Elderly care building function topology relationship adaptation generation method based on large language model

Abstract

The application relates to the technical field of building design and artificial intelligence, and discloses a method for generating an aging-adapted function topological relationship of a pension building based on a large language model, which comprises the following steps: S1, constructing a function topological relationship database of a pension building; S2, converting the function topological relationship of the pension building case in S1; S3, building a large language model training environment, and calling an API-key to start the large language model; S4, defining reasoning key rules based on the function requirements of a target pension building project, fine-tuning prompt parameters, and enabling the large language model to generate a function topological relationship prediction graph; S5, performing visual processing on the prediction result of S4 by using a Matplotlib library, and distinguishing the display styles of function nodes and connection relationships; S6, selecting the function topological relationship prediction graph meeting the requirements in S5, dividing function partitions, arranging rooms, and adjusting details to complete the plan scheme design of the pension building. The application improves the intelligent level of the pension building design by introducing the large language model and topological relationship reasoning.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of architectural design and artificial intelligence, specifically to a method for generating age-friendly functional topology relationships for elderly care buildings based on a large language model. Background Technology

[0002] As the aging population continues to grow, the demand for professional and high-quality elderly care spaces is becoming increasingly urgent. Traditional elderly care building designs often focus on the number of beds and economies of scale, neglecting the diverse needs of the elderly in areas such as safety, social interaction, and mental health, thus impacting their quality of life and well-being. Against this backdrop, the design philosophy of elderly care buildings is shifting from a "scale-oriented" to a "quality-oriented" approach, emphasizing "age-friendly design" and "healthy aging," and focusing on creating spatial environments that support the elderly's independent living, social interaction, and psychological comfort.

[0003] At the same time, the construction industry is undergoing a wave of digital transformation, with artificial intelligence technology providing new methods and tools for architectural design. Traditional design methods rely on manual sketches and experience-based judgment, making it difficult to systematically handle complex functional relationships and spatial organization. However, AI technologies such as large language models possess powerful natural language understanding and reasoning capabilities. By learning from large amounts of case data, they can identify topological relationships between functions, assisting architects in generating and optimizing spatial layouts.

[0004] Currently, research on applying large language models to the generative design of senior living buildings is still lacking, especially in the area of ​​age-friendly reasoning based on functional topological relationships, which lacks systematic and interpretable methods. Existing technologies mostly focus on morphological generation or parametric design, failing to deeply integrate the specific characteristics of senior living buildings (such as nursing circulation, centralized layout of social spaces, and independent bedrooms) for intelligent generation. Therefore, it is necessary to construct a generative design framework that integrates topology, age-friendly design principles, and the reasoning capabilities of large language models to improve the scientific rigor, efficiency, and humanistic care in the design of senior living buildings. Summary of the Invention

[0005] To address the aforementioned shortcomings in existing technologies, this invention provides an age-friendly method for generating functional topology relationships in elderly care buildings based on a large language model. This method enables intelligent prediction and generation of functional topology relationships in elderly care buildings through a large language model, promoting the deep integration of artificial intelligence and architectural design, and providing data-driven and rule-guided design support for age-friendly buildings.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: a method for generating age-friendly functional topology relationships in elderly care buildings based on a large language model, comprising the following steps:

[0007] S1. Construct a database of functional topology relationships for senior living buildings, collect cases of senior living buildings, and classify the general category and sub-category of senior living building functions according to building design specifications and the requirements of the target senior living building project. Define the connection relationships of three types of functional nodes: door connection, spatial connection and vertical connection.

[0008] S2 transforms the functional topology of the elderly care building case in S1, simplifies the building plan space and marks the functions and connections using Rhino modeling, extracts the building plan space data using Grasshopper battery pack, and outputs a functional connection table and an area and floor data table.

[0009] S3: Set up a large language model training environment and call the API-key to start the large language model;

[0010] S4. Based on the functional requirements of the target elderly care building project, define key inference rules and fine-tune the prompt parameter to enable the large language model to generate a functional topology prediction graph.

[0011] S5 uses the Matplotlib library to visualize the prediction results of S4, distinguishing the display styles of functional nodes and connection relationships;

[0012] S6: Select the functional topology prediction diagram that meets the requirements from S5, divide the functional areas and arrange the rooms, and adjust the details to complete the floor plan design of the elderly care building.

[0013] Furthermore, in the above-mentioned method for generating age-friendly functional topology relationships for elderly care buildings based on a large language model, the functional categories in S1 include living rooms, cultural and recreational and fitness rooms, rehabilitation and medical rooms, management and service rooms, and transportation spaces, with 103 functional subcategories. In S1, door connection refers to the connection between functional subcategories using doors as a spatial barrier, spatial connection refers to the connection between functional subcategories without visual obstruction, and vertical connection refers to the connection between functional subcategories on different floors.

[0014] Furthermore, the above-mentioned method for generating age-friendly functional topology relationships in elderly care buildings based on a large language model, specifically S2, involves: using Rhino modeling software and combining it with convex space segmentation, simplifying the concave space of the building case into convex space, using millimeters as the modeling unit, dividing each floor plan into a set of convex spaces using closed polylines, labeling the functions of the convex spaces, setting the entrance as the starting space, and connecting the convex spaces using three types of connections: single-line connection by door, spatial connection, and vertical connection, while maintaining vertical coordinate correspondence in the Z-axis direction for multi-story buildings; using the Grasshopper battery pack, extracting text labels and floor numbers using the Human plugin, extracting the coordinates and area of ​​the convex spaces using the Point and Geometry batteries, and then using the LunchBoxML plugin to output the connection relationship table and two types of table data: area and floor data, completing the machine-recognizable data format conversion.

[0015] Furthermore, in the above-mentioned method for generating age-friendly functional topology relationships for elderly care buildings based on a large language model, the underlying architecture of the large language model in S3 is a Transformer neural network. This network includes an encoder and a decoder. The encoder processes the input tokens and generates embeddings through a self-attention mechanism. The decoder integrates the encoder output information using a self-attention mechanism and a cross-attention mechanism. Finally, the output embeddings are converted into probabilities through a softmax layer to achieve the prediction and generation of functional topology relationships.

[0016] Furthermore, in the above-mentioned method for generating age-friendly functional topology relationships for elderly care buildings based on a large language model, the key inference rules in S4 include the total number of building floors, the area of ​​each floor, the distribution of functional subclass floors, the connection restrictions of functional nodes, the age-friendly layout pattern, the distribution of social spaces, and the naming and coordinate rules of functional nodes.

[0017] Furthermore, the key inference rules in S4 of the above-mentioned method for generating age-friendly functional topology relationships in elderly care buildings based on a large language model are as follows: the building has a total of four floors, with each floor having an area of ​​2300 square meters; all functional subclasses are interconnected; in multi-story buildings, elevators and stairs are set in the same location on each floor, and elevators and stairs are connected by corridors; vertical connections are only allowed between stairs or elevators; all functional subclasses are included in the prediction data and their number is not reduced; bedrooms cannot be connected in series, but can only be connected by corridors; if it is a multi-story building, the number of functional subclasses is evenly distributed on each floor; a circular layout pattern is adopted, social spaces are centrally distributed, social spaces are connected to other spaces, social spaces are distributed at the intersection of corridors, and nurse stations can be arranged in zones or in a concentrated manner; functional subclasses with the same name are distinguished by numbers.

[0018] Furthermore, in the above-mentioned method for generating age-friendly functional topology relationships for elderly care buildings based on a large language model, the specific distinction in display style in S5 is as follows: different RGB color values ​​are assigned to different functional nodes, and the connection relationships are displayed according to the rule that door connections are black, spatial connections are blue, and vertical connections are red.

[0019] The beneficial effects of this invention are as follows: This invention combines topology with a large language model, using the "node-relationship-node" triple as the original language of machine learning. With the help of the core mechanisms of self-attention and cross-attention in the Transformer architecture, it achieves accurate prediction and generation of functional topological relationships in elderly care buildings. At the same time, through Rhino modeling, Grasshopper battery extraction and other technical means, it completes the effective transformation and classification of building case data, constructs a scientific dataset, and provides accurate data support for design. Compared with the traditional design method that relies on design sketches and experience-driven design, it greatly improves the scientificity and efficiency of scheme generation and reduces the blindness in the design process.

[0020] This invention incorporates age-friendly inference rules and constraints through prompt fine-tuning, enabling the large language model to better understand and respond to age-friendly design requirements such as central spatial arrangement and concentrated distribution of social spaces. The generated functional topology diagram provides a reliable foundation for subsequent floor plan refinement. Simultaneously, it clarifies the role of the large language model as a design assistant, achieving collaborative work between artificial intelligence and architects. The complete process constructed by this invention, from data collection, transformation, model training to scheme refinement, provides strong support for the digital transformation of the field of elderly care building design.

[0021] This approach also responds to the need for a shift in elderly care building design from "scale and quantity-oriented" to "age-friendly and healthy aging" in the context of deepening aging. It addresses the problem of traditional designs neglecting the diverse physical and mental needs of the elderly, such as safety, social interaction, and mental health. By deeply integrating building functions with age-friendly concepts, it helps elderly care buildings evolve from simple living spaces into comprehensive spaces where the elderly can realize their self-worth and enjoy their later years, fully reflecting society's respect for the dignity, independence, and well-being of the elderly. Attached Figure Description

[0022] Figure 1 A visualization of the functional topology of a circular layout example;

[0023] Figure 2 For predicting the node connection graph;

[0024] Figure 3 This is a floor plan;

[0025] Figure 4 This is a second-floor plan;

[0026] Figure 5 This is a three-story floor plan;

[0027] Figure 6 This is a four-story floor plan. Detailed Implementation

[0028] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0029] This example uses the Ningbo Medical Rehabilitation Center project. The project is located in Tongjia Village, Zhuangqiao Street, Jiangbei District, Ningbo City (the site west of Ningbo Minkang Hospital). The total land area is approximately 4,000 square meters, designated for welfare facilities. The site is currently farmland, bordered by Ningbo Minkang Hospital to the east, a river to the west, and farmland to the north and south. The project has a construction scale of approximately 9,000 square meters, with an average bed area of ​​approximately 30 square meters. It requires the construction of living quarters, service rooms, recreational rooms, management rooms, and circulation spaces. The aim is to provide comprehensive services for the elderly, including medical care, rehabilitation training, and high-quality nursing care. The design must adhere to the "home" concept, creating a warm and comfortable age-friendly space.

[0030] I. Construction of Functional Topology Relationship Database for Senior Living Buildings

[0031] We collected 57 case studies of senior living buildings as the basic database. The case studies came from professional architecture platforms such as ArchDaily and Gooood, ensuring that the case studies cover senior living building types of different sizes and functional layouts, and meet the data diversity requirements for model training.

[0032] Based on the "Architectural Design Standard for Elderly Care Facilities" JGJ450-2018 and the functional requirements of the Ningbo Medical Rehabilitation Center construction project, the functions of elderly care buildings are divided into a two-level system: "General Functional Category - Sub-Functional Category". The general functional category includes five types: living rooms, recreational and fitness rooms, rehabilitation and medical rooms, management and service rooms, and circulation spaces. The sub-functional categories are further subdivided into 103 types, specifically including elderly people's rooms, treatment rooms, physiotherapy rooms, emergency rooms, nursing rooms, pharmacies, activity rooms, recreational rooms, barbershops, laundry rooms, dining rooms, workshops, multi-functional rooms, offices, staff rooms, changing rooms, storage rooms, gardens, atriums, rest rooms, terraces, corridors, entrance halls, and rear halls. The number and requirements of each sub-functional category are referenced in the project's functional task table (as shown in Table 1, including 105 bedrooms, 4 public restrooms, 4 assisted bathrooms, and 4 care stations, etc.).

[0033] Table 1:

[0034]

[0035] The connection relationships between the three functional nodes are clearly defined as follows:

[0036] Door connection: A connection method between functional subclasses that uses "doors" as a spatial barrier, such as the connection between a bedroom and a corridor.

[0037] Spatial connection: The connection between functional subclasses without visual obstruction, such as the connection between a corridor and a public living room.

[0038] Vertical connection: The connection method between functional subclasses on different floors, which is only allowed between staircases or elevators, such as the connection between the first floor staircase and the second floor staircase.

[0039] II. Functional Topology Relationship Transformation

[0040] The Rhino modeling software was used to remap convex spaces in 57 architectural cases, with the modeling unit set to millimeters. Combining the convex space segmentation method in spatial syntax, concave spaces in the building plan were simplified into convex spaces. Each floor plan was divided into a set of convex spaces using closed polylines, and the functional subclass corresponding to each convex space was labeled. The "entrance" was set as the starting space, and single lines were used to connect the convex spaces according to three types: "door connection," "space connection," and "vertical connection." For multi-story senior living building cases, the vertical coordinates of the functional nodes on each floor were maintained along the Z-axis to ensure the accuracy of the vertical connection relationships.

[0041] The Grasshopper battery pack is used to extract architectural floor plan data. The specific process is as follows:

[0042] The Human plugin is used to extract the text labels (function names) and floor numbers of each convex space.

[0043] The coordinates (X, Y, Z axes) and area data of each convex space are extracted using Point and Geometry cells.

[0044] The extracted data is converted into a machine-readable tabular format using the LunchBoxML plugin, ultimately outputting two types of tabular data:

[0045] The functional connection relationships are shown in Table 2, which records the functions, coordinates, and connection relationship information of the convex space, including fields such as "conversion point", "gate start point", "gate end point", "space start point", "space end point", "vertical start point", and "vertical end point". Example data is such as the conversion point {2.6405e+7,-1.1908e+7,0} corresponding to the gate start point {2.6418e+7,-1.1872e+7,0} and the gate end point {2.6411e+7,-1.1872e+7,0}.

[0046] Table 2:

[0047]

[0048]

[0049] The area and floor data are shown in Table 3. The records include the area and floor information of the functional nodes, including fields such as "total number of floors", "total area", "function", "conversion point", "floor", and "area". For example, the function "toilet" is located on the 2nd floor with an area of ​​19.453255㎡ and conversion point coordinates {25970.539681,-11829.335338,0}; the function "bedroom" is located on the 3rd floor with an area of ​​28.774585㎡ and conversion point coordinates {25970.483871,-11905.5393,0}, etc.

[0050] Table 3:

[0051]

[0052] III. Setting up a large language model training environment

[0053] Googlecolab was chosen as the training platform. The hardware configuration was an Intel i9-10980XE CPU and an NVIDIA RTX 3090 GPU. The software environment configuration was as follows: CudaRuntimeAPI V11.2.67 (for GPU matrix operations), CudaDriverAPI V11.5 (GPU driver), and Python 3.7 (project runtime environment).

[0054] Model Launch and Invocation: The underlying architecture of the large language model is a Transformer neural network, which includes an encoder and a decoder. The large language model is invoked via an API key. First, the input functional node information (such as functional name, coordinates, area, etc.) is converted into tokens. The encoder processes these tokens using a self-attention mechanism, extracting contextual information from other tokens and generating embeddings. The decoder integrates the encoder's output information using both self-attention and cross-attention mechanisms to ensure the accuracy of functional connectivity relationships. Finally, a softmax layer converts the output embeddings into probabilities, enabling the prediction and generation of functional topological relationships.

[0055] IV. Large Language Model Prompt Fine-tuning and Topological Relationship Prediction

[0056] Based on the requirements of the Ningbo Medical Rehabilitation Center construction project, the functional node requirements for model input are clarified. The overall functional categories remain the same: living quarters, service rooms, cultural and recreational rooms, management rooms, and transportation spaces. The number and location of functional subcategories must meet the requirements of the project task list (such as 105 bedrooms and 4 care stations).

[0057] Key inference rules defined: Based on the project characteristics and age-friendly design principles, key inference rules are defined as follows:

[0058] Basic building parameters: The building has four floors, each with a floor area of ​​approximately 2,300 square meters, for a total floor area of ​​approximately 9,200 square meters.

[0059] Connection constraint rules: All functional subclasses must be interconnected; multi-story buildings have elevators and stairs in the same location on each floor, and the elevators and stairs are connected by corridors; vertical connections are only allowed between stairs or elevators; bedrooms cannot be connected in series, but can only be connected by corridors, and each bedroom retains only one connection; if it is a multi-story building, the number of functional subclasses is evenly distributed on each floor.

[0060] Age-friendly layout rules: adopt a "ring" layout pattern (because the project site is square and surrounded by farmland and rivers, the ring layout can optimize the landscape view and spatial flow); social spaces (such as living rooms and dining rooms) should be concentrated and connected with other spaces; social spaces should be distributed at the intersection of corridors; nurse stations can be arranged in zones or in a concentrated manner.

[0061] Naming and Coordinate Rules: Functional subclasses with the same name are distinguished by numbers, such as Corridor 1 and Corridor 2; the coordinates of functional nodes need to be reasonably allocated in combination with the site boundary and building layout to ensure that each space location meets the usage requirements.

[0062] Prompt parameter fine-tuning and prediction generation: Based on project site information and inference rules, the initial prompt text is designed as follows: "You are an architect, and you need to generate corresponding connection triples according to the architectural cases (57 elderly care building cases) and the given architectural functional nodes (functional requirements of Ningbo Medical Rehabilitation Center). Your output format is: <Node 1>"<x,y,floor,area> Node 2<x,y,floor,area> ;Link method>\n<Node 3<x,y,floor,area> Node 4<x,y,floor,area> The linking method > ... and must adhere to the following rules: the building has four floors, each with an area of ​​approximately 2300 square meters; all functional subcategories are interconnected; in multi-story buildings, elevators and staircases are located at the same position on each floor; vertical connections are only allowed between staircases or elevators; bedrooms are only connected by corridors and are not linked; a circular layout is adopted, with social spaces centrally distributed and connected by spaces; functional subcategories with the same name are distinguished by numbers. Through multiple fine-tuning of the prompt parameters (such as adding details like "corridor width approximately 3.5 meters" and "a landscaped courtyard in the center"), the large language model continuously optimizes its output, ultimately generating a functional topology prediction diagram that meets the project's requirements. "Vertical connections" are represented by corresponding connections between staircases or elevators; corridors and rooms are connected by "doors"; corridors are distributed in a circular pattern, with rooms arranged around the corridors; and corridors and living rooms are connected by "spaces," conforming to age-friendly layout rules.

[0063] V. Visualization of Prediction Results

[0064] The Matplotlib library is used to visualize the functional topology prediction results generated by the large language model. The specific settings are as follows:

[0065] Functional node display: Different RGB color values ​​are assigned to different types of functional nodes (such as light pink for bedrooms, light yellow for living rooms, and light blue for corridors). The node size is set according to the area of ​​the functional space (the larger the area, the larger the node size) to ensure intuitive differentiation of each functional type.

[0066] Connection relationship display: The colors and styles of the connecting lines are set according to the rules of "door connection - black solid line," "space connection - blue solid line," and "vertical connection - red dashed line," clearly presenting different spatial connection methods. The final result is a functional connection topology view of a "ring" layout, as shown below. Figure 1 As shown, this diagram can intuitively display the location, area, and connection relationship of each functional node, making it easier for architects to quickly determine the rationality of the topological relationship.

[0067] VI. Detailed Design of the Floor Plan

[0068] Select the functional topology prediction graph with better prediction performance, such as... Figure 2 As shown, functional zoning is carried out based on the characteristics of the project site (adjacent to Minkang Hospital on the east side and a river on the west side):

[0069] Vertical zoning: The first floor consists of rehabilitation and service rooms (such as treatment rooms, emergency rooms, kitchens, and restaurants), with a central landscaped courtyard; the second to fourth floors are mainly residential rooms (bedrooms), along with living rooms, nurse stations, and other service and management rooms.

[0070] Horizontal zoning: The central area of ​​each floor is arranged with service and management rooms such as living rooms, nurse stations, and offices, while a circular traffic corridor and elderly living rooms are arranged around the center; the western area near the river is prioritized for activity rooms, training rooms, etc., to make full use of the landscape resources.

[0071] Room Layout and Detail Adjustments: A 7.2 x 7.2 meter column grid is used for the floor plan layout, with a corridor width of 3.5 meters. The rooms are arranged according to the quantity and area requirements of each function in the project's functional task list (e.g., 105 bedrooms, 4 public restrooms, etc.), and detailed adjustments are made accordingly.

[0072] First Floor Plan: The dining room is located in the center of the floor plan, with the nursing station situated at the intersection of the two end corridors; Activity Room 1, Activity Room 2, and Training Rooms 1-3 are located on the west side of the floor plan; the kitchen and offices are located on the south side of the floor plan; in addition to the dining room, an outdoor courtyard space is located in the central area, forming a landscaped view. Figure 3 As shown.

[0073] Second Floor Plan: The living room is located in the center of the floor plan, surrounded by four corridors that connect the living room and other rooms; an open resting space is designed at the intersection of the corridors; the rooms face the west-facing landscape, and the internal atrium is enclosed by a glass curtain wall structure to enhance natural light and visual communication. Figure 4 As shown.

[0074] The third and fourth floors: The living room remains in the center of the floor plan. To increase the number of rooms, in addition to rooms facing the west side with views, several rooms are also arranged facing the courtyard. Balconies and a nurses' station are located at the intersection of corridors, forming visual corridors. Figures 5-6 As shown.

[0075] Entrance and Traffic Organization: In conjunction with the surrounding roads (adjacent to Minkang Hospital on the east side, with a road width of 9 meters), the main entrance of the building is located in the southeast corner of the site, with a pedestrian plaza provided; the vehicle entrance is located in the northeast corner of the site, and the road on the right serves as a fire lane, ensuring clear and convenient traffic flow and meeting the travel needs of the elderly.

[0076] VII. Verification of Implementation Results

[0077] The method described in this embodiment successfully generated the functional topology diagram and floor plan design scheme for the Ningbo Medical Rehabilitation Center project. The scheme meets the following requirements:

[0078] Functional completeness: It covers 103 functional subcategories required by the project, and the number of each function meets the requirements of the task list (such as 105 bedrooms, 4 care stations, etc.). The functional layout is reasonable and meets the diverse needs of the elderly in medical care, rehabilitation, living and social activities.

[0079] Age-friendly features: The layout adopts a circular design with a central space arrangement. Social spaces are concentrated and connected to other spaces. Bedrooms are only connected by corridors and are not connected in series. Elevators and stairs are located in the same position on each floor, which meets the principles of age-friendly design and ensures the safety and convenience of the elderly.

[0080] Data accuracy: The coordinates and area data of functional nodes are consistent with the actual needs of the project (e.g., the area of ​​each floor is about 2300㎡ and the total building area is about 9200㎡). The connection relationship conforms to the preset rules. The prediction results match the manual refinement plan with a high degree of accuracy, which greatly improves the design efficiency and reduces the deviation caused by experience judgment in traditional design.

[0081] This embodiment fully verifies the feasibility and practicality of the age-friendly generation method of functional topology relationship of elderly care buildings based on large language model, and provides a practical path that can be promoted for the digitalization and intelligentization of elderly care building design.

Claims

1. A method for generating age-friendly functional topology relationships in elderly care buildings based on a large language model, characterized in that, Includes the following steps: S1. Construct a database of functional topology relationships for senior living buildings, collect cases of senior living buildings, and classify the general category and sub-category of senior living building functions according to building design specifications and the requirements of the target senior living building project. Define the connection relationships of three types of functional nodes: door connection, spatial connection and vertical connection. S2 transforms the functional topology of the elderly care building case in S1, simplifies the building plan space and marks the functions and connections using Rhino modeling, extracts the building plan space data using Grasshopper battery pack, and outputs a functional connection table and an area and floor data table. S3: Set up a large language model training environment and call the API-key to start the large language model; S4. Based on the functional requirements of the target elderly care building project, define key inference rules and fine-tune the prompt parameter to enable the large language model to generate a functional topology prediction graph. S5 uses the Matplotlib library to visualize the prediction results of S4, distinguishing the display styles of functional nodes and connection relationships; S6: Select the functional topology prediction diagram that meets the requirements from S5, divide the functional areas and arrange the rooms, and adjust the details to complete the floor plan design of the elderly care building.

2. The method for generating age-friendly functional topology relationships for elderly care buildings based on a large language model according to claim 1, characterized in that, The functional categories mentioned in S1 include living rooms, entertainment and fitness rooms, rehabilitation and medical rooms, management and service rooms, and transportation spaces, with 103 functional subcategories. The door connection mentioned in S1 refers to the connection method between functional subcategories using doors as a spatial barrier. The spatial connection refers to the connection method between functional subcategories without visual obstruction. The vertical connection refers to the connection method between functional subcategories on different floors.

3. The method for generating age-friendly functional topology relationships for elderly care buildings based on a large language model according to claim 1, characterized in that, S2 specifically involves: using Rhino modeling software and combining it with convex space segmentation, simplifying the concave space of the building case into convex space, using millimeters as the modeling unit, dividing each floor plan into a set of convex spaces using closed polylines, labeling the functions of the convex spaces, setting the entrance as the starting space, and connecting the convex spaces using three types of connections: single-line connection by door, spatial connection, and vertical connection, while maintaining vertical coordinate correspondence in the Z-axis direction for multi-story buildings; using the Grasshopper battery pack, extracting text labels and floor numbers using the Human plugin, and extracting the coordinates and area of ​​the convex spaces using the Point and Geometry batteries, and then using the LunchBoxML plugin to output the connection relationship table and the area and floor data tables, completing the data format conversion that can be recognized by the machine.

4. The method for generating age-friendly functional topology relationships for elderly care buildings based on a large language model according to claim 1, characterized in that, The underlying architecture of the large language model described in S3 is a Transformer neural network, which includes an encoder and a decoder. The encoder processes the input tokens and generates embeddings through a self-attention mechanism. The decoder integrates the encoder output information using a self-attention mechanism and a cross-attention mechanism. Finally, the output embeddings are converted into probabilities through a softmax layer to achieve functional topological relation prediction and generation.

5. The method for generating age-friendly functional topology relationships for elderly care buildings based on a large language model according to claim 1, characterized in that, The key inference rules described in S4 include the total number of building floors, the area of ​​each floor, the distribution of functional sub-class floors, the connection restrictions of functional nodes, the age-friendly layout pattern, the distribution of social spaces, and the naming and coordinate rules of functional nodes.

6. The method for generating age-friendly functional topology relationships for elderly care buildings based on a large language model according to claim 5, characterized in that, The key inference rules described in S4 are as follows: the building has a total of four floors, with each floor having an area of ​​2,300 square meters; all functional subclasses are interconnected; in multi-story buildings, elevators and stairs are set in the same location on each floor, and elevators and stairs are connected by corridors; vertical connections are only allowed between stairs or elevators; all functional subclasses are included in the prediction data and their number is not reduced; bedrooms cannot be connected in series, but can only be connected by corridors; if it is a multi-story building, the number of functional subclasses is evenly distributed on each floor; a circular layout pattern is adopted, social spaces are centrally distributed, social spaces are connected to other spaces, social spaces are distributed at the intersection of corridors, and nurse stations can be arranged in zones or in a centralized manner; Subclasses with the same name are distinguished by numbers.

7. The method for generating age-friendly functional topology relationships for elderly care buildings based on a large language model according to claim 1, characterized in that, The specific distinction of display styles described in S5 is as follows: different RGB color values ​​are assigned to different functional nodes, and the connection relationship is displayed according to the rule of black for door connections, blue for spatial connections, and red for vertical connections.

Images