Building space multi-agent design method and system for parameter-component-scene collaborative optimization

Abstract

The invention discloses a parameter-component-scene collaborative optimization building space multi-agent design method, which comprises the following steps of: under the condition that an initial scene and a personalized optimization strategy are given, constructing a multi-dimensional parameter matrix containing scene and component information, and mutually coordinating a plurality of agents consisting of a design agent, an evaluation agent and a supervision agent, through iterative generation of a multi-dimensional parameter matrix and a scene rendering picture, collaborative optimization of scene layout and component parameters is carried out to form a three-dimensional scene and a parameterized component, and industrial customization or parameterized purchase is carried out on an optimization result. The method has the advantages that an evaluation-supervision-design coordinated multi-agent system is constructed, building space multi-parameter optimization under different styles and initial states is realized from cognition through image-text evaluation and parameter output, and personalized customization and factory manufacturing requirements are met.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of artificial intelligence and civil engineering, and in particular to a multi-agent design method and system for architectural space with parameter-component-scene collaborative optimization. Background Technology

[0002] With the increasing demands for humanization and humanization in urban public spaces and interior design, how to quickly respond to users' personalized needs through interaction during the design process, while meeting the parametric industrial customization needs of the designed products, has become a key challenge. The application of traditional design methods and related artificial intelligence technologies still has the following shortcomings: (1) Relying on design based on experience and static specifications, the scene is difficult to meet subjective needs such as visual comfort and objective needs such as component layout. At the same time, it is necessary to manually adjust design parameters and rely on later manual or simple index evaluation, resulting in low iteration efficiency, strong subjectivity and dynamic design requirements; (2) The progress of artificial intelligence methods in spatial intelligence has enabled simple design based on images and text, but it is mostly based on pixels and text as units. The optimization results are mostly aimed at visualization and cannot be broken down into physical components. It lacks the physical parameter information required for industrial production. At the same time, the above methods lead to high-dimensional complexity of the parameter space, making it difficult to optimize globally. In terms of matching design process, although existing multimodal large models, such as Qwen-VL, can be used for image evaluation, the existing solutions only use them as single-point scoring tools, without constructing an "evaluation-execution" closed loop, lacking a semantic feedback mechanism, and unable to drive parameter iteration optimization.

[0003] In recent years, with breakthroughs in reinforcement learning and large language models in the field of agent decision-making, such as large models acting as policy generators, under the guidance of designers and parameterized expression, a convergent, interpretable, and generalizable closed-loop optimization system has been constructed, providing a solid foundation for solving personalized matching in the preceding design process and industrial customization of designed products. Therefore, this application provides a method and system for collaborative design and adaptive optimization from component parameters to spatial scenes using large models and multiple agents, applicable to intelligent design and optimization of building structures and spatial comfort. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a multi-agent design method and system for architectural space with parameter-component-scene collaborative optimization. It constructs a multi-agent system with mutual coordination of evaluation, supervision and design. Starting from cognition, it realizes multi-parameter optimization of architectural space in different styles and initial states through graphic evaluation and parameter output, so as to meet the needs of personalized customization and factory manufacturing.

[0005] To address the aforementioned technical problems, this invention provides a multi-agent design method for architectural spaces that coordinates parameter-component-scene optimization. Given an initial scene and a personalized optimization strategy, a multi-dimensional parameter matrix containing scene and component information is constructed. Multiple agents—a design agent, an evaluation agent, and a supervision agent—collaborate to iteratively generate scene layout and component parameters using the multi-dimensional parameter matrix and scene rendering images, forming a 3D scene and parameterized components. The optimization results are then used for industrial customization or parametric procurement. The specific steps include: Step S1: Initialize the scene optimization strategy: The initial scene is set by providing a specific spatial distribution of components and their default parameter values. The initial optimization strategy is determined by providing keywords or images that are semantically consistent with the optimization target, guiding subsequent multi-agents to generate scenes with the same style. At the same time, the optimization target is determined and the parameter target range is set. Step S2: Generate / optimize the parameterized matrix. Based on the component position matrix and default parameter values ​​given in Step S1, or based on the component parameters and component position matrix provided by the design agent, construct a multi-dimensional parameter matrix containing scene and component information. Step S3: Use the scene rendering API tool to perform spatial scene rendering on the multi-dimensional parameter matrix and generate scene images; Step S4: Evaluate the agent through multi-dimensional and multi-index evaluation. Based on different evaluation indicators and dimensions such as color, hue, lighting, scene atmosphere, and spatial distribution, a large language model is used. Through rule definition, one or more agents are set up to understand the input multi-dimensional parameter matrix and rendering map according to the evaluation indicators and different style requirements. The agent provides semantic suggestions for optimizing the rendering map and the parameter distribution range of relevant parameters. The relevant evaluation conclusions and parameter suggestion values ​​are saved in the log and aligned with the corresponding input map and parameters. Step S5: The supervisory agent distinguishes the results of each evaluation agent according to the optimization objective and outputs the optimization strategy. The relevant strategy generation method includes assigning weights to different evaluation agents and calculating the weighted average value. The value is determined by manual preset, given by an existing large model, or by reinforcement learning based on the dataset in the log file generated by the agent. Finally, the parameter setting range of each parameter and the optimization direction of each index are output. Step S6: The design agent modifies the design scheme based on the suggestions provided by the supervised agent, optimizes and adjusts parameters using a large language model, automatically adjusts the multi-dimensional parameter matrix, and determines whether it meets the predetermined target value range. Finally, it outputs complete executable parameters, and subsequently calls the scene rendering API tool to visualize the parameters as scene images. During the iteration process, the optimization stops if the requirements are met through manual review or conditional judgment; otherwise, it proceeds to the next round of optimization iteration. Step S7: Perform industrial-scale customization or procurement based on the optimized component parameters.

[0006] The multi-agent system consists of a design agent, an evaluation agent, and a supervision agent.

[0007] The evaluation agent is configured using a large language model and rules to define one or more agents based on different evaluation indicators and dimensions, such as color, hue, lighting, scene atmosphere, and spatial distribution. These agents interpret the input multidimensional parameter matrix and rendering image according to the evaluation indicators and different style requirements, and provide semantic suggestions for optimizing the rendering image and the parameter distribution range of relevant parameters. The relevant evaluation conclusions and parameter suggestion values ​​are synchronously saved in the log and aligned with the corresponding input image and parameters.

[0008] The supervisory agent is responsible for differentiating the results of each evaluation agent based on the optimization objective and outputting optimization strategies. The relevant strategy generation method includes assigning weights to different evaluation agents and calculating a weighted average value. The value is determined by manual preset, given by an existing large model, or by reinforcement learning based on the dataset in the log file generated by the agent. Finally, the parameter setting range of each parameter and the optimization direction of each indicator are output.

[0009] The design agent optimizes and adjusts parameters using a large language model based on the design modification suggestions provided by the supervisory agent, automatically adjusts the multi-dimensional parameter matrix and determines whether it meets the predetermined target value range, and finally outputs complete executable parameters. It then connects to the scene rendering module to generate the updated scene and render the image.

[0010] The collaborative optimization of scene layout and component parameters includes setting an initial scene and initial optimization strategy, generating a multi-dimensional parameter matrix and rendering graph, and multi-agent collaborative optimization. The loop stops when the predetermined optimization objective is met. Wherein: The optimization objective refers to determining that the parameters and scores meet the predetermined target value range. When the parameter changes and historical trajectory oscillations are monitored during the iterative optimization process and the relevant convergence and target values ​​are found, the optimization iteration is stopped. The initial scenario includes a specific spatial distribution of components and their default parameter values; The initial optimization strategy refers to a strategy that combines evaluation indicators and the semantics of the optimization target, and guides style by providing optimization keywords or images. The multidimensional parameter matrix refers to a multidimensional matrix constructed by integrating the position coordinates and parameters of components, which contains information on the spatial layout and parameters of components. The multidimensional matrix consists of a component parameter matrix and a scene layout position matrix. The spatial position of the corresponding component in the position matrix is ​​encoded and mapped to the component parameter matrix. The row vectors of the component parameter matrix represent component types, and the column vectors represent component parameters. The types of components include ceilings, columns, walls, floors, decorations, electromechanical equipment, and light sources; The component parameters include spatial location, component size, light intensity, hue, color temperature, color, material, and density parameters required for scene optimization, and are consistent with industrial customization.

[0011] This invention also provides a multi-agent design system for architectural space with parameter-component-scene collaborative optimization, including a scene and optimization strategy initialization input module, a scene parameterization expression module, and a multi-agent optimization module; wherein: The scenario and optimization strategy initialization input module supports inputting the initial layout of the scenario, component types and their definition parameters, keywords and images reflecting specific optimization target semantics; The scene parameterization expression module enables the input of component layout spatial position and parameters to generate a multi-dimensional parameter matrix containing scene and component information. The multi-agent optimization module coordinates the evaluation agent, the supervisory agent, and the design agent to optimize parameters and automatically generate evaluation logs.

[0012] The one or more technical solutions proposed in this invention have at least the following beneficial effects: (1) By adjusting the component layout and parametric expression of the scene, combined with multi-agent iterative optimization, the scene and entity components and their parameters are optimized synchronously. The design results are directly used for component-based customized production and procurement, thus connecting the design with the front-end industrial production. (2) A multi-agent system of evaluation, supervision and design was constructed. Starting from cognition, the system achieves multi-parameter optimization of architectural space in different styles and initial states through graphic evaluation and parameter output, so as to meet the personalized needs of designers and users. Attached Figure Description

[0013] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram illustrating the principle of a specific embodiment of the present invention; Figure 2 This is a flowchart of a specific embodiment of the present invention. Detailed Implementation

[0014] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0015] This invention provides a multi-agent design method for architectural space with parameter-component-scene collaborative optimization. Under the given initial scene and personalized optimization strategy, a multi-dimensional parameter matrix containing scene and component information is constructed. The multi-agent system, consisting of a design agent, an evaluation agent, and a supervision agent, collaborates with each other to perform collaborative optimization of scene layout and component parameters through iterative generation of the multi-dimensional parameter matrix and scene rendering images, forming a three-dimensional scene and parameterized components. The optimization results can then be industrially customized or procured parametrically.

[0016] like Figure 1 As shown, given the scenario initialization (7) and optimization strategy (8), the input is the iterative scenario parameterized expression (4), and the output is the scenario graph (5) and multi-dimensional parameter matrix (6). Through the collaborative efforts of multiple agents—design agent 1, evaluation agent 2, and supervision agent 3—the scenario, components, and parameters are simultaneously optimized and digitally output. Wherein: The evaluation agent 1 uses a large language model and rules to define one or more agents based on different evaluation indicators and dimensions such as color, hue, lighting, scene atmosphere, and spatial distribution. The agents understand the input multidimensional parameter matrix and rendering image according to the evaluation indicators and different style requirements, and provide semantic suggestions for rendering image optimization and parameter distribution ranges of relevant parameters. The relevant evaluation conclusions and parameter suggestion values ​​are synchronously saved in the log and aligned with the corresponding input image and parameters. The supervisory agent 2 is to output optimization strategies based on the optimization objective, distinguish the results of each evaluation agent, and the relevant strategy generation method includes assigning weights to different evaluation agents and calculating the weighted average value. The value is determined by manual preset, given by an existing large model, or by reinforcement learning based on the dataset in the log file generated by the agent. Finally, the parameter setting range of each parameter and the optimization direction of each indicator are output. The design agent 3 optimizes and adjusts parameters using a large language model based on the design modification suggestions provided by the supervisory agent, automatically adjusts the multi-dimensional parameter matrix and determines whether it meets the predetermined target value range, and finally outputs complete executable parameters. It then connects to the scene rendering module to generate the updated scene and render the image.

[0017] The synchronous optimization of the scene, components, and parameters includes the following steps: setting an initial scene 7 and optimization strategy 8; generating a multi-dimensional parameter matrix 6; scene rendering and scene graph generation 5; evaluation agent 1, supervision agent 2, and design agent 3 conducting parameter iteration and cyclic optimization; and stopping optimization when the pre-set optimization target is met. Wherein: The component type 401 includes ceilings, columns, walls, floors, decorations, electromechanical equipment, and light sources; The component parameters include spatial location, component size, light intensity, hue, color temperature, color, material, and density parameters required for scene optimization, and are consistent with industrial customization; the spatial layout defines the spatial location of components according to different scenes, forming a position matrix 402. The multidimensional parameter matrix 403 refers to a multidimensional matrix constructed by integrating the position coordinates and parameters of components, which contains information on the spatial layout and parameters of components. The multidimensional matrix consists of a component parameter matrix and a scene layout position matrix. The spatial position of the corresponding component in the position matrix is ​​encoded and mapped to the component parameter matrix. The row vectors of the component parameter matrix represent component types, and the column vectors represent component parameters. The initialization scenario 7 includes a specific spatial distribution of components and their default parameter values; The initial optimization strategy 8 combines evaluation indicators and the semantics of the optimization target, and provides optimization keywords or images to determine the initial strategy. The optimization objective is for the parameters and scores to meet a predetermined target value range. Based on the parameter changes and historical trajectory oscillation monitoring during the iterative optimization process, the optimization iteration stops when the relevant convergence and target value are reached.

[0018] This invention also provides a multi-agent design system for architectural space with parameter-component-scene collaborative optimization, including a scene and optimization strategy initialization input module, a scene parameterization expression module, and a multi-agent optimization module; wherein: The scenario and optimization strategy initialization input module supports inputting the initial layout of the scenario, component types and their definition parameters, keywords and images reflecting specific optimization target semantics; The scene parameterization expression module enables the input of component layout spatial position and parameters to generate a multi-dimensional parameter matrix containing scene and component information. The multi-agent optimization module coordinates the evaluation agent, the supervisory agent, and the design agent to optimize parameters and automatically generate evaluation logs.

[0019] like Figure 2 As shown, this invention provides a multi-agent design method for architectural space with parameter-component-scene collaborative optimization, including the following specific steps: Step S1: Initialize the scene optimization strategy: The initial scenario is set by providing a specific spatial distribution of components and their default parameter values. The initial optimization strategy is determined by providing keywords or images that are consistent with the semantics of the optimization target. For example, keywords may be "tech style" or "punk style", or images of different porcelain styles such as Jun kiln or Ru kiln may be provided to guide subsequent multi-agents to generate scenarios of the same style. At the same time, the optimization target is determined and the target range of parameters is set. Step S2: Generate / optimize the parameterized matrix. Based on the component position matrix and default parameter values ​​given in Step S1, or based on the component parameters and component position matrix provided by the design agent, construct a multi-dimensional parameter matrix containing scene and component information. Step S3: Use the scene rendering API tool to perform spatial scene rendering on the multi-dimensional parameter matrix and generate scene images; Step S4: Evaluate Agent 2 by performing multi-dimensional and multi-index evaluation. Based on different evaluation indicators and dimensions such as color, hue, lighting, scene atmosphere, and spatial distribution, a large language model is used. Through rule definition, one or more agents are set up to understand the input multi-dimensional parameter matrix and rendering map according to the evaluation indicators and different style requirements. The agents provide semantic suggestions for optimizing the rendering map and the parameter distribution range of related parameters. The relevant evaluation conclusions and parameter suggestion values ​​are synchronously saved in the log and aligned with the corresponding input map and parameters. Step S5: Supervisory agent 3 distinguishes the results of each evaluation agent according to the optimization objective and outputs the optimization strategy; the relevant strategy generation method includes assigning weights to different evaluation agents and calculating the weighted average value, the value of which is determined by manual preset, given by an existing large model, or by reinforcement learning based on the dataset in the log file generated by the agent, and finally outputs the parameter setting range of each parameter and the optimization direction of each indicator; Step S6: The design agent modifies the design scheme based on the suggestions provided by the supervised agent, optimizes and adjusts parameters using a large language model, automatically adjusts the multi-dimensional parameter matrix, and determines whether it meets the predetermined target value range. Finally, it outputs complete executable parameters, and subsequently calls the scene rendering API tool to visualize the parameters as scene images. During the iteration process, the optimization stops if the requirements are met through manual review or conditional judgment; otherwise, it proceeds to the next round of optimization iteration. Step S7: Perform industrial-scale customization or procurement based on the optimized component parameters.

[0020] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-agent design method for architectural space with parameter-component-scene collaborative optimization, characterized in that: Given an initial scenario and personalized optimization strategies, a multi-dimensional parameter matrix containing scenario and component information is constructed. Multiple agents, consisting of a design agent, an evaluation agent, and a supervision agent, collaborate to iteratively generate the multi-dimensional parameter matrix and scenario rendering images, thereby optimizing the scenario layout and component parameters to form a three-dimensional scenario and parameterized components. The optimization results are then used for industrial customization or parameterized procurement.

2. The multi-agent design method for architectural space collaborative optimization of parameters, components, and scenes according to claim 1 includes the following specific steps: Step S1: Initialize the scene optimization strategy: Set the initial scene by providing a specific spatial distribution of components and their default parameter values. The initial optimization strategy is determined by providing keywords or images that are semantically consistent with the optimization target, guiding subsequent multi-agents to generate scenes of the same style; at the same time, determine the optimization target and set the parameter target range; Step S2: Generate / optimize the parameterized matrix: Based on the component position matrix and default parameter values ​​given in Step S1, or based on the component parameters and component position matrix provided by the design agent, construct a multi-dimensional parameter matrix containing scene and component information; Step S3: Use the scene rendering API tool to perform spatial scene rendering on the multi-dimensional parameter matrix and generate scene images; Step S4: Evaluate the agent through multi-dimensional and multi-index evaluation. Based on different evaluation indicators and dimensions such as color, hue, lighting, scene atmosphere, and spatial distribution, a large language model is used. Through rule definition, one or more agents are set up to understand the input multi-dimensional parameter matrix and rendering map according to the evaluation indicators and different style requirements. The agent provides semantic suggestions for optimizing the rendering map and the parameter distribution range of relevant parameters. The relevant evaluation conclusions and parameter suggestion values ​​are saved in the log and aligned with the corresponding input map and parameters. Step S5: The supervising agent, based on the optimization objective, differentiates the results of each evaluation agent and outputs an optimization strategy. The strategy generation methods include assigning weights to different evaluation agents and calculating a weighted average. The value of this average is determined through manual pre-setting, provision by an existing large model, or reinforcement learning based on a dataset from the agent's log files. Finally, the parameter setting ranges for each parameter and the optimization direction for each indicator are output. Step S6: The design agent, based on the design modification suggestions provided by the supervising agent, uses a large language model to optimize and adjust parameters, automatically adjusting the multi-dimensional parameter matrix and determining whether it meets the predetermined target value range. Finally, the complete executable parameters are output, and subsequently, the scene rendering API tool is called to visualize the parameters as scene images. During the iteration process, optimization stops if the requirements are met through manual review or conditional judgment; otherwise, the next round of optimization iteration begins. Step S7: Perform industrial-scale customization or procurement based on the optimized component parameters.

3. The multi-agent design method for architectural space with parameter-component-scene collaborative optimization according to claim 1 or 2, characterized in that: The evaluation agent is configured using a large language model and rules to define one or more agents based on different evaluation indicators and dimensions, such as color, hue, lighting, scene atmosphere, and spatial distribution. These agents interpret the input multidimensional parameter matrix and rendering image according to the evaluation indicators and different style requirements, and provide semantic suggestions for optimizing the rendering image and the parameter distribution range of relevant parameters. The relevant evaluation conclusions and parameter suggestion values ​​are synchronously saved in the log and aligned with the corresponding input image and parameters.

4. The multi-agent design method for architectural space with parameter-component-scene collaborative optimization according to claim 1 or 2, characterized in that: The supervisory agent is responsible for differentiating the results of each evaluation agent based on the optimization objective and outputting optimization strategies. The relevant strategy generation method includes assigning weights to different evaluation agents and calculating a weighted average value. The value is determined by manual preset, given by an existing large model, or by reinforcement learning based on the dataset in the log file generated by the agent. Finally, the parameter setting range of each parameter and the optimization direction of each indicator are output.

5. The multi-agent design method for architectural space with parameter-component-scene collaborative optimization according to claim 1 or 2, characterized in that: The design agent optimizes and adjusts parameters using a large language model based on the design modification suggestions provided by the supervisory agent, automatically adjusts the multi-dimensional parameter matrix and determines whether it meets the predetermined target value range, and finally outputs complete executable parameters. It then connects to the scene rendering module to generate the updated scene and render the image.

6. The multi-agent design method for architectural space with parameter-component-scene collaborative optimization according to claim 1, characterized in that: The collaborative optimization of scene layout and component parameters includes setting an initial scene and initial optimization strategy, generating a multi-dimensional parameter matrix and rendering graph, and multi-agent collaborative optimization. When the predetermined optimization target is met, the loop stops.

7. The multi-agent design method for architectural space with parameter-component-scene collaborative optimization according to claim 6, characterized in that: The optimization objective refers to determining that the parameters and scores meet the predetermined target value range. When the parameter changes and historical trajectory oscillations are monitored during the iterative optimization process and the relevant convergence and target values ​​are found, the optimization iteration is stopped.

8. The multi-agent design method for architectural space with parameter-component-scene collaborative optimization according to claim 6, characterized in that: The initial scenario includes a specific spatial distribution of components and their default parameter values; the initial optimization strategy refers to a strategy that combines evaluation indicators and optimization target semantics, and provides style guidance by providing optimization keywords or images.

9. The multi-agent design method for architectural space with parameter-component-scene collaborative optimization according to claim 6, characterized in that: The multidimensional parameter matrix refers to a multidimensional matrix constructed by integrating the position coordinates and parameters of the components, which contains information on the spatial layout and parameters of the components. The multidimensional matrix consists of a component parameter matrix and a scene layout position matrix. The spatial position of the corresponding component in the position matrix is ​​encoded and mapped to the component parameter matrix. The component parameter matrix has row vectors representing component types and column vectors representing component parameters. The component types include ceilings, columns, walls, floors, decorations, electromechanical equipment, and light sources. The component parameters include spatial location, component size, light intensity, hue, color temperature, color, material, and density parameters required for scene optimization, which are consistent with industrial customization.

10. A multi-agent design system for architectural space with parameter-component-scene collaborative optimization, characterized in that: The system includes a scene and optimization strategy initialization input module, a scene parameterization expression module, and a multi-agent optimization module. The scene and optimization strategy initialization input module supports inputting the initial scene layout, component types and their defined parameters, keywords reflecting specific optimization objectives, and images. The scene parameterization expression module generates a multi-dimensional parameter matrix containing scene and component information by inputting the spatial location and parameters of the component layout. The multi-agent optimization module coordinates an evaluation agent, a supervisory agent, and a design agent to optimize parameters and automatically generates evaluation logs.

Images