Building group geometric model generation method and system based on building module unitization

Abstract

The invention relates to the technical field of digital building design, and discloses a building group geometric model generation method and system based on building module unitization. Through a recursive unit cluster generation mechanism, automatic construction from a standardized module to a composite building body with rich spatial and morphological logic is realized, and the diversity and complexity of modular design are improved; by means of an environment interactive intelligent agent layout model, the building group can dynamically respond and optimize specific environment factors of a site, and the overall lighting, ventilation and terrain adaptability performance is improved; according to the integrated multi-objective optimization process, function connection, structure efficiency, construction cost and aesthetics evaluation based on deep learning are considered cooperatively, and a generated result is balanced between engineering feasibility and visual coordination; finally, the method and the system can automatically produce a fine model and material data which can be directly used for BIM collaborative and digital construction, and the efficiency and the integration degree of the whole process from design to production are improved.

Description

Technical Field

[0001] This invention relates to the field of digital architectural design technology, specifically to a method and system for generating geometric models of building complexes based on modular building units. Background Technology

[0002] With the acceleration of industrialization and digitalization in the construction industry, modular building design has become an important direction for modern architectural development due to its advantages in improving construction efficiency, ensuring project quality, controlling costs, and promoting sustainable development. Meanwhile, the widespread adoption of Building Information Modeling (BIM) and parametric design tools has provided a technological foundation for handling complex building forms and performance optimization.

[0003] In existing technologies, the design of modular buildings typically follows this pattern: designers first select functional units (such as residential modules, transportation core modules, etc.) from a standard module library, and then, based on design specifications, functional flow lines, and experience, arrange and combine them manually or semi-automatically within a given site to ultimately form a layout scheme for individual buildings or groups of buildings. In recent years, some studies have attempted to introduce algorithms to assist in layout, such as using gridding methods for module arrangement or employing multi-objective optimization algorithms to compare and select from a limited number of layout schemes.

[0004] However, current mainstream modular combination methods remain relatively simple and linear, often involving the direct stacking and splicing of modules on a two-dimensional plane or three-dimensional grid. This method struggles to automatically generate complex architectural forms with rich diversity and organic connections in form, space, and structure. The resulting designs are often rigid and simplistic, heavily reliant on the designer's personal experience for later adjustments and optimizations, failing to fully leverage the potential of computers in form exploration and generation. Furthermore, existing technologies typically use individual modules as the basic unit of layout, lacking an intermediate-scale "growth" or "aggregation" process. This results in architectural complexes that lack a sense of hierarchy and rhythm on a macro level, and on a micro level, the spatial and structural logic between modules is not tightly coupled, making it difficult to meet the demands of projects with higher requirements for architectural aesthetics and interior space quality. Most modular layout algorithms primarily consider rigid specifications such as functional zoning, sunlight spacing, and floor area ratio, but lack the ability to refine and dynamically respond to the site's unique environmental physical fields (such as dynamic sunlight, prevailing wind direction, and topographic relief). The relationship between the layout scheme and the environment is static and passive, rather than a dynamic and proactive optimization process, making it difficult to maximize the use of natural energy (light and wind), thus limiting the improvement of the overall environmental performance of the building complex. At the same time, morphological generation, structural feasibility, environmental performance simulation, and aesthetic evaluation are usually carried out in stages in different software or workflows, lacking an integrated platform to introduce multi-source feedback in real time during the generation process. This results in performance optimization often being an ex-post correction, which is inefficient and makes it difficult to guarantee the synergistic optimality of the final scheme in multiple dimensions.

[0005] Therefore, there is an urgent need for a method and system that can break through the limitations of existing modular design and achieve efficient, intelligent, and performance-driven automated generation from module libraries to complex building complexes, thereby improving design quality and efficiency and promoting the development of intelligent building design. To this end, a method and system for generating geometric models of building complexes based on building module units is proposed. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method and system for generating geometric models of building complexes based on modular building units, thereby resolving the problems described in the background.

[0007] To achieve the above objectives, the present invention provides the following technical solution: The first aspect is a method for generating geometric models of building clusters based on modular building units, which includes the following steps: S1: Input and Initialization: Obtain multi-dimensional constraint data of the parameterized building module unit library and the target site. Each module in the unit library has a defined structural interface, performance parameters and adjustable morphological parameters. The constraint data includes at least terrain, environmental physical field vector data, planning boundaries and indicators. S2: Recursive Unit Cluster Generation: Based on the unit library provided in step S1, at least one module that meets the functional type of the task is selected from the library as the initial seed according to the preset generation task; starting from the seed module, according to the preset structural compatibility rules and spatial adjacency probability table, and around the optimization goal of maximizing internal passage efficiency and minimizing structural span, other modules are recursively selected and aggregated to iteratively generate one or more composite building unit clusters that have completeness and diversity in function, structure and form; S3: Interactive Cluster Layout in the Environment: The unit clusters generated in step S2 are used as discrete layout agents and imported into the dynamic simulation environment constructed by the site constraint data obtained in step S1. In this environment, the terrain elevation is transformed into a supporting plane, and the solar radiation and wind environment vector fields are transformed into forces that drive the agents to adjust their position and orientation. This allows each unit cluster to achieve a stable layout state with optimal overall lighting, ventilation and terrain adaptability through autonomous movement and morphological fine-tuning. S4: Global Connectivity and Form Optimization: The optimal connection path is automatically calculated and generated between unit clusters that have completed the adaptive layout, and connection structures such as corridors, stairs or platforms are created accordingly. At the same time, a pre-trained aesthetic style discrimination model is introduced to evaluate the overall form, outline and solid-void relationship of the building complex, and use this feedback to drive further fine-tuning of the layout to ensure that the generated results are visually harmonious and unified. S5: Model Output and Integration: Integrate the results of the above steps to generate the final high-precision 3D geometric model of the building complex, the BIM model with engineering attributes, and the module material statistics list, and output them to the downstream engineering design or prefabrication production system.

[0008] Preferably, the recursive unit cluster generation in step S2 employs an optimization strategy combined with Monte Carlo tree search, specifically: In each growth decision, the current state of the unit cluster is taken as the root node, and possible module connection operations are taken as child nodes for expansion. The long-term benefits of each child node are evaluated by simulating several steps of random growth. These benefits are determined by the streamline efficiency, structural stability index, and morphological complexity within the unit cluster. Based on the evaluation results, the optimal or probabilistically optimal growth direction is selected for actual connection, and the unit cluster state is updated. This process is repeated until the termination condition is met.

[0009] Preferably, the interactive cluster layout of the environment described in step S3 is specifically as follows: Each unit cluster is modeled as a rigid body intelligent agent with mass, volume, and orientation attributes; The solar radiation vector field is transformed into a repulsive potential field to avoid excessive shading between clusters; the wind direction vector field is transformed into a guiding potential field to optimize the wind path. The intelligent agent moves under the combined force of the combined potential field, the boundary repulsion force, and the inter-cluster repulsion force. The stable equilibrium state of the system is solved through physical simulation, and this is used as the final layout.

[0010] Preferably, the recursive unit cluster generation process in step S2 integrates a performance pre-evaluation and screening mechanism while or after iteratively generating multiple candidate unit clusters: The mechanism quantifies the structural rationality, functional completeness, and morphological aesthetics of one or more candidate unit clusters generated; based on the preset scoring threshold, it automatically selects qualified unit clusters to form a set, and uses this set as the input object for the interactive cluster layout in step S3.

[0011] Preferably, the method of using the deep learning-based aesthetic evaluation model in step S4 is as follows: A discriminator of a generative adversarial network is pre-trained so that it can output a probability score for the input architectural perspective view that conforms to a specific aesthetic style; During the optimization process, the bird's-eye view and multi-angle perspective view of the current layout are input into the discriminator, and the obtained score is used as one of the optimization goals to guide the layout adjustment algorithm to evolve towards a higher aesthetic score.

[0012] Secondly, a system for generating geometric models of building clusters based on modular building units, used to implement the method for generating geometric models of building clusters based on modular building units described in the first aspect, the system comprising: The data input and preprocessing module is used to manage the parameterized module unit library and integrate and process multi-source site constraint data to generate a standardized environmental potential energy field. The intelligent unit cluster generation module has a built-in recursive growth algorithm and optimization strategy to generate diverse composite unit clusters from the module library; The dynamic environment interaction layout module has a built-in intelligent agent simulation engine, which is used to execute the autonomous layout and morphological adaptation of unit clusters in the environmental potential energy field. The global optimization and connection design module is used to automatically generate connection structures and integrate an aesthetic evaluation model to collaboratively optimize the overall form. The results integration and output module is used to generate the final model, perform conflict detection, and output it to an external platform.

[0013] Preferably, the intelligent unit cluster generation module further includes an online evolution unit; The online evolutionary unit continuously collects evaluation data on the actual performance of each unit cluster from the downstream layout and optimization module, and uses this data to dynamically adjust the adjacency rule weights and probability growth function parameters in the recursive growth algorithm, enabling the system to self-optimize and iterate, and improve the applicability of the generated unit clusters.

[0014] Preferably, the dynamic environment interaction layout module provides a real-time visual simulation interface; During the layout process, users can dynamically adjust the intensity parameters of the environmental potential energy field, set position locks or motion constraints for specific unit clusters, or manually intervene to adjust the local layout. The system will recalculate the equilibrium state in real time based on user input and update the visualization results.

[0015] Preferably, the global optimization and connectivity design module includes a multi-objective optimizer; The optimizer simultaneously considers the total length of the connection path, structural feasibility, construction cost, and the score provided by the aesthetic evaluation model. It employs the Pareto front algorithm to provide designers with a set of optimal connection schemes that achieve different trade-offs among multiple objectives.

[0016] Preferably, the system further includes a collaborative design management module; This module supports multi-user online collaboration, allowing designers with different roles to remotely edit and adjust unit cluster generation rules, layout preference parameters, and aesthetic style goals. The system automatically integrates all inputs, generates a consistent overall model, and records version iteration history.

[0017] Compared with the prior art, the present invention has the following beneficial effects: This invention achieves automated construction from standardized modules to composite building complexes with rich spatial and morphological logic through a recursive unit cluster generation mechanism, enhancing the diversity and complexity of modular design. Utilizing an interactive intelligent body layout model, the building complex can dynamically respond to and optimize specific site environmental factors, improving overall lighting, ventilation, and terrain adaptability. An integrated multi-objective optimization process considers functional connectivity, structural efficiency, construction costs, and deep learning-based aesthetic evaluation in a synergistic manner, resulting in a balance between engineering feasibility and visual harmony. Ultimately, this method and system can automatically generate detailed models and material data that can be directly used for BIM collaboration and digital construction, improving efficiency and integration throughout the entire process from design to production.

[0018] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0019] Figure 1 This is a diagram showing the overall module architecture and data flow of the system of this invention; Figure 2 This is a schematic diagram of the recursive unit cluster generation and filtering process of the present invention; Figure 3 This is a schematic diagram of the physical simulation of the interactive cluster layout in the environment of this invention; Figure 4 This is a schematic diagram of the global connection generation and multi-objective optimization process of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figures 1-4The method and system for generating geometric models of building clusters based on modular building units in this invention starts from a standardized module library and site data; then, a diverse range of composite unit clusters are recursively grown using the Monte Carlo tree search algorithm; these unit clusters are placed as intelligent agents in a simulated environment, and their layout is optimized through dynamic interaction with sunlight, wind field, and terrain; the connection structure is automatically generated and multi-objective collaborative optimization is performed to balance function, structure, and aesthetics; the output is an integrated digital model that can be directly used for design and production.

[0022] I. System Overall Architecture and Initialization After the system of this invention starts, the data input and preprocessing module first performs initialization operations. This module is responsible for managing a structured, parametric building module unit library. Each module in the library... Defined as a tuple containing the following data: in: For the parameterized model of the 3D geometric boundary representation (B-Rep) of the module, its key dimensions (such as length) ,Width ,high () represents a variable that can be adjusted using a slider; For a set of structural interfaces, Each interface face Its spatial location and normal vector were defined. A list of connection types (such as "structural load-bearing", "pipeline connection", "pedestrian passage") and compatible connection module types; Performance parameters, including but not limited to estimated weight thermal performance coefficient Daylight factor wait; For example, the adjustable range of morphological parameters. .

[0023] Simultaneously, this module loads multi-dimensional constraint data of the target site. The terrain data is processed into a Digital Elevation Model (DEM) function. ,in For site planar coordinates. Environmental physical field data are processed into spatial vector fields: solar radiation field is represented as a function of time. and location Changing radiation intensity vector The prevailing wind direction is represented by the wind speed vector. The planning boundary is defined by polygon B, and planning indicators include the maximum floor area ratio. Building density wait.

[0024] This module standardizes and parameterizes discrete, multi-source input data, providing a unified and computable data foundation for all subsequent algorithms. The result is the construction of a digital sandbox containing all design elements and constraints, ensuring that the generation process begins within a precise engineering and environmental context.

[0025] II. Intelligent Generation and Filtering of Recursive Unit Clusters After initialization, the intelligent unit cluster generation module begins operation. The user specifies the generation task through the interface, such as "Generate residential unit clusters". The module filters all modules in the library whose function tag contains "residential" and randomly selects one (such as a standard bedroom module). As the initial seed, the state of the unit cluster at this time is... Only contains .

[0026] Next, the module initiates a recursive growth algorithm based on Monte Carlo tree search. This process takes place within a decision tree, where each node... Represents a specific cluster state (i.e., a set of aggregated modules and their spatial relationships). Starting from the root node (initial cluster) Each iteration begins with the following sub-steps, which are controlled by the module's built-in optimization strategy: Select: From the current node Initially, child nodes are selected recursively until a scalable node is reached. The selection criterion uses a variant of the Upper Confidence Bound (UTC) formula: in: In the state All possible valid join operations (i.e., on a certain interface of the cluster) Connect a compatibility module above A set of actions, where each action 'a' represents a specific operation that connects to a module at a certain interface; For Action (i.e., the historical average return score for executing the connection); Access status Total number of times; In the state Take action The number of times; To explore constants, balance utilization (selecting high-scoring actions) with exploration (trying actions that have not been fully explored).

[0027] This formula intelligently guides the search direction, favoring connection methods that have yielded high returns in the past while leaving room for exploring new combinations.

[0028] Extension: When the selected node If the cluster is not in a terminated state (e.g., the cluster size is not exceeded) and has not been fully explored, randomly select a legal action that has not been tried before. (Connect a new module), execute this action, and create a new child node. , representing a new cluster state.

[0029] Simulation: From the new node We begin with a round of randomized simulated growth. This involves completely randomizing subsequent connection operations to quickly generate a complete candidate cluster of units. Then, the performance pre-evaluation and screening mechanism is immediately invoked to quantitatively score the candidate cluster. Scoring function Taking into account multiple indicators: in: For internal traffic efficiency, the average shortest path length between all functional spaces within the cluster is calculated and normalized. The maximum structural span ratio is the ratio of the distance between the two farthest load-bearing points within the cluster to the benchmark span, which is used to evaluate the structural rationality. The morphological aesthetics score is obtained by real-time analysis of the axonometric views of the clusters using a lightweight convolutional neural network (CNN), which evaluates their proportions and the relationship between solid and void. The conflict volume between modules is used to detect the severity of geometric interference between modules; , , , These are the weighting coefficients for each indicator. This represents the candidate cluster of units being evaluated.

[0030] The role of this mechanism is to introduce real-time feedback during the growth process, ensuring that the search direction develops towards clusters with reasonable structure, smooth function and good morphology.

[0031] Backtracking: The final score obtained in the simulation As a benefit of this round Along from Backtrack the path to the root node and update the visit count of all nodes along the path. and action value . The update formula is: The above four steps are repeated for a preset number of iterations (e.g., 10,000 times). After the loop ends, starting from the root node, the historical average return is selected. The cluster of cells corresponding to the highest-scoring action path is used as the main output of this generation. The system runs this generation process in parallel multiple times, or retains multiple high-scoring candidate clusters in a single round of MCTS. All these candidate clusters must be quantitatively scored through the aforementioned performance pre-evaluation and screening mechanism (integrated in the simulation steps), and then screened according to a preset scoring threshold. Finally, the qualified clusters form a set. This set will serve as the input object for the subsequent interactive cluster layout process.

[0032] The module and process mimic the logic of organic growth, generating complex, diverse, and performance-optimized intermediate-scale building units from atomic modules. The result is the creation of an "architectural vocabulary" that combines functional inherent logic with rich morphology, rather than simply stacking "building blocks."

[0033] III. Dynamic Simulation of Interactive Cluster Layout in the Environment After obtaining the set of unit clusters, the dynamic environment interaction layout module begins to operate. This module incorporates a physics-driven agent simulation engine. Each unit cluster... Instantiated as a rigid body intelligent agent Its attributes include: mass (based on estimated volume and material density), bounding box volume, and current centroid position. and orientation .

[0034] The site data is here transformed into an environmental potential field that drives the agent's movement: Terrain Adaptation Field: Intelligent Agent Subjected to a force perpendicular to its direction, its center of mass... of Coordinates tend to ,in The foundation elevation height was calculated based on the cluster morphology, which enabled the building to conform to the terrain.

[0035] Solar repulsion potential field: To optimize daylighting and avoid excessive inter-cluster shading. In the simulation time step... For each intelligent agent Calculate the solar repulsion force it experiences. This force originates from other intelligent agents. The shadowing effect it causes. Define a simplified potential function. Its negative gradient The direction of the repulsive force is the direction of the force, and the magnitude of the force is related to the direction of the repulsive force. exist The component in the occlusion direction is proportional.

[0036] Wind direction guiding potential field: To optimize natural ventilation, the wind direction vector field Transformed into guiding force. Intelligent agent. The guiding force Its lateral projected area is proportional to the component of its area along the wind direction, and its direction is perpendicular to the wind direction, aiming to adjust the orientation of the cluster. To create gaps that guide the airflow.

[0037] In addition, there is a bounding box-based mutual exclusion force between intelligent agents to prevent geometric interpenetration; and they are constrained by the planning boundary B to ensure that the layout is within the red line.

[0038] At the start of the simulation, all agents They are placed within the field randomly or according to heuristic rules. The physics engine operates at each time step. Calculate the net force acting on each agent. And update its position and orientation according to Newton's laws of motion:

[0039] in For intelligent agents The resultant torque; For intelligent agents The velocity vector at time t.

[0040] The system solves the problem through numerical integration until the kinetic and potential energy changes of all agents tend to stabilize, reaching a state of force equilibrium. The positions of the agents at this point are... and orientation The set is the final layout.

[0041] This module transforms the abstract goal of "environmental optimization" into a computable physical simulation problem. The result is that the layout of building complexes spontaneously forms a spatial configuration highly coordinated with terrain, sunlight, and wind direction, much like a natural object finding a stable position in a complex force field. Simultaneously, through a real-time visual simulation interface, designers can observe this dynamic process and directly drag intelligent entities and adjust potential field strength parameters (such as increasing the weight of solar repulsion forces) to achieve interactive layout fine-tuning. The system will then instantly recalculate the equilibrium.

[0042] IV. Global Connectivity Generation and Multi-Objective Morphology Optimization After obtaining a stable cluster layout, the global optimization and connectivity design module begins to handle the relationships between clusters. First, it automatically generates the connectivity structure based on graph theory methods. Each unit cluster... center of mass Treating nodes as graph nodes, calculate the Euclidean distance between all pairs of nodes. The connectivity requirements are predefined into a connectivity weight matrix based on the functional attributes of the clusters (e.g., clusters with strong "commonness" require more connectivity). .

[0043] The module solves a variant of the Steiner tree problem to minimize the total weight cost of connection paths while satisfying full or specified graph connectivity. The generated optimal or suboptimal connection paths are instantiated as 3D connection structures (such as corridors). Its cross-sectional dimensions are automatically calculated based on the preset pedestrian load.

[0044] Next, the module initiates a multi-objective optimizer to collaboratively optimize the overall shape. The optimizer considers three objectives: connectivity economy ,in and These are the unit price coefficients for materials and labor, respectively.

[0045] Structural feasibility ,in It is the calculated maximum stress at the interface between the connecting structure and the key module under the current layout and load conditions, obtained through simplified finite element analysis. It is the allowable stress of the material.

[0046] Aesthetic rating .here, This is the output of a pre-trained aesthetic style discrimination model. This model is the discriminator part of a generative adversarial network (GAN) pre-trained on a large architectural image dataset and fine-tuned on images of specific styles (such as "modern minimalism" and "eco-settlement"). It takes multiple angle renderings of the current layout as input and outputs a scalar between 0 and 1, representing the probability that the layout conforms to the target style. The lower the score, the higher the aesthetic score.

[0047] Because these objectives often conflict (e.g., the shortest connection may be aesthetically unappealing), the optimizer employs a Pareto front algorithm (such as NSGA-II). This algorithm maintains a population of solutions, generating new solutions through selection, crossover, and mutation, and selecting the next generation based on non-dominated sorting and crowding distance. After multiple generations of evolution, it outputs a Pareto optimal set, where each solution represents a different trade-off between connection economy, structural feasibility, and aesthetic value.

[0048] This module aims to integrate discrete building units into a cohesive whole, seeking a globally optimal balance between engineering and aesthetics. The result is the automatic generation of multiple alternative building complex designs, each deeply optimized in terms of technical feasibility and visual appeal, for decision-makers to choose from.

[0049] V. Model Integration, Collaboration, and Output After the designer selects a final solution from the Pareto front, the results integration and output module begins its work. This module integrates all data: parametric module instances (and their adjusted morphological parameters), unit cluster composition relationships, and precise cluster poses. Connection structure geometry It generates automatically: High-precision 3D geometric models: can be used for visualization and rendering.

[0050] BIM model with engineering attributes: Each module and connection structure is assigned the correct IFC entity type and attributes (such as material and fire rating).

[0051] Module Material List: Accurately lists the model number of each module. The number of times it is used, its specific dimensions and volume under its current morphological parameters, are used to facilitate factory prefabrication.

[0052] During the generation process, the collaborative design management module supports multi-user online parallel work. For example, structural engineers can log in to the system and directly adjust the weight parameters of the structural stability objective in the unit cluster generation algorithm; landscape architects can modify the calculated intensity and direction preference of the solar radiation field in the environmental interactive simulation; and lead designers can switch the style benchmark referenced by the aesthetic evaluation model in the interface (such as switching from "modern minimalism" to "ecological settlement"). All users' adjustment commands are integrated into the same core model in real time for calculation. The system automatically integrates opinions from different disciplines and generates a unified and coordinated design scheme, while fully recording the historical versions of each parameter adjustment and model iteration, forming a traceable design evolution log.

[0053] The purpose of this module and process is to close the data flow from digital generation to digital construction. The result is a complete digital deliverable that can be directly used for design refinement, engineering analysis, cost accounting, and automated factory processing, achieving design-production integration.

[0054] VI. System Online Evolution Throughout the system's operational cycle, the online evolutionary unit within the intelligent unit cluster generation module continuously operates. It collects feedback data from downstream modules, such as: unit cluster types that frequently collide or struggle to achieve a stable layout during environmental interaction simulations; and structural feasibility in multi-objective optimization. A consistently poor connectivity pattern; or cluster morphological features that consistently receive low scores from aesthetic discrimination models.

[0055] This feedback data is quantified as a penalty signal, used to dynamically adjust key parameters in the recursive growth algorithm. For example, if a certain module connection combination... If structural problems occur frequently, the online evolutionary unit will lower the probability value of that combination in the spatial adjacency probability table. At the same time, it also fine-tunes the payoff function in Monte Carlo tree search. Weighting coefficients This allows the system to automatically avoid discovered problem patterns when generating cell clusters in the future, tending to produce cell clusters with better overall performance.

[0056] The purpose of this unit is to endow the system with the ability to learn from historical project experience. The result is that as the system is used more frequently, the overall applicability and quality of its generated results can iteratively improve, demonstrating a certain degree of domain-adaptive intelligence.

[0057] In summary, this embodiment details the entire technological chain of this invention, from parametric input, intelligent intermediate generation, environmentally responsive layout, global collaborative optimization to digital delivery. Through the organic combination of mechanisms such as recursive growth, intelligent agent simulation, multi-objective optimization, and online learning, this invention can automatically transform basic building modules into complex building complex schemes that are deeply coupled with the site and consider both engineering and aesthetics.

Claims

1. A method for generating geometric models of building complexes based on modular building units, characterized in that, Includes the following steps: S1: Input and Initialization: Obtain multi-dimensional constraint data of the parameterized building module unit library and the target site. Each module in the unit library has a defined structural interface, performance parameters and adjustable morphological parameters. The constraint data includes at least terrain, environmental physical field vector data, planning boundaries and indicators. S2: Recursive unit cluster generation: Based on the unit library provided in step S1, select at least one module that conforms to the functional type of the task from the library as the initial seed according to the preset generation task. Starting with this seed module, based on the preset structural compatibility rules and spatial adjacency probability table, and with the optimization goal of maximizing internal passage efficiency and minimizing structural span, other modules are recursively selected and aggregated to iteratively generate one or more composite building unit clusters that are complete and diverse in function, structure and form. S3: Interactive Cluster Layout in the Environment: The unit clusters generated in step S2 are used as discrete layout agents and imported into the dynamic simulation environment constructed by the site constraint data obtained in step S1. In this environment, the terrain elevation is transformed into a supporting plane, and the solar radiation and wind environment vector fields are transformed into forces that drive the agents to adjust their position and orientation. This allows each unit cluster to achieve a stable layout state with optimal overall lighting, ventilation and terrain adaptability through autonomous movement and morphological fine-tuning. S4: Global Connectivity and Form Optimization: The optimal connection path is automatically calculated and generated between unit clusters that have completed the adaptive layout, and connection structures such as corridors, stairs or platforms are created accordingly. At the same time, a pre-trained aesthetic style discrimination model is introduced to evaluate the overall form, outline and solid-void relationship of the building complex, and use this feedback to drive further fine-tuning of the layout to ensure that the generated results are visually harmonious and unified. S5: Model Output and Integration: Integrate the results of the above steps to generate the final high-precision 3D geometric model of the building complex, the BIM model with engineering attributes, and the module material statistics list, and output them to the downstream engineering design or prefabrication production system.

2. The method for generating geometric models of building complexes based on modular building units according to claim 1, characterized in that, The recursive unit cluster generation described in step S2 employs an optimization strategy combined with Monte Carlo tree search, specifically as follows: In each growth decision, the current state of the unit cluster is taken as the root node, and possible module connection operations are taken as child nodes for expansion. The long-term benefits of each child node are evaluated by simulating several steps of random growth. These benefits are determined by the streamline efficiency, structural stability index, and morphological complexity within the unit cluster. Based on the evaluation results, the optimal or probabilistically optimal growth direction is selected for actual connection, and the unit cluster state is updated. This process is repeated until the termination condition is met.

3. The method for generating geometric models of building complexes based on modular building units according to claim 1, characterized in that, The specific process of the interactive cluster layout in step S3 is as follows: Each unit cluster is modeled as a rigid body intelligent agent with mass, volume, and orientation attributes; The solar radiation vector field is transformed into a repulsive potential field to avoid excessive shading between clusters; the wind direction vector field is transformed into a guiding potential field to optimize the wind path. The intelligent agent moves under the combined force of the combined potential field, the boundary repulsion force, and the inter-cluster repulsion force. The stable equilibrium state of the system is solved through physical simulation, and this is used as the final layout.

4. The method for generating geometric models of building complexes based on modular building units according to claim 1, characterized in that, The recursive unit cluster generation process described in step S2 integrates a performance pre-evaluation and screening mechanism while or after iteratively generating multiple candidate unit clusters: The mechanism quantifies the structural rationality, functional completeness, and morphological aesthetics of one or more candidate unit clusters generated; based on the preset scoring threshold, it automatically selects qualified unit clusters to form a set, and uses this set as the input object for the interactive cluster layout in step S3.

5. The method for generating geometric models of building complexes based on modular building units according to claim 1, characterized in that, The method of using the deep learning-based aesthetic evaluation model described in step S4 is as follows: A discriminator of a generative adversarial network is pre-trained so that it can output a probability score for the input architectural perspective view that conforms to a specific aesthetic style; During the optimization process, the bird's-eye view and multi-angle perspective view of the current layout are input into the discriminator, and the obtained score is used as one of the optimization goals to guide the layout adjustment algorithm to evolve towards a higher aesthetic score.

6. A system for generating geometric models of building complexes based on modular building units, characterized in that: A system for implementing the building group geometric model generation method based on building module unitization as described in any one of claims 1-5, the system comprising: The data input and preprocessing module is used to manage the parameterized module unit library and integrate and process multi-source site constraint data to generate a standardized environmental potential energy field. The intelligent unit cluster generation module has a built-in recursive growth algorithm and optimization strategy to generate diverse composite unit clusters from the module library; The dynamic environment interaction layout module has a built-in intelligent agent simulation engine, which is used to execute the autonomous layout and morphological adaptation of unit clusters in the environmental potential energy field. The global optimization and connection design module is used to automatically generate connection structures and integrate an aesthetic evaluation model to collaboratively optimize the overall form. The results integration and output module is used to generate the final model, perform conflict detection, and output it to an external platform.

7. The building complex geometric model generation system based on building module unitization according to claim 6, characterized in that, The intelligent unit cluster generation module also includes an online evolution unit; The online evolutionary unit continuously collects evaluation data on the actual performance of each unit cluster from the downstream layout and optimization module, and uses this data to dynamically adjust the adjacency rule weights and probability growth function parameters in the recursive growth algorithm, enabling the system to self-optimize and iterate, and improve the applicability of the generated unit clusters.

8. The building complex geometric model generation system based on building module unitization according to claim 6, characterized in that, The dynamic environment interactive layout module provides a real-time visual simulation interface; During the layout process, users can dynamically adjust the intensity parameters of the environmental potential energy field, set position locks or motion constraints for specific unit clusters, or manually intervene to adjust the local layout. The system will recalculate the equilibrium state in real time based on user input and update the visualization results.

9. The building complex geometric model generation system based on building module unitization according to claim 6, characterized in that, The global optimization and connectivity design module includes a multi-objective optimizer; The optimizer simultaneously considers the total length of the connection path, structural feasibility, construction cost, and the score provided by the aesthetic evaluation model. It employs the Pareto front algorithm to provide designers with a set of optimal connection schemes that achieve different trade-offs among multiple objectives.

10. The building complex geometric model generation system based on building module unitization according to claim 6, characterized in that, The system also includes a collaborative design management module; This module supports multi-user online collaboration, allowing designers with different roles to remotely edit and adjust unit cluster generation rules, layout preference parameters, and aesthetic style goals. The system automatically integrates all inputs, generates a consistent overall model, and records version iteration history.

Images