Architectural design planning scheme generation method and device based on artificial intelligence, and medium
By constructing a Generative Adversarial Network (GAN) model and a hierarchical generation strategy, combined with a Large Language Model (LLM), architectural design data is automatically processed, solving the problem of excessively long design processes in existing technologies and achieving efficient and rational architectural design planning.
Patent Information
- Application Number
- CN202511451412.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-11
- Publication Date
- 2025-12-16
AI Technical Summary
In the current architectural design process, repeated revisions and verifications are too time-consuming, especially when dealing with complex building shapes and massive amounts of drawing data. This is inefficient, prone to human error, and makes it difficult to respond quickly to market changes.
An AI-based architectural design and planning scheme generation method is adopted. By constructing a Generative Adversarial Network (GAN) model and a hierarchical generation strategy, combined with a Large Language Model (LLM) and a conditional generation module, architectural design data is automatically processed and multi-source information collaboratively driven design and planning drawings are generated.
Significantly improves design efficiency, reduces manual review costs, ensures the rationality of the functional layout and the coordination of element proportions in the generated drawings, enhances the generalization ability of the model, and shortens the design cycle.
Smart Images
Figure CN121145666A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application relates to an intelligent building design technology, in particular to a building design planning scheme generation method, device and medium based on artificial intelligence. BACKGROUND
[0002] In the field of building design, with the wide application of computer technology in the field of building, numerous auxiliary design research and practice have emerged as the times require. In the direction of three-dimensional visualization, professional modeling software such as SketchUp can quickly generate scene effect pictures of multiple perspectives, multiple lighting conditions and multiple material effects based on building models by means of parameterized adjustment and rendering engine technology, which significantly improves the visualization efficiency of design schemes. In terms of visual creativity, computer vision technologies such as image synthesis and image translation can automatically complete material replacement, texture synthesis and object addition based on existing scene images, realize style transfer of building scenes, and provide multiple creative choices for design schemes. However, in the existing technology, the time-consuming of repeated modification and verification in the process of intelligent building design is too long, which has become a major bottleneck for the development of the industry. From the conceptual design to the delivery of construction drawings, each link needs to be audited and adjusted for multiple rounds. Especially when facing complex building shapes, spatial relationships and massive drawing data, the efficiency is low and human errors are easy to occur, which prolongs the design cycle, causes the project progress to be out of control, and makes it difficult to quickly respond to market changes. SUMMARY
[0003] The technical problem to be solved by the application is to provide a building design planning scheme generation method, device and medium based on artificial intelligence to solve the defects in the prior art.
[0004] The technical scheme adopted by the application to solve the technical problem is: a building design planning scheme generation method based on artificial intelligence, comprising the following steps: Step 1, obtaining building design historical data from a database, wherein the building design historical data includes historical building building regulation documents, specification standard documents, design planning drawing data, building geographic information data and user building design requirement data; Step 2, cleaning the building design historical data, including removing duplicate, error and incomplete data records; Then, the building regulation documents, specification standard documents, geographic information data and user building design requirement data are generated into structured description text and XML files through a large language model LLM; annotating the design planning drawing data with the structured description text and the XML file; Step 3, performing flipping, rotating, cropping and color adjusting operations on the images of the annotated design planning drawing data to obtain a second data set after data augmentation; Step 4, constructing a generative adversarial network (GAN) model, training the GAN model using the second data set, generating a plurality of expanded first design planning drawing data using the trained GAN model, and merging the expanded first design planning drawing data into the second data set to obtain a third data set; Step 5, constructing a building design planning drawing generation model, The input of the building design planning drawing generation model is structured description text and XML files generated from geographic information data, building regulation documents, specification standard documents, and user building design requirement data, and the output is design planning drawing data. Step 6, training the building design planning drawing generation model using the third data set to obtain a trained building design planning drawing generation model; Step 7, obtaining user input building design requirement data, geographic information data, building regulation documents, and specification standard documents, and inputting the building design planning drawing generation model to generate a plurality of design planning drawing data.
[0005] According to the above scheme, in step 1, the design planning drawing data includes design plan, elevation, section, effect, technical parameter, and project construction data; the building geographic information data includes terrain data, topography data, climate data, and surrounding environment data of the building; The user building design requirement data includes functional requirement, aesthetic preference, and budget data.
[0006] According to the above scheme, in step 2, the structured description text and XML file are generated, including the following: Extracting mandatory provisions in building regulation documents and technical parameter indicators in specification standard documents to form structured description text; Converting geographic information data into structured description text including terrain features, climate zones, and surrounding building types; Parsing user requirement data to obtain XML files including functional requirement relationship matrix, aesthetic keyword vector, and budget constraint conditions.
[0007] According to the above scheme, in step 4, the generative adversarial network (GAN) model generates a plurality of expanded first design planning drawing data by inputting different random noise vectors into the generator.
[0008] According to the above scheme, in step 4, the generator of the GAN model adopts U-Net architecture, and the discriminator adopts PatchGAN architecture.
[0009] According to the above scheme, in step 5, the model includes: A conditional generation module is configured to convert the numerical vector after parsing the user demand into a semantic feature vector by using a text encoder of the CLIP. An image generation module is configured to generate an architectural design image meeting the demand from random noise by a step-by-step denoising process under the guidance of the semantic feature vector; the image generation module is a neural network of a UNet architecture. A conditional adjustment module is configured to perform noise disturbance on the semantic feature vector to enhance the generalization ability of the model to the demand.
[0010] According to the above scheme, the image generation module adopts a hierarchical generation module and a CFG (Classifier-Free Guidance) control strategy; wherein, The hierarchical generation module is configured to first generate a functional layout skeleton and then refine architectural element details; the CFG control is used to generate the number of images, and a plurality of design planning drawings are generated by adjusting the guidance coefficient.
[0011] The application further provides an electronic device, at least one processor; and A memory in communication connection with the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the above-mentioned architectural design planning scheme generation method.
[0012] The application further provides a machine-readable storage medium, the machine-readable storage medium stores instructions for causing a machine to execute the above-mentioned architectural design planning scheme generation method.
[0013] The application has the following beneficial effects: 1. The hierarchical generation strategy and the architectural element consistency loss are adopted to ensure the rationality of the functional layout and the proportionality between elements of the generated drawings. 2. The design area violating the mandatory provisions can be automatically identified to reduce the cost of manual review, effectively integrate the text semantics and visual features, realize the collaborative driving design generation of multi-source information, and significantly improve the design efficiency.
[0014] 3. The application combines the traditional image enhancement and the GAN model dual expansion mode to solve the problem of insufficient training samples and effectively improve the generalization ability of the model. BRIEF DESCRIPTION OF DRAWINGS
[0015] The application will be further described below with reference to the accompanying drawings and embodiments, wherein: Figure 1 is a method flowchart of an embodiment of the application. DETAILED DESCRIPTION
[0016] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.
[0017] As shown in Figure 1 A building design planning scheme generation method based on artificial intelligence, comprising the following steps: Step 1, obtaining building design historical data from a database, the building design historical data including historical building building regulation documents, specification standard documents, design planning drawing data, building geographic information data and user building design requirement data; Among them, the design planning drawing data includes design plan, elevation, section, effect, technical parameter, project construction data; the building geographic information data includes topographic data, geomorphic data, climate data, surrounding environment data of the building; The user building design requirement data includes functional requirement, aesthetic preference, budget; Step 2, cleaning the building design historical data, including removing duplicate, error, incomplete data records; Then, the building regulation documents, specification standard documents, geographic information data, user building design requirement data are generated into structured description text and XML file through large language model LLM; The design planning drawing data is labeled with the structured description text and the XML file; In this embodiment, the LLM model adopts GPT-4 or Llama-3 model, and the structured description text and XML file are generated, including the following: Extracting mandatory provisions in building regulation documents and technical parameter indicators in specification standard documents to form structured description text; The geographic information data is converted into structured description text including topographic features, climate zoning and surrounding building types; Analyzing user requirement data to obtain XML file including functional requirement relationship matrix, style (aesthetic) keyword vector and budget constraint condition.
[0018] For example, building regulations documents (PDF / Word), standard documents (tables / text), geographic information data (GIS format / CSV), and user requirement data (questionnaires / text) are uniformly converted into plain text format, and core content is extracted using regular expressions and natural language processing (NLP) techniques. A regulatory knowledge graph is constructed: the building regulations system is organized, and key information such as clause type (mandatory / recommended), applicable scenarios, and penalties are labeled. Named entity recognition (NER) and rule matching technologies are used to locate keywords such as "must," "prohibited," and "should" in regulatory documents through an LLM model, extracting mandatory clauses and generating a structured list. A structured output template (such as JSON format) is designed, including fields such as "topographic features (mountainous / plain)," "climate zones (severe cold / hot summer / cold winter)," and "surrounding building types (residential / commercial)." Combined with GIS spatial analysis tools, quantitative data such as topographic slope and sunshine duration are extracted and integrated with the qualitative descriptions generated by the LLM to form a complete environmental feature description.
[0019] To utilize XML files with hierarchical tags to clearly classify and organize various elements in architectural design (such as walls, doors, windows, beams, and columns), transforming previously scattered and unstructured architectural data into a standardized format, attribute parameters precisely define the dimensions, materials, performance, and other characteristics of each element. In a more preferred embodiment of this invention, the generated XML file includes hierarchical architectural element tags, attribute parameters, and relational constraints.
[0020] Step 3: Flip, rotate, crop, and adjust the color of the labeled design planning drawing data to obtain the second dataset after data augmentation. In this embodiment, the design planning drawing image is horizontally or vertically flipped, with rotations of 90°, 180°, and 270°. Random cropping is used to maintain the integrity of architectural elements while introducing different perspectives. Image brightness is adjusted by ±20%, contrast by ±15%, and saturation by ±10% to simulate different drawing styles. Step S3, data augmentation, expands the limited training samples into a large-scale, diverse dataset through geometric transformations and color perturbations. Experiments show that the dataset augmented by this method can improve the model's convergence speed by more than 30% within the same training cycle, while reducing the generalization error on unseen drawings by 15%-20%.
[0021] Step 4: Construct a Generative Adversarial Network (GAN) model. Train the GAN model using the second dataset. Using the trained GAN model, generate multiple augmented first design planning drawing data by inputting different random noise vectors into the generator. Merge the augmented first design planning drawing data into the second dataset to obtain the third dataset. In the Generative Adversarial Network (GAN) model, the generator adopts the U-Net architecture, and the discriminator adopts the PatchGAN architecture. Features are extracted from the input structured descriptive text and XML files to form feature vectors. The feature vectors are added to the random noise input model to generate image data. The generated image data is compared with the real data in the first dataset in the discriminator, and the design planning drawing data that meets the conditions is output as the expanded first design planning drawing data. Step 5: Construct a model for generating architectural design and planning drawings; The architectural design and planning drawing generation model is based on the Stable Diffusion architecture. Its inputs include structured descriptive text and XML files generated from geographic information data, building regulations documents, standard documents, and user architectural design requirements data; the output is the design and planning drawing data. The model includes: The condition generation module is used to convert the numerical vector of user requirements after parsing into a semantic feature vector through CLIP's text encoder. The image generation module is used to gradually generate architectural design images that meet the requirements from random noise, guided by semantic feature vectors; the image generation module is a neural network with a UNet architecture. The image generation module includes a layered generation module. This module first generates a functional layout skeleton, then refines architectural element details. CFG (Content Graphing and Graphical Functions) is used to control the number of generated images, and multiple design plans are generated layer by layer by adjusting guidance coefficients. The skeleton generation stage first determines the core layout of the building, such as functional zoning and circulation organization, including the active and quiet zones of residential buildings and pedestrian flow in commercial buildings. Then, details of elements such as walls, doors, windows, and equipment are refined based on the skeleton. The layered generation strategy, combined with CFG's batch processing capabilities, simultaneously outputs multiple differentiated schemes, such as different orientations for apartment types and various facade styles. For example, after inputting the requirement of "modern style residential," the model simultaneously generates 3 to 5 design drawings with different layouts, materials, and colors by adjusting CFG coefficients and skeleton parameters, improving efficiency by more than 5 times compared to traditional single-scheme generation. This "macro to micro" generation logic aligns with architects' design thinking and avoids functional conflicts caused by the randomness of traditional generation models. For example, when generating a hospital scheme, the spatial relationships between outpatient, inpatient, and medical technology departments are first determined, and then the location of the nurse station and the size of the wards are refined, improving the layout rationality by approximately 40%. Layered generation uses semantic masks to constrain the relationship between the skeleton and details, ensuring that the functional layout is not disrupted during refinement. For example, after marking the "stairwell" area in the skeleton, the refinement stage automatically generates a staircase structure that meets the evacuation width requirements. The proportional relationships of architectural elements, such as the width-to-depth ratio and window-to-wall ratio, are determined simultaneously during skeleton generation. During detail generation, a consistency loss function enforces adherence to preset proportions, preventing issues such as excessively thick walls or misaligned door and window dimensions. Experimental data shows that this mechanism increases the element proportion compliance rate of generated drawings from 75% using traditional methods to over 95%.
[0022] The condition adjustment module slightly perturbs the semantic feature vector (such as adding Gaussian noise) to enhance the model's ability to generalize to requirements. Loss function: The loss function adopts a combination of perceptual loss (measuring the difference in high-rise features between the generated image and the real image), L2 loss (ensuring pixel detail accuracy), and constraint loss for building codes.
[0023] Step 6: Train the architectural design and planning drawing generation model using the third dataset to obtain the trained architectural design and planning drawing generation model; Step 7: Obtain the user-input architectural design requirements data, geographic information data, building regulations documents, and standard documents. After parsing and converting them into numerical vectors or semantic representations using LLM, input them into the architectural design planning drawing generation model to generate multiple design planning drawing data and obtain architectural design planning schemes.
[0024] The present invention also provides an electronic device, comprising at least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which enables the at least one processor to perform the above-described artificial intelligence-based architectural design and planning scheme generation method.
[0025] The present invention provides a machine-readable storage medium storing instructions that cause a machine to execute the artificial intelligence-based architectural design and planning scheme generation method described above.
[0026] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.
Claims
1. A method for generating architectural design and planning schemes based on artificial intelligence, characterized in that, Includes the following steps: Step 1: Obtain historical architectural design data from the database. The historical architectural design data includes architectural regulations documents, standard documents, design planning drawings, architectural geographic information data, and user architectural design requirements data for historical buildings. Step 2: Clean the architectural design history data, including removing duplicate, erroneous, and incomplete data records; Then, the building regulations documents, standard documents, geographic information data, and user building design requirements data are used to generate structured description text and XML files through the Large Language Model (LLM). The design planning drawing data is annotated using the structured description text and the XML file; Step 3: Perform image processing on the image of the annotated design planning drawing data to obtain the second dataset after data augmentation; Step 4: Construct a Generative Adversarial Network (GAN) model. Train the GAN model using the second dataset. Use the trained GAN model to generate multiple expanded first design planning drawing data. The expanded first design planning drawing data is merged into the second dataset to obtain the third dataset; Step 5: Construct a model for generating architectural design and planning drawings; The inputs for generating architectural design and planning drawings models are structured descriptive text and XML files generated from geographic information data, building regulations documents, standard documents, and user architectural design requirements data. The output is design planning drawing data; Step 6: Train the architectural design and planning drawing generation model using the third dataset to obtain the trained architectural design and planning drawing generation model; Step 7: Obtain the architectural design requirements data, geographic information data, building regulations documents, and standard documents input by the user; input the architectural design planning drawing generation model to generate multiple design planning drawing data; and obtain the architectural design planning scheme.
2. The method for generating architectural design and planning schemes based on artificial intelligence according to claim 1, characterized in that, In step 1, the design planning drawing data includes design floor plans, elevations, sections, renderings, technical parameters, and project construction data; Building geographic information data includes topographic data, geomorphological data, climate data, and surrounding environmental data of the building's location; User architectural design requirements data includes functional requirements, aesthetic preferences, and budget data.
3. The method for generating architectural design and planning schemes based on artificial intelligence according to claim 1, characterized in that, Step 2, in generating the structured description text and XML file, includes the following: Extract mandatory clauses from building regulations and technical parameters from standards and specifications to create structured descriptive text. Transform geographic information data into structured descriptive text that includes topographic features, climate zones, and surrounding building types; Parse user requirement data to obtain an XML file that includes a functional requirement relationship matrix, aesthetic keyword vectors, and budget constraints.
4. The method for generating architectural design and planning schemes based on artificial intelligence according to claim 1, characterized in that, In step 3, image processing includes operations such as flipping, rotating, cropping, and color adjustment.
5. The method for generating architectural design and planning schemes based on artificial intelligence according to claim 1, characterized in that, In step 4, a Generative Adversarial Network (GAN) model is generated by inputting different random noise vectors into the generator to generate multiple expanded first design planning drawing data.
6. The method for generating architectural design and planning schemes based on artificial intelligence according to claim 1, characterized in that, In step 5, the model includes: The condition generation module is used to convert the numerical vector of user requirements after parsing into a semantic feature vector through CLIP's text encoder. The image generation module is used to gradually generate architectural design images that meet the requirements from random noise through a progressive denoising process, guided by semantic feature vectors. The condition adjustment module is used to apply noise perturbation to the semantic feature vector.
7. The method for generating architectural design and planning schemes based on artificial intelligence according to claim 6, characterized in that, The image generation module employs a layered generation module and a CFG control strategy; wherein... The layered generation module is used to first generate the functional layout skeleton and then refine the architectural element details; CFG is used to control the number of generated images, and multiple design planning drawings are generated layer by layer by adjusting the guidance coefficient.
8. A method for generating architectural design and planning schemes based on artificial intelligence according to claim 6, characterized in that, The image generation module is a neural network based on the UNet architecture.
9. An electronic device, characterized in that, include: One or more processors; as well as Storage device for storing one or more programs. Wherein, when the one or more programs are executed by the one or more processors, the one or more processors perform the method according to any one of claims 1 to 8.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 8.