A method and system for optimizing design of urban block morphology based on generating large models

Abstract

This invention proposes a method and system for urban block morphology optimization design decision-making based on a generative large model. The method includes: collecting multimodal green performance data affecting local heat island intensity and wind environment performance at the urban block scale and preprocessing it; converting the preprocessed data into RDF triples based on entities, relations, and events to construct a multimodal knowledge graph; modularly clustering the knowledge in the multimodal knowledge graph to obtain structured subgraphs, improving knowledge retrieval efficiency based on the structured subgraphs, and obtaining a building green performance data vector database to establish an external knowledge base to enhance the large language model; using knowledge representation to enhance the large language model to provide guidance and receive feedback for the machine learning model executing urban block morphology optimization design, forming an edge-cloud collaborative architecture of large and small models. This method and system can improve the multimodal data processing capabilities of traditional single machine learning methods, and enhance the efficiency and accuracy of optimization design decisions.

Description

Technical Field

[0001] This invention belongs to the field of green building performance optimization design technology, and in particular relates to a method and system for urban block morphology optimization design decision based on a large-scale generated model. Background Technology

[0002] Significant progress has been made both domestically and internationally in research on green building performance simulation based on physical and machine learning models. Most existing green building performance simulations are based on the assumption of uniform mixing, while the microclimate at the urban block scale introduces non-uniform boundary conditions. Therefore, improving the accuracy of green building performance optimization requires in-depth research into non-uniform environmental prediction and microclimate coupling using multimodal data.

[0003] In recent years, large language models (LLMs) have demonstrated outstanding multimodal understanding and generation capabilities. These models can handle inputs and outputs from multiple modalities, including text, images, video, and audio, and include understanding and generating multimodal content, modality-specific generation, and mimicking human modality transitions. These capabilities are driving the development of large multimodal language models. However, researchers have also discovered the illusion problem when using LLMs. Existing research shows that injecting knowledge graph information into LLMs in different forms can significantly improve the model's illusion problem. The core idea of ​​this method is to use knowledge graph retrieval to obtain triples of knowledge related to entities explicitly mentioned in the input question, thereby providing LLMs with rich factual information.

[0004] For green building performance simulation, machine learning-based building performance simulation has significant advantages in terms of speed and reduced hardware requirements. However, it lacks generalization ability, struggles to handle multimodal data, and is highly specialized, making widespread adoption difficult. Therefore, combining generative design and big data technologies can improve the accuracy of optimization design and promote efficient decision-making in green building performance optimization design. Summary of the Invention

[0005] The purpose of this invention is to solve the problems in the prior art by proposing a method and system for urban block morphology optimization design decision-making based on a large generative model.

[0006] This invention is achieved through the following technical solution: This invention proposes a method for optimizing the design decision of urban block morphology based on a large-scale generative model, the method comprising the following steps:

[0007] S1. Collect multimodal green performance data affecting the intensity of local heat island and wind environment performance at the urban block scale, and perform preprocessing.

[0008] Step S1 includes the following steps:

[0009] S11. Collect multimodal green performance data;

[0010] S12. Generate an embedded representation that integrates multimodal information;

[0011] S13. Extract information from the data;

[0012] S2. The preprocessed multimodal green performance data is converted into RDF triples based on entities, relations, and events to construct a multimodal knowledge graph;

[0013] Step S2 includes the following steps:

[0014] S21. Convert the extracted information into RDF triples;

[0015] S22. Constructing a multimodal knowledge graph based on RDF triples;

[0016] S3. Modularly cluster the knowledge in the multimodal knowledge graph and obtain structured subgraphs. Improve knowledge retrieval efficiency based on the structured subgraphs and obtain a building green performance data vector database to establish an external knowledge base to enhance the large language model.

[0017] Step S3 includes the following steps:

[0018] S31. Perform modular clustering of knowledge in the knowledge graph and obtain structured subgraphs;

[0019] S32. Improve knowledge retrieval efficiency based on structured subgraphs;

[0020] S33. Obtain a vector database of building green performance data;

[0021] S4. The knowledge representation enhances the large language model to provide guidance and feedback to the machine learning model that performs urban block morphology optimization design, forming an edge-cloud collaborative architecture that coordinates large and small models.

[0022] Step S4 includes the following steps:

[0023] S41. The cloud-side big language model receives the designer's optimization design task instructions and related information, thereby understanding the designer's needs and providing simulation instructions to the edge model.

[0024] S42. The edge model combines cloud-side instructions with relevant engineering scenario model files and relevant urban street environment parameters to conduct urban street morphology optimization simulation and provides simulation result sample data.

[0025] S43. Analyze the simulation results sample data using a large language model and provide decision-making guidance.

[0026] Furthermore, in step S11, a multimodal green performance database affecting the intensity of local heat islands and wind environment performance at the urban block scale is constructed by dynamically acquiring data from CIM, BIM or IoT integrated platforms, and the data modality type and corresponding data format requirements are determined in the data parsing process.

[0027] In step S12, CLIP or MMV cross-modal alignment algorithms and multimodal Transformer are used to process building green performance data of different modalities, and these data are uniformly converted into high-dimensional vectors that retain their multimodal data features. They are then embedded in a shared vector space to achieve information fusion and alignment, retaining and strengthening the correlation between building green performance data of different modalities, so as to assist in understanding the semantic relationship of multimodal data and enable the model to consider information of multiple modalities at the same time when processing multimodal tasks.

[0028] In step S13, the information extraction includes entity extraction, relation extraction, and event extraction. Entity extraction refers to using named entity recognition methods based on the BiLSTM-CRF deep learning model and utilizing contextual information to extract specific building green performance objects or concepts from an unstructured multi-scale building green performance dataset. Relation extraction refers to using relation classification models or dependency parsing methods to identify the relationships between the aforementioned building green performance entities. Event extraction is detected through the ACE event extraction framework and identifies building green performance event trigger words and related information based on a deep learning model.

[0029] Furthermore, in step S21, the building green performance data is transformed into "entity-relationship-entity" RDF triples using the RDF toolkit RDFLib of Python, and the triples are stored in a graph database for subsequent processing and query operations.

[0030] In step S22, the extracted entities are aligned with existing entities in the knowledge base to construct a relationship graph between entities, and a graph matching algorithm is used to link the entities. For records in the knowledge base that point to the same entity, the distance calculation of entities and relationships embedded in the vector space is merged using a clustering algorithm or a DeepWalk algorithm. An embedding model is used to calculate relationship similarity, and a graph neural network is used to integrate the relationships between the same entities from different data sources. The fused data is then used for knowledge enhancement through reasoning to improve the accuracy of information extraction and the reasoning ability of the knowledge graph.

[0031] Furthermore, in step S31, the knowledge graph is segmented into different communities by identifying interconnected node groups through a community detection algorithm; and by identifying and adding missing relationships or entities through a graph embedding algorithm, the completeness of the knowledge graph is enhanced, and knowledge reasoning is assisted.

[0032] In step S32, a hierarchical clustering method is used to extract summaries at multiple scales, including city, street, building, and space, and to perform hierarchical analysis of the community; the structured subgraph and its summaries are converted into high-dimensional vector representations; and the FAISS or HNSW algorithm is used to assist in high-dimensional vector retrieval, thereby improving the efficiency of machine learning tasks.

[0033] In step S33, an enhanced base model is constructed using the obtained building green performance data vector database.

[0034] Furthermore, in step S41, the large language model, as a question-and-answer assistant in the architectural vertical field, opens up intelligent question-and-answer service functions to designers. The input end receives the designer's optimization design task instructions and related multimodal data input, and then understands the designer's needs and transforms them into simulation instructions to provide to the end-side model.

[0035] In step S42, the end-side model is a machine learning model. Machine learning-based building performance simulation has significant advantages in improving speed and reducing hardware requirements.

[0036] In step S43, through the edge-cloud collaborative architecture, the large language model on the cloud side combines the natural language instructions and related information given by the designer to generate optimization design simulation instructions. The edge-side machine learning model executes the urban street morphology optimization simulation and gives the simulation result sample data, which is fed back to the cloud-side large language model for further analysis and decision guidance.

[0037] This invention also proposes a decision-making system for optimizing urban block morphology based on a generative large model, the system comprising:

[0038] Acquisition Module: Integrates remote sensing satellite and UAV platforms to monitor the urban thermal environment at two scales, collect urban surface temperature data and urban physical environment information, study the intensity and distribution characteristics of the urban heat island effect in different areas, and analyze the impact of human activity intensity;

[0039] Search module: Based on the search task, the search content of the designer is converted into prompt words, the search results are limited to the vertical field of architecture, and the search results are determined by a large language model;

[0040] Decision module: The big language model inputs the retrieval results into the machine learning model to further complete the simulation of urban street morphology optimization. The machine learning model can feed back the simulation results to the big language model for further decision-making. Edge-cloud collaboration assists in optimization design.

[0041] Interactive module: Establishes a digital twin model of urban street morphology, providing a visual representation of building layout and height based on simulation results, and intuitively demonstrating the impact of building morphology on urban heat island intensity and street microclimate.

[0042] Furthermore, in the acquisition module, at the city scale, the urban thermal environment is monitored based on remote sensing satellites equipped with MODIS, collecting urban surface temperature data and a series of remote sensing satellite images; at the street scale, the DJI FlyCart 30 medium-sized multi-rotor drone platform integrates infrared thermal imaging sensors, temperature and humidity sensors, and ultrasonic wind speed sensors to further collect information on temperature difference, air humidity, air velocity, air flow, urban surface roughness, urban underlying surface morphology, and urban building morphology, and analyze the microclimate of urban streets.

[0043] The designer's search content includes natural language text and image data. The search results of the large language model are used to generate street morphology design schemes to improve the intensity of local heat islands and wind environment performance, so as to further assist the machine learning model in optimization simulation tasks.

[0044] Furthermore, the decision-making module is an edge-cloud collaborative architecture that combines an edge-side machine learning model with a cloud-side large language model. In combination with typical engineering scenarios of urban street form optimization design, the large language model uses cloud computing power to transform the design decision-making mode based on subjective experience into a computational design decision-making mode under the fusion of vertical large model knowledge; the edge-side machine learning model leverages the low latency and real-time processing characteristics of the edge to form a base model collaborative architecture of "cloud-based intelligent decision-making and edge real-time response".

[0045] The present invention also proposes an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the urban block morphology optimization design decision method based on a large generative model.

[0046] The present invention also proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the urban block morphology optimization design decision method based on a large generative model.

[0047] The beneficial effects of this invention are:

[0048] This invention provides a method and system for urban block morphology optimization design decision-making based on generative large models. The method and system can improve the multimodal data processing capabilities of traditional single machine learning methods, enhancing the efficiency and accuracy of optimization design decisions. By combining generative design and large language model technology, a question-answering assistant for the building vertical domain is constructed, which is beneficial for guiding efficient decision-making in the optimization design of green building performance and provides strong support for the rapid and efficient simulation of green building performance. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0050] Figure 1 This is a flowchart of a method for optimizing urban block morphology design based on a large generative model, according to the present invention. Detailed Implementation

[0051] 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.

[0052] Combination Figure 1 This invention proposes a method for optimizing urban block morphology design decisions based on a large generative model. The method includes the following steps:

[0053] S1. Collect multimodal green performance data affecting the intensity of local heat island and wind environment performance at the urban block scale, and perform preprocessing.

[0054] Step S1 includes the following steps:

[0055] S11. Collect multimodal green performance data;

[0056] S12. Generate an embedded representation that integrates multimodal information;

[0057] S13. Extract information from the data;

[0058] In step S11, a multimodal green performance database affecting the intensity of local heat islands and wind environment performance at the urban block scale is constructed by dynamically acquiring data from CIM, BIM or IoT integrated platforms, and the data modality type and corresponding data format requirements are determined in the data parsing process.

[0059] In step S12, CLIP or MMV cross-modal alignment algorithms and multimodal Transformer are used to process building green performance data of different modalities, and these data are uniformly converted into high-dimensional vectors that retain their multimodal data features. They are then embedded into a shared vector space to achieve information fusion and alignment, preserving and strengthening the correlation between building green performance data of different modalities, such as the correlation between images and text descriptions, to assist in understanding the semantic relationships of multimodal data, so that the model can consider information from multiple modalities at the same time when processing multimodal tasks.

[0060] In step S13, the information extraction includes entity extraction, relation extraction, and event extraction. Entity extraction refers to applying the Named Entity Recognition (NER) method based on the BiLSTM-CRF deep learning model and utilizing contextual information to extract specific building green performance objects or concepts from an unstructured multi-scale building green performance dataset. Relation extraction refers to using relation classification models such as OpenIE and BERT or dependency parsing methods to identify the relationships between the aforementioned building green performance entities. Event extraction typically involves the spatiotemporal dimension parameters of multiple building green performance entities, which are detected using the ACE event extraction framework, and the trigger words and related information for building green performance events are identified based on deep learning models such as BiLSTM-CRF and BERT.

[0061] S2. The preprocessed multimodal green performance data is transformed into RDF triples based on entities, relations, and events to construct a multimodal knowledge graph; specifically: the preprocessed multimodal green performance data is integrated and associated with multi-source heterogeneous data using knowledge graph technology to form a structured knowledge system, and a comprehensive model is performed on urban morphology, climate background, geographical location, and intensity of human activities to construct a multimodal knowledge graph of urban heat island intensity and wind environment influencing factors;

[0062] Step S2 includes the following steps:

[0063] S21. Convert the extracted information into RDF triples;

[0064] S22. Constructing a multimodal knowledge graph based on RDF triples;

[0065] In step S21, the green performance data of buildings is transformed into RDF triples of "entity-relationship-entity" using the RDF toolkit RDFLib of Python, and the triples are stored in the graph database for subsequent processing and query operations.

[0066] In step S22, the extracted entities are aligned with existing entities in the knowledge base to construct a relationship graph between entities, and graph matching algorithms such as PageRank and Random Walk are used to link entities. For records in the knowledge base that point to the same entity, the distance calculation of entities and relationships embedded in the vector space is merged using clustering algorithms such as K-means and DBSCAN or DeepWalk. Relationship similarity is calculated using embedding models such as Word2Vec and GloVe, and the relationships between the same entities from different data sources are integrated using graph neural networks such as GCN and RGCNs. The fused data is then used for knowledge enhancement through reasoning to improve the accuracy of information extraction and the reasoning ability of the knowledge graph.

[0067] S3. Modularly cluster the knowledge in the multimodal knowledge graph to obtain structured subgraphs. Improve knowledge retrieval efficiency based on the structured subgraphs. Obtain a building green performance data vector database to build an external knowledge base to enhance the large language model. Obtain design information from designers for optimizing the urban block morphology. Based on the design information, use the large language model combined with the multimodal knowledge graph to generate interpretable analysis results to assist the machine learning model in simulating urban block morphology optimization.

[0068] Step S3 includes the following steps:

[0069] S31. Perform modular clustering of knowledge in the knowledge graph and obtain structured subgraphs;

[0070] S32. Improve knowledge retrieval efficiency based on structured subgraphs;

[0071] S33. Obtain a vector database of building green performance data;

[0072] In step S31, the knowledge graph is divided into different communities by identifying interconnected node groups through a community detection algorithm; and missing relationships or entities are identified and added through a graph embedding algorithm to enhance the integrity of the knowledge graph and assist in knowledge reasoning.

[0073] In step S32, hierarchical clustering methods such as hierarchical clustering and k-means clustering are used to extract summaries at multiple scales, including city, street, building, and space, and to perform hierarchical analysis of communities; structured subgraphs and their summaries are converted into high-dimensional vector representations using algorithms such as Node2Vec, DeepWalk, and GraphSAGE; and high-dimensional vector retrieval is assisted by FAISS or HNSW algorithms to improve the efficiency of machine learning tasks such as clustering and retrieval.

[0074] In step S33, an enhanced base model is constructed using the obtained building green performance data vector database.

[0075] S4. The knowledge representation enhances the large language model to provide guidance and feedback to the machine learning model that performs urban block morphology optimization design, forming an edge-cloud collaborative architecture that coordinates large and small models.

[0076] Step S4 includes the following steps:

[0077] S41. The cloud-side large language model receives the designer's optimization design task instructions and related inputs, understands user needs, and provides simulation instructions to the edge-side model.

[0078] S42. The edge model combines cloud-side instructions with relevant engineering scenario model files and relevant urban street environment parameters to conduct urban street morphology optimization simulation and provides simulation result sample data.

[0079] S43. Analyze the simulation results sample data using a large language model and provide decision-making guidance.

[0080] In step S41, the large language model, as a question-and-answer assistant in the vertical field of architecture, opens up intelligent question-and-answer service functions to designers. The input end receives the designer's optimization design task instructions and related multimodal data input, and then understands the designer's needs and transforms them into simulation instructions to provide to the end-side model.

[0081] In step S42, the end-side model is a machine learning model. Machine learning-based building performance simulation has significant advantages in improving speed and reducing hardware requirements.

[0082] In step S43, through the edge-cloud collaborative architecture, the large language model on the cloud side combines the natural language instructions and related information given by the designer to generate optimization design simulation instructions. The edge-side machine learning model executes the urban street morphology optimization simulation and gives the simulation result sample data, which is fed back to the cloud-side large language model for further analysis and decision guidance.

[0083] This invention also proposes a decision-making system for optimizing urban block morphology based on a generative large model, the system comprising:

[0084] Acquisition Module: Integrates remote sensing satellite and UAV platforms to monitor the urban thermal environment at two scales, collect urban surface temperature data and urban physical environment information, study the intensity and distribution characteristics of the urban heat island effect in different areas, and analyze the impact of human activity intensity;

[0085] Search module: Based on the search task, the search content of the designer is converted into prompt words, the search results are limited to the vertical field of architecture, and the search results are determined by a large language model;

[0086] Decision module: The big language model inputs the retrieval results into the machine learning model to further complete the simulation of urban street morphology optimization. The machine learning model can feed back the simulation results to the big language model for further decision-making. Edge-cloud collaboration assists in optimization design.

[0087] Interactive module: Establishes a digital twin model of urban street morphology, providing a visual representation of building layout and height based on simulation results, and intuitively demonstrating the impact of building morphology on urban heat island intensity and street microclimate.

[0088] The multimodal green performance data includes numerical values ​​such as temperature difference, wind speed, building density, and building height; textual data such as government documents, engineering documents, and standards; and images such as low-altitude aerial photographs of urban blocks. The data acquisition method includes dynamic acquisition of data through integrated platforms such as CIM, BIM, and IoT. The data preprocessing method includes using cross-modal alignment algorithms such as CLIP and MMV and multimodal Transformer processing to uniformly convert the data into high-dimensional vectors that retain their multimodal data characteristics, and then extracting information through entity extraction, relation extraction, and event extraction, retaining the entity information and relation information required for knowledge graph construction.

[0089] In the acquisition module, at the city scale, the urban thermal environment is monitored by remote sensing satellites equipped with MODIS (Medium Resolution Imaging Spectroradiometer), collecting urban surface temperature data and a series of remote sensing satellite images. At the street scale, the DJI FlyCart 30 medium-sized multi-rotor drone platform integrates infrared thermal imaging sensors, temperature and humidity sensors, and ultrasonic wind speed sensors to further collect information on temperature difference, air humidity, air velocity, air flow, urban surface roughness, urban underlying surface morphology, and urban building morphology, analyzing the microclimate of urban blocks. Furthermore, 3D reconstruction technology can be used to establish digital twin models of urban block morphology, providing visualization support for decision-making.

[0090] The content retrieved by the designer includes natural language text and image data, such as urban environmental information, urban and building morphology spatial data, and low-altitude aerial photographs of urban blocks collected by the acquisition module, including design task text, government documents, and standards. The large language model retrieval results generate street morphology design schemes to improve the intensity of local heat islands and wind environment performance, including design parameters such as building height, building form, and building location, to further assist the machine learning model in optimizing and simulating tasks.

[0091] The decision-making module is an edge-cloud collaborative architecture that combines an edge-side machine learning model with a cloud-side large language model. It is combined with typical engineering scenarios of urban street form optimization design. The large language model uses cloud computing power to transform the design decision-making mode based on subjective experience into a computational design decision-making mode under the knowledge fusion of vertical large models. The edge-side machine learning model leverages the low latency and real-time processing characteristics of the edge to form a base model collaborative architecture of "cloud-based intelligent decision-making and edge real-time response".

[0092] In the interactive module, the urban block digital twin system can improve the visualization and intelligence of optimization design. Utilizing UAV 3D reconstruction technology, it can display high-precision 3D models of urban blocks through a digital sand table, simulate the architectural form and layout of urban blocks, and further analyze the local heat island intensity and wind environment climate of urban blocks to achieve visualized intelligent decision support.

[0093] The present invention also proposes an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the urban block morphology optimization design decision method based on a large generative model.

[0094] The present invention also proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the urban block morphology optimization design decision method based on a large generative model.

[0095] The memory in this application embodiment can be volatile memory or non-volatile memory, or it can include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DRRAM). It should be noted that the memory used in the methods described in this invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0096] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., high-density digital video discs (DVDs)), or semiconductor media (e.g., solid-state disks (SSDs)).

[0097] During implementation, each step of the above method can be completed by an integrated logic circuit of the hardware in the processor or by instructions in the form of software. The steps of the method disclosed in conjunction with the embodiments of the present application can be directly embodied as being executed by a hardware processor, or can be executed by a combination of hardware and software modules in the processor. The software module can be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in conjunction with its hardware. To avoid repetition, it will not be described in detail here.

[0098] It should be noted that the processor in the embodiments of the present application can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method embodiment can be completed by an integrated logic circuit of the hardware in the processor or by instructions in the form of software. The above processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component. The various methods, steps, and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in the embodiments of the present application can be directly embodied as being executed by a hardware decoding processor, or can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium mature in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, or electrically erasable programmable memory, registers, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0099] The above provides a detailed description of the urban block morphology optimization design decision method and system based on a large generative model proposed in this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for optimizing urban block morphology design decisions based on a generative large model, characterized in that, The method comprises the following steps: S1. Collect multimodal green performance data affecting the intensity of local heat island and wind environment performance at the urban block scale, and perform preprocessing. Step S1 includes the following steps: S11. Collect multimodal green performance data; S12. Generate an embedded representation that integrates multimodal information; S13. Extract information from the data; S2. The preprocessed multimodal green performance data is converted into RDF triples based on entities, relations, and events to construct a multimodal knowledge graph; Step S2 includes the following steps: S21. Convert the extracted information into RDF triples; S22. Constructing a multimodal knowledge graph based on RDF triples; S3. Modularly cluster the knowledge in the multimodal knowledge graph and obtain structured subgraphs. Improve knowledge retrieval efficiency based on the structured subgraphs and obtain a building green performance data vector database to establish an external knowledge base to enhance the large language model. Step S3 includes the following steps: S31. Perform modular clustering of knowledge in the knowledge graph and obtain structured subgraphs; S32. Improve knowledge retrieval efficiency based on structured subgraphs; S33. Obtain a vector database of building green performance data; In step S31, the knowledge graph is divided into different communities by identifying interconnected node groups through a community detection algorithm; and missing relationships or entities are identified and added through a graph embedding algorithm to enhance the integrity of the knowledge graph and assist in knowledge reasoning. In step S32, a hierarchical clustering method is used to extract summaries at multiple scales, including city, street, building, and space, and to perform hierarchical analysis of the community; the structured subgraph and its summaries are converted into high-dimensional vector representations; and the FAISS or HNSW algorithm is used to assist in high-dimensional vector retrieval, thereby improving the efficiency of machine learning tasks. In step S33, an enhanced base model is constructed using the obtained building green performance data vector database. S4. By enhancing the large language model through the knowledge base, the machine learning model that performs urban block morphology optimization design is guided and receives feedback, forming an edge-cloud collaborative architecture that coordinates large and small models. Step S4 includes the following steps: S41. The cloud-side big language model receives the designer's optimization design task instructions and related information, thereby understanding the designer's needs and providing simulation instructions to the edge model. S42. The edge model combines cloud-side instructions with relevant engineering scenario model files and relevant urban street environment parameters to perform urban street morphology optimization simulation and provides simulation result sample data. S43. Analyze the simulation results sample data using a large language model and provide decision-making guidance.

2. The method according to claim 1, characterized in that, In step S11, a multimodal green performance database affecting the intensity of local heat islands and wind environment performance at the urban block scale is constructed by dynamically acquiring data from CIM, BIM or IoT integrated platforms, and the data modality type and corresponding data format requirements are determined in the data parsing process. In step S12, CLIP or MMV cross-modal alignment algorithms and multimodal Transformer are used to process building green performance data of different modalities, and these data are uniformly converted into high-dimensional vectors that retain their multimodal data features. They are then embedded in a shared vector space to achieve information fusion and alignment, retaining and strengthening the correlation between building green performance data of different modalities, so as to assist in understanding the semantic relationship of multimodal data and enable the model to consider information of multiple modalities at the same time when processing multimodal tasks. In step S13, the information extraction includes entity extraction, relation extraction, and event extraction. Entity extraction refers to applying named entity recognition methods based on the BiLSTM-CRF deep learning model and utilizing contextual information to extract specific building green performance objects or concepts from an unstructured multi-scale building green performance dataset. Relation extraction refers to using relation classification models or dependency parsing methods to identify the relationships between the aforementioned building green performance entities. The event extraction is detected using the ACE event extraction framework and identified trigger words and related information for building green performance events based on a deep learning model.

3. The method according to claim 1, characterized in that, In step S21, the green performance data of buildings is transformed into "entity-relationship-entity" RDF triples using the RDF toolkit RDFLib of Python, and the triples are stored in the graph database for subsequent processing and query operations. In step S22, the extracted entities are aligned with existing entities in the knowledge base to construct a relationship graph between entities, and a graph matching algorithm is used to link the entities; for records in the knowledge base that point to the same entity, the distance calculation of entities and relationships embedded in the vector space is merged by clustering algorithm or DeepWalk algorithm; the relationship similarity is calculated using an embedding model, and the relationship between the same entities from different data sources is integrated using a graph neural network; The fused data is augmented through reasoning to improve the accuracy of information extraction and the reasoning ability of the knowledge graph.

4. The method according to claim 1, characterized in that, In step S41, the large language model, as a question-and-answer assistant in the vertical field of architecture, opens up intelligent question-and-answer service functions to designers. The input end receives the designer's optimization design task instructions and related multimodal data input, and then understands the designer's needs and transforms them into simulation instructions to provide to the end-side model. In step S42, the end-side model is a machine learning model. Machine learning-based building performance simulation has significant advantages in improving speed and reducing hardware requirements. In step S43, through the edge-cloud collaborative architecture, the large language model on the cloud side combines the natural language instructions and related information given by the designer to generate optimization design simulation instructions. The edge-side machine learning model executes the urban street morphology optimization simulation and gives the simulation result sample data, which is fed back to the cloud-side large language model for further analysis and decision guidance.

5. A decision-making system for optimizing urban block morphology based on a generative large model, characterized in that, The system is used to implement the urban block morphology optimization design decision method according to any one of claims 1-4, and the system includes: Acquisition Module: Integrates remote sensing satellite and UAV platforms to monitor the urban thermal environment at two scales, collect urban surface temperature data and urban physical environment information, study the intensity and distribution characteristics of the urban heat island effect in different areas, and analyze the impact of human activity intensity; Search module: Based on the search task, the search content of the designer is converted into prompt words, the search results are limited to the vertical field of architecture, and the search results are determined by a large language model; Decision module: The big language model inputs the retrieval results into the machine learning model to further complete the simulation of urban street morphology optimization. The machine learning model can feed back the simulation results to the big language model for further decision-making. Edge-cloud collaboration assists in optimization design. Interactive module: Establishes a digital twin model of urban street morphology, providing a visual representation of building layout and height based on simulation results, and intuitively demonstrating the impact of building morphology on urban heat island intensity and street microclimate.

6. The system according to claim 5, characterized in that, In the acquisition module, at the city scale, the urban thermal environment is monitored based on remote sensing satellites equipped with MODIS, collecting urban surface temperature data and a series of remote sensing satellite images; at the street scale, the DJI FlyCart 30 medium-sized multi-rotor drone platform integrates infrared thermal imaging sensors, temperature and humidity sensors and ultrasonic wind speed sensors to further collect information on temperature difference, air humidity, air velocity, air flow, urban surface roughness, urban underlying surface morphology and urban building morphology physical environment, and analyze the microclimate of urban streets. The designer's search content includes natural language text and image data. The search results of the large language model are used to generate street morphology design schemes to improve the intensity of local heat islands and wind environment performance, so as to further assist the machine learning model in optimization simulation tasks.

7. The system according to claim 5, characterized in that, The decision-making module is an edge-cloud collaborative architecture that combines an edge-side machine learning model with a cloud-side large language model. It is combined with typical engineering scenarios of urban street form optimization design. The large language model uses cloud computing power to transform the design decision-making mode based on subjective experience into a computational design decision-making mode under the knowledge fusion of vertical large models. The edge-side machine learning model leverages the low latency and real-time processing characteristics of the edge to form a base model collaborative architecture of "cloud-based intelligent decision-making and edge real-time response".

8. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-4.

9. A computer-readable storage medium for storing computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the method according to any one of claims 1-4.

Images