Building structure design method and device based on large language model and multi-level graph representation

Abstract

The invention provides a building structure design method and device based on a large language model and multi-level graph characterization, the method is achieved through planning of a building structure design planner, the building structure design planner is obtained based on large language model supervision and fine tuning, and the method comprises the following steps: on the basis of a pre-trained building structure designer, the multi-level graph characterization is carried out; constructing a multi-level map according to the initial building design drawing and the basic parameter information of the building structure design; the building structure designer is constructed based on a large language model, and the multi-level atlas comprises a building contour level atlas, a building space level atlas, a building component level atlas, a potential structure level atlas, a structural component level atlas and a component size level atlas; and obtaining a target building structure design drawing according to the multi-level map. According to the method, the multi-level graph representation of the building design information is combined with the semantic reasoning capability of the large language model, so that the general intelligent design of the building structure is realized, and the design efficiency and the design scheme quality are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent design of building structures and the intersection of artificial intelligence technologies, and in particular to a method and device for building structure design based on a large language model and multi-level graph representation. Background Art

[0002] In the field of modern building structural design, traditional methods rely primarily on the expertise and experience of engineers. This process involves multiple aspects, including spatial layout, component design, structural system selection, component placement, and sizing. These tasks are often highly repetitive and require significant time and effort, resulting in overall inefficiency. Especially for newcomers, this design approach is not only difficult to master quickly but also has a high learning curve, hindering the development of intelligent architectural design.

[0003] To improve efficiency, recent attempts to apply artificial intelligence to building structural design have emerged. For example, some studies have proposed generative design methods specifically for optimizing the layout or size of specific structural components. However, these current generative design methods mostly focus on a single step in the design process, such as simply arranging beams and columns or optimizing cross-sectional dimensions. They lack a unified framework to comprehensively consider the overall aspects of building layout and structural design, making it difficult to ensure the overall coordination and safety of the final design.

[0004] At the same time, the rise of large language models (LLMs) has provided new solutions for complex design tasks. LLMs excel at processing and generating natural language information, and can carry and understand vast corpora of specialized knowledge, effectively guiding design decisions. However, despite their impressive performance in understanding and integrating complex knowledge, LLMs struggle to directly address architectural structural design problems involving complex geometric and topological relationships.

[0005] Therefore, how to solve the problem that existing building structure design methods lack integration and are difficult to directly handle the design of building structures with complex geometric topological relationships is an important issue that needs to be urgently addressed in the intersection of intelligent building structure design and artificial intelligence. Summary of the Invention

[0006] The present invention provides a method and device for building structure design based on a large language model and multi-level graph representation, which is used to overcome the defects of existing building structure design methods, such as lack of integration and difficulty in directly processing building structure designs with complex geometric topological relationships, and realize universal automatic generation of building structure designs.

[0007] On the one hand, the present invention provides a building structure design method based on a large language model and a multi-level graph representation, which is planned and implemented through a building structure design planner, and the building structure design planner is obtained by supervised fine-tuning based on a large language model; the building structure design method includes: based on a pre-trained building structure designer, constructing a multi-level graph according to initial building design drawings and basic parameter information of the building structure design; wherein, the building structure designer is constructed based on a large language model, and the multi-level graph includes a building outline level graph, a building space level graph, a building component level graph, a potential structure level graph, a structural component level graph, and a component size level graph; according to the multi-level graph, a target building structure design drawing is obtained.

[0008] Furthermore, the pre-trained architectural structure designer constructs a multi-level atlas according to the initial architectural design drawings and basic parameter information of the architectural structure design, including: determining the current stage according to the initial architectural design drawings; extracting the architectural structure information in the initial architectural design drawings to obtain the level atlas corresponding to the current stage; based on the pre-trained architectural structure designer, obtaining the level atlas corresponding to the next stage according to the level atlas corresponding to the current stage and the basic parameter information; obtaining the multi-level atlas according to the level atlases corresponding to multiple different stages; wherein the level atlas corresponding to the current stage is one of the building outline level atlas, the building space level atlas, the building component level atlas, the potential structure level atlas and the structural component level atlas.

[0009] Furthermore, different stages correspond to different building structure designers, and the hierarchical graph corresponding to the current stage includes nodes and edges, where edges represent building components and nodes represent intersections of building components; accordingly, the pre-trained building structure designer obtains the hierarchical graph corresponding to the next stage according to the hierarchical graph corresponding to the current stage and the basic parameter information, including: numbering the nodes and edges in the hierarchical graph corresponding to the current stage, where each edge contains the numbers of the two nodes connected to it; arranging the basic parameter information, the numbers of the nodes, the numbers of the edges and the features of the hierarchical graph corresponding to the current stage in sequence and encoding them to obtain the serialized features of the current stage; inputting the serialized features of the current stage into the pre-trained building structure designer of the current stage to obtain the hierarchical graph corresponding to the next stage; wherein the basic parameter information includes the function of the building, the site information of the building, the seismic design conditions of the building and the basic shape parameters of the building.

[0010] Furthermore, the multiple different stages include a building outline stage, a building space layout stage, a building component layout stage, a structural system stage, a structural layout stage, and a component size stage; wherein the building outline stage corresponds to the building outline hierarchy map, the building space layout stage corresponds to the building space hierarchy map, the building component layout stage corresponds to the building component hierarchy map, the structural system stage corresponds to the potential structure hierarchy map, the structural layout stage corresponds to the structural component hierarchy map, and the component size stage corresponds to the component size hierarchy map.

[0011] Furthermore, the characteristics of the building outline hierarchical map include building outline coordinates; the characteristics of the building space hierarchical map include building space coordinates, building space types, building space areas and building space adjacency relationships; the characteristics of the building component hierarchical map include building component coordinates, building component types, building component connection relationships and building component sizes; the characteristics of the potential structure hierarchical map include arrangable structure positions and arrangable structure types; the characteristics of the structural component hierarchical map include structural component coordinates, structural component types, structural component connection relationships; the characteristics of the structural size hierarchical map include structural component sizes.

[0012] Furthermore, it also includes: when the current stage is the structural layout stage or the component size stage, obtaining the retrieval enhancement result of the hierarchical graph corresponding to the current stage in the preset knowledge base; based on a pre-trained empirical rule evaluator, obtaining the empirical rule evaluation result according to the serialization characteristics of the current stage and the retrieval enhancement result; according to the empirical rule evaluation result, determining to output or regenerate the hierarchical graph corresponding to the current stage; wherein, the empirical rule evaluator is constructed based on a large language model.

[0013] Furthermore, it also includes: when the current stage is the component size stage, extracting the position parameters and design parameters of the building structure components from the hierarchical atlas corresponding to the current stage; generating a three-dimensional structural analysis model based on the position parameters and design parameters, as well as the basic parameter information; performing mechanical analysis verification on the three-dimensional structural analysis model to obtain a structural index verification result; and determining to output or regenerate the hierarchical atlas corresponding to the current stage based on the structural index verification result.

[0014] In a second aspect, the present invention also provides a building structure design device based on a large language model and a multi-level graph representation, applying any of the above-mentioned building structure design methods based on a large language model and a multi-level graph representation, including: a multi-level graph construction module, for constructing a multi-level graph based on a pre-trained building structure designer according to the initial building design drawings and basic parameter information of the building structure design; wherein, the building structure designer is constructed based on a large language model, and the multi-level graph includes a building outline level graph, a building space level graph, a building component level graph, a potential structure level graph, a structural component level graph and a component size level graph; a target building structure design drawing acquisition module, for acquiring a target building structure design drawing according to the multi-level graph.

[0015] In a third aspect, the present invention also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the architectural structure design method based on a large language model and multi-level graph representation as described above is implemented.

[0016] In a fourth aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described methods for designing building structures based on a large language model and multi-level graph representation.

[0017] The architectural structure design method provided by the present invention, based on a large language model and multi-level graph representation, constructs a multi-level graph based on initial architectural design drawings and basic architectural structure design parameter information using a pre-trained architectural structure designer. The architectural structure designer is constructed based on the large language model, and the multi-level graph includes a building outline level graph, a building space level graph, a building component level graph, a potential structure level graph, a structural component level graph, and a component size level graph. The target architectural structure design drawing is obtained based on the multi-level graph. By combining the multi-level graph representation of architectural design information with the semantic reasoning capabilities of the large language model, this method achieves universal intelligent design of architectural structures, significantly improving design efficiency and the quality of design solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1It is a flowchart of a building structure design method based on a large language model and multi-level graph representation provided by an embodiment of the present invention.

[0020] Figure 2 Schematic diagram of a multi-level graph provided by an embodiment of the present invention.

[0021] Figure 3 It is a schematic diagram of a building structure design at the structural arrangement stage provided by an embodiment of the present invention.

[0022] Figure 4 4 is a schematic diagram of the operation of the empirical rule evaluator provided by an embodiment of the present invention.

[0023] Figure 5 It is a working diagram of the structural index checker provided by an embodiment of the present invention.

[0024] Figure 6 It is a schematic diagram of the overall process of the building structure design method based on a large language model and multi-level graph representation provided by an embodiment of the present invention.

[0025] Figure 7 It is a structural schematic diagram of an architectural structure design device based on a large language model and multi-level graph representation provided by an embodiment of the present invention.

[0026] Figure 8 It is a schematic diagram of the physical structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0027] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0028] It's important to note that to improve the efficiency of traditional building structural design, some studies have proposed generative design methods specifically for the placement or sizing of specific structural components. However, existing generative design methods often operate independently, lacking unity and integration, making them incapable of supporting universal design needs. Specifically, existing generative design methods typically focus on a single step in the design process (e.g., placing beams and columns or optimizing cross-sectional dimensions), lacking a comprehensive understanding of building layout and structural design, making it difficult to ensure the overall coordination, safety, and reliability of the final solution.

[0029] The rise of large language models (LLMs) offers new possibilities for intelligent planning and knowledge integration for complex design tasks. LLMs excel at understanding and generating natural language and can carry a vast database of specialized knowledge to guide design decisions. However, LLMs alone are unable to directly handle architectural structural design problems involving complex geometric and topological relationships.

[0030] In view of this, an embodiment of the present invention provides a building structure design method based on a large language model and multi-level graph representation. The building structure design method is planned and implemented by a building structure design planner, and the building structure design planner is fine-tuned based on the large language model. Specifically, Figure 1 A flow chart of a building structure design method based on a large language model and multi-level graph representation provided by an embodiment of the present invention is shown.

[0031] like Figure 1 As shown, the method includes: S110, based on a pre-trained building structure designer, constructing a multi-level graph according to the initial building design drawings and basic parameter information of the building structure design; wherein, the building structure designer is constructed based on a large language model, and the multi-level graph includes a building outline level graph, a building space level graph, a building component level graph, a potential structure level graph and a structural component level graph; S120, obtaining a target building structure design drawing according to the multi-level graph.

[0032] The following will describe steps S110 - S120 and related steps in detail.

[0033] S110, based on a pre-trained building structure designer, construct a multi-level graph according to the initial building design drawings and basic parameter information of the building structure design; wherein, the building structure designer is constructed based on a large language model, and the multi-level graph includes a building outline level graph, a building space level graph, a building component level graph, a potential structure level graph, a structural component level graph, and a component size level graph.

[0034] It's easy to understand that the architectural structure design method based on a large language model and multi-level graph representation provided in this embodiment of the present invention includes a specially designed architectural structure designer for performing architectural structure design at different stages. In other words, different architectural structure designers correspond to different stages. The architectural structure designer is constructed based on the large language model, and prompt word engineering can be used to enable the large language model to perform architectural structure design.

[0035] Confirming the current stage and invoking the corresponding architectural structure designer can be accomplished through the architectural structure design planner designed in this embodiment. In other words, the architectural structure design method based on a large language model and multi-level graph representation provided in this embodiment is implemented using the architectural structure design planner. The architectural structure design planner is also constructed based on the large language model, utilizing prompt word engineering to imbue the large language model with thinking and planning capabilities, thereby completing the planning of structural design tasks and invoking the corresponding architectural structure designer.

[0036] Specifically, first of all, before conducting building structure design, it is necessary to obtain initial building design drawings. The initial building design drawings provide key information about the overall layout, space allocation, exterior shape, etc. of the building, which plays an important role in subsequent structural design.

[0037] Initial architectural design drawings include, but are not limited to, preliminary design drawings, building plans, elevations, sections, building details, and equipment layout drawings. Preliminary design drawings provide the basic form, functional zoning, and spatial relationships of the building, from which one can understand the overall framework and purpose of the building. Building plans, elevations, and sections detail the floor plan layout, exterior shape, and internal structural details of each floor of the building, helping to determine the location, size, and form of structural supports. Building details include detailed construction of specific parts or components, which can provide a basis for accurately calculating load distribution and selecting appropriate structural solutions. If the building design includes the installation requirements of special equipment, the corresponding equipment layout drawings can help understand the requirements of these special equipment on the building structure.

[0038] It should be noted that the initial architectural design drawings correspond to different stages, such as the architectural outline stage, the architectural space layout stage, the architectural component layout stage, the structural system stage, the structural layout stage or the component size stage. The information contained in the initial architectural design drawings of different stages is different, and the number of hierarchical maps to be constructed in different stages is also different. This will be elaborated in detail in the embodiments below.

[0039] After obtaining the initial architectural design drawings, the architectural structure information in the initial architectural design drawings is extracted to obtain a hierarchical map corresponding to the current stage. This hierarchical map may be a building outline hierarchical map, a building space hierarchical map, a building component hierarchical map, a potential structure hierarchical map or a structural component hierarchical map.

[0040] Next, the current hierarchical graph is fused with the basic parameter information of the building structure design to form a serialized feature. This serialized feature is then used as the input of the current building structure designer to obtain the output, which is the next hierarchical graph. This hierarchical graph may be a building space hierarchical graph, a building component hierarchical graph, a potential structure hierarchical graph, or a structural component hierarchical graph.

[0041] Therefore, the generation of the next-stage hierarchical map depends on the current-stage hierarchical map and the basic parameter information of the building structure design. Following this logic, we can repeatedly obtain hierarchical maps for all stages, i.e., multi-level maps. These multi-level maps refer to independent hierarchical maps for building outlines, building spaces, building components, potential structures, structural components, and component dimensions.

[0042] S120: Obtain a target building structure design drawing according to the multi-level atlas.

[0043] It is easy to understand that the building outline hierarchy map is used to describe the overall shape, height, volume and relationship of the building with the surrounding environment. The building space hierarchy map defines the spatial layout inside the building, including functional zoning, streamline design, etc. The building component hierarchy map involves non-structural components in the building (such as doors, windows, partitions) and some components that have both architectural and structural functions. The potential structure hierarchy map provides conceptual thinking about the building structural system, including possible structural forms (such as frame structure, shear wall structure). The structural component hierarchy map details the arrangement of specific structural components (such as beams, columns, and slabs). The component size hierarchy map details the size of specific structural components.

[0044] Based on the multi-level atlas, starting from the overall building outline, the work is gradually refined to the systematic work of spatial layout, component arrangement, and structural design. After integration in a logical sequence, the target building structural design drawing can be obtained. Modern design software can be used to improve the integration efficiency.

[0045] In this embodiment, a pre-trained architectural structure designer constructs a multi-level graph based on initial architectural design drawings and basic architectural structure design parameter information. The architectural structure designer is built based on a large language model, and the multi-level graph includes a building outline level graph, a building space level graph, a building component level graph, a potential structure level graph, a structural component level graph, and a component size level graph. Based on the multi-level graph, the target architectural structure design drawing is obtained. By combining the multi-level graph representation of architectural design information with the semantic reasoning capabilities of a large language model, this method achieves universal intelligent design of architectural structures, significantly improving design efficiency and the quality of design solutions.

[0046] On the basis of the above embodiments, the construction process of the multi-level graph will be described in detail below.

[0047] Based on a pre-trained building structure designer, a multi-level atlas is constructed according to the initial building design drawings and basic parameter information of the building structure design, including: determining the current stage according to the initial building design drawings; extracting the building structure information in the initial building design drawings to obtain the level atlas corresponding to the current stage; based on the pre-trained building structure designer, according to the level atlas corresponding to the current stage and the basic parameter information, obtaining the level atlas corresponding to the next stage; obtaining a multi-level atlas according to the level atlas corresponding to multiple different stages; wherein the level atlas corresponding to the current stage is one of the building outline level atlas, the building space level atlas, the building component level atlas, the potential structure level atlas and the structural component level atlas.

[0048] It is easy to understand that after obtaining the initial architectural design drawings, the building structure design planner can determine the current stage corresponding to the initial architectural design drawings and the specific structural design method based on the initial architectural design drawings. The current stage can be any one of the building outline stage, the building space layout stage, the building component layout stage, the structural system stage, the structural layout stage, and the component size stage.

[0049] Specifically, the current stage can be determined based on the building structure information contained in the initial architectural design drawings. For example, if the initial architectural design drawings only include the building outline, the current stage is the building outline level atlas stage.

[0050] Subsequently, by digitally extracting the building structure information from the initial architectural design drawings, the corresponding hierarchical map for the current stage can be obtained. For example, if the initial architectural design drawings only include the building outline, the corresponding hierarchical map for the current stage is the building outline hierarchical map.

[0051] In detail, if the current stage is the building outline stage, the corresponding hierarchical map is the building outline hierarchical map; if the current stage is the building space layout stage, the corresponding hierarchical map is the building space hierarchical map; if the current stage is the building component arrangement stage, the corresponding hierarchical map is the building component hierarchical map; if the current stage is the structural system stage, the corresponding hierarchical map is the potential structure hierarchical map; if the current stage is the structural arrangement stage, the corresponding hierarchical map is the structural component hierarchical map; if the current stage is the component size stage, the corresponding hierarchical map is the component size hierarchical map.

[0052] Furthermore, based on the pre-trained architectural structure designer, the hierarchical graph corresponding to the next stage is obtained according to the hierarchical graph and basic parameter information corresponding to the current stage. Specifically, the nodes and edges in the hierarchical graph corresponding to the current stage are numbered, with each edge containing the numbers of the two nodes connected to it. The basic parameter information, node numbers, edge numbers, and features of the hierarchical graph corresponding to the current stage are arranged in sequence and encoded to obtain the serialized features of the current stage. The serialized features of the current stage are input into the pre-trained architectural structure designer for the current stage to obtain the hierarchical graph corresponding to the next stage.

[0053] The basic parameter information of the building structure includes, but is not limited to, the function of the building, the site information of the building, the seismic design conditions of the building, and the basic shape parameters of the building.

[0054] Each hierarchical graph is composed of different nodes and edges. Edges represent line segments (i.e., building components), and nodes represent the intersections of line segments (i.e., the intersections of building components). The characteristics of the hierarchical graph corresponding to the current stage refer to the characteristics of the nodes and edges in the hierarchical graph.

[0055] If the current stage is the building outline stage, the corresponding hierarchical map of the current stage is the building outline hierarchical map, and the characteristics of the building outline hierarchical map include the building outline coordinates.

[0056] If the current stage is the space layout stage, the corresponding hierarchical map of the current stage is the building space hierarchical map. The characteristics of the building space hierarchical map include building space coordinates, building space type, building space area and building space adjacency.

[0057] If the current stage is the component arrangement stage, the corresponding hierarchical map of the current stage is the building component hierarchical map. The characteristics of the building component hierarchical map include building component coordinates, building component types, building component connection relationships, and building component sizes.

[0058] If the current stage is the structural system stage, the corresponding hierarchical map of the current stage is the potential structural hierarchical map. The characteristics of the potential structural hierarchical map include the arrangable structural positions and the arrangable structural types.

[0059] If the current stage is the structural layout stage, the corresponding hierarchical map of the current stage is the structural component hierarchical map. The characteristics of the structural component hierarchical map include structural component coordinates, structural component types, and structural component connection relationships.

[0060] If the current stage is the component size stage, the corresponding hierarchical map of the current stage is the component size hierarchical map, and the characteristics of the component size hierarchical map include the size of the structural components.

[0061] The basic parameter information of the building structure, the node numbers and edge numbers in the hierarchical graph corresponding to the current stage, and the features of the hierarchical graph corresponding to the current stage are arranged in order and encoded. The encoding method here can be direct encoding or self-encoding, which is not specifically limited here.

[0062] After obtaining the hierarchical atlas corresponding to the next stage, the hierarchical atlas corresponding to the next stage is used as the hierarchical atlas corresponding to the current stage, and the step of "based on the pre-trained building structure designer, according to the hierarchical atlas corresponding to the current stage and basic parameter information, obtaining the hierarchical atlas corresponding to the next stage" is continued. This cycle is repeated, and finally, hierarchical atlases corresponding to multiple different stages, that is, multi-level atlases, can be obtained.

[0063] It should be noted that the different stages of building structure design are, from simple to complex, the building space layout stage, the building component layout stage, the structural system stage, the structural layout stage, and the component size stage. The generation of the hierarchical atlas corresponding to each "next stage" depends on the hierarchical atlas corresponding to the "current stage".

[0064] Figure 2 FIG. 1 shows a schematic diagram of a multi-level atlas provided by an embodiment of the present invention. Figure 2 As shown, the multi-level atlas in the embodiment of the present invention includes a hierarchical atlas of the building outline layer (i.e., the building outline hierarchical atlas), a hierarchical atlas of the building space layer (i.e., the building space hierarchical atlas), a hierarchical atlas of the building component layer (i.e., the building component hierarchical atlas), a hierarchical atlas of the potential structure layer (i.e., the potential structure hierarchical atlas), and a hierarchical atlas of the structural component layer (i.e., the structural component hierarchical atlas). In addition, the multi-level atlas should also include a component size hierarchical atlas (i.e., the component size hierarchical atlas) generated based on the structural component hierarchical atlas. Figure 2 not shown).

[0065] The information contained in the initial building design drawings varies depending on the current stage, but at least the building outline information needs to be provided.

[0066] In a specific embodiment, taking the initial architectural design drawings at the component sizing stage as an example, the drawings already contain information on the building outline level map, the building space level map, the building component level map, the potential structure level map, and the structural component level map. Only the dimensional design task in the structural component layer needs to be completed.

[0067] First, the initial parameters required for the design (such as floor height, usage function, etc.) are read, and the drawing information is digitally extracted to obtain the outer contour of the building and the boundaries of each space. Polygons represent the building outline and spatial areas; the positions of building components such as walls, doors, and windows are identified; the positions of structural components such as beams, columns, and shear walls are identified; and based on the extracted information, the following are generated: Figure 2 The multi-level map shown.

[0068] Specifically, at the building outline level, nodes and edges are created to represent the building's exterior wall outlines. At the building space level, spatial nodes are created for the corners of each space (functional areas such as living rooms, bedrooms, and kitchens), with the spatial outlines serving as edges. At the building component level, corresponding nodes and edges are created for each identified wall segment and each door and window, describing the location and adjacency of the building components. At the potential component level, potential structural components are derived based on the location of the building components and the topological relationships of the building plan. The program generates candidate structural components according to preset rules at locations where structural support may be required. For example, candidate line segments for possible beam or shear wall locations are generated from the intersections of all wall and space boundaries with the building outline or adjacent components. Intersections between spaces and walls are marked as nodes for possible column locations. All these candidate structural elements constitute the nodes and edges of the potential structural layer. At the structural component level, corresponding nodes and edges are created for the identified beams, columns, and shear walls, describing the location and adjacency of the structural components.

[0069] In another specific embodiment, taking the initial architectural design drawings in the architectural space layout stage as an example, the drawings only include information on the building outline hierarchical map. It is also necessary to use architectural structure designers at different stages to generate architectural space hierarchical maps, building component hierarchical maps, potential structure hierarchical maps, structural component hierarchical maps, and component size hierarchical maps to obtain a multi-level map.

[0070] Specifically, the building structure design planner first invokes the building structure designer from the spatial layout phase to generate a spatial layout and obtain a building spatial hierarchy map. It then invokes the building structure designer from the building component placement phase to generate a building layout and obtain a building component hierarchy map. Based on its understanding of the building layout and function, it then determines the appropriate structural system type. In this embodiment, due to the high number of floors, the building plan layout containing numerous vertical and horizontal walls, and the regular spatial layout, a reinforced concrete frame-shear wall structural system is determined to be more suitable. The rationale is provided and a specific layout strategy is generated, such as: "Place shear walls along the building perimeter and important partition walls, install frame columns at building space intersections, and arrange beams between columns and shear walls along the upper sides of the building spaces." Simultaneously, based on this layout strategy and experience, the building structure designer from the structural system phase is invoked to generate a potential structure and obtain a potential structural hierarchy map. The building structure generator from the structural placement phase is then invoked to generate a structural layout and obtain a structural component hierarchy map. Finally, the building structure designer from the component sizing phase is invoked to generate structural component dimensions and obtain a component dimension hierarchy map.

[0071] In another specific embodiment, taking the initial architectural design drawings in the building component layout stage as an example, the drawings only include information on the building outline hierarchical map and the building space hierarchical map. It is also necessary to use architectural structure designers at different stages to generate building component hierarchical maps, potential structure hierarchical maps, structural component hierarchical maps, and component size hierarchical maps to obtain multi-level maps.

[0072] Specifically, the building structure designer in the building component layout phase is first invoked to generate the building layout and obtain a building component hierarchy map. Then, based on an understanding of the building layout and function, the appropriate structural system type is determined. Simultaneously, based on layout strategies and experience, the building structure designer in the structural system phase is invoked to generate a potential structure and obtain a potential structural hierarchy map. The building structure generator in the structural layout phase is then invoked to generate the structural layout and obtain a structural component hierarchy map. Finally, the building structure designer in the component sizing phase is invoked to generate structural component sizes and obtain a component size hierarchy map.

[0073] In other embodiments, Figure 3 A schematic diagram of a building structure design at the structural arrangement stage provided by an embodiment of the present invention is shown.

[0074] like Figure 3 As shown in the figure, taking the structural layout phase of building design as an example, an evaluation is made for each candidate location based on the nodes in the potential structural layer, determining whether to place a structural component at that location, the component type, and the length of the component. Taking into account factors such as the layout strategy provided by the large language model, the building function of the node location, and the span size, a layout plan is output, namely the structural component hierarchical map.

[0075] For example, potential component nodes along the building's perimeter and space partitions are determined to require shear walls; potential component nodes at space corners are determined to require frame columns; and potential components under the ceiling in the middle of the space are determined to require beams connecting adjacent columns or walls. Potential component nodes located at door openings are determined not to have any structural components installed to avoid compromising passage. Through this process, the Building Structural Designer assigns clear structural elements (columns, beams, shear walls, or blank spaces) to all potential structural nodes in the entire diagram (potential structural hierarchy map), forming a preliminary structural layout plan, known as the structural component hierarchy map.

[0076] The implementation process in other stages is similar. Specifically: (1) The building structure designer in the building space layout stage generates a preliminary space layout plan based on the building function requirements, usable area, traffic flow, lighting, ventilation and evacuation requirements, and obtains the building space hierarchy map; (2) The building structure designer in the building component layout stage further determines the layout of building components (walls, doors and windows) and obtains the building component hierarchy map; (3) The building structure generator in the structural system stage determines the possible location of the structure according to the building information and structure type and obtains the potential structure hierarchy map; (4) The building structure generator in the component size stage determines the cross-sectional dimensions for each structural component that has been arranged (such as beams, columns, shear walls, etc.) and obtains the component size hierarchy map.

[0077] It should be noted that the architectural structure designers at different stages are all implemented through prompt word engineering and supervised fine-tuning methods. The training dataset is constructed using data pairs before and after the design of different stages. The large language model is fine-tuned through supervised fine-tuning, and the prompt words of this stage are introduced to construct architectural structure designers for multiple stages.

[0078] In this embodiment, the current stage is determined based on the initial architectural design drawings, and the architectural structure information in the initial architectural design drawings is extracted to obtain the hierarchical graph corresponding to the current stage. Then, based on the hierarchical graph and basic parameter information corresponding to the current stage, a hierarchical graph corresponding to the next stage is obtained based on the pre-trained architectural structure designer. Furthermore, based on the hierarchical graphs corresponding to multiple different stages, a multi-level graph is obtained, and thus, the target architectural structure design drawing is obtained based on the multi-level graph. This method combines the multi-level graph representation of architectural design information with the semantic reasoning capabilities of a large language model to achieve universal intelligent design of architectural structures, significantly improving design efficiency and design solution quality.

[0079] On the basis of the above embodiments, the compliance assessment process of the multi-level graph will be described in detail below.

[0080] When the current stage is the structural layout stage or the component size stage, the retrieval enhancement result of the hierarchical graph corresponding to the current stage in the preset knowledge base is obtained; based on the pre-trained empirical rule evaluator, the empirical rule evaluation result is obtained according to the serialization characteristics of the current stage and the retrieval enhancement result; according to the empirical rule evaluation result, the output or regeneration of the hierarchical graph corresponding to the current stage is determined; wherein, the empirical rule evaluator is constructed based on a large language model.

[0081] It is easy to understand that compliance checks are required for the hierarchical graphs at specific stages, where the specific stages refer to the structural layout stage and the component size stage. In this embodiment, the current stage refers to the structural layout stage or the component size stage.

[0082] Specifically, this embodiment constructs an empirical rule evaluator by adopting a prompt word engineering method on a large language model, and uses the empirical rule evaluator to perform compliance checks on the hierarchical graph of the current stage. Figure 4 FIG. 4 shows a working diagram of the empirical rule evaluator provided by an embodiment of the present invention.

[0083] like Figure 4 As shown, this embodiment uses the current national "Code for Design of Building Structures" and related professional specifications as the preset knowledge base for the empirical rule evaluator. The hierarchical map corresponding to the current stage is used as the search keyword to search in the preset knowledge base to obtain the search enhancement results. The search enhancement results are then input into the pre-trained empirical rule evaluator along with the serialized features corresponding to the current stage. The structural layout and dimensions are checked item by item to obtain the empirical rule evaluation results. Among them, the structural layout and dimensions are checked item by item, for example: whether the spacing between beams and columns meets the maximum spacing requirements, whether the column grid layout is regular, and whether the shear wall length is compliant.

[0084] If a component is found not to meet the requirements of the specification (for example, the reinforcement ratio of a span beam exceeds the standard), the system will record the problem and add it to the feedback item, and finally return it to the building structure design planner. The building structure design planner will re-call the current stage of the building structure designer and regenerate the corresponding hierarchical atlas until it passes the verification of the empirical rule evaluator.

[0085] On the basis of the above embodiments, the structural index verification process of the multi-level graph will be described in detail below.

[0086] When the current stage is the component size stage, the position parameters and design parameters of the building structure components are extracted from the hierarchical atlas corresponding to the current stage; a three-dimensional structural analysis model is generated based on the position parameters and design parameters, as well as basic parameter information; a mechanical analysis verification is performed on the three-dimensional structural analysis model to obtain a structural index verification result; and based on the structural index verification result, the hierarchical atlas corresponding to the current stage is determined to be output or regenerated.

[0087] It is easy to understand that the structural index verification needs to be performed for the hierarchical map at a specific stage, where the specific stage refers to the component size stage. The current stage in this embodiment refers to the component size stage.

[0088] Specifically, this embodiment uses a structure indicator verifier to perform structure indicator verification on the hierarchical graph of the current stage. Figure 5 The figure shows a working diagram of the structural index checker provided by the embodiment of the present invention.

[0089] like Figure 5As shown, when performing structural index verification, the location parameters and design parameters of the building structure are first extracted from the hierarchical atlas corresponding to the current stage. Combined with the basic parameter information of the overall building structure design, parametric modeling can be used to obtain a three-dimensional structural analysis model, such as a finite element analysis model. Subsequently, structural analysis software such as Yingjianke and PKPM is used to perform mechanical analysis and verification on the structure in the three-dimensional structural analysis model under vertical loads and horizontal seismic effects to obtain the structural index verification results. Among them, the structural index verification results include the mechanical indicators of each building component, such as internal forces and deformations.

[0090] The structural indicator verification results are then compared with the predetermined structural indicator values. If the analysis reveals that the overall structural performance does not meet the requirements (for example, the lateral displacement of a certain floor exceeds the limit or the bearing capacity of an individual component is insufficient), the relevant information is recorded and returned to the building structure design planner. The building structure design planner then re-invokes the current building structure designer and regenerates the corresponding hierarchical diagram until the structural indicator verifier verifies the structure.

[0091] In this embodiment, a unified intelligent design framework is constructed by calling a building structure designer to implement building structure design, calling an empirical rule evaluator to implement compliance checking, and calling a structural index verifier to implement structural mechanical performance verification. This overcomes the fragmentation and inefficiency problems of existing technologies and realizes the universal automatic generation of building structure design.

[0092] In some other embodiments, Figure 6 The figure shows the overall process diagram of the architectural structure design method based on a large language model and multi-level graph representation provided by an embodiment of the present invention.

[0093] like Figure 6 As shown, the overall building structure design includes six stages, namely the building outline stage, the building space layout stage (i.e. Figure 6 Architectural space design in the building), building component layout stage (i.e. Figure 6 Architectural layout design in the structural system stage (i.e. Figure 6 structural type design in the process), structural layout stage, and component size stage.

[0094] In the building outline stage, building space layout stage, building component arrangement stage and structural system stage, based on the building structure designer corresponding to each stage, the hierarchical graph corresponding to each stage can be generated using the method of the above embodiment.

[0095] During the structural layout phase, after the building structure designer generates the structural component hierarchy diagram, it must be verified for compliance by the empirical rule evaluator. If non-compliant, feedback is sent to the building structure design planner, which then calls the corresponding building structure designer to regenerate the structural component hierarchy diagram and perform another compliance check. This cycle continues until the generated structural component hierarchy diagram passes the empirical rule evaluator's compliance check.

[0096] In the component size stage, after its building structure designer generates a component size hierarchy map, it is necessary not only to verify the compliance of the component size hierarchy map through the empirical rule evaluator, but also to verify the structural indicators of the component size hierarchy map through the structural indicator verifier.

[0097] If there are any non-compliance issues or structural indicators that do not meet the standards, these issues will be fed back to the Building Structural Design Planner, which will then call the corresponding Building Structural Designer to regenerate the component size hierarchy diagram and perform another compliance check and structural indicator verification. This cycle continues until the generated component size hierarchy diagram passes the compliance check of the empirical rule evaluator and the structural indicator verification of the structural indicator verifier.

[0098] Finally, a multi-level atlas is obtained, which can be used to restore the original information of the building structure, visualize the design results at different stages, and further integrate to obtain the target building structure design drawing.

[0099] It is worth mentioning that the architectural structure design method based on a large language model and multi-level graph representation described in the above embodiments is planned and implemented through an architectural structure design planner, which is obtained through supervised fine-tuning based on the large language model. Specifically, the architectural structure designer, empirical rule evaluator, and structural index verifier mentioned in the architectural structure design process are all called by the architectural structure design planner. In other words, the architectural structure design method provided by the embodiments of the present invention determines which design method to use at which stage. This process is intelligent and not manually planned.

[0100] Corresponding to the architectural structure design method based on a large language model and multi-level graph representation described in the above embodiments, the present invention also proposes an architectural structure design device based on a large language model and multi-level graph representation.

[0101] Specifically, Figure 7 A schematic structural diagram of an architectural structure design device based on a large language model and multi-level graph representation provided by an embodiment of the present invention is shown.

[0102] like Figure 7As shown, the device includes: a multi-level graph construction module 710, which is used to construct a multi-level graph based on a pre-trained building structure designer according to the initial building design drawings and basic parameter information of the building structure design; wherein, the building structure designer is constructed based on a large language model, and the multi-level graph includes a building outline level graph, a building space level graph, a building component level graph, a potential structure level graph, a structural component level graph and a component size level graph; a target building structure design drawing acquisition module 720, which is used to obtain a target building structure design drawing according to the multi-level graph.

[0103] In this embodiment, a multi-level graph construction module 710 constructs a multi-level graph based on a pre-trained architectural structure designer, initial architectural design drawings, and basic architectural structure design parameter information. The architectural structure designer is constructed based on a large language model, and the multi-level graph includes a building outline level graph, a building space level graph, a building component level graph, a potential structure level graph, a structural component level graph, and a component size level graph. A target architectural structure design drawing acquisition module 720 acquires the target architectural structure design drawing based on the multi-level graph. By combining the multi-level graph representation of architectural design information with the semantic reasoning capabilities of the large language model, this device achieves universal intelligent design of architectural structures, significantly improving design efficiency and design solution quality.

[0104] It should be noted that the architectural structure design device based on a large language model and multi-level graph representation provided in the embodiment of the present invention can be referenced to each other with the architectural structure design method based on a large language model and multi-level graph representation described in the above embodiments, and will not be repeated here.

[0105] Figure 8 An example of a physical structure diagram of an electronic device is shown below. Figure 8As shown, the electronic device may include: a processor 810, a communications interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communications interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 may invoke logic instructions in the memory 830 to execute a building structure design method based on a large language model and a multi-level graph representation. The method includes: constructing a multi-level graph based on initial building design drawings and basic parameter information of the building structure design based on a pre-trained building structure designer; wherein the building structure designer is constructed based on the large language model, and the multi-level graph includes a building outline level graph, a building space level graph, a building component level graph, a potential structure level graph, a structural component level graph, and a component size level graph; and obtaining a target building structure design drawing based on the multi-level graph.

[0106] Furthermore, the logic instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0107] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the architectural structure design method based on a large language model and a multi-level graph representation provided by the above-mentioned methods, the method comprising: based on a pre-trained architectural structure designer, constructing a multi-level graph according to the initial architectural design drawings and basic parameter information of the architectural structure design; wherein, the architectural structure designer is constructed based on a large language model, and the multi-level graph includes a building outline level graph, a building space level graph, a building component level graph, a potential structure level graph, a structural component level graph, and a component size level graph; according to the multi-level graph, obtaining a target architectural structure design drawing.

[0108] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0109] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0110] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A building structure design method based on a large language model and multi-level graph representation, characterized in that: The building structure design method is implemented by planning through a building structure design planner, which is obtained by fine-tuning the supervision of a large language model; The building structure design method comprises: Based on a pre-trained building structure designer, a multi-level graph is constructed according to the initial building design drawings and basic parameter information of the building structure design; wherein the building structure designer is constructed based on a large language model, and the multi-level graph includes a building outline level graph, a building space level graph, a building component level graph, a potential structure level graph, a structural component level graph, and a component size level graph; According to the multi-level atlas, a target building structure design drawing is obtained.

2. The architectural structure design method based on a large language model and multi-level graph representation according to claim 1 is characterized in that: The pre-trained building structure designer constructs a multi-level graph based on the initial building design drawings and basic parameter information of the building structure design, including: Determine the current stage based on the initial architectural design drawings; Extracting architectural structure information from the initial architectural design drawings to obtain a hierarchical map corresponding to the current stage; Based on the pre-trained building structure designer, obtaining the hierarchical graph corresponding to the next stage according to the hierarchical graph corresponding to the current stage and the basic parameter information; According to the hierarchical maps corresponding to the multiple different stages, the multi-level map is obtained; Among them, the hierarchical map corresponding to the current stage is one of the building outline hierarchical map, the building space hierarchical map, the building component hierarchical map, the potential structure hierarchical map and the structural component hierarchical map.

3. The architectural structure design method based on a large language model and multi-level graph representation according to claim 2 is characterized in that: Different stages correspond to different building structure designers. The hierarchical graph corresponding to the current stage includes nodes and edges, where edges represent building components and nodes represent intersections of building components. Accordingly, the pre-trained building structure designer obtains the hierarchical graph corresponding to the next stage according to the hierarchical graph corresponding to the current stage and the basic parameter information, including: Numbering the nodes and edges in the hierarchical graph corresponding to the current stage, where each edge contains the numbers of the two nodes connected to it; Arranging the basic parameter information, the node numbers, the edge numbers, and the features of the hierarchical graph corresponding to the current stage in sequence and encoding them to obtain the serialized features of the current stage; Inputting the serialized features of the current stage into a pre-trained building structure designer of the current stage to obtain a hierarchical graph corresponding to the next stage; The basic parameter information includes the function of the building, the site information of the building, the seismic design conditions of the building, and the basic shape parameters of the building.

4. The architectural structure design method based on a large language model and multi-level graph representation according to claim 3 is characterized in that: The multiple different stages include the building outline stage, the building space layout stage, the building component layout stage, the structural system stage, the structural layout stage and the component size stage; wherein, The building outline stage corresponds to the building outline hierarchical atlas, the building space layout stage corresponds to the building space hierarchical atlas, the building component arrangement stage corresponds to the building component hierarchical atlas, the structural system stage corresponds to the potential structure hierarchical atlas, the structural arrangement stage corresponds to the structural component hierarchical atlas, and the component size stage corresponds to the component size hierarchical atlas.

5. The architectural structure design method based on a large language model and multi-level graph representation according to claim 3 is characterized in that: The features of the building outline level atlas include building outline coordinates; The characteristics of the building space hierarchical map include building space coordinates, building space types, building space areas and building space adjacency relationships; The characteristics of the building component hierarchical atlas include building component coordinates, building component types, building component connection relationships, and building component sizes; The characteristics of the potential structure hierarchy map include arrangable structure positions and arrangable structure types; The characteristics of the structural component hierarchical map include structural component coordinates, structural component types, and structural component connection relationships; The structural size hierarchy map is characterized by structural component sizes.

6. The architectural structure design method based on a large language model and multi-level graph representation according to claim 4 is characterized in that: Also includes: When the current stage is the structure layout stage or the component size stage, obtaining a search enhancement result of the hierarchical graph corresponding to the current stage in a preset knowledge base; Based on a pre-trained empirical rule evaluator, obtaining an empirical rule evaluation result according to the serialization features of the current stage and the retrieval enhancement result; Determine output or regenerate the hierarchical graph corresponding to the current stage according to the empirical rule evaluation result; Wherein, the empirical rule evaluator is constructed based on a large language model.

7. The architectural structure design method based on a large language model and multi-level graph representation according to claim 4 is characterized in that: Also includes: When the current stage is the component size stage, extracting location parameters and design parameters of the building structure components from the hierarchical atlas corresponding to the current stage; generating a three-dimensional structural analysis model according to the position parameters and design parameters, as well as the basic parameter information; Performing mechanical analysis and calculation on the three-dimensional structural analysis model to obtain structural index calculation results; According to the structural indicator verification result, determine to output or regenerate the hierarchical map corresponding to the current stage.

8. A building structure design device based on a large language model and multi-level graph representation, applying the building structure design method based on a large language model and multi-level graph representation according to any one of claims 1 to 7, characterized in that: include: A multi-level graph construction module is used to construct a multi-level graph based on a pre-trained building structure designer, according to the initial building design drawings and basic parameter information of the building structure design; wherein the building structure designer is constructed based on a large language model, and the multi-level graph includes a building outline level graph, a building space level graph, a building component level graph, a potential structure level graph, a structural component level graph, and a component size level graph; The target building structure design drawing acquisition module is used to acquire the target building structure design drawing according to the multi-level atlas.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the architectural structure design method based on a large language model and multi-level graph representation is implemented as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the architectural structure design method based on a large language model and multi-level graph representation as described in any one of claims 1 to 7 is implemented.

Images