components. In the geometric topology graph, these connection relationships are abstracted as directed or undirected edges connecting graph nodes. For example, when two beams are connected to the same column, the direct connection relationship between each beam and column can be determined based on the geometric intersection relationship, shared boundary relationship, or comparison result of minimum distance and preset geometric tolerance between the beams and columns; a connection edge is established between the graph nodes corresponding to the two beams only when the two beams themselves also meet the corresponding direct geometric connection judgment conditions. This can be achieved by detecting the intersection of component geometries, shared boundaries (such as faces, edges, points), or proximity within a certain spatial tolerance range. For example, when the geometric boundaries of two components coincide in three-dimensional space, a direct connection relationship can be determined. For candidate component pairs judged solely by minimum distance, in addition to the minimum distance being less than a preset geometric tolerance, at least one of the following can be used for auxiliary judgment: component end region, preset candidate connection region, component type, or existing object identifier. When the corresponding auxiliary connection conditions are met, a direct connection relationship is determined between the two components, and connecting edges are added between the corresponding graph nodes. A geometric topology graph is an abstract data structure used to represent the graph theory model of components and their connections in a geometric model. This graph can be stored using data structures such as adjacency matrices or adjacency lists. The construction process involves traversing all components in the geometric model, converting them into graph nodes, and then traversing all component pairs, adding graph edges according to their connection relationships. The construction of this graph provides a structured data foundation for subsequent topology analysis and entity recognition.

[0064] In step S33, after constructing the geometric topology graph, the graph nodes and their adjacent graph nodes in the geometric topology graph are traversed to obtain each geometric entity in the geometric model and its connection relationships. Specifically, depth-first search or breadth-first search can be used. Starting from any unvisited graph node, the graph nodes connected to it are visited sequentially according to the adjacency relationship of the graph node. After a connected region is traversed, the traversal continues from other unvisited graph nodes until all graph nodes in the geometric topology graph are visited. In this way, the subsequent entity recognition can cover all connected geometric entities in the geometric topology graph, and is not limited to a single topological path. Based on the candidate geometric entities obtained in step S31 and the connection relationships between the candidate geometric entities determined in step S32, the traversed geometric entities are classified and identified. For geometric entities that retain object identification associations during the geometric model generation process, their initial entity type can be determined according to their corresponding rod instances, node instances, or module unit instances, and verified in combination with geometric features and connection relationships; for geometric entities that do not retain corresponding object identifications, they are identified according to preset node features and edge features. Preset node features and edge features are rules or attributes used to distinguish different types of geometric entities. Node features can include the geometric dimensions of the component (e.g., slenderness ratio, cross-sectional dimensions), material type, number of connections, and layer level. Edge features can include connection type (e.g., rigid connection, hinged connection), connection strength, and connection method (e.g., welding, bolted connection). These features can be stored in a rule base. For example, a component with a large slenderness ratio may be identified as a rod, while an area connecting multiple rods with relatively concentrated geometric dimensions may be identified as a node domain. Identification refers to classifying and characterizing geometric entities based on preset features. Segmentation refers to determining the precise geometric boundaries of these identified entities. For example, when traversing candidate geometric entities or locally connected areas that are connected to multiple rods, the system can determine whether the area is a beam-column node domain based on its connection features and its own geometric features. If the preset node domain identification conditions are met, its geometric boundaries are determined based on the size range of the node domain and its transition relationship with adjacent rods. Members are typically slender linear components, node domains are areas connecting multiple members, and modular units are prefabricated assemblies containing multiple members and node domains. This identification and segmentation is iterative, ensuring that all relevant entities are accurately extracted and defined.

[0065] refer to Figure 3 In some embodiments, to ensure that the geometric parameters of the geometric entity and the physical parameters of the physical analysis object can be accurately and automatically mapped and synchronized during the model update process, so as to ensure the consistency between the analysis model and the geometric model, the specific steps of the aforementioned step S5 further include: Step S51: Based on the updated multi-level structural information model, read the updated geometric parameters of each geometric entity and the updated physical parameters of each physical analysis object. This step aims to obtain the latest data after the model update, ensuring that subsequent model generation is based on accurate and real-time information. Specifically, the system can query records associated with the updated multi-level structural information model from the database and extract the geometric attributes of each geometric entity (such as rods, node domains, and module units), including but not limited to their spatial coordinates, dimensions, and topological connections. Simultaneously, the system will also read the physical attributes of each physical analysis object (such as rod analysis objects, shell analysis objects, node analysis objects, or connection objects), such as material type, elastic modulus, Poisson's ratio, section properties, connection stiffness, and applied loads and boundary conditions. Alternatively, the system can maintain an in-memory data structure that synchronizes the multi-level structural information model state in the database in real time. When the model is updated, the latest geometric and physical parameters can be efficiently obtained directly from this in-memory structure.

[0066] Step S52: Based on the mapping relationship table, the updated geometric parameters of each geometric entity are mapped to the corresponding physical analysis object. This step is crucial in connecting the geometric model and the physical analysis model, accurately converting geometric modifications into parameters required for physical analysis. The mapping relationship table can be pre-established and stored in a database, containing the correspondence between geometric entities and physical analysis objects, as well as rules or algorithms for converting geometric parameters into physical parameters. When the same geometric entity has multiple mapping records at different analysis scales in the mapping relationship table, the system synchronizes the updated shared geometric parameters and the physical parameters associated with the geometric entity to the corresponding physical analysis object according to the attribute mapping rules associated with each mapping record. Specific analysis idealization and discretization operations are performed after the target analysis scale is determined. For example, at the rod system analysis scale, for a geometric rod entity, the mapping relationship table can specify its corresponding rod system physical analysis object and define how to determine the cross-sectional area and moment of inertia based on the rod's cross-sectional shape and how to transfer the corresponding material properties. After determining the rod system analysis scale, the corresponding geometric entity is then subjected to reference axis extraction and discretization processing according to step S62 to generate the corresponding beam element or rod element. During execution, the system iterates through the mapping table, calculates and updates the physical parameters of the corresponding physical analysis object based on the updated geometric parameters of each geometric entity and the mapping rules. Furthermore, the mapping table can also contain more complex transformation logic, such as dynamically calculating the cross-sectional physical properties based on geometric contour parameters by calling specific calculation modules and assigning them to the corresponding physical analysis object.

[0067] Step S53: Based on the updated geometric and physical parameters of the physical analysis object, and according to the preset analysis model generation rules, the system automatically generates or updates the corresponding physical analysis model. This step aims to automatically construct a physical analysis model usable for structural analysis from the updated physical analysis object data. The system generates or updates the physical analysis model by calling the built-in finite element preprocessing module or external finite element analysis software interface based on information such as the node positions, connection relationships, material properties, cross-sectional properties, loads, and boundary conditions corresponding to the physical analysis object. For cases requiring different levels of analysis precision, the system first determines the target analysis scale type according to step S61, then performs corresponding analysis idealization and discretization on the corresponding geometric entities according to step S62, and assigns the updated material properties, cross-sectional properties, connection properties, loads, and boundary conditions to the corresponding analysis object, thereby forming a rod model, shell model, or solid model. This ensures that the generated physical analysis model remains consistent with the updated geometric model and multi-level structural information model.

[0068] In some embodiments, reference Figure 4 The step of creating a multi-level structured information model in the database in step S2 further includes: Step S201: Register the project in the database and create a structural system record at the structural system layer; As an example, step S201 aims to provide a top-level data container and management entry point for the entire building structural information model. Registering a project means assigning a unique identifier and storage space to the current design task in the database management system, ensuring that all relevant data can be uniformly managed and retrieved. Creating a structural system record at the structural system layer provides a macro-level description of the entire building structure, including basic information such as project name, design stage, overall structural type, total number of floors, and total height. This step can be achieved by creating a "Project Table" and a "Structural System Table" in the database. Upon receiving the design input text and parsing the preliminary structural design parameters, the system automatically inserts a new project record in the project table and a structural system record associated with that project in the structural system table, containing the overall structural parameters extracted from the design input text. Alternatively, this can be achieved using an object-oriented database or a graph database, treating the project and structural system as top-level objects or nodes, assigning them unique object IDs or node IDs, and storing their attribute information, thereby establishing the logical starting point for the entire multi-level structural information model.

[0069] Step S202: Based on the structural design parameters, create each module unit instance in the module unit layer. Each module unit instance is recorded by referencing the module unit template ID and its spatial location parameters. As an example, step S202 is responsible for instantiating specific modular units at the modular unit level. A modular unit is the basic building block of a modular building, such as a room module or a staircase module. Creating a modular unit instance means determining the required type and quantity of modular units based on structural design parameters (such as building floor plan, functional zoning, etc.) and generating an independent record for each modular unit in the database. These instances do not store all detailed geometric and physical information of the modular unit; instead, they inherit its general attributes by referencing a predefined "modular unit template ID," while recording its "spatial position parameters" (such as coordinates, rotation angle, and size adjustment factor) within the overall structure, thereby achieving lightweight data storage and efficient reuse. The system can maintain a "modular unit template library," which stores the general attributes and geometric topology information of various standard modular units. When a modular unit instance needs to be created, the system selects a suitable template from the template library based on the structural design parameters and creates a new record in the "modular unit instance table." This record contains the template ID, instance ID, and the instance's position, orientation, and possible size scaling parameters in three-dimensional space. In addition, a parametric modeling engine can be used to dynamically generate the geometric and topological information of the module unit based on the design parameters, store it as a new module unit template, and then create instances by referencing the new template ID and spatial location parameters.

[0070] Step S203: In each of the module unit instances, create each member instance in the member layer, and record each member instance by referencing the member template ID and its spatial position parameters; As an example, step S203 further refines the internal structure of the module unit by creating specific member instances at the member layer. Members are linear structural components, such as beams, columns, and supports. Creating member instances within a module unit instance means that the system identifies and instantiates each member constituting the module unit based on its type and internal structural layout. Similar to module unit instances, member instances inherit common attributes such as cross-sectional type and material by referencing a predefined "member template ID," and record their "spatial position parameters" (such as starting point coordinates, ending point coordinates, and local coordinate system information) within the module unit or the overall structure to achieve data simplification and consistency. The database can contain a "member template library," storing the common attributes of different types of members (such as H-beams and square steel columns). When the system processes a module unit instance, it creates a record for each member in the "member instance table" based on the definition of the module unit, including the ID of its module unit, the member template ID, and its geometric definition in three-dimensional space. Alternatively, a pre-defined rule engine or geometry parser can be used to automatically identify linear components from the geometric definition of the module unit and record them as rod instances, while matching or generating corresponding rod template IDs and spatial position parameters.

[0071] Step S204: At the connection point of each of the aforementioned member instances, create a node instance in the node layer, determine the connection attribute in the connection layer, and associate the connection attribute with the corresponding node instance; As an example, step S204 focuses on the connection relationships between structural members, creating node instances at the node layer and defining connection attributes at the connection layer. A node is a point where multiple members intersect or where a member connects to the foundation; it is a critical location for force transmission. Creating node instances at the connection points of member instances means that the system automatically identifies the connection locations between member ends, member intersection areas, or between members and the foundation based on the spatial location and topological relationship of the member instances, and generates a unique node instance record for each location. Simultaneously, at the connection layer, the system determines and records the connection attributes corresponding to the node instances. These connection attributes may include at least one of the following: connection type, connection template identifier, connection stiffness, bearing capacity parameter, and structural attributes. These connection attributes can originate from design input text or user-specified information, or can be obtained by querying a preset connection template library or connection attribute library. For connection stiffness, bearing capacity, or other mechanical properties that need to be determined based on structural calculation results, the corresponding calculation results can be obtained by calling a preset engineering calculation module or an external structural analysis / design software interface, and these calculation results are written into the database as connection attributes. In practical implementation, the system can traverse the endpoint coordinates and topological connections of each member instance, identify overlapping member ends or those within a preset spatial tolerance range, and create node instances. Subsequently, based on the connection type, member type, and preset connection rules corresponding to the node, the system matches the corresponding connection template or connection attribute record in the connection attribute table, and associates the connection attribute with the node instance through object identification or database association. When the connection attribute is obtained by an external engineering calculation module, the system can also record the node instance or connection object identification corresponding to the calculation result for subsequent physical analysis model generation and parameter update.

[0072] Step S205: Based on the cross-sectional requirements of each member instance, generate or reference a cross-sectional template in the cross-sectional layer, and associate the cross-sectional template with the corresponding member instance.

[0073] As an example, step S205 is used to establish cross-section template associations for each member instance at the cross-section layer. The "cross-section requirements" can originate from the cross-section type or size parameters explicitly given in the design input text, user-specified cross-section parameters, existing cross-section definitions in the preset member templates, or cross-section selection results output by the external structural analysis or design module. The system queries the cross-section template library for cross-section templates that match the corresponding cross-section type and size parameters based on the cross-section requirements. If a matching cross-section template exists, the system directly references the corresponding cross-section template ID and establishes an association with the member instance. When no corresponding template exists in the cross-section template library, the system can call the cross-section generation module to create a new parameterized cross-section template based on the cross-section type and geometric size parameters already determined in the cross-section requirements, and then store it in the cross-section template library and associate it with the corresponding member instance. For cases where it is necessary to further determine or optimize the cross-section dimensions based on structural response, bearing capacity requirements, or other design constraints, the system can call the preset engineering calculation module or external structural analysis / design software to obtain the corresponding cross-section parameters or selection results, and then generate or match a cross-section template based on the returned cross-section parameters.

[0074] By introducing a standardized multi-level structural information model creation process, this application effectively solves the problems of standardization in converting abstract design parameters into specific database records and the accuracy and consistency of logical relationships between component instances at different levels during model construction. Specifically, by registering projects and creating structural system records in the database, a clear top-level entry point and management foundation are provided for the entire model, avoiding data management chaos. At the module unit and member layers, instances are created by referencing template IDs and their spatial location parameters, significantly reducing data redundancy, achieving lightweight storage of model data, and ensuring the uniformity of components of the same specifications in different locations, which is crucial for the mass production and reuse of modular buildings. Furthermore, node instances are created at the connection points of member instances, and connection attributes are determined and associated with the corresponding node instances. This explicitly embeds the physical connection logic between structural components into the model, greatly improving the integrity and accuracy of structural topological relationships and providing reliable input for subsequent structural analysis. Simultaneously, based on the cross-sectional requirements of the member instances, cross-sectional templates are generated or referenced at the cross-sectional layer and associated with the corresponding member instances, realizing the dynamic binding of cross-sectional attributes and member entities. This makes the management of cross-sectional information more flexible and easier to maintain, and can adapt to the cross-sectional adjustment requirements brought about by design changes. Through the above-mentioned progressive and standardized instantiation steps, this application systematically transforms the initial design parameters into structured database records. This not only solves problems such as data redundancy, hierarchical logic confusion, and non-standard component references in the model construction process, but also achieves efficient model reuse and consistency management through the template reference mechanism. This lays a solid and accurate data foundation for subsequent parameter dependency graph construction, linkage updates, and multi-scale analysis, enabling the entire building structure information model generation method based on a large language model to respond to design changes more efficiently and accurately, and support information flow throughout the entire lifecycle.

[0075] In some embodiments, reference Figure 5 The method further includes steps A1 to A4 for generating irregularly shaped cross-section components: Step A1: Receive the strip sequence input, wherein each strip unit in the strip sequence includes a width parameter, a thickness parameter, and a bending angle parameter.

[0076] This step aims to describe complex non-standard cross-sections in a structured and parametric manner. By decomposing the irregular cross-section into a series of strip elements with well-defined geometric properties, a data foundation is laid for subsequent automated processing. The strip sequence input can be interactively defined by the user through a graphical user interface (GUI), specifying the properties of each strip element, for example, by inputting numerical values ​​or dragging sliders to set width, thickness, and bending angle. Alternatively, the input can be received by parsing a predefined text file (such as JSON or XML format) or through an application programming interface (API) call, containing an ordered list of parameters for the strip elements. It should be noted that, in the embodiments of this application, "irregular cross section" refers to a non-standard cross section form formed by combining one or more strip units according to preset width parameters, thickness parameters and bending angle parameters, and its cross section profile can be parametrically described by the strip sequence.

[0077] Step A2: Based on the bending angle and width of each strip unit in the strip sequence, connect each strip unit sequentially to generate the geometric profile of the irregular cross section.

[0078] This step transforms the parameterized strip description into a concrete two-dimensional geometry. By sequentially connecting the strip elements, the system can accurately construct the boundary of the irregular cross-section. For example, computational geometry libraries (such as OpenCASCADE or Parasolid) can be used to perform the sequential connection operation of strip elements, calculating the vertex coordinates and edge lines on the cross-sectional profile based on the width and bending angle of each strip. Alternatively, a custom geometric algorithm can be developed that iteratively calculates the geometric position of the next strip element based on the end point and orientation of the previous strip element, combined with the width and bending angle of the current strip element, thereby gradually drawing the complete geometric profile.

[0079] Step A3: Calculate the physical parameters of the irregular cross-section based on its geometric profile. The physical parameters include the cross-sectional area, centroid position, and principal moment of inertia.

[0080] This step links geometric information with the physical properties required for structural engineering, which are crucial for subsequent structural analysis. For arbitrary geometric profiles, numerical integration methods (e.g., based on Green's theorem) can be used to accurately calculate their cross-sectional area, centroid coordinates, and moment of inertia. Another approach is to decompose the complex geometric profile into multiple simpler geometric shapes (such as rectangles and triangles), then calculate the physical parameters of each simple shape separately, and finally superimpose them using methods such as the parallel axis theorem to obtain the total physical parameters of the entire irregular cross-section.

[0081] Step A4: Store the geometric contour and physical parameters of the irregular cross-section as a cross-section template in the database.

[0082] This step ensures the reusability and standardized management of irregular cross-section members. By storing defined and calculated irregular cross-sections as templates, they can be efficiently referenced and reused in subsequent designs, avoiding redundant modeling and calculations. A dedicated cross-section template table can be established in the database, containing fields for storing geometric profile data (e.g., in the form of coordinate point sequences, SVG paths, or parametric description strings) and calculated physical parameters such as cross-sectional area, centroid location, and principal moments of inertia. Each cross-section template can be assigned a unique identifier for reference in the member layers of a multi-level structural information model.

[0083] Through the above technical solution, this application effectively addresses the shortcomings of existing technologies in parametric generation, expression, calculation, and reuse when dealing with non-standard, complex, irregularly shaped cross-section components. This method decomposes complex irregular cross-sections into parameterizable plate-strip sequences, achieving automated construction from abstract parameters to precise geometric contours, greatly simplifying the definition process of irregular cross-sections. Simultaneously, it automatically calculates the physical parameters of the cross-section based on the generated geometric contours, ensuring the accuracy of structural analysis and achieving a deep integration of geometric information and mechanical properties. Furthermore, storing these irregular cross-sections as templates in a database allows designers to efficiently reuse pre-defined complex components, significantly improving design efficiency and model consistency. This parametric, automated, and template-based approach not only enhances the flexibility of structural design, enabling designers to quickly explore and apply various complex cross-sectional forms, but also improves the automation level of the entire modeling process, providing strong support for irregularly shaped components in building structural information model generation methods based on large language models.

[0084] Furthermore, in some embodiments, when generating the updated physical analysis model based on the updated multi-level structural information model and mapping relationship table in step S5, the target analysis scale can be further determined according to the specific analysis task, and the updated geometric model can be analyzed, idealized, and discretized according to the determined target analysis scale to generate a physical analysis model of the corresponding scale. Specifically, this includes: Step S61: Obtain the target analysis scale type, wherein the target analysis scale type includes at least one of the following: rod model, shell model, and solid model; Step S62: Based on the target analysis scale type, determine the physical analysis object corresponding to the target analysis scale type from the mapping relationship table, and perform analysis idealization and discretization processing on the updated geometric model corresponding to the target analysis scale type to generate a physical analysis model of the corresponding scale.

[0085] The "target analysis scale type" refers to the system determining the required model refinement level for structural analysis based on user needs, analysis objectives, or preset rules. For example, users can select this level through a graphical user interface (GUI), specifying whether to perform rod analysis, shell analysis, or solid analysis via drop-down menus, checkboxes, or input boxes. Alternatively, the system can automatically determine the appropriate level based on the specific analysis task. For instance, for overall structural stability analysis, the system might default to a rod model; while for stress concentration analysis of local connection nodes, the system might automatically select a solid model. The target analysis scale type covers three different refinement levels of physical models commonly used in structural analysis. The rod model (1D, one-dimensional) simplifies structural components to one-dimensional rod elements and is mainly used for overall structural force analysis, displacement, and internal force calculations, offering high computational efficiency. The shell model (2D, two-dimensional) simplifies components such as plates and walls to two-dimensional shell elements and is used to analyze thin-walled structures, floor slabs, shear walls, etc., capable of capturing in-plane and out-of-plane deformations. Solid models (3D) model components as three-dimensional solid elements, which can be used for detailed analysis of complex nodes, local three-dimensional stress states and contact problems, and can provide richer local three-dimensional response information, but usually have a large computational load.

[0086] After obtaining the target analysis scale type, the system queries the mapping table based on the geometric entity identifier and the target analysis scale type to determine the physical analysis object corresponding to the current target analysis scale and the associated analysis idealization rules. Then, in combination with the type of each geometric entity, the system performs corresponding analysis idealization processing on the updated geometric model.

[0087] When the target analysis scale is a rod system model, for rod entities such as beams, columns, and supports, their reference axes are extracted based on the geometric position and axial direction of the rods, and the cross-sectional parameters and material parameters associated with the rods are assigned to the rod system analysis object corresponding to the reference axis. Based on the connection attributes recorded in the connection layer, corresponding analysis nodes and connection relationships are established at the ends of the rods or at the connection positions of different rods to maintain the component connection characteristics in the original geometric model.

[0088] When the target analysis scale is a shell model, for thin-walled geometric entities whose thickness-to-in-plane feature size ratio is no greater than a preset thin-walled threshold and whose relative boundary surfaces can be identified, a reference surface or mid-surface located in the middle of its thickness direction is extracted from the thin-walled geometric entity. Specifically, for thin-walled geometric entities with corresponding boundary surfaces set relative to each other along the thickness direction, the reference surface or mid-surface is generated at the middle position between the two boundary surfaces based on the local thickness direction between the two boundary surfaces; for cases where the two boundary surfaces have curvature changes, the reference surface or mid-surface is determined based on the corresponding position established along the local thickness direction. The thickness and material parameters of the original geometric entity are used as the physical properties of the reference surface or mid-surface. For adjacent reference surfaces or mid-surfaces, their boundaries are trimmed, extended, or connected according to the topological connection relationship in the original geometric model, so that adjacent shell analysis objects maintain the connection relationship corresponding to the original structure. For complex geometric entities that cannot establish a stable correspondence between the two boundary surfaces or cannot reliably generate reference surfaces or mid-surfaces, their three-dimensional entity representation is retained without dimensionality reduction.

[0089] When the target analysis scale is a solid model, the three-dimensional volume geometry of the corresponding geometric entities is preserved, and the common boundaries or connection regions between different entities are determined based on the connection relationships in the original geometric model. For tiny edges, tiny faces, or local details in the geometric model whose size is smaller than the preset geometric feature threshold and are not used to define load locations, constraint locations, or component connection relationships, they can be merged or simplified; geometric features that affect loads, boundary conditions, or connection relationships are preserved.

[0090] After completing the above idealization analysis, the corresponding analytical geometry is discretized according to the target analysis scale. Specifically, the rod model generates one-dimensional elements along the reference axis; the shell model generates two-dimensional mesh elements on the reference surface or mid-surface; and the solid model generates three-dimensional mesh elements within the three-dimensional volume region. The mesh size can be determined based on a preset global mesh size, and local refinement is performed in component connection areas, areas with large geometric curvature changes, or areas with small local geometric dimensions. After the finite element mesh is generated, a quality check is performed on the generated mesh. Jacobi correlation indicators, element shape quality indicators, minimum side length, or other preset mesh quality or size indicators can be used to evaluate the finite element mesh. When elements do not meet the preset mesh quality requirements, while maintaining the corresponding geometric boundaries and topological connections, node position adjustments, local mesh optimization, or re-meshing are performed on the local area, and a quality check is performed again. If the mesh quality requirements still cannot be met after a preset number of optimizations, the corresponding geometric region is marked as a mesh anomaly region and the corresponding status is output.

[0091] Furthermore, for anomalous regions in the finite element model, geometric mesh anomalies and analytical singularities can be distinguished and processed accordingly. For mesh anomalies caused by tiny edges, narrow surfaces, coincident boundaries, or other local geometric details, the corresponding geometric features can be merged, simplified, or locally re-meshed without changing the main geometric features of the structure, load application locations, boundary conditions, and connectivity relationships. For regions that may generate stress singularities, such as sharp concave corners, concentrated load application points, and local concentrated constraints, the goal is not to continuously refine the mesh to eliminate local peaks. Instead, these regions are marked, and depending on the analysis purpose, at least one of the following processing methods is adopted, such as local geometric smoothing, converting concentrated loads into equivalent distributed loads, or local fine analysis, provided that preset equivalent modeling conditions are met. When the above processing would change the actual structural geometry, load definitions, or boundary conditions, the original analysis conditions are retained, and the corresponding singularity region identifier is output for use in subsequent analysis result evaluation.

[0092] Through the steps S61 to S62 above, the system not only achieves dynamic synchronization of geometric information and physical analysis information, but also solves the problem of model construction for complex structures at different analysis depths by using conversion logic for different scale types. This ensures the real-time performance and consistency of the analysis model after geometric changes, greatly reduces manual intervention, and improves analysis efficiency and accuracy.

[0093] Please see Figure 6 , Figure 6 This application provides a schematic diagram of an embodiment of a computer device for executing a method for generating a building structure information model based on a large language model. The computer device of this embodiment includes: at least one processor 10, a memory 11, and a computer program 12 stored in the memory 11 and executable on the at least one processor 10. When the processor 10 executes the computer program 12, it implements the steps in the embodiment of the building structure information model method based on a large language model of this application.

[0094] The computer device may be a desktop computer, laptop, smartphone, PDA, or other computing device that is equipped with drawing tools software.

[0095] Figure 6 The computer device shown may include, but is not limited to, processor 10 and memory 11. Those skilled in the art will understand that... Figure 6 The computer device 1 is merely an example and does not constitute a limitation on the computer device 1. It may include more or fewer components than shown in the figure, or combine certain components, or different components, such as input / output devices, network access devices, etc.

[0096] Furthermore, embodiments of this application also propose a system for generating a building structure information model based on a large language model, comprising: The interaction module is used to receive design input text; the interaction module can be presented through a human-computer interaction interface.

[0097] A large language model engine is used to perform semantic parsing on the design input text and extract structural design parameters; The database is used to store multi-level structural information models, which are arranged in a progressive combination hierarchy from local components to the overall system, including node layer, section layer, connection layer, rod layer, module unit layer and structural system layer in sequence. The parameter dependency graph module is used to construct parameter dependency graphs between the levels of the multi-level structure information model. The parameter dependency graph includes parameter nodes and directed edges connecting the parameter nodes. The directed edges are used to represent the dependency relationship between one parameter and another parameter. The path in the parameter dependency graph from one parameter node to another parameter node via a directed edge constitutes the dependency path. The topology identification and mapping module is used to perform topology identification based on the geometric model generated by the multi-level structural information model, identify each geometric entity in the geometric model, the geometric entity includes rods, node domains and module units, and establish a mapping relationship between each identified geometric entity and the corresponding physical analysis object, and generate a mapping relationship table. The linkage update module is used to update all target parameters that have direct or indirect dependencies with the modified parameter in sequence along the dependency path according to the parameter dependency graph when a modification instruction for any parameter in the multi-level structure information model is received, so as to obtain the updated multi-level structure information model. The model generation module is used to automatically generate updated geometric and physical analysis models based on the updated multi-level structural information model and the mapping relationship table.

[0098] It should be noted that the system described in this embodiment can represent a software architecture for generating building structure information models. Its specific implementation has been described in the foregoing method embodiments and will not be repeated here.

[0099] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention and should not be construed as limiting the specific implementation of the invention to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the inventive concept, and all such modifications and substitutions should be considered within the scope of protection of the present invention. That is, the processing control optimization method for intelligent manufacturing proposed in this invention can be applied to other intelligent manufacturing processes. In any scenario involving limited computing power and storage memory, the neural network architecture search method mentioned in this invention can be used to construct parameter prediction models that run on computing devices that meet these constraints.

Claims

1. A method for generating a building structure information model based on a large language model, characterized in that, The method is applied to a computer device, including: The design input text in response to the input is semantically parsed by a large language model to extract structural design parameters from the design input text; Based on the structural design parameters, a multi-level structural information model is created in the database; and a parameter dependency graph is constructed between the various levels of the multi-level structural information model. The parameter dependency graph includes parameter nodes and directed edges connecting the parameter nodes. The directed edges are used to represent the dependency relationship between one parameter and another parameter. The path in the parameter dependency graph from one parameter node to another parameter node via a directed edge constitutes a dependency path. The multi-level structural information model, according to the progressive combination hierarchy from local components to the overall system, includes at least the node layer, section layer, connection layer, rod layer, module unit layer, and structural system layer in sequence. A geometric model is generated based on the multi-level structural information model. Topological identification is performed on the geometric model to identify each geometric entity in the geometric model. The geometric entities include rods, node domains, and module units. For each identified geometric entity, a mapping relationship is established between each geometric entity and its corresponding physical analysis object, and a mapping relationship table is generated. When the computer receives a modification instruction for any parameter in the multi-level structure information model, it updates all target parameters that have direct or indirect dependencies with the modified parameter sequentially along the dependency path according to the parameter dependency graph, so as to obtain the updated multi-level structure information model. Based on the updated multi-level structural information model and the mapping relationship table, an updated geometric model and physical analysis model are generated.

2. The method as described in claim 1, characterized in that, The step of performing topological identification on the geometric model includes: Obtain the geometric identifier and boundary topology information of each candidate geometric entity from the geometric model, and treat each candidate geometric entity as a graph node; Based on the comparison results of the geometric intersection relationship, shared boundary relationship or minimum distance with the preset geometric tolerance between the candidate geometric entities, the connection relationship between the candidate geometric entities is determined, and the connection relationship is used as graph edges to construct a geometric topology graph; Traverse the geometric topology graph and identify and determine each rod, node domain, and module unit in the geometric model based on at least one of the object identifier, geometric features, and connection features of each candidate geometric entity.

3. The method as described in claim 1, characterized in that, The step of generating the updated geometric model and physical analysis model based on the updated multi-level structural information model and the mapping relationship table includes: Based on the updated multi-level structural information model, read the updated geometric parameters of each geometric entity and the updated physical parameters of each physical analysis object; Based on the mapping table, the updated geometric parameters of each geometric entity are mapped to the corresponding physical analysis object; Based on the updated geometric and physical parameters of the physical analysis object, the corresponding physical analysis model is automatically generated or updated according to the preset analysis model generation rules.

4. The method as described in claim 1, characterized in that, The steps for creating a multi-level structured information model in the database include: Register the project in the database and create a structural system record at the structural system layer; Based on the structural design parameters, each module unit instance is created in the module unit layer, and each module unit instance is recorded by referencing the module unit template ID and its spatial location parameters; In each of the module unit instances, each member instance is created in the member layer, and each member instance is recorded by referencing the member template ID and its spatial position parameters; At the connection points of each of the aforementioned member instances, a node instance is created in the node layer, and a connection attribute is determined in the connection layer, and the connection attribute is associated with the corresponding node instance; Based on the cross-sectional requirements of each member instance, a cross-sectional template is generated or referenced in the cross-sectional layer, and the cross-sectional template is associated with the corresponding member instance.

5. The method as described in claim 1, characterized in that, The method further includes: Receive a strip sequence input, wherein each strip unit in the strip sequence includes a width parameter, a thickness parameter, and a bending angle parameter; Based on the bending angle and width of each strip unit in the strip sequence, the strip units are connected sequentially to generate the geometric profile of the irregular cross section. Based on the geometric profile of the irregular cross-section, calculate the cross-sectional physical parameters of the irregular cross-section, including the cross-sectional area, centroid position, and principal moment of inertia; The geometric contour and physical parameters of the irregular cross-section are stored in the database as a cross-section template.

6. The method as described in claim 1, characterized in that, The step of generating an updated physical analysis model based on the updated multi-level structure information model and the mapping relationship table includes: Obtain the target analysis scale type, which includes at least one of the following: rod model, shell model, and solid model; The mapping table can record one or more mapping records for the same geometric entity, and each mapping record includes at least a geometric entity identifier, a physical analysis object identifier, and a corresponding analysis scale type. Based on the target analysis scale type, the physical analysis object corresponding to the target analysis scale type is determined from the mapping table, and the updated geometric model is subjected to analysis idealization and discretization processing corresponding to the target analysis scale type to generate a physical analysis model of the corresponding scale.

7. The method as described in claim 1, characterized in that, The construction of parameter dependency graphs between the various levels of the multi-level structured information model includes: The dependencies between parameters in the multi-level structural information model are identified according to preset parameter calculation rules. These rules include input parameters, output parameters, and calculation relationships used to determine the output parameters based on the input parameters. The parameters in the multi-level structural information model are used as parameter nodes. For each input parameter in the parameter calculation rules, directed edges are established pointing from the parameter node corresponding to the input parameter to the parameter node corresponding to the output parameter. These directed edges are then associated with the corresponding parameter calculation rules to construct a directed acyclic graph, resulting in the parameter dependency graph. The direction of the directed edges points from the dependent parameter to the dependent parameter. The parameter nodes in the directed acyclic graph are sorted according to the topological order of their dependencies to generate the parameter update sequence of the parameter dependency graph.

8. A system for generating building structure information models based on a large language model, characterized in that, include: The interaction module is used to receive design input text; A large language model engine is used to perform semantic parsing on the design input text and extract structural design parameters; The database is used to store multi-level structural information models, which are arranged in a progressive combination hierarchy from local components to the overall system, including node layer, section layer, connection layer, rod layer, module unit layer and structural system layer in sequence. The parameter dependency graph module is used to construct parameter dependency graphs between the levels of the multi-level structure information model. The parameter dependency graph includes parameter nodes and directed edges connecting the parameter nodes. The directed edges are used to represent the dependency relationship between one parameter and another parameter. The path in the parameter dependency graph from one parameter node to another parameter node via a directed edge constitutes the dependency path. The topology identification and mapping module is used to perform topology identification based on the geometric model generated by the multi-level structural information model, identify each geometric entity in the geometric model, the geometric entity includes rods, node domains and module units, and establish a mapping relationship between each identified geometric entity and the corresponding physical analysis object, and generate a mapping relationship table. The linkage update module is used to update all target parameters that have direct or indirect dependencies with the modified parameter in sequence along the dependency path according to the parameter dependency graph when a modification instruction for any parameter in the multi-level structure information model is received, so as to obtain the updated multi-level structure information model. The model generation module is used to automatically generate updated geometric and physical analysis models based on the updated multi-level structural information model and the mapping relationship table.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method for generating a building structure information model based on a large language model as described in any one of claims 1 to 7.

10. A computer program product, said computer program product storing a computer program, characterized in that, When the computer program is executed by a computer device, it implements the steps of the method for generating a building structure information model based on a large language model as described in any one of claims 1 to 7.

Images